Delphi World - это проект, являющийся сборником статей и малодокументированных возможностей  по программированию в среде Delphi. Здесь вы найдёте работы по следующим категориям: delphi, delfi, borland, bds, дельфи, делфи, дэльфи, дэлфи, programming, example, программирование, исходные коды, code, исходники, source, sources, сорцы, сорсы, soft, programs, программы, and, how, delphiworld, базы данных, графика, игры, интернет, сети, компоненты, классы, мультимедиа, ос, железо, программа, интерфейс, рабочий стол, синтаксис, технологии, файловая система...
Фильтрация, регрессия, работа с массивом и серией

Автор: Lookin
WEB-сайт: http://delphibase.endimus.com

{ **** UBPFD *********** by delphibase.endimus.com ****
>> Фильтрация, регрессия, работа с массивом и серией

Модуль предназначен для выполнения процедур:
- фильтрации
- регрессии
- операций с массивами
- операций с сериями

Зависимости: Math, TeEngine, Graphics, SysUtils, Dialogs
Автор:       lookin, lookin@mail.ru, Екатеринбург
Copyright:   lookin
Дата:        30 апреля 2002 г.
***************************************************** }

unit FilterRegressionArraySeries;

interface

uses Math, TeEngine, Graphics, SysUtils, Dialogs;

type
  TIntegerArray = array of integer;
type
  TExIntegerArray = array of TIntegerArray;
type
  TDoubleArray = array of double;
type
  TExDoubleArray = array of TDoubleArray;
type
  TStringArray = array of string;
type
  TExStringArray = array of TStringArray;

procedure ArrayExpanding(var ValueArray: TDoubleArray; ExpandCoef: integer);
procedure ArrayLengthening(var ValueArray: TDoubleArray; SplitValue: integer);
procedure ArrayShortening(var ValueArray: TDoubleArray; SplitValue: integer);
procedure CubicSplineSmoothing(var ValueArray: TDoubleArray; Dsc: double;
  Coef: integer);
procedure SevenPointNonLinearSmoothing(var ValueArray: TDoubleArray;
  Dsc: double; Coef: integer);
procedure FourierAnalysis(var ValueArray: TDoubleArray; NumGarmonics: integer);
procedure DoArraySmoothing(var ValueArray: TDoubleArray; FilterType: integer;
  Dsc: double; SplitCoef, ExpandCoef: integer;
  CycledFilter: boolean);

procedure LinearRegression(ValueArray, ArgumentArray: TDoubleArray;
  SourceSeries, DestSeries: TChartSeries;
  var MainCoef, FreeCoef: double; SeriesColor: TColor;
  var Hint: string);
procedure HyperbolicRegression(ValueArray, ArgumentArray: TDoubleArray;
  SourceSeries, DestSeries: TChartSeries;
  var MainCoef, FreeCoef: double;
  SeriesColor: TColor; var Hint: string);
procedure PowerRegression(ValueArray, ArgumentArray: TDoubleArray;
  SourceSeries, DestSeries: TChartSeries;
  var MainCoef, FreeCoef: double; SeriesColor: TColor;
  var Hint: string);
procedure PolynomialRegression(ValueArray, ArgumentArray: TDoubleArray;
  SourceSeries, DestSeries: TChartSeries;
  PolyDegree: integer; var ArrayCoefs: TDoubleArray;
  SeriesColor: TColor; var Hint: string);
procedure ExponentRegression(ValueArray, ArgumentArray: TDoubleArray;
  SourceSeries, DestSeries: TChartSeries;
  var MainCoef, FreeCoef: double; SeriesColor: TColor;
  var Hint: string; Warning: boolean);
procedure ExponentialRegression(ValueArray, ArgumentArray: TDoubleArray;
  SourceSeries, DestSeries: TChartSeries;
  var MainCoef, FreeCoef: double; SeriesColor: TColor;
  var Hint: string; Warning: boolean);
procedure ExpPowerRegression(ValueArray, ArgumentArray: TDoubleArray;
  SourceSeries, DestSeries: TChartSeries;
  var MainCoef, FreeCoef: double; SeriesColor: TColor;
  var Hint: string; Warning: boolean);

procedure CheckArrayBounds(var CArray: TDoubleArray; var FromPoint, ToPoint:
  integer);
procedure CheckSeriesBounds(CSeries: TChartSeries; var FromPoint, ToPoint:
  integer);
procedure ArrayFromArray(var SourceArray, DestArray: TDoubleArray;
  FromPoint, ToPoint, Discrete: integer; Derivative: boolean);
procedure ArrayFromSeries(var ValueArray: TDoubleArray; DataSeries:
  TChartSeries;
  FromPoint, ToPoint, Discrete: integer; Derivative: boolean);
procedure SeriesFromArray(var ValueArray: TDoubleArray; DataSeries:
  TChartSeries;
  FromPoint, ToPoint, Discrete: integer; Derivative: boolean);
function DerivFromArray(var SourceArray: TDoubleArray; FromPoint, ToPoint,
  Discrete: integer; Extremum: string;
  var Position: integer): double;
function DerivFromSeries(DataSeries: TChartSeries; FromPoint, ToPoint,
  Discrete: integer; Extremum: string;
  var Position: integer): double;
function ValueFromSeries(DataSeries: TChartSeries; FromPoint, ToPoint: integer;
  Extremum: string; var Position: integer): double;
function ValueFromArray(var SourceArray: TDoubleArray; FromPoint, ToPoint:
  integer;
  Extremum: string; var Position: integer): double;
function CalculateAreaOfArray(var SourceArray: TDoubleArray;
  FromPoint, ToPoint, Method: integer;
  BindToZero: boolean): double;
function CalculateAreaOfSeries(DataSeries: TChartSeries; FromPoint, ToPoint,
  Method: integer; BindToZero: boolean): double;
procedure LinearTrendExclusion(var ValueArray: TDoubleArray);

