CN116186548A

CN116186548A - Power load prediction model training method and power load prediction method

Info

Publication number: CN116186548A
Application number: CN202310482932.3A
Authority: CN
Inventors: 罗剑洪; 卢雪明; 欧阳家淦; 张家杰; 陈建松
Original assignee: Guangzhou Sanjing Electric Co Ltd
Current assignee: Guangzhou Sanjing Electric Co Ltd
Priority date: 2023-05-04
Filing date: 2023-05-04
Publication date: 2023-05-30
Anticipated expiration: 2043-05-04
Also published as: CN116186548B

Abstract

The application relates to a power load prediction model training method and a power load prediction method, wherein the power load prediction model training is based on the following steps: and acquiring load historical data and influence factor data, performing screening processing on the influence factor data according to the relativity of the load historical data and the influence factor data to acquire training factor data, and finally taking the training factor data and the load historical data as training data to train the power load prediction model. Based on the above, the accuracy of the power load prediction can be improved by the correlation screening of the load history data and the influence factor data, and the method is particularly suitable for the power load prediction based on the influence factor data with discrete variable characteristics.

Description

Power load prediction model training method and power load prediction method

Technical Field

The application relates to the technical field of photovoltaic power generation prediction, in particular to a power load prediction model training method and a power load prediction method.

Background

In recent years, along with the rapid increase of the economic level of China, the power industry is continuously developed. Electric load prediction is increasingly important for safe and economic operation of an electric power system, and becomes the basis of economic dispatch, energy storage management, future energy contracts and power plant maintenance plans. The accuracy of the power load prediction is directly related to the operation of the power system. If the predicted value is too large, the power production is excessive, so that energy waste and production machine loss are caused; and the predicted value is smaller, so that the electric power production is insufficient, the life of people is influenced, and the economic loss is caused.

Currently, many studies are being made on power load prediction. According to the predicted time range, it is largely classified into ultra-short term, medium term and long term. The ultra-short-term load prediction refers to load prediction shorter than one day, and is mainly used for real-time power dispatching and daytime dispatching; the short-term load prediction refers to load prediction from one day to one week, and is used for daytime operation of the power system, such as energy transaction and power system safety research; mid-load forecast refers to a forecast of several weeks to one year for fuel supply scheduling and infrastructure adjustment; long-term load prediction is typically a prediction over a year for long-term power system planning. Nowadays, short-term load prediction becomes an indispensable part of various power supply systems, and the requirements for the precision thereof are becoming higher and higher.

Among other things, the electrical load is affected by various environmental factors, including temperature, humidity, date type, season, etc. In particular, in the application scene of the photovoltaic power station, due to the characteristics of photovoltaic power generation, the power load prediction has more important significance on the operation and maintenance of the photovoltaic power generation.

The traditional power load prediction method mainly comprises a regression analysis method, a time sequence method, an exponential smoothing method and the like. Regression analysis predicts future load values by building a mathematical model reflecting causal relationships. The time series method establishes a mathematical model by performing curve fitting and parameter estimation on historical load data. The exponential smoothing method is implemented by an exponentially weighted combination. Although these methods can explicitly express the relationship between independent variables and dependent variables, they have limitations in dealing with complex nonlinear systems, and have poor predictive performance of power loads that are characteristic of non-stationary time series.

In summary, it can be seen that the above disadvantages still exist in the conventional power load prediction method.

Disclosure of Invention

Based on this, it is necessary to provide a power load prediction model training method and a power load prediction method for overcoming the defects existing in the conventional power load prediction method.

At least one embodiment of the present disclosure provides a power load prediction model training method, including the steps of:

acquiring load history data and influence factor data; wherein the influencing factor data corresponds to the load history data;

according to the correlation degree of the load historical data and the influence factor data, screening the influence factor data to obtain training factor data;

and training the electric power load prediction model by taking the training factor data and the load history data as training data.

According to the power load prediction model training method, the load historical data and the influence factor data are obtained, screening processing is carried out on the influence factor data according to the correlation degree of the load historical data and the influence factor data, training factor data are obtained, and finally the training factor data and the load historical data are used as training data to train the power load prediction model. Based on the above, the accuracy of the power load prediction can be improved by the correlation screening of the load history data and the influence factor data, and the method is particularly suitable for the power load prediction based on the influence factor data with discrete variable characteristics.

In one embodiment, the method further comprises the steps of:

testing the accuracy of the power load prediction model according to the training factor data and the load history data;

and (5) adjusting screening treatment according to the precision.

In one embodiment, the influencing factor data includes a digitized result of temperature, somatosensory temperature, humidity, precipitation, wind speed, light intensity, holidays, weather type, and/or season.

