CN104933489A - Wind power real-time high precision prediction method based on adaptive neuro-fuzzy inference system - Google Patents

Abstract

A wind power real-time high precision prediction method based on an adaptive neuro-fuzzy inference system belongs to the wind power technical field. The method is characterized in that firstly, an experiment scheme is designed to collect and process data, establishing a multi-step rolling prediction mode, then constructing a wind power prediction model based on an adaptive neuro-fuzzy inference system (ANFIS), and finally evaluating the prediction precision. In this way, a wind power multi-step rolling real-time prediction method that is based on data, has a self-learning capability and can meet the requirement of real-time prediction precision is achieved. The wind power real-time high precision prediction method based on an adaptive neuro-fuzzy inference system provides a prediction result that is quite close to a real value and is effective, is high in prediction precision, and is highly operable.

Description

Based on the wind power real-time high-precision Forecasting Methodology of Adaptive Neuro-fuzzy Inference

Technical field

The present invention relates to a kind of wind power real-time high-precision Forecasting Methodology based on Adaptive Neuro-fuzzy Inference, belong to technical field of wind power.

Background technology

Wind energy is free, so people wish that the electric energy produced by wind energy is accepted by electrical network as much as possible.But wind has randomness and intermittence, the available power supply that it produces is unknown, will bring stern challenge when allowing a large amount of wind-electricity integration to electric system.Wind power prediction plays key player in these challenges of reply.Electric system must have powerful dispatching, increases the power swing problem that Wind turbines brings newly, thus realize improving constantly of wind-powered electricity generation permeability with reason herein.Therefore, develop perfervid period at new forms of energy, short-term wind-electricity prediction is very important technical research, by carrying out the accurately predicting of short-term to wind energy turbine set generated energy, can alleviate electric system frequency modulation, peak regulation pressure, improves wind-powered electricity generation and receives ability.Wind power real-time multistep rail vehicle roller test-rig requires the wind power prediction data reporting following 15 minutes to 4 hours of rolling for every 15 minutes, like this for wind power real-time estimate, need every day to carry out 96 predictions, the data volume of each forecast is 16 (i.e. l=16).Therefore how to predict for the real-time high-precision of wind power under multistep rolling forecast pattern the technical barrier just becoming and be badly in need of solving.So, by data acquisition and process, set up multistep rolling forecast pattern, build the wind power prediction model based on Adaptive Neuro-fuzzy Inference (ANFIS) and evaluate precision of prediction, it is necessary for inventing a kind of wind power real-time high-precision Forecasting Methodology based on Adaptive Neuro-fuzzy Inference.

Summary of the invention

In order to overcome a difficult problem for the real-time high-precision prediction of wind power under multistep rolling forecast pattern, the invention provides a kind of wind power real-time high-precision Forecasting Methodology based on Adaptive Neuro-fuzzy Inference, should based on the wind power real-time high-precision Forecasting Methodology first contrived experiment scheme of Adaptive Neuro-fuzzy Inference, sampling and processing is carried out to data, set up multistep rolling forecast pattern, build the wind power prediction model based on Adaptive Neuro-fuzzy Inference (ANFIS), finally precision of prediction is evaluated, reach and a kind of (non-intuitive is any given) based on data is provided, there is self-learning ability, meet the object of the wind power multistep rolling real-time predicting method of real-time estimate accuracy requirement.

Technical scheme based on the wind power real-time high-precision Forecasting Methodology of Adaptive Neuro-fuzzy Inference comprises the following steps:

(1) data acquisition and process

Gather the actual wind power data at each Wind turbine of wind energy turbine set every 15 minutes intervals, for multistep rolling forecast pattern, known historical data is divided into two parts, a part is as training set (being divided into input and output), and another part is as input data during prediction; First to the input in training set, export data and form initial fuzzy inference system, the membership function etc. of the membership function of input variable and input variable, fuzzy rule, output variable and output variable connects by this initial fuzzy inference system.When forming initial fuzzy inference system, the present invention adopts subtraction clustering algorithm, effectively can avoid the shot array problem that artificial setting structure method produces; Based on this, the Parameter Learning Algorithm (back-propagation algorithm of back-propagation algorithm or least square method) using comparative maturity in neural network to the input in training set, export the form parameter that data learn to adjust the membership function of variable in fuzzy inference system, to make this model can constantly approach given training set data, final formed have data study adjustment capability, can be used to the fuzzy inference system that carries out predicting.