procedure ColorizeSeries(DataSeries: TChartSeries; NewColor: TColor);
procedure SetXInterval(DataSeries: TChartSeries; XInterval: double);
procedure SetSeriesAxis(DataSeries: TChartSeries; NewAxis: TVertAxis);

var
  rv, rsmooth, smootha: TDoubleArray;

implementation

//Нелинейный фильтр по 7 точкам

procedure SevenPointNonLinearSmoothing(var ValueArray: TDoubleArray;
  Dsc: double; Coef: integer);
var
  j, k, i: integer;
  resv: array of array of double;
begin
  if (Coef = 0) or (Coef = 1) then
    Exit;
  SetLength(resv, Coef, (Length(ValueArray) div Coef));
  for j := 0 to Coef - 1 do
    for i := 0 to Length(resv[0]) - 1 do
      resv[j][i] := ValueArray[i * Coef + j];
  for k := 0 to Coef - 1 do
    for j := 0 to Length(resv[0]) - 1 do
    begin
      if j = 0 then
        resv[k][j] := (39 * ValueArray[j * Coef + k] +
          8 * ValueArray[(j + 1) * Coef + k] - 4 * (ValueArray[(j + 2) * Coef +
            k] +
          ValueArray[(j + 3) * Coef + k] - ValueArray[(j + 4) * Coef + k]) +
          ValueArray[(j + 5) * Coef + k] - 2 * ValueArray[(j + 6) * Coef + k]) /
            42;
      if j = 1 then
        resv[k][j] := (8 * ValueArray[j * Coef + k] +
          19 * ValueArray[(j + 1) * Coef + k] + 16 * ValueArray[(j + 2) * Coef +
            k] +
          6 * ValueArray[(j + 3) * Coef + k] - 4 * ValueArray[(j + 4) * Coef + k]
            -
          7 * ValueArray[(j + 5) * Coef + k] + 4 * ValueArray[(j + 6) * Coef +
            k]) / 42;
      if j = 2 then
        resv[k][j] := (-4 * ValueArray[j * Coef + k] +
          16 * ValueArray[(j + 1) * Coef + k] + 19 * ValueArray[(j + 2) * Coef +
            k] +
          12 * ValueArray[(j + 3) * Coef + k] + 2 * ValueArray[(j + 4) * Coef +
            k] -
          4 * ValueArray[(j + 5) * Coef + k] + ValueArray[(j + 6) * Coef + k]) /
            42;
      if (j > 2) and (j < Length(resv[0]) - 3) then
        resv[k][j] :=
          (7 * ValueArray[j * Coef + k] + 6 * (ValueArray[(j - 1) * Coef + k] +
          ValueArray[(j + 1) * Coef + k]) + 3 * (ValueArray[(j - 2) * Coef + k]
            +
          ValueArray[(j + 2) * Coef + k]) - 2 * (ValueArray[(j - 3) * Coef + k]
            +
          ValueArray[(j + 3) * Coef + k])) / 21;
      if j = Length(resv[0]) - 3 then
        resv[k][j] := (-4 * ValueArray[j * Coef + k] +
          16 * ValueArray[(j - 1) * Coef + k] + 19 * ValueArray[(j - 2) * Coef +
            k] +
          12 * ValueArray[(j - 3) * Coef + k] + 2 * ValueArray[(j - 4) * Coef +
            k] -
          4 * ValueArray[(j - 5) * Coef + k] + ValueArray[(j - 6) * Coef + k]) /
            42;
      if j = Length(resv[0]) - 2 then
        resv[k][j] := (8 * ValueArray[j * Coef + k] +
          19 * ValueArray[(j - 1) * Coef + k] + 16 * ValueArray[(j - 2) * Coef +
            k] +
          6 * ValueArray[(j - 3) * Coef + k] - 4 * ValueArray[(j - 4) * Coef + k]
            -
          7 * ValueArray[(j - 5) * Coef + k] + 4 * ValueArray[(j - 6) * Coef +
            k]) / 42;
      if j = Length(resv[0]) - 1 then
        resv[k][j] := (39 * ValueArray[j * Coef + k] +
          8 * ValueArray[(j - 1) * Coef + k] - 4 * ValueArray[(j - 2) * Coef + k]
            -
          4 * ValueArray[(j - 3) * Coef + k] - 4 * ValueArray[(j - 4) * Coef + k]
            +
          ValueArray[(j - 5) * Coef + k] - 2 * ValueArray[(j - 6) * Coef + k]) /
            42;
    end;
  for j := Coef to Length(resv[0]) - Coef do
    for k := 0 to Coef - 1 do
      ValueArray[j * Coef + k] := resv[k][j];
end;

//Фильтр с кубическими сплайнами

procedure CubicSplineSmoothing(var ValueArray: TDoubleArray; Dsc: double;
  Coef: integer);
var
  j, k, i, N: integer;
  vresv, resv: array of array of double;
  maxv: array of double;
  av, h, mi, mj, v1, v2: double;
begin
  if (Coef = 0) or (Coef = 1) then
    Exit;
  N := Length(ValueArray);
  SetLength(resv, Coef, N);
  h := Coef * Dsc;
  for k := 0 to Coef - 1 do
    for j := 0 to (N div Coef) - 2 do
    begin
      if j = 0 then
      begin
        mi := (4 * ValueArray[(j + 1) * Coef + k] -
          ValueArray[(j + 2) * Coef + k] - 3 * ValueArray[j * Coef + k]) / 2;
        mj := (ValueArray[(j + 2) * Coef + k] - ValueArray[j * Coef + k]) / 2;
      end;
      if j = (N div Coef) - 2 then
      begin
        mi := (ValueArray[(j + 1) * Coef + k] - ValueArray[(j - 1) * Coef + k])
          / 2;
        mj := (3 * ValueArray[(j + 1) * Coef + k] + ValueArray[(j - 1) * Coef +
          k] -
          4 * ValueArray[j * Coef + k]) / 2;
      end;
      if (j > 0) and (j < ((N div Coef) - 2)) then
      begin
        mi := (ValueArray[(j + 1) * Coef + k] - ValueArray[(j - 1) * Coef + k])
          / 2;
        mj := (ValueArray[(j + 2) * Coef + k] - ValueArray[j * Coef + k]) / 2;
      end;
      for i := j * Coef to (j + 1) * Coef do
      begin
        v1 := ((j + 1) * Coef + k) * Dsc - (i + k) * Dsc;
        v2 := (i + k) * Dsc - (j * Coef + k) * Dsc;
        resv[k][i + k] := (Sqr(v1) * (2 * v2 + h) * ValueArray[j * Coef + k] +
          Sqr(v2) * (2 * v1 + h) * ValueArray[(j + 1) * Coef + k] +
          (Sqr(v1) * v2 * mi + Sqr(v2) * (-v1) * mj) / 2) / h / h / h;
      end;
    end;
  for j := Coef to N - 1 - Coef do
  begin
    av := 0;
    for k := 0 to Coef - 1 do
      av := av + resv[k][j];
    av := av / Coef;
    ValueArray[j] := av;
  end;
end;