In one embodiment, the load history data includes a load characteristic index over a set period.

In one embodiment, the load characteristic index includes a maximum load, a minimum load, an average load, a load factor, a peak-to-valley difference, and/or a peak-to-valley difference rate.

In one embodiment, the process of performing filtering processing on the influence factor data according to the correlation degree between the load history data and the influence factor data to obtain training factor data includes the steps of:

calculating the correlation degree between the load history data and the influence factor data;

and reserving influence factor data with the correlation degree larger than a set threshold value as training factor data.

In one embodiment, the process of adjusting the screening process according to the accuracy includes the steps of:

if the precision is relatively improved, the set threshold is reduced, otherwise, the set threshold is improved.

In one embodiment, a process for calculating the correlation of load history data and influence factor data includes the steps of:

determining a discrete space of influence factor data; wherein the discrete space is composed of a plurality of values;

classifying the load historical data according to the value of the influence factor data to obtain a classification result;

and carrying out correlation calculation according to the classification result.

In one embodiment, the power load prediction model is built based on an LSTNet network.

At least one embodiment of the present disclosure provides a power load prediction model training apparatus, including:

the data acquisition module is used for acquiring load history data and influence factor data; wherein the influencing factor data corresponds to the load history data;

the data screening module is used for performing screening processing on the influence factor data according to the correlation degree of the load historical data and the influence factor data to obtain training factor data;

and the data training module is used for taking the training factor data and the load history data as training data to train the power load prediction model.

According to the power load prediction model training device, the load historical data and the influence factor data are obtained, screening processing is carried out on the influence factor data according to the correlation degree of the load historical data and the influence factor data, training factor data are obtained, and finally the training factor data and the load historical data are used as training data to train the power load prediction model. Based on the above, the accuracy of the power load prediction can be improved by the correlation screening of the load history data and the influence factor data, and the method is particularly suitable for the power load prediction based on the influence factor data with discrete variable characteristics.

At least one embodiment of the present disclosure also provides a model training apparatus, including:

one or more memories non-transitory storing computer-executable instructions;

one or more processors configured to execute computer-executable instructions, wherein the computer-executable instructions, when executed by the one or more processors, implement a power load prediction model training method according to any embodiment of the present disclosure.

According to the model training device, the load historical data and the influence factor data are obtained, screening processing is carried out on the influence factor data according to the correlation degree of the load historical data and the influence factor data, training factor data are obtained, and finally the training factor data and the load historical data are used as training data to train the electric power load prediction model. Based on the above, the accuracy of the power load prediction can be improved by the correlation screening of the load history data and the influence factor data, and the method is particularly suitable for the power load prediction based on the influence factor data with discrete variable characteristics.

At least one embodiment of the present disclosure also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer-executable instructions that, when executed by a processor, implement a power load prediction model training method according to any embodiment of the present disclosure.

The non-transitory computer readable storage medium acquires the load history data and the influence factor data, performs screening processing on the influence factor data according to the correlation degree of the load history data and the influence factor data to obtain training factor data, and finally trains the electric power load prediction model by taking the training factor data and the load history data as training data. Based on the above, the accuracy of the power load prediction can be improved by the correlation screening of the load history data and the influence factor data, and the method is particularly suitable for the power load prediction based on the influence factor data with discrete variable characteristics.

At least one embodiment of the present disclosure also provides a power load prediction method.

A method of power load prediction comprising the steps of:

acquiring influence factor data;

and inputting the influence factor data into the power load prediction model of any embodiment to obtain a power load prediction result output by the power load prediction model.

According to the power load prediction method, the load historical data and the influence factor data are obtained, the influence factor data are screened according to the correlation degree of the load historical data and the influence factor data, the training factor data are obtained, and finally the training factor data and the load historical data are used as training data to train a power load prediction model. Based on the above, the accuracy of the power load prediction can be improved by the correlation screening of the load history data and the influence factor data, and the method is particularly suitable for the power load prediction based on the influence factor data with discrete variable characteristics.

At least one embodiment of the present disclosure provides an electrical load prediction apparatus, comprising:

the data acquisition module is used for acquiring influence factor data;

and the data output module is used for inputting the influence factor data into the power load prediction model of any embodiment to obtain a power load prediction result output by the power load prediction model.

According to the power load prediction device, the load historical data and the influence factor data are obtained, the screening processing is carried out on the influence factor data according to the correlation degree of the load historical data and the influence factor data, the training factor data are obtained, and finally the training factor data and the load historical data are used as training data to train a power load prediction model. Based on the above, the accuracy of the power load prediction can be improved by the correlation screening of the load history data and the influence factor data, and the method is particularly suitable for the power load prediction based on the influence factor data with discrete variable characteristics.