(2) multistep rolling forecast pattern is set up

When carrying out wind power prediction, the actual value P (t-n Δ t) of all moment wind powers in general known modeling territory, n=0,1,2...N, therefore the historical data quantity in modeling territory is N+1, the wind power needing prediction is P (t+l Δ t), l=1,2...L, L is the step number of multi-step prediction, order represent the wind power prediction value under rolling multi-step prediction pattern, have under rolling multi-step prediction mode:

${\hat{P}}_r(t+\mathrm{l\Δt})=f(P(t-(N-l+1)\mathrm{\Δt}),...,P(t-\mathrm{\Δt}),P\left(t\right),{\hat{P}}_r(t+\mathrm{\Δt}),...,{\hat{P}}_r(t+(1-1)\mathrm{\Δt}))---\left(1\right)$

The mapping relations that wherein the selected Forecasting Methodology of f representative is corresponding.

(3) wind power prediction model based on Adaptive Neuro-fuzzy Inference (ANFIS) is set up

Whole modeling process divides 5 layers to carry out, and uses O _{k, i}represent the output of i-th node of kth layer.

1st layer: in this layer, each node i is represented (this layer parameter is variable) by node function:

$O_{1,i}=\left\{\begin{array}{cc}{\mathrm{\μ}}_{\mathrm{Ai}}\left(x_1\right)& i=\mathrm{1,2}\\ {\mathrm{\μ}}_{B(i-2)}\left(x_2\right)& i=\mathrm{3,4}\end{array}\right.---\left(2\right)$

Wherein: x ₁(or x ₂) be the input of node i, A _i(or B _i-2) be the language amount relevant to this node function value, as " greatly " or " little " etc.In other words, O _{1, i}fuzzy set A (A=A ₁, A ₂, B ₁, B ₂) membership function, usually can select Gaussian function (used herein) and bell shaped function etc.

2nd layer: the node of this layer represents with ∏ in Fig. 1 of Figure of description, is multiplied by input signal, its product is fuzzy rule excitation density wi, it can be used as the output of the 2nd layer:

Q _2，i＝w _i＝μ _Ai(x ₁)μ _Bi(x ₂)，i＝1，2 (3)

3rd layer: the node of this layer represents with N in the drawings, what this layer carried out is excitation density normalized, and its output is:

$O_{3,i}=\stackrel{\‾}{w_i}=\frac{w_i}{w_1+w_2},i=\mathrm{1,2}---\left(4\right)$

4th layer: this layer each node is self-adaptation node, should calculate the contribution of every rule, and its output is:

$Q_{4,i}=\stackrel{\‾}{w_i}{f}_i=\stackrel{\‾}{w_i}(p_i{x}_1+q_i{x}_2+r_i),i=\mathrm{1,2}---\left(5\right)$

Wherein, p _i, q _iand r _ibe consequent parameter.

5th layer: the final output calculating strictly all rules, the total output namely calculating all input signals is:

$O_{5,i}=z=\underset{i}{\mathrm{\Σ}}\stackrel{\‾}{w_i}{f}_i=\frac{\underset{i}{\mathrm{\Σ}}{w}_i{f}_i}{\underset{i}{\mathrm{\Σ}}{w}_i}---\left(6\right)$

Therefore using the input of wind power historical data known in modeling territory as ANFIS forecast model, thus the wind power prediction value predicting territory can be obtained.

ANFIS forecast model is integrated with the independent learning ability of artificial neural network and the language performance function of fuzzy reasoning, being suitable for being applied in the short-term wind power prediction field utilizing historical data to predict Future Data, therefore can obtaining higher precision of prediction when carrying out wind power real-time multistep rolling forecast.

(4) predicted exactitude evaluation

Predicted exactitude evaluation adopts National Energy Board in the administrative provisions to existing wind energy turbine set wind power real-time estimate forecast, and the index of proposition carries out error assessment, thus realizes the evaluation predicted the outcome.The concrete calculating formula of index is as follows:

Per day prediction Plan Curve accuracy rate r ₁:

$r_{1i}=[1-{\sqrt{\frac{1}{16}\underset{k=1}{\overset{16}{\mathrm{\Σ}}}\left(\frac{P_{\mathrm{Mi}}^k-P_{\mathrm{Pi}}^k}{\mathrm{Cap}}\right)}}^2]\×100\%---\left(7\right)$

$r_1=\frac{1}{96}\underset{i=1}{\overset{96}{\mathrm{\Σ}}}{r}_{1i}---\left(8\right)$

In formula, r _1iit is the accuracy rate of i-th real-time prediction; it is the wind power actual value in kth moment in i-th real-time prediction; be the wind power prediction value in kth moment in i-th real-time estimate, Cap is the start capacity of wind energy turbine set.