//Гармонический синтез Фурье

procedure FourierAnalysis(var ValueArray: TDoubleArray; NumGarmonics: integer);
var
  i, j, N: integer;
  yn, ap, bp: double;
  AFCoef, BFCoef: TDoubleArray;
begin
  N := Length(ValueArray);
  SetLength(AFCoef, NumGarmonics);
  SetLength(BFCoef, NumGarmonics);
  AFCoef[0] := Sum(ValueArray) / N;
  BFCoef[0] := 0;
  for i := 1 to NumGarmonics - 1 do
  begin
    AFCoef[i] := 0;
    BFCoef[i] := 0;
    for j := 0 to N - 1 do
    begin
      AFCoef[i] := AFCoef[i] + ValueArray[j] * cos(Pi * i * j * 2 / N);
      BFCoef[i] := BFCoef[i] + ValueArray[j] * sin(Pi * i * j * 2 / N);
    end;
    AFCoef[i] := AFCoef[i] * 2 / N;
    BFCoef[i] := BFCoef[i] * 2 / N;
  end;
  for j := 0 to N - 1 do
  begin
    yn := 0;
    ap := 0;
    bp := 0;
    for i := 1 to NumGarmonics - 1 do
    begin
      ap := ap + AFCoef[i] * cos(2 * Pi * i * (j / N));
      bp := bp + BFCoef[i] * sin(2 * Pi * i * (j / N));
    end;
    yn := AFCoef[0] + ap + bp;
    ValueArray[j] := yn;
  end;
end;

//Общая процедура вызова нужного фильтра

procedure DoArraySmoothing(var ValueArray: TDoubleArray; FilterType: integer;
  Dsc: double; SplitCoef, ExpandCoef: integer; CycledFilter: boolean);
var
  j: integer;
begin
  smoothA := nil;
  rsmooth := ValueArray;
  ArrayExpanding(rsmooth, ExpandCoef);
  ArrayLengthening(smoothA, SplitCoef);
  if FilterType = 1 then
    if CycledFilter then
      for j := 2 to SplitCoef do
        SevenPointNonLinearSmoothing(smoothA, Dsc, j)
    else
      SevenPointNonLinearSmoothing(smoothA, Dsc, SplitCoef);
  if FilterType = 2 then
    CubicSplineSmoothing(smoothA, Dsc, SplitCoef);
  ArrayShortening(smoothA, SplitCoef);
  ValueArray := smoothA;
end;

//Расширение массива заданным числом точек справа и слева

procedure ArrayLengthening(var ValueArray: TDoubleArray; SplitValue: integer);
var
  sv, N, i: integer;
  bv, ev: double;
begin
  N := Length(ValueArray);
  sv := 10 * SplitValue;
  bv := 0;
  ev := 0;
  for i := 0 to 9 do
    bv := bv + ValueArray[i];
  bv := bv / 10;
  for i := N - 1 downto N - 10 do
    ev := ev + ValueArray[i];
  ev := ev / 10;
  SetLength(ValueArray, N + sv);
  for i := N - 1 downto 0 do
    ValueArray[i + trunc(sv / 2)] := ValueArray[i];
  for i := trunc(sv / 2) - 1 downto 0 do
    ValueArray[i] := bv;
  for i := N + trunc(sv / 2) to N + sv - 1 do
    ValueArray[i] := ev;
end;

//Сокращение массива заданным числом точек справа и слева

procedure ArrayShortening(var ValueArray: TDoubleArray; SplitValue: integer);
var
  sv, N, i: integer;
begin
  N := Length(ValueArray);
  sv := 10 * SplitValue;
  for i := 0 to N - sv - 1 do
    ValueArray[i] := ValueArray[i + trunc(sv / 2)];
  SetLength(ValueArray, N - sv);
end;

//Расширение массива заданным числом точек между 2-мя соседними

procedure ArrayExpanding(var ValueArray: TDoubleArray; ExpandCoef: integer);
var
  i, k, N, sub: integer;
  diap: double;
begin
  N := Length(ValueArray);
  sub := ExpandCoef - 1;
  SetLength(smoothA, N * ExpandCoef - sub);
  for i := 0 to N - 1 do
  begin
    smoothA[i * ExpandCoef] := ValueArray[i];
    if i <> 0 then
    begin
      diap := (smoothA[i * ExpandCoef] - smoothA[(i - 1) * ExpandCoef]);
      for k := 0 to ExpandCoef - 1 do
        smoothA[(i - 1) * ExpandCoef + k] :=
          smoothA[(i - 1) * ExpandCoef] + diap * (k / ExpandCoef);
    end;
  end;
end;

