CN115115119A - OA-GRU short-term power load prediction method based on grey correlation - Google Patents

Abstract

The invention discloses a short-term power load prediction method based on OA-GRU (office automation-general) associated with gray, which comprises the following steps of: acquiring daily load influence factor data and power load data, calculating Spearman values among the daily load influence factor data load data, acquiring the weight of each influence factor, dividing historical daily influence factor characteristic data and daily influence factor characteristic data to be predicted, performing cluster analysis, determining a classification standard according to the power load data, and determining a classical domain and a nodal domain; calculating gray relevance of each factor, sequencing the gray relevance from big to small, setting a threshold, taking the historical days meeting the threshold as a similar day set, constructing an OA-GRU model, optimizing parameters of a gate control circulation unit GRU by using a skyhawk optimizer, inputting similar day data in the OA-GRU model, and performing load prediction on a day to be predicted. The method can be used for forecasting by combining factors such as historical load, weather and date type, and can realize short-term forecasting on the day to be forecasted.

Description

OA-GRU short-term power load prediction method based on grey correlation

Technical Field

The invention relates to the technical field of power load prediction, in particular to a short-term power load prediction method based on OA-GRU (office automation-general rule) of grey correlation.

Background

The load situation is related to the construction problem of the power system, the load prediction is not carried out, the power supply amount of the supply side cannot be adjusted in time according to the prediction situation, and the waste of electric energy and the power failure can be caused. The power load prediction is an important link of safe and stable operation of the power system, power generation enterprises can make an optimal power generation plan according to an accurate prediction result, and meanwhile, a power dispatching department can make reasonable planning for normal operation of the system according to the prediction result, so that effective management of the power system is realized.

Disclosure of Invention

The embodiment of the invention provides a short-term power load prediction method based on OA-GRU (office automation-group unit) associated with gray, which comprises the following steps:

acquiring daily load influence factor characteristic data and power load data;

calculating Spearman values between the daily load influence factors and the power load data to obtain the weight of each influence factor;

dividing historical daily influence factor characteristic data and to-be-predicted daily influence factor characteristic data, performing cluster analysis, determining a classification standard according to the power load data, and determining a classical domain and a nodal domain; calculating the grey correlation degrees of all factors, sequencing the grey correlation degrees from large to small, setting a threshold value, and taking the history days meeting the threshold value as a similar day set;

optimizing parameters of a gate control circulation unit GRU by using a skyhawk optimizer, and constructing an OA-GRU model;

inputting the characteristic data of the influencing factors of the similar days in the OA-GRU model, and carrying out load prediction on the day to be predicted.

And further, the daily load influence factor data comprises:

season, air pressure, time, holidays, working days, highest daily temperature, lowest daily temperature, average daily temperature, relative daily humidity.

And the method further comprises a step of preprocessing the load data, which comprises the following steps:

firstly, randomly selecting a plurality of characteristic values from a data set to form a characteristic space; secondly, randomly dividing values between the maximum value and the minimum value in the selected features to form partitions, and constructing an isolated tree; then, forming the constructed isolated tree into iForest; finally, an anomaly score is calculated for each point. And setting corresponding threshold values according to the characteristics of the load data so as to judge whether the data is abnormal or not and correct the abnormal data. The correction method is to use a multi-order Lagrange interpolation method to calculate the average value of the abnormal data close to the normal data for substitution.

Further, the calculation formula of the weight of each influence factor includes:

in the formula: d _i Represents the difference in the bit order values of the ith data pair, and n represents the total number of observed samples.

Further, determining a classical domain and a nodal domain, comprising:

wherein, i ═ 1., m, C ═ C ₁ ,c ₂ ,...,c _n ) ^T Is a factor attribute, c ₁ ,c ₂ ,...,c _n Is I _i N different factor attributes, X ═ X (X) _i1 ,X _i2 ,...,X _in ) ^T Representing magnitude, X, over various factors _i1 ,X _i2 ,...,X _in Are respectively I _i About factor attribute c ₁ ,c ₂ ,...,c _n Value range of (A), X _ij ＝＜a _ij ,b _ij > (j ═ 1.. times, n) is a class i sample with respect to factor c _j The value range of (a);

and further, constructing a node domain matter element, wherein a calculation formula comprises:

wherein X _p1 ,X _p2 ,...,X _pn P about the factor attribute c, respectively ₁ ,c ₂ ,...,c _n The value range of (a).