Per day prediction Plan Curve qualification rate r ₂:

$r_{2i}=\frac{1}{16}\underset{k=1}{\overset{16}{\mathrm{\Σ}}}{B}_i^k\×100\%---\left(9\right)$

$(1-\frac{|P_{\mathrm{Mi}}^k-P_{\mathrm{Pi}}^k|}{\mathrm{Cap}})\×100\%\≥85\%,B_i^k=1$

$(1-\frac{|P_{\mathrm{Mi}}^k-P_{\mathrm{Pi}}^k|}{\mathrm{Cap}})\&times;100\%<85\%,B_i^k=0$

$r_2=\frac{1}{96}\underset{i=1}{\overset{96}{\mathrm{\&Sigma;}}}{r}_{2i}---\left(10\right)$

In formula, r _2iit is the qualification rate of i-th real-time estimate.

Whole day predicts the outcome root-mean-square error r ₃:

$r_3=\sqrt{\frac{1}{96\&times;16}\underset{i=1}{\overset{96}{\mathrm{\&Sigma;}}}\underset{k=1}{\overset{16}{\mathrm{\&Sigma;}}}{\left(\frac{P_{\mathrm{Mi}}^k-P_{\mathrm{Pi}}^k}{\mathrm{Cap}}\right)}^2}\&times;100\%---\left(11\right)$

As accuracy rate r ₁, qualification rate r ₂larger, whole day predicts the outcome root-mean-square error r ₃more hour, precision of prediction is higher.

The beneficial effect that the present invention is based on the wind power real-time high-precision Forecasting Methodology of Adaptive Neuro-fuzzy Inference is: (1) ANFIS forecast model is integrated with the independent learning ability of artificial neural network and the language performance function of fuzzy reasoning, the closely actual value that predicts the outcome of ANFIS forecast model, is suitable for being applied in the short-term wind power prediction field utilizing historical data to predict Future Data; (2) what adopt when forming initial fuzzy system architecture is the method for subtractive clustering, effectively avoids the shot array problem that artificial setting structure method produces; (3) precision of prediction of the method is high, no matter be single unit or whole field unit, its accuracy rate, qualification rate are all higher, whole day predicts the outcome root-mean-square error also all lower than 20%, meet the requirements, describe the validity of ANFIS forecast model, whole day particularly in the unit total wind power prediction evaluation of result index of 267, the whole field root-mean-square error that predicts the outcome is 7.03%, and predicting the outcome root-mean-square error well below the whole day of National Energy Board to wind power real-time estimate precision must lower than the requirement of 20%.Wind power real-time high-precision Forecasting Methodology based on Adaptive Neuro-fuzzy Inference predicts the outcome closely actual value, and predict the outcome effectively, precision of prediction is high, workable.

Accompanying drawing explanation

Fig. 1 is the ANFIS model structure of the wind power real-time high-precision Forecasting Methodology based on Adaptive Neuro-fuzzy Inference.

Fig. 2 is Power Output for Wind Power Field curve on the sunny side and the prediction curve comparison diagram under other Forecasting Methodologies of wind energy turbine set wind power prediction curve and reality on the sunny side under the Forecasting Methodology based on the wind power real-time high-precision Forecasting Methodology of Adaptive Neuro-fuzzy Inference, in figure: mark dashed curve is the prediction output power of ANFIS model, block curve is real output, wherein ordinate is output power value, abscissa representing time, wherein each scale represents 15 minutes.

Embodiment

Drawings and Examples are utilized to be described in detail to the wind power real-time high-precision Forecasting Methodology based on Adaptive Neuro-fuzzy Inference of the present invention below.