//Линейная регрессия

procedure LinearRegression(ValueArray, ArgumentArray: TDoubleArray;
  SourceSeries,
  DestSeries: TChartSeries; var MainCoef, FreeCoef: double;
  SeriesColor: TColor; var Hint: string);
var
  b0, b1, xsum, ysum, pxy, xsqua: double;
  y, x: array of double;
  i, N: integer;
  s: string;
begin
  if ValueArray <> nil then
    N := Length(ValueArray)
  else
    N := SourceSeries.XValues.Count;
  pxy := 0;
  xsqua := 0;
  SetLength(x, N);
  SetLength(y, N);
  for i := 0 to N - 1 do
  begin
    if ValueArray <> nil then
    begin
      y[i] := ValueArray[i];
      x[i] := ArgumentArray[i];
    end
    else
    begin
      y[i] := SourceSeries.YValues.Value[i];
      x[i] := SourceSeries.XValues.Value[i];
    end;
    pxy := pxy + x[i] * y[i];
    xsqua := xsqua + x[i] * x[i];
  end;
  xsum := Sum(x);
  ysum := Sum(y);
  b1 := (xsum * ysum - N * pxy) / (xsum * xsum - N * xsqua);
  b0 := (ysum - b1 * xsum) / N;
  MainCoef := b1;
  FreeCoef := b0;
  if DestSeries <> nil then
    for i := 0 to N - 1 do
      if ValueArray <> nil then
        DestSeries.AddXY(ArgumentArray[i],
          b1 * ArgumentArray[i] + b0, '', SeriesColor)
      else
        DestSeries.AddXY(SourceSeries.XValues.Value[i],
          b1 * SourceSeries.XValues.Value[i] + b0, '', SeriesColor);
  if b0 < 0 then
    s := ''
  else
    s := '+ ';
  Hint := Format('%0.3f', [b1]) + '*X ' + s + Format('%0.3f', [b0]);
  x := nil;
  y := nil;
end;

//Гиперболическая регрессия

procedure HyperbolicRegression(ValueArray, ArgumentArray: TDoubleArray;
  SourceSeries, DestSeries: TChartSeries; var MainCoef, FreeCoef: double;
  SeriesColor: TColor; var Hint: string);
var
  b0, b1, ax, ysum, axsqua, dxy: double;
  y, x: array of double;
  i, N: integer;
  s: string;
begin
  if ValueArray <> nil then
    N := Length(ValueArray)
  else
    N := SourceSeries.XValues.Count;
  axsqua := 0;
  ax := 0;
  dxy := 0;
  SetLength(x, N);
  SetLength(y, N);
  for i := 0 to N - 1 do
  begin
    if ValueArray <> nil then
    begin
      y[i] := ValueArray[i];
      x[i] := ArgumentArray[i];
    end
    else
    begin
      y[i] := SourceSeries.YValues.Value[i];
      x[i] := SourceSeries.XValues.Value[i];
    end;
    if x[i] = 0 then
    begin
      MessageDlg('Hyperbolic regression inapplicable...',
        mtWarning, [mbOk], 0);
      Hint := 'No equation';
      MainCoef := 0;
      FreeCoef := 0;
      Exit;
    end;
    dxy := dxy + y[i] / x[i];
    ax := ax + 1 / x[i];
    axsqua := axsqua + 1 / (x[i] * x[i]);
  end;
  ysum := Sum(y);
  b1 := (dxy - (ysum * ax) / N) / (axsqua - (ax * ax) / N);
  b0 := (ysum - b1 * ax) / N;
  MainCoef := b1;
  FreeCoef := b0;
  if DestSeries <> nil then
    for i := 0 to N - 1 do
      if ValueArray <> nil then
        DestSeries.AddXY(ArgumentArray[i],
          b1 / ArgumentArray[i] + b0, '', SeriesColor)
      else
        DestSeries.AddXY(SourceSeries.XValues.Value[i],
          b1 / SourceSeries.XValues.Value[i] + b0, '', SeriesColor);
  if b0 < 0 then
    s := ''
  else
    s := '+ ';
  Hint := Format('%0.3f', [b1]) + '/X ' + s + Format('%0.3f', [b0]);
  x := nil;
  y := nil;
end;

//Степенная регрессия

procedure PowerRegression(ValueArray, ArgumentArray: TDoubleArray;
  SourceSeries, DestSeries: TChartSeries; var MainCoef, FreeCoef: double;
  SeriesColor: TColor; var Hint: string);
var
  b0, b1, lnx, lny, xlnsqua, plnxy: double;
  y, x: array of double;
  i, N: integer;
begin
  if ValueArray <> nil then
    N := Length(ValueArray)
  else
    N := SourceSeries.XValues.Count;
  lnx := 0;
  lny := 0;
  xlnsqua := 0;
  plnxy := 0;
  SetLength(x, N);
  SetLength(y, N);
  for i := 0 to N - 1 do
  begin
    if ValueArray <> nil then
    begin
      y[i] := ValueArray[i];
      x[i] := ArgumentArray[i];
    end
    else
    begin
      y[i] := SourceSeries.YValues.Value[i];
      x[i] := SourceSeries.XValues.Value[i];
    end;
    if (x[i] <= 0) or (y[i] <= 0) then
    begin
      MessageDlg('Power regression inapplicable...', mtWarning, [mbOk], 0);
      Hint := 'No equation';
      MainCoef := 0;
      FreeCoef := 0;
      Exit;
    end;
    lnx := lnx + ln(x[i]);
    lny := lny + ln(y[i]);
    plnxy := plnxy + ln(x[i]) * ln(y[i]);
    xlnsqua := xlnsqua + ln(x[i]) * ln(x[i]);
  end;
  b1 := (lnx * lny - N * plnxy) / (lnx * lnx - N * xlnsqua);
  b0 := exp((lny - b1 * lnx) / N);
  MainCoef := b1;
  FreeCoef := b0;
  if DestSeries <> nil then
    for i := 0 to N - 1 do
      if ValueArray <> nil then
        DestSeries.AddXY(ArgumentArray[i],
          Power(ArgumentArray[i], b1) * b0, '', SeriesColor)
      else
        DestSeries.AddXY(SourceSeries.XValues.Value[i],
          Power(SourceSeries.XValues.Value[i], b1) * b0, '', SeriesColor);
  Hint := Format('%0.3f', [b0]) + '*X^' + Format('%0.3f', [b1]);
  x := nil;
  y := nil;
end;

