public static void ForecastAndExport_v3(string forecast_tag, string data_tag, int forecastPerspective = 2)
{
string tag_name = forecast_tag;
double[] vMax, vMin;
DateTime fdate = DateTime.Today.AddDays(forecastPerspective); // дата прогноза
SOM model = CreateSomModel(data_tag, out vMax, out vMin);
Dictionary <DateTime, double> solution = new Dictionary <DateTime, double>(); // прогноз
TagDbContext context = new TagDbContext();
var data = context.TagValues
.Where(t => t.Tag.TagName.Trim() == data_tag)
.OrderByDescending(v => v.DateTime)
.Select(v => new { v.DateTime.Hour, Day = SqlFunctions.DatePart("weekday", v.DateTime) + 1, Value = v.Value })
.Take(24 + 3 + 1)
.AsEnumerable()
.Reverse()
.ToArray();
for (int index = 0; index < 24; index++)
{
var vector = new double[] {
(double)(vMax[0] - (double)data[index].Hour) / (vMax[0] - vMin[0]),
(double)(vMax[1] - (double)data[index].Day) / (vMax[1] - vMin[1]),
(double)(vMax[2] - data[index + 1].Value) / (vMax[2] - vMin[2]),
(double)(vMax[3] - data[index + 2].Value) / (vMax[3] - vMin[3]),
(double)(vMax[4] - data[index + 3].Value) / (vMax[4] - vMin[4])
};
var value = model.Compute(vector)[0];
solution[fdate.AddHours(data[index].Hour)] = value;
}
ExportToDB(tag_name, solution, context);
}
static SOM CreateSomModel_v2(string tagname)
{
int samples = 1500;
var context = new TagDbContext();
double[] data = context.TagValues
.Where(t => t.Tag.TagName == tagname)
.OrderByDescending(v => v.DateTime)
.Select(v => v.Value)
.Take(samples + retrospective + 1)
.AsEnumerable()
.Reverse()
.ToArray();
var inputs = new Vector[samples];
for (int i = 0; i < samples; i++)
{
inputs[i] = new Vector();
inputs[i].Input = data.Skip(i).Take(retrospective).ToArray();
// set output
inputs[i].Output = new double[] { data[retrospective + i + 1] };
}
SOM model = new SOM(5, 1, 15, 15);
model.Train(inputs, 500);
return(model);
}
static SOM CreateSomModel(string tagname, out double[] vMax, out double[] vMin)
{
int retro = 3;
int samples = 1500;
var context = new TagDbContext();
var data = context.TagValues
.Where(t => t.Tag.TagName.Trim() == tagname)
.OrderByDescending(v => v.DateTime)
.Select(v => new { v.DateTime.Hour, Day = SqlFunctions.DatePart("weekday", v.DateTime) + 1, Value = v.Value })
.Take(samples + retro + 1)
.AsEnumerable()
.Reverse()
.ToArray();
vMax = new double[] { data.Max(d => d.Hour), data.Max(d => (double)d.Day), data.Max(d => d.Value), data.Max(d => d.Value), data.Max(d => d.Value) };
vMin = new double[] { data.Min(d => d.Hour), data.Min(d => (double)d.Day), data.Min(d => d.Value), data.Min(d => d.Value), data.Min(d => d.Value) };
var inputs = new Vector[samples];
for (int index = 0; index < samples; index++)
{
inputs[index] = new Vector();
inputs[index].Input = new double[] {
(double)(vMax[0] - (double)data[index].Hour) / (vMax[0] - vMin[0]),
(double)(vMax[1] - (double)data[index].Day) / (vMax[1] - vMin[1]),
(double)(vMax[2] - data[index + 1].Value) / (vMax[2] - vMin[2]),
(double)(vMax[3] - data[index + 2].Value) / (vMax[3] - vMin[3]),
(double)(vMax[4] - data[index + 3].Value) / (vMax[4] - vMin[4])
};
inputs[index].Output = new double[] { data[index].Value };
}
SOM model = new SOM(5, 1, 15, 15);
model.Train(inputs, 500);
return(model);
}