Embodiment one

Wind power real-time high-precision Forecasting Methodology based on Adaptive Neuro-fuzzy Inference of the present invention, it is characterized in that, it comprises the following steps:

(1) data acquisition and process

Gather that wind energy turbine set on August 1st, 2012 on the sunny side, data sampling was spaced apart No. 91 Wind turbines wind powers of 15min and the total wind power of 267, whole field unit to August 30.Subordinate function is Gaussian function; Number is two; Frequency of training is 2000 times; What training sample was chosen is the wind power actual value of August 5 to August 7;

(2) multistep rolling forecast pattern is set up

When carrying out wind power prediction, the actual value P (t-n Δ t) of all moment wind powers in general known modeling territory, n=0,1,2...N, therefore the historical data quantity in modeling territory is N+1, the wind power needing prediction is P (t+l Δ t), l=1,2...L, L is the step number of multi-step prediction, order represent the wind power prediction value under rolling multi-step prediction pattern, have under rolling multi-step prediction mode:

${\hat{P}}_r(t+\mathrm{l\&Delta;t})=f(P(t-(N-l+1)\mathrm{\&Delta;t}),...,P(t-\mathrm{\&Delta;t}),P\left(t\right),{\hat{P}}_r(t+\mathrm{\&Delta;t}),...,{\hat{P}}_r(t+(1-1)\mathrm{\&Delta;t}))---\left(1\right)$

The mapping relations that wherein the selected Forecasting Methodology of f representative is corresponding.

(3) wind power prediction model based on Adaptive Neuro-fuzzy Inference (ANFIS) is set up

The structure of input ANFIS model is as shown in Fig. 1 of Figure of description, and whole modeling process divides 5 layers to carry out, and uses O _{k, i}represent the output of i-th node of kth layer.

1st layer: in this layer, each node i is represented (this layer parameter is variable) by node function:

$O_{1,i}=\left\{\begin{array}{cc}{\mathrm{\&mu;}}_{\mathrm{Ai}}\left(x_1\right)& i=\mathrm{1,2}\\ {\mathrm{\&mu;}}_{B(i-2)}\left(x_2\right)& i=\mathrm{3,4}\end{array}\right.---\left(2\right)$

Wherein: x _i(or x ₂) be the input of node i, A _i(or B _i-2) be the language amount relevant to this node function value, as " greatly " or " little " etc.In other words, O _{1, i}fuzzy set A (A=A ₁, A ₂, B ₁, B ₂) membership function, usually can select Gaussian function (used herein) and bell shaped function etc.

2nd layer: the node of this layer represents with ∏ in Fig. 1 of Figure of description, is multiplied by input signal, its product is fuzzy rule excitation density w _i, it can be used as the output of the 2nd layer:

O _2，i＝w _i＝μ _Ai(x ₁)μ _Bi(x ₂)，i＝1，2 (3)

3rd layer: the node of this layer represents with N in the drawings, what this layer carried out is excitation density normalized, and its output is:

$O_{3,i}=\stackrel{\&OverBar;}{w_i}=\frac{w_i}{w_1+w_2},i=\mathrm{1,2}---\left(4\right)$

4th layer: this layer each node is self-adaptation node, should calculate the contribution of every rule, and its output is:

$Q_{4,i}=\stackrel{\&OverBar;}{w_i}{f}_i=\stackrel{\&OverBar;}{w_i}(p_i{x}_1+q_i{x}_2+r_i),i=\mathrm{1,2}---\left(5\right)$

Wherein, p _i, q _iand r _ibe consequent parameter.

5th layer: the final output calculating strictly all rules, the total output namely calculating all input signals is:

$O_{5,i}=z=\underset{i}{\mathrm{\&Sigma;}}\stackrel{\&OverBar;}{w_i}{f}_i=\frac{\underset{i}{\mathrm{\&Sigma;}}{w}_i{f}_i}{\underset{i}{\mathrm{\&Sigma;}}{w}_i}---\left(6\right)$

Therefore using the input of wind power historical data known in modeling territory as ANFIS forecast model, thus the wind power prediction value predicting territory can be obtained.

(4) predicted exactitude evaluation

Predicted exactitude evaluation adopts National Energy Board in the administrative provisions to existing wind energy turbine set wind power real-time estimate forecast, and the index of proposition carries out error assessment, thus realizes the evaluation predicted the outcome.

Table 1 and table 2 are for August 8, respectively to No. 91 units and 267, whole field unit wind power under each Forecasting Methodology, predicted exactitude evaluation indicator-specific statistics table when predicting according to specific embodiments.