//Полиномиальная регрессия

procedure PolynomialRegression(ValueArray, ArgumentArray: TDoubleArray;
  SourceSeries, DestSeries: TChartSeries; PolyDegree: integer;
  var ArrayCoefs: TDoubleArray; SeriesColor: TColor; var Hint: string);
var
  bcoef, dcoef: TDoubleArray;
  ccoef: array of TDoubleArray;
  i, j, k, N: integer;
  polynom: double;
begin
  if ValueArray <> nil then
    N := Length(ValueArray)
  else
    N := SourceSeries.XValues.Count;
  Hint := '';
  ArrayCoefs := nil;
  SetLength(ccoef, PolyDegree + 1);
  for i := 0 to Length(ccoef) - 1 do
    SetLength(ccoef[i], PolyDegree + 1);
  SetLength(dcoef, PolyDegree + 1);
  SetLength(bcoef, PolyDegree + 1);
  for i := 0 to Length(dcoef) - 1 do
  begin
    dcoef[i] := 0;
    for j := 0 to N - 1 do
    begin
      if ValueArray <> nil then
        dcoef[i] := dcoef[i] +
          Power(ArgumentArray[j], i) * ValueArray[j]
        else
        dcoef[i] := dcoef[i] + Power(SourceSeries.XValues.Value[j], i) *
          SourceSeries.YValues.Value[j];
    end;
    for j := 0 to Length(ccoef) - 1 do
    begin
      ccoef[i][j] := 0;
      for k := 0 to N - 1 do
      begin
        if ValueArray <> nil then
          ccoef[i][j] :=
            ccoef[i][j] + Power(ArgumentArray[k], i + j)
          else
          ccoef[i][j] := ccoef[i][j] + Power(SourceSeries.XValues.Value[k], i +
            j);
      end;
    end;
  end;
  for i := 0 to Length(ccoef) - 2 do
    for j := i + 1 to Length(ccoef) - 1 do
    begin
      ccoef[j][i] := -ccoef[j][i] / ccoef[i][i];
      dcoef[j] := dcoef[j] + ccoef[j][i] * dcoef[i];
      for k := i + 1 to Length(ccoef) - 1 do
        ccoef[j][k] := ccoef[j][k] + ccoef[j][i] * ccoef[i][k];
    end;
  bcoef[Length(bcoef) - 1] := dcoef[Length(dcoef) - 1] /
    ccoef[Length(bcoef) - 1][Length(bcoef) - 1];
  for i := Length(ccoef) - 2 downto 0 do
  begin
    for j := i + 1 to Length(ccoef) - 1 do
      bcoef[i] := bcoef[i] + bcoef[j] * ccoef[i][j];
    bcoef[i] := (dcoef[i] - bcoef[i]) / ccoef[i][i];
  end;
  SetLength(ArrayCoefs, Length(bcoef));
  for i := 0 to Length(bcoef) - 1 do
    ArrayCoefs[i] := bcoef[i];
  if DestSeries <> nil then
    for i := 0 to N - 1 do
    begin
      polynom := 0;
      if ValueArray <> nil then
      begin
        for j := 0 to PolyDegree do
          polynom := polynom + bcoef[j] * Power(ArgumentArray[i], j);
        DestSeries.AddXY(ArgumentArray[i], polynom, '', SeriesColor);
      end
      else
      begin
        for j := 0 to PolyDegree do
          polynom := polynom +
            bcoef[j] * Power(SourceSeries.XValues.Value[i], j);
        DestSeries.AddXY(SourceSeries.XValues.Value[i],
          polynom, '', SeriesColor);
      end;
    end;
  for j := PolyDegree downto 0 do
    Hint := Hint + Format('%0.3f', [bcoef[j]]) + '*X^' + IntToStr(j);
  dcoef := nil;
  bcoef := nil;
  ccoef := nil;
end;

//Показательная регрессия

procedure ExponentRegression(ValueArray, ArgumentArray: TDoubleArray;
  SourceSeries, DestSeries: TChartSeries; var MainCoef, FreeCoef: double;
  SeriesColor: TColor; var Hint: string; Warning: boolean);
var
  i, N: integer;
  x, y: array of double;
  lgy, xsum, xsqua, a, b, lga, xlgy, lgb: double;
begin
  if ValueArray <> nil then
    N := Length(ValueArray)
  else
    N := SourceSeries.XValues.Count;
  lgy := 0;
  xsqua := 0;
  xlgy := 0;
  SetLength(x, N);
  SetLength(y, N);
  for i := 0 to N - 1 do
  begin
    if ValueArray <> nil then
    begin
      y[i] := ValueArray[i];
      x[i] := ArgumentArray[i];
    end
    else
    begin
      y[i] := SourceSeries.YValues.Value[i];
      x[i] := SourceSeries.XValues.Value[i];
    end;
    if y[i] <= 0 then
    begin
      if Warning then
        MessageDlg('Exponent regression inapplicable',
          mtWarning, [mbOk], 0);
      Hint := 'No equation';
      MainCoef := 0;
      FreeCoef := 0;
      Exit;
    end;
    lgy := lgy + Log10(y[i]);
    xsqua := xsqua + x[i] * x[i];
    xlgy := xlgy + x[i] * Log10(y[i]);
  end;
  xsum := Sum(x);
  lgb := (xlgy - (lgy * xsum) / N) / (xsqua - (xsum * xsum) / N);
  lga := (lgy - lgb * xsum) / N;
  b := Power(10, lgb);
  a := Power(10, lga);
  MainCoef := b;
  FreeCoef := a;
  if DestSeries <> nil then
    for i := 0 to N - 1 do
      if ValueArray <> nil then
        DestSeries.AddXY(ArgumentArray[i],
          a * Power(b, ArgumentArray[i]), '', SeriesColor)
      else
        DestSeries.AddXY(SourceSeries.XValues.Value[i],
          a * Power(b, SourceSeries.XValues.Value[i]), '', SeriesColor);
  Hint := 'Exponent regression equation: Y = ' +
    Format('%0.5f', [a]) + ' * (' + Format('%0.5f', [b]) + ' ^ X)';
  x := nil;
  y := nil;
end;