And step one, calculating the grey correlation degree of each factor, including:

determining response variables as reference sequences x _o Covariates as comparison sequences x _i (i＝1,2,..,n)；

Calculating the absolute value of the difference between the reference sequence and the comparison sequence _i (k)＝|x _o (k)-x _i (k) And calculating Δ _i (k) Maximum and minimum values of (1), wherein _i (k) Representing the absolute value of the difference of the reference variable and the ith comparison sequence;

calculating a gray correlation coefficient

Wherein, Delta _min And Δ _max Are each Δ _i (k) Beta is a resolution coefficient, beta is more than 0 and less than 1;

calculating the degree of correlation of gray

And further, optimizing the parameters of the gate control loop unit GRU by using a skyhawk optimizer, wherein the step comprises the following steps of:

setting the number of eagle populations and the frequency of eagle populations; and setting the value ranges of two hyper-parameters (the number of hidden layer GRU units and the learning rate) of the GRU, and initializing the population randomly. Exploring and developing parameters alpha and delta; initializing a population position X, initial population fitness and optimal individuals.

And constructing a GRU network model, and sequentially assigning the individual position information of the eagle to the number of hidden layer units and the learning rate.

According to the AO-GRU model, taking the average absolute average percentage error of the prediction model as the current fitness value of the eagle individual when t is<At T, AO begins to cycle: expanding the exploration stage, calculating the average position of the population, and updating the position X of the population ₁ (t + 1); narrowing down the exploration phase and updating the population position X ₂ (t + 1); expanding development stage, updating population position X ₃ (t + 1); reducing development stages, updatingGroup position X ₄ (t + 1); and calculating the fitness of the updated population to obtain the current optimal individual position and the fitness.

And calculating the minimum fitness value of the eagle individual every time of operation, comparing the fitness of the current best individual with the fitness of the best individual found in the t generation by comparison, reserving the better individual position, selecting the minimum fitness value as the optimal value, and storing the number of hidden layer units and the learning rate corresponding to the optimal value.

When the maximum iteration times of the skyhawk optimizer are met, stopping the parameter optimization process of the GRU model, and outputting the optimal fitness value of the skyhawk and the number of parameter hidden layer units and the learning rate corresponding to the optimal value; otherwise, the iteration is continued.

The embodiment of the invention provides a short-term power load prediction method based on OA-GRU associated with gray, which has the following beneficial effects compared with the prior art:

by adopting the method of similar days, some power load influence factors can be eliminated, thereby achieving higher prediction precision. And determining the weight of each influencing factor by using Spearman and a correlation coefficient method. And combining the eagle optimizer with the gating cycle unit to optimize partial parameters of the GRU network.

Drawings

FIG. 1 is a flowchart of a method for predicting a short-term power load based on a gray-associated OA-GRU according to an embodiment of the present invention;

FIG. 2 is a flowchart of an OA-GRU short-term power load prediction method based on gray correlated OA-GRU according to an embodiment of the present invention.

Detailed Description

The present invention or utility model will be further explained by the following specific examples.

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Referring to fig. 1-2, an embodiment of the present invention provides a method for predicting a short-term power load based on a gray-correlated OA-GRU, including:

acquiring daily load influence factor characteristic data and power load data; calculating Spearman values between the daily load influence factors and the power load data to obtain the weight of each influence factor; dividing historical daily influence factor characteristic data and to-be-predicted daily influence factor characteristic data, performing cluster analysis, determining a classification standard according to the power load data, and determining a classical domain and a nodal domain; calculating the grey relevance of each factor, sequencing the grey relevance from big to small, setting a threshold value, and taking the historical days meeting the threshold value as a similar day set; optimizing parameters of a gate control circulation unit GRU by using a skyhawk optimizer, and constructing an OA-GRU model; inputting the characteristic data of the influencing factors of the similar days in the OA-GRU model, and carrying out load prediction on the day to be predicted.

Example (b):

1. acquiring data such as load, weather, date type, etc., and preprocessing the data

Firstly, randomly selecting a plurality of characteristic values from a data set to form a characteristic space; secondly, randomly dividing values between the maximum value and the minimum value in the selected features to form partitions, and constructing an isolated tree; then, forming the constructed isolated tree into iForest; finally, an anomaly score is calculated for each point. And setting corresponding threshold values according to the characteristics of the load data so as to judge whether the data is abnormal or not and correct the abnormal data. The correction method is to use a multi-order Lagrange interpolation method to calculate the mean value of the neighbor normal data of the abnormal data for substitution.

2. Spearman feature selection

As a sample for calculating the gray association degree, a day characteristic factor is selected: season, air pressure, time, holidays, working days, highest daily temperature, lowest daily temperature, average daily temperature, relative daily humidity. Calculating Spearman values between each feature and the load; the Spearman threshold is set to be 0.6, namely influence factors with rho values larger than 0.6 are selected.

3. Partitioning data into training and test sets

The historical data is used as a training set, and the day data to be predicted is used as a test set.