Table 1 No. 91 unit wind power prediction evaluation of result indexs (August 8)

267, the whole field of table 2 unit total wind power prediction evaluation of result index (August 8)

National Energy Board of the People's Republic of China (PRC) requires as the whole day root-mean-square error that predicts the outcome should be less than 20% to wind power real-time estimate in " wind farm power prediction forecast management tentative method " within 2011, to issue, known by table 1 and table 2, when being predicted by ANFIS model, no matter be single unit or whole field unit, its accuracy rate, qualification rate are all higher, whole day predicts the outcome root-mean-square error also all lower than 20%, meets the requirements, and describes the validity of ANFIS forecast model.

No. 91 units (whole day predicts 96 times altogether, each prediction 16 points) in the 29th prediction on August 8, contrast as shown in Figure 2 based on the predicted value under ANFIS Forecasting Methodology and actual value.Therefrom can find out, in this prediction, the closely actual value that predicts the outcome of ANFIS forecast model.

More than show and describe ultimate principle of the present invention and principal character and advantage of the present invention.The technician of the industry should be appreciated that; the present invention is not restricted to the described embodiments; what describe in above-described embodiment and instructions just illustrates principle of the present invention; without departing from the spirit and scope of the present invention; the present invention also has various changes and modifications; these changes and improvements all fall in the claimed scope of the invention, and application claims protection domain is defined by its equivalent of appending claims.

Claims (1)

1. based on the wind power real-time high-precision Forecasting Methodology of Adaptive Neuro-fuzzy Inference, it is characterized in that: the wind power real-time high-precision Forecasting Methodology based on Adaptive Neuro-fuzzy Inference comprises following data acquisition and process, set up multistep rolling forecast pattern, set up wind power prediction model based on Adaptive Neuro-fuzzy Inference (ANFIS) and predicted exactitude evaluation 4 steps; Based on the precision of prediction of the wind power real-time high-precision Forecasting Methodology of Adaptive Neuro-fuzzy Inference, the single unit whole day root-mean-square error that predicts the outcome is 14.01%, whole day in the unit total wind power prediction evaluation of result index of 267, the whole field root-mean-square error that predicts the outcome is 7.03%, meet the whole day of National Energy Board to the wind power real-time estimate precision root-mean-square error that predicts the outcome and lower than the requirement of 20%, the validity of ANFIS forecast model must be described; 4 step particular contents based on the wind power real-time high-precision Forecasting Methodology of Adaptive Neuro-fuzzy Inference are as follows:

(1) data acquisition and process

Gather the actual wind power data at each Wind turbine of wind energy turbine set every 15 minutes intervals, historical data is divided into two parts, a part is as training set (being divided into input and output), and another part is as input data during prediction;

(2) multistep rolling forecast pattern is set up

When carrying out wind power prediction, the actual value P (t-n Δ t) of all moment wind powers in general known modeling territory, n=0,1,2...N, therefore the historical data quantity in modeling territory is N+1, the wind power needing prediction is P (t+l Δ t), l=1,2...L, L is the step number of multi-step prediction, order represent the wind power prediction value under rolling multi-step prediction pattern, have under rolling multi-step prediction mode:

${\hat{P}}_r(t+\mathrm{l\&Delta;t})=f(P(t-(N-l+1)\mathrm{\&Delta;t}),...,P(t-\mathrm{\&Delta;t}),P\left(t\right),{\hat{P}}_r(t+\mathrm{\&Delta;t}),...{\hat{P}}_r(t+(l-1)\mathrm{\&Delta;t}))---\left(1\right)$

The mapping relations that wherein the selected Forecasting Methodology of f representative is corresponding;

(3) the whole modeling process of wind power prediction model set up based on Adaptive Neuro-fuzzy Inference (ANFIS) divides 5 layers to carry out, and uses O _{k, i}represent the output of i-th node of kth layer;

1st layer: in this layer, each node i is represented (this layer parameter is variable) by node function:

$O_{1,i}=\left\{\begin{array}{cc}{\mathrm{\&mu;}}_{\mathrm{Ai}}\left(x_1\right)& i=\mathrm{1,2}\\ {\mathrm{\&mu;}}_{B(i-2)}\left(x_2\right)& i=\mathrm{3,4}\end{array}\right.---\left(2\right)$