//Экспоненциальная регрессия

procedure ExponentialRegression(ValueArray, ArgumentArray: TDoubleArray;
  SourceSeries, DestSeries: TChartSeries; var MainCoef, FreeCoef: double;
  SeriesColor: TColor; var Hint: string; Warning: boolean);
var
  i, N: integer;
  x, y: array of double;
  lny, xsum, xsqua, xlny, b0, b1: double;
begin
  MainCoef := 0;
  FreeCoef := 0;
  if ValueArray <> nil then
    N := Length(ValueArray)
  else
    N := SourceSeries.XValues.Count;
  lny := 0;
  xsqua := 0;
  xlny := 0;
  SetLength(x, N);
  SetLength(y, N);
  for i := 0 to N - 1 do
  begin
    if ValueArray <> nil then
    begin
      y[i] := ValueArray[i];
      x[i] := ArgumentArray[i];
    end
    else
    begin
      y[i] := SourceSeries.YValues.Value[i];
      x[i] := SourceSeries.XValues.Value[i];
    end;
    if y[i] <= 0 then
    begin
      if Warning then
        MessageDlg('Exponential regression inapplicable',
          mtWarning, [mbOk], 0);
      Hint := 'No equation';
      MainCoef := 0;
      FreeCoef := 0;
      Exit;
    end;
    lny := lny + Ln(y[i]);
    xsqua := xsqua + x[i] * x[i];
    xlny := xlny + x[i] * Ln(y[i]);
  end;
  xsum := Sum(x);
  b1 := (xsum * lny - N * xlny) / (xsum * xsum - N * xsqua);
  b0 := exp((lny - b1 * xsum) / N);
  MainCoef := b1;
  FreeCoef := b0;
  if DestSeries <> nil then
    for i := 0 to N - 1 do
      if ValueArray <> nil then
        DestSeries.AddXY(ArgumentArray[i],
          b0 * Exp(b1 * ArgumentArray[i]), '', SeriesColor)
      else
        DestSeries.AddXY(SourceSeries.XValues.Value[i],
          b0 * Exp(b1 * SourceSeries.XValues.Value[i]), '', SeriesColor);
  Hint := 'Exponential regression equation: Y = ' +
    Format('%0.5f', [b0]) + ' * exp(' + Format('%0.5f', [b1]) + ' * X)';
  x := nil;
  y := nil;
end;

//Степенно-экспоненциальная регрессия

procedure ExpPowerRegression(ValueArray, ArgumentArray: TDoubleArray;
  SourceSeries, DestSeries: TChartSeries; var MainCoef, FreeCoef: double;
  SeriesColor: TColor; var Hint: string; Warning: boolean);
var
  i, N: integer;
  x, y: array of double;
  matr: array[0..3] of double;
  lny, xsum, xsqua, xlny, b0, b1: double;
begin
  MainCoef := 0;
  FreeCoef := 0;
  if ValueArray <> nil then
    N := Length(ValueArray)
  else
    N := SourceSeries.XValues.Count;
  lny := 0;
  xsqua := 0;
  xlny := 0;
  SetLength(x, N);
  SetLength(y, N);
  for i := 0 to N - 1 do
  begin
    if ValueArray <> nil then
    begin
      y[i] := ValueArray[i];
      x[i] := ArgumentArray[i];
    end
    else
    begin
      y[i] := SourceSeries.YValues.Value[i];
      x[i] := SourceSeries.XValues.Value[i];
    end;
    if y[i] <= 0 then
    begin
      if Warning then
        MessageDlg('Exponent-Power regression inapplicable',
          mtWarning, [mbOk], 0);
      Hint := 'No equation';
      MainCoef := 0;
      FreeCoef := 0;
      Exit;
    end;
    lny := lny + Ln(y[i]);
    xsqua := xsqua + x[i] * x[i];
    xlny := xlny + x[i] * Ln(y[i]);
  end;
  xsum := Sum(x);
  b1 := (xsum * lny - N * xlny) / (xsum * xsum - N * xsqua);
  b0 := exp((lny - b1 * xsum) / N);
  MainCoef := b1;
  FreeCoef := b0;
  if DestSeries <> nil then
    for i := 0 to N - 1 do
      if ValueArray <> nil then
        DestSeries.AddXY(ArgumentArray[i],
          b0 * Exp(b1 * ArgumentArray[i]), '', SeriesColor)
      else
        DestSeries.AddXY(SourceSeries.XValues.Value[i],
          b0 * Exp(b1 * SourceSeries.XValues.Value[i]), '', SeriesColor);
  Hint := 'Exponent-Power regression equation: Y = ' +
    Format('%0.5f', [b0]) + ' * exp(' + Format('%0.5f', [b1]) + ' * X)';
  x := nil;
  y := nil;
end;

//Общая процедура проверки массива

procedure CheckArrayBounds(var CArray: TDoubleArray; var FromPoint, ToPoint:
  integer);
begin
  if FromPoint < 0 then
    FromPoint := 0;
  if (ToPoint <= 0) or (ToPoint > Length(CArray) - 1) then
    ToPoint := Length(CArray) - 1;
  if FromPoint > ToPoint then
    ToPoint := FromPoint;
end;

//Общая процедура проверки серии

procedure CheckSeriesBounds(CSeries: TChartSeries; var FromPoint, ToPoint:
  integer);
begin
  if FromPoint < 0 then
    FromPoint := 0;
  if (ToPoint <= 0) or (ToPoint > CSeries.XValues.Count - 1) then
    ToPoint := CSeries.XValues.Count - 1;
  if FromPoint > ToPoint then
    ToPoint := FromPoint;
end;