At least one embodiment of the present disclosure also provides a load prediction apparatus, including:

one or more memories non-transitory storing computer-executable instructions;

one or more processors configured to execute computer-executable instructions, wherein the computer-executable instructions, when executed by the one or more processors, implement a power load prediction method according to any embodiment of the present disclosure.

According to the load prediction device, the load historical data and the influence factor data are obtained, screening processing is carried out on the influence factor data according to the relativity of the load historical data and the influence factor data, training factor data are obtained, and finally the training factor data and the load historical data are used as training data to train the electric power load prediction model. Based on the above, the accuracy of the power load prediction can be improved by the correlation screening of the load history data and the influence factor data, and the method is particularly suitable for the power load prediction based on the influence factor data with discrete variable characteristics.

At least one embodiment of the present disclosure also provides another non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer-executable instructions that, when executed by a processor, implement a power load prediction method according to any embodiment of the present disclosure.

Drawings

FIG. 1 is a flowchart of a power load prediction model training method according to an embodiment;

FIG. 2 is a flowchart of another embodiment power load prediction model training method;

FIG. 3 is a LSTNet model diagram;

FIG. 4 is an outlier processing result diagram;

FIG. 5 is a graph showing the comparison of the predicted results and actual loads of two models of the example;

FIG. 6 is a graph comparing prediction errors of two models of an embodiment;

FIG. 7 is a block diagram of a power load prediction model training device according to an embodiment;

FIG. 8 is a flow chart of a method of predicting electrical loads according to an embodiment;

FIG. 9 is a block diagram of a power load prediction device module according to an embodiment;

FIG. 10 is a schematic block diagram of a model training apparatus provided in accordance with at least one embodiment of the present disclosure;

FIG. 11 is a schematic diagram of a non-transitory computer-readable storage medium provided by at least one embodiment of the present disclosure.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present disclosure more apparent, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present disclosure. It will be apparent that the described embodiments are some, but not all, of the embodiments of the present disclosure. All other embodiments, which can be made by one of ordinary skill in the art without the need for inventive faculty, are within the scope of the present disclosure, based on the described embodiments of the present disclosure.

Unless defined otherwise, technical or scientific terms used in this disclosure should be given the ordinary meaning as understood by one of ordinary skill in the art to which this disclosure belongs. The terms "first," "second," and the like, as used in this disclosure, do not denote any order, quantity, or importance, but rather are used to distinguish one element from another. The word "comprising" or "comprises", and the like, means that elements or items preceding the word are included in the element or item listed after the word and equivalents thereof, but does not exclude other elements or items. The terms "connected" or "connected," and the like, are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "upper", "lower", "left", "right", etc. are used merely to indicate relative positional relationships, which may also be changed when the absolute position of the object to be described is changed.

In order to keep the following description of the embodiments of the present disclosure clear and concise, the present disclosure omits a detailed description of some known functions and known components.

At least one embodiment of the present disclosure provides a power load prediction model training method.

Fig. 1 is a flowchart of an embodiment of a power load prediction model training method, as shown in fig. 1, and the embodiment of the power load prediction model training method includes steps S100 to S102:

s100, acquiring load history data and influence factor data; wherein the influencing factor data corresponds to the load history data;

s101, screening the influence factor data according to the correlation degree of the load history data and the influence factor data to obtain training factor data;

s102, training the electric power load prediction model by taking training factor data and load history data as training data.

Wherein, the load history data and the influence factor data are both history data. And corresponding data acquisition is carried out according to the data record of the power load operation system (such as a photovoltaic power station) operated in the historical period. The load history data is a data result/a numeric result of the power load, and the influence factor data is a data result/a numeric result of the influence factor. The influencing factors can influence the power load of the power system, including environmental factors or time factors.

The load history data corresponds to the influence factor data, namely, the influence factor data is used as an independent variable, and the load history data is used as a corresponding relation of the dependent variable. The influencing factors have different characteristics, some have continuous variable characteristics such as temperature, humidity, precipitation, etc., and others have discrete variable characteristics such as weather type, season, etc., while the electrical load itself is a continuous variable. Therefore, the technical difficulty of predicting the existence of a data layer for the power load according to the influence factor data.

In one embodiment, the influencing factor data comprises a numerical result of temperature, somatosensory temperature, humidity, precipitation, wind speed, illumination intensity, holidays, weather type and/or season according to the characteristics of the power system. The generation mode of the numerical result comprises normalization or discrete data processing. For example, whether the result of the numeralization of holidays includes: "0" and "1", "0" represent non-holidays, and "1" represents holidays. It should be noted that the selection of the above-mentioned influencing factor data can be flexibly adjusted according to the type of the power system. This embodiment is a specific choice of power system of the type that is a photovoltaic power plant and does not represent a unique limitation on the influencing factor data.