4. Similar days selected by grey correlation analysis

Step 1: determining a classical domain and a segment domain according to the classification criteria

Wherein, i ═ 1., m, C ═ C ₁ ,c ₂ ,...,c _n ) ^T Is a factor attribute, c ₁ ,c ₂ ,...,c _n Is I _i N different factor attributes, X ═ X (X) _i1 ,X _i2 ,...,X _in ) ^T Representing magnitude, X, over various factors _i1 ,X _i2 ,...,X _in Are respectively I _i About factor attribute c ₁ ,c ₂ ,...,c _n Value range of (A), X _ij ＝＜a _ij ,b _ij > (j ═ 1.. times, n) is a class i sample with respect to factor c _j The value range of (a);

constructing a region-saving matter element according to the classification standard and various factors

Wherein X _p1 ,X _p2 ,...,X _pn P about the factor attribute c, respectively ₁ ,c ₂ ,...,c _n The value range of (a).

Step 2: calculating grey correlation degree of each factor

Determining response variables as reference sequences x _o Covariates as comparison sequences x _i (i＝1,2,..,n)；

Calculating the absolute value of the difference between the reference sequence and the comparison sequence _i (k)＝|x _o (k)-x _i (k) And calculating Δ _i (k) A maximum value and a minimum value of (1), whereinΔ _i (k) Representing the absolute value of the difference of the reference variable and the ith comparison sequence;

calculating a gray correlation coefficient

Wherein, Delta _min And delta _max Are each Δ _i (k) Beta is a resolution coefficient, beta is more than 0 and less than 1;

calculating the degree of correlation of gray

And step 3: and sorting the gray relevance of the feature vectors of each historical day from large to small, setting a threshold, and selecting samples meeting the threshold as a similar day set. The present application sets the threshold to 0.6.

5. Establishing short-term load prediction model based on optimization of gated cyclic unit by skyhawk optimizer

The specific steps of establishing the prediction model are as follows:

step 1: initializing relevant parameters of GRU and eagle optimizer

Setting the number of eagle populations and the frequency of eagle populations; and setting the value ranges of two hyper-parameters (the number of hidden layer GRU units and the learning rate) of the GRU, and initializing the population randomly. Exploring and developing parameters alpha and delta; initializing a population position X, initial population fitness and optimal individuals.

Step 2: and constructing a GRU network model, and sequentially assigning the individual position information of the eagle to the number of hidden layer units and the learning rate.

And step 3: according to the AO-GRU model, taking the average absolute average percentage error of the prediction model as the current fitness value of the eagle individual when t is<At T, AO begins to cycle: expanding the exploration stage, calculating the average position of the population, and updating the position X of the population ₁ (t + 1); narrowing down the exploration phase and updating the population position X ₂ (t + 1); expanding development stage, updating population position X ₃ (t + 1); reducing development stage and updating population position X ₄ (t + 1); calculating the fitness of the updated population to obtain the current best fitnessIndividual location and fitness.

And 4, step 4: and calculating the minimum fitness value of the eagle individual every time of operation, comparing the fitness of the current best individual with the fitness of the best individual found in the t generation by comparison, reserving the better individual position, selecting the minimum fitness value as the optimal value, and storing the number of hidden layer units and the learning rate corresponding to the optimal value.

And 5: when the maximum iteration times of the hawk optimizer are met, stopping the parameter optimization process of the GRU model, and outputting the optimal fitness value of the hawk and the number of parameter hidden layer units and the learning rate corresponding to the optimal value; otherwise, continuing the iteration and repeatedly executing the step 3.

6. Prediction daily load of OA-GRU model

And performing load prediction on the day to be predicted by using the number of the GRU units of the trained optimal hidden layer and the learning rate and using the GRU model and the similar day set as a training set.

7. Model performance assessment

The Mean Absolute Percentage Error MAPE (Mean Absolute Percentage Error) and the Mean Absolute Error (Mean Absolute Percentage Error) are adopted as Error standards and are respectively expressed as

In the formula: n is the number of predicted time points, y _t And

respectively, an actual value and a predicted value corresponding to the time t.

It is obvious to a person skilled in the art that the invention/utility model is not restricted to details of the above-described exemplary embodiments, but that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention/utility model being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.