//Извлечение массива из массива

procedure ArrayFromArray(var SourceArray, DestArray: TDoubleArray;
  FromPoint, ToPoint, Discrete: integer; Derivative: boolean);
var
  i: integer;
begin
  DestArray := nil;
  if SourceArray = nil then
    DestArray := nil
  else
  begin
    CheckArrayBounds(SourceArray, FromPoint, ToPoint);
    if Discrete = 0 then
      Discrete := 1;
    if Derivative = false then
    begin
      SetLength(DestArray, ((ToPoint - FromPoint) div Discrete) + 1);
      for i := 0 to Length(DestArray) - 1 do
        DestArray[i] :=
          SourceArray[i * Discrete + FromPoint];
    end
    else
    begin
      SetLength(DestArray, ((ToPoint - FromPoint) div Discrete));
      for i := 1 to Length(DestArray) do
        DestArray[i - 1] :=
          (SourceArray[i * Discrete + FromPoint] -
          SourceArray[i * Discrete + FromPoint - 1]) / Discrete;
    end;
  end;
end;

//Извлечение массива из серии

procedure ArrayFromSeries(var ValueArray: TDoubleArray; DataSeries:
  TChartSeries;
  FromPoint, ToPoint, Discrete: integer; Derivative: boolean);
var
  i: integer;
begin
  if DataSeries = nil then
    ValueArray := nil
  else
    with DataSeries do
    begin
      CheckSeriesBounds(DataSeries, FromPoint, ToPoint);
      if Discrete = 0 then
        Discrete := 1;
      if Derivative = false then
      begin
        SetLength(ValueArray, ((ToPoint - FromPoint) div Discrete) + 1);
        for i := 0 to Length(ValueArray) - 1 do
          ValueArray[i] :=
            YValues.Value[i * Discrete + FromPoint];
      end
      else
      begin
        SetLength(ValueArray, ((ToPoint - FromPoint) div Discrete));
        for i := 1 to Length(ValueArray) do
          ValueArray[i - 1] :=
            (YValues.Value[i * Discrete + FromPoint] - YValues.Value[i * Discrete
              + FromPoint - 1]) /
            (XValues.Value[i * Discrete + FromPoint] -
            XValues.Value[i * Discrete + FromPoint - 1]);
      end;
    end;
end;

//Извлечение серии из массива

procedure SeriesFromArray(var ValueArray: TDoubleArray; DataSeries:
  TChartSeries;
  FromPoint, ToPoint, Discrete: integer; Derivative: boolean);
var
  i, n: integer;
begin
  if DataSeries = nil then
    Exit
  else
    with DataSeries do
    begin
      Clear;
      CheckArrayBounds(ValueArray, FromPoint, ToPoint);
      if Discrete = 0 then
        Discrete := 1;
      if Derivative = false then
      begin
        n := ((ToPoint - FromPoint) div Discrete) + 1;
        for i := 0 to n - 1 do
          DataSeries.AddXY(i, ValueArray[i * Discrete + FromPoint],
            '', DataSeries.SeriesColor);
      end
      else
      begin
        n := (ToPoint - FromPoint) div Discrete;
        for i := 1 to n do
          DataSeries.AddXY(i - 1, (ValueArray[i * Discrete + FromPoint] -
            ValueArray[i * Discrete + FromPoint - 1]) / Discrete,
            '', DataSeries.SeriesColor);
      end;
    end;
end;

//Извлечение производной из массива

function DerivFromArray(var SourceArray: TDoubleArray; FromPoint, ToPoint,
  Discrete: integer; Extremum: string; var Position: integer): double;
var
  i: integer;
  d: double;
begin
  DerivFromArray := 0;
  if SourceArray = nil then
    DerivFromArray := 0
  else
  begin
    CheckArrayBounds(SourceArray, FromPoint, ToPoint);
    if Discrete = 0 then
      Discrete := 1;
    SetLength(rv, (ToPoint - FromPoint) div Discrete);
    for i := 1 to Length(rv) do
      rv[i - 1] := (SourceArray[i * Discrete + FromPoint] -
        SourceArray[i * Discrete + FromPoint - 1]) / Discrete;
    if Extremum = 'max' then
      d := MaxValue(rv);
    if Extremum = 'min' then
      d := MinValue(rv);
    if Extremum = 'mean' then
      d := Mean(rv);
    for i := 0 to Length(rv) - 1 do
      if rv[i] = d then
      begin
        Position := i;
        break;
      end;
    DerivFromArray := d;
  end;
end;

//Извлечение производной из серии

function DerivFromSeries(DataSeries: TChartSeries; FromPoint, ToPoint,
  Discrete: integer; Extremum: string; var Position: integer): double;
var
  i: integer;
  d: double;
begin
  DerivFromSeries := 0;
  if DataSeries = nil then
    DerivFromSeries := 0
  else
    with DataSeries do
    begin
      CheckSeriesBounds(DataSeries, FromPoint, ToPoint);
      if Discrete = 0 then
        Discrete := 1;
      SetLength(rv, (ToPoint - FromPoint) div Discrete);
      for i := 1 to Length(rv) do
        rv[i - 1] := (YValues.Value[i * Discrete + FromPoint] -
          YValues.Value[i * Discrete + FromPoint - 1]) / (XValues.Value[i *
            Discrete + FromPoint] -
          XValues.Value[i * Discrete + FromPoint - 1]);
      if Extremum = 'max' then
        d := MaxValue(rv);
      if Extremum = 'min' then
        d := MinValue(rv);
      if Extremum = 'mean' then
        d := Mean(rv);
      for i := 0 to Length(rv) - 1 do
        if rv[i] = d then
        begin
          Position := i;
          break;
        end;
      DerivFromSeries := d;
    end;
end;

//Извлечение величины из серии

function ValueFromSeries(DataSeries: TChartSeries; FromPoint, ToPoint: integer;
  Extremum: string; var Position: integer): double;
var
  i: integer;
  d: double;
begin
  if DataSeries = nil then
    ValueFromSeries := 0
  else
    with DataSeries do
    begin
      CheckSeriesBounds(DataSeries, FromPoint, ToPoint);
      SetLength(rv, ToPoint - FromPoint);
      for i := 0 to Length(rv) - 1 do
        rv[i] := YValues.Value[FromPoint + i];
      if Extremum = 'max' then
        d := MaxValue(rv);
      if Extremum = 'min' then
        d := MinValue(rv);
      if Extremum = 'mean' then
        d := Mean(rv);
      for i := 0 to Length(rv) - 1 do
        if rv[i] = d then
        begin
          Position := i;
          break;
        end;
      ValueFromSeries := d;
    end;
end;