Wherein: x ₁(or x ₂) be the input of node i, A _i(or B _i-2) be the language amount relevant to this node function value, as " greatly " or " little " etc.; In other words, O _{1, i}fuzzy set A (A=A ₁, A ₂, B ₁, B ₂) membership function, usually can select Gaussian function (used herein) and bell shaped function etc.;

2nd layer: the node of this layer represents with ∏ in Fig. 1 of Figure of description, is multiplied by input signal, its product is fuzzy rule excitation density w _i, it can be used as the output of the 2nd layer:

O _2，i＝w _i＝μ _Ai(x ₁)μ _Bi(x ₂)，i＝1，2 (3)

3rd layer: the node of this layer represents with N in the drawings, what this layer carried out is excitation density normalized, and its output is:

$O_{3,i}=\stackrel{\&OverBar;}{w_i}=\frac{w_i}{w_1+w_2},i=\mathrm{1,2}---\left(4\right)$

4th layer: this layer each node is self-adaptation node, should calculate the contribution of every rule, and its output is:

$O_{4,i}=\stackrel{\&OverBar;}{w_i}{f}_i=\stackrel{\&OverBar;}{w_i}(p_i{x}_1+q_i{x}_2+r_i),i=\mathrm{1,2}---\left(5\right)$

Wherein, p _i, q _iand r _ibe consequent parameter;

5th layer: the final output calculating strictly all rules, the total output namely calculating all input signals is:

$O_{5,i}=z=\underset{i}{\mathrm{\&Sigma;}}\stackrel{\&OverBar;}{w_i}{f}_i=\frac{\underset{i}{\mathrm{\&Sigma;}}{w}_i{f}_i}{\underset{i}{\mathrm{\&Sigma;}}{w}_i}---\left(6\right)$

Therefore using the input of wind power historical data known in modeling territory as ANFIS forecast model, thus the wind power prediction value predicting territory can be obtained;

(4) predicted exactitude evaluation

Predicted exactitude evaluation adopts National Energy Board in the administrative provisions to existing wind energy turbine set wind power real-time estimate forecast, and the index of proposition carries out error assessment, thus realizes the evaluation predicted the outcome; The concrete calculating formula of index is as follows:

Per day prediction Plan Curve accuracy rate r ₁:

$r_{1i}=[1-\sqrt{\frac{1}{16}\underset{k=1}{\overset{16}{\mathrm{\&Sigma;}}}{\left(\frac{P_{\mathrm{Mi}}^k-P_{\mathrm{Pi}}^k}{\mathrm{Cap}}\right)}^2}]\&times;100\%---\left(7\right)$

$r_1=\frac{1}{96}\underset{i=1}{\overset{96}{\mathrm{\&Sigma;}}}{r}_{1i}---\left(8\right)$

In formula, r _1iit is the accuracy rate of i-th real-time prediction; it is the wind power actual value in kth moment in i-th real-time prediction; be the wind power prediction value in kth moment in i-th real-time estimate, Cap is the start capacity of wind energy turbine set;

Per day prediction Plan Curve qualification rate r ₂:

$r_{2i}=\frac{1}{16}\underset{k=1}{\overset{16}{\mathrm{\&Sigma;}}}{B}_i^k\&times;100\%---\left(9\right)$

$(1-\frac{|P_{\mathrm{Mi}}^k-P_{\mathrm{Pi}}^k|}{\mathrm{Cap}})\&times;100\%\&GreaterEqual;85\%,B_i^k=1$

$(1-\frac{|P_{\mathrm{Mi}}^k-P_{\mathrm{Pi}}^k|}{\mathrm{Cap}})\&times;100\%<85\%,B_i^k=0$

$r_2=\frac{1}{96}\underset{i=1}{\overset{96}{\mathrm{\&Sigma;}}}{r}_{2i}---\left(10\right)$

In formula, r _2iit is the qualification rate of i-th real-time estimate;

Whole day predicts the outcome root-mean-square error r ₃:

$r_3=\sqrt{\frac{1}{96\&times;16}\underset{i=1}{\overset{96}{\mathrm{\&Sigma;}}}\underset{k=1}{\overset{16}{\mathrm{\&Sigma;}}}{\left(\frac{P_{\mathrm{Mi}}^k-P_{\mathrm{Pi}}^2}{\mathrm{Cap}}\right)}^2}\&times;100\%---\left(11\right)$

As accuracy rate r ₁, qualification rate r ₂larger, whole day predicts the outcome root-mean-square error r ₃more hour, precision of prediction is higher.