Claims (7)

1. A method for predicting a short-term power load based on OA-GRU associated with gray color, comprising:

acquiring daily load influence factor characteristic data and power load data;

calculating Spearman values between the daily load influence factors and the power load data to obtain the weight of each influence factor;

dividing historical daily influence factor characteristic data and to-be-predicted daily influence factor characteristic data, performing cluster analysis, determining a classification standard according to the power load data, and determining a classical domain and a nodal domain; calculating the grey relevance of each factor, sequencing the grey relevance from big to small, setting a threshold value, and taking the historical days meeting the threshold value as a similar day set;

optimizing parameters of a gate control circulation unit GRU by using a skyhawk optimizer, and constructing an OA-GRU model;

inputting the characteristic data of the influencing factors of the similar days in the OA-GRU model, and carrying out load prediction on the day to be predicted.

2. The method of claim 1, wherein the daily load influencing factors comprise:

season, air pressure, time, holidays, working days, highest daily temperature, lowest daily temperature, average daily temperature, relative daily humidity.

3. The method of claim 1, further comprising preprocessing the power load data, comprising:

firstly, randomly selecting a plurality of characteristic values from a data set to form a characteristic space; secondly, randomly dividing values between the maximum value and the minimum value in the selected features to form partitions, and constructing an isolated tree; then, forming the constructed isolated tree into iForest; finally, an anomaly score is calculated for each point. And setting corresponding threshold values according to the characteristics of the load data so as to judge whether the data is abnormal or not and correct the abnormal data. The correction method is to use a multi-order Lagrange interpolation method to calculate the average value of the abnormal data close to the normal data for substitution.

4. The method of claim 1, wherein the calculation formula of the weight of each influence factor comprises:

in the formula: d _i Represents the difference in the bit order values of the ith data pair, and n represents the total number of observed samples.

5. The method of claim 4, wherein the step of determining the classical domain and the nodal domain according to the classification criteria comprises:

wherein, i ═ 1., m, C ═ C ₁ ,c ₂ ,...,c _n ) ^T Is a factor attribute, c ₁ ,c ₂ ,...,c _n Is I _i N different factor attributes, X ═ X (X) _i1 ,X _i2 ,...,X _in ) ^T Representing magnitude, X, over various factors _i1 ,X _i2 ,...,X _in Are respectively I _i About factor attribute c ₁ ,c ₂ ,...,c _n Value range of (A), X _ij ＝＜a _ij ,b _ij > (j ═ 1.. times, n) is a class i sample with respect to factor c _j The value range of (a);

constructing a region-saving matter element according to the classification standard and various factors

Wherein, X _p1 ,X _p2 ,...,X _pn P about the factor attribute c, respectively ₁ ,c ₂ ,...,c _n The value range of (a).

6. The method for predicting the OA-GRU in accordance with claim 5, wherein the step of calculating the gray correlation degree of each factor comprises:

determining response variables as reference sequences x _o Covariates as comparison sequences x _i (i＝1,2,..,n)；

Calculating the absolute value of the difference between the reference sequence and the comparison sequence _i (k)＝|x _o (k)-x _i (k) And calculating Δ _i (k) Maximum and minimum values of (1), wherein _i (k) Representing the absolute value of the difference of the reference variable and the ith comparison sequence;

calculating a gray correlation coefficient

Wherein, Delta _min And delta _max Are each Δ _i (k) Beta is a resolution coefficient, and beta is more than 0 and less than 1;

calculating the degree of correlation of gray

7. The method of claim 6, wherein the step of optimizing parameters of gated loop units GRUs with a skyhawk optimizer comprises:

setting the number of eagle populations and the frequency of eagle populations; and setting the value ranges of two hyper-parameters (the number of hidden layer GRU units and the learning rate) of the GRU, and initializing the population randomly. Exploring and developing parameters alpha and delta; initializing a population position X, initial population fitness and optimal individuals.

And constructing a GRU network model, and sequentially assigning the individual position information of the eagle to the number of hidden layer units and the learning rate.

According to the AO-GRU model, taking the average absolute average percentage error of the prediction model as the current fitness value of the eagle individual, when t<At T, AO begins to cycle: expanding the exploration stage, calculating the average position of the population, and updating the position X of the population ₁ (t + 1); narrowing down the exploration phase and updating the population position X ₂ (t + 1); expanding development stage, updating population position X ₃ (t + 1); reducing the development stage and updating the population position X ₄ (t + 1); and calculating the fitness of the updated population to obtain the current optimal individual position and the fitness.

And calculating the minimum fitness value of the eagle individual every time of operation, comparing the fitness of the current best individual with the fitness of the best individual found in the t generation by comparison, reserving the better individual position, selecting the minimum fitness value as the optimal value, and storing the number of hidden layer units and the learning rate corresponding to the optimal value.

When the maximum iteration times of the skyhawk optimizer are met, stopping the parameter optimization process of the GRU model, and outputting the optimal fitness value of the skyhawk and the number of parameter hidden layer units and the learning rate corresponding to the optimal value; otherwise, the iteration is continued.

Images