//Извлечение величины из массива

function ValueFromArray(var SourceArray: TDoubleArray; FromPoint,
  ToPoint: integer; Extremum: string; var Position: integer): double;
var
  i: integer;
  d: double;
begin
  if SourceArray = nil then
    ValueFromArray := 0
  else
  begin
    CheckArrayBounds(SourceArray, FromPoint, ToPoint);
    SetLength(rv, ToPoint - FromPoint);
    for i := 0 to Length(rv) - 1 do
      rv[i] := SourceArray[FromPoint + i];
    if Extremum = 'max' then
      d := MaxValue(rv);
    if Extremum = 'min' then
      d := MinValue(rv);
    if Extremum = 'mean' then
      d := Mean(rv);
    for i := 0 to Length(rv) - 1 do
      if rv[i] = d then
      begin
        Position := i;
        break;
      end;
    ValueFromArray := d;
  end;
end;

//Вычисление площади под кривой, получаемой данными из массива

function CalculateAreaOfArray(var SourceArray: TDoubleArray;
  FromPoint, ToPoint, Method: integer; BindToZero: boolean): double;
var
  i: integer;
  sq, subv: double;
begin
  if SourceArray = nil then
    CalculateAreaOfArray := 0
  else
  begin
    CheckArrayBounds(SourceArray, FromPoint, ToPoint);
    sq := 0;
    if BindToZero then
      subv :=
        (SourceArray[ToPoint] + SourceArray[FromPoint]) / 2
      else
      subv := 0;
    for i := FromPoint to ToPoint - 1 do
    begin
      if Method = 1 then
        sq := sq + Abs(SourceArray[i] - subv) +
          (Abs(SourceArray[i + 1] - subv) - Abs(SourceArray[i] - subv)) / 2;
      if Method = 2 then
        sq := sq + Abs(SourceArray[i] - subv) +
          (Abs(SourceArray[i + 1] - subv) - Abs(SourceArray[i] - subv)) / 2 - 1
            / (48 * Power(0.5, 1.5));
      if Method = 3 then
        if (i mod 2) = 1 then
          sq := sq + 2 * Abs(SourceArray[i] - subv);
      if Method = 4 then
        if (i mod 2) = 1 then
          sq := sq + 2 * Abs(SourceArray[i] - subv) - 1 / (96 * Power(0.5,
            1.5));
    end;
    CalculateAreaOfArray := sq;
  end;
end;

//Вычисление площади под кривой, получаемой данными из серии

function CalculateAreaOfSeries(DataSeries: TChartSeries; FromPoint, ToPoint,
  Method: integer; BindToZero: boolean): double;
var
  i: integer;
  sq, subv: double;
begin
  if DataSeries = nil then
    CalculateAreaOfSeries := 0
  else
    with DataSeries do
    begin
      CheckSeriesBounds(DataSeries, FromPoint, ToPoint);
      sq := 0;
      if BindToZero then
        subv := (YValues.Value[ToPoint] +
          YValues.Value[FromPoint]) / 2
      else
        subv := 0;
      for i := FromPoint to ToPoint - 1 do
      begin
        if Method = 1 then
          sq := sq + Abs(YValues.Value[i] - subv) +
            (Abs(YValues.Value[i + 1] - subv) - Abs(YValues.Value[i] - subv)) /
              2;
        if Method = 2 then
          sq := sq + Abs(YValues.Value[i] - subv) +
            (Abs(YValues.Value[i + 1] - subv) -
            Abs(YValues.Value[i] - subv)) / 2 - 1 / (48 * Power(0.5, 1.5));
        if Method = 3 then
          if (i mod 2) = 1 then
            sq := sq + 2 * Abs(YValues.Value[i] - subv);
        if Method = 4 then
          if (i mod 2) = 1 then
            sq := sq + 2 * Abs(YValues.Value[i] - subv) - 1 / (96 * Power(0.5,
              1.5));
      end;
      CalculateAreaOfSeries := sq;
    end;
end;

//Исключение линейной составляющей

procedure LinearTrendExclusion(var ValueArray: TDoubleArray);
var
  i, N: integer;
  b0, b1, nx: double;
begin
  N := Length(ValueArray);
  nx := 0;
  for i := 0 to N - 1 do
    nx := nx + (i + 1) * ValueArray[i];
  b0 := (2 * (2 * N + 1) * Sum(ValueArray) - 6 * nx) / (N * (N - 1));
  b1 := (12 * nx - 6 * (N + 1) * Sum(ValueArray)) / (N * (N - 1) * (N + 1));
  for i := 0 to N - 1 do
  begin
    ValueArray[i] := ValueArray[i] - (i * b1);
  end;
end;

//Расцветка серии

procedure ColorizeSeries(DataSeries: TChartSeries; NewColor: TColor);
var
  i: integer;
begin
  for i := 0 to DataSeries.XValues.Count - 1 do
    DataSeries.ValueColor[i] := NewColor;
end;

//Задание нового приращения по оси X

procedure SetXInterval(DataSeries: TChartSeries; XInterval: double);
var
  i: integer;
begin
  for i := 0 to DataSeries.XValues.Count - 1 do
    DataSeries.XValues.Value[i] := DataSeries.XValues.Value[i] * XInterval;
end;

//Привязка серии к новой оси

procedure SetSeriesAxis(DataSeries: TChartSeries; NewAxis: TVertAxis);
begin
  DataSeries.VertAxis := NewAxis;
end;

end.
Проект Delphi World © Выпуск 2002 - 2024
Автор проекта: USU Software
Вы можете выкупить этот проект.