In one embodiment, the load history data includes a load characteristic index over a set period. And adjusting the granularity of the power load prediction according to the length of the selected set period. For example, if the set period is 1 day or less, the ultra-short-term power load prediction is performed; setting the period to be 1 day to 1 week, and then carrying out short-term power load prediction; setting the period to be 1 week to 1 year, and then predicting the middle-stage power load; and if the set period is greater than 1 year, the long-term power load prediction is performed.

The load characteristic index is a characteristic index of the electric load.

As a preferred implementation of the disclosed embodiments, in combination with the subsequent preferred implementation, and suitable for (ultra) short-term power load prediction in the field of photovoltaic power plants, the load characteristic indicators include maximum load, minimum load, average load, load factor, peak-to-valley difference and/or peak-to-valley difference ratio. Taking a setting period of 1 day as an example, the load characteristic index includes a daily maximum load, a daily minimum load, a daily average load, a daily load rate, a daily peak-valley difference, and/or a daily peak-valley difference rate.

The maximum load is the maximum load value of the load data recorded in the set period, the minimum load is the minimum load value of the load data recorded in the set period, the average load is the average value of the electricity consumption in the set period, the load rate is the ratio of the average load to the maximum load (representing the balance of load distribution), the peak-valley difference is the difference between the maximum load and the minimum load, and the peak Gu Chalv is the ratio of the peak-valley difference to the maximum load.

Meanwhile, due to the continuous and discrete characteristics of the influence factor data, the load history data and the influence factor data have correlation differences, so that the influence factor data need to be screened, and the influence factor data with higher correlation is reserved and used as training factor data.

In one embodiment, the relevance of the load history data and the influence factor data can be determined through normalization or subjective weighting, a relevant set threshold value is determined according to an empirical value, and the influence factor data with the relevance greater than the set threshold value is reserved.

As a preferred implementation manner, the embodiment of the disclosure discloses a correlation calculation manner of a preferred implementation manner, which is applied to power load prediction with obvious discrete characteristics of influencing factors in a photovoltaic power station scene. Fig. 2 is a flowchart of another embodiment of a power load prediction model training method, as shown in fig. 2, in step S101, according to the correlation between load history data and influence factor data, a filtering process is performed on the influence factor data, and a process of obtaining training factor data is performed, including step S200 and step S201:

s200, calculating the correlation degree between the load history data and the influence factor data;

s201, reserving influence factor data with the correlation degree larger than a set threshold value as training factor data.

The process of calculating the correlation between the load history data and the influence factor data in step S200 includes the steps of:

The following explains the above steps with a specific application example:

for influencing factor data with discrete features

The discrete space is determined as:

；

wherein,,

representing influence factor data->

There is->

The values of the two different values are:

。

According to the data of influencing factors

Is->

Different values of the load history data are divided into +.>

A set of different classes, expressed as:

；

wherein,,

indicate->

Load history data of class, influence factor data corresponding to the load history data of class->

The value of (2) is +.>

。

Based on daily load characteristic indexes of power load, the classification of load history data is completed by adopting a K-means clustering algorithm, and the specific clustering flow is as follows (1) - (5):

(1) Determining the input and output of a K-means clustering algorithm:

the algorithm input is the cluster number of the clusters

Data samples of load characteristic indicators:

；

wherein,,

is>

Is a matrix containing 6 attributes: daily maximum load, daily minimum load, daily average load, daily load rate, daily peak-to-valley difference, and daily peak-to-valley difference rate.

(2) And selecting an initial clustering center. At the position of

Is selected at random->

The individual samples are taken as initial cluster centers, and the cluster center set is expressed as:

；/>

(3) Class classification. Calculated according to the following formula

Is associated with->

The Euclidean distance of each cluster center in the cluster, and dividing the samples into the class nearest to the cluster center to form +.>

And (3) clustering:

；

wherein,,

indicate->

Sample No. H>

Personal attributes (i.e.)>

Indicate->

First of clustering centers>

And attributes.

(4) Updating the cluster center-calculating the mean of the samples in each class according to the following formula, taking the mean as a new cluster center:

；

wherein,,

indicate->

First->

Sample number->

Indicate->

Sample set of individual clusters, +.>

Representation->

Is a number of samples of (a).

(5) Repeating the step (3) and the step (4) until all the cluster centers remain unchanged. At the same time, the square error function is calculated according to the following formula

Will also converge to a constant value (minimum):

；

wherein,,

representation->

And->

Distance of cluster centers.

The above operations divide the sample data of the load characteristic index into

Different classes. For each class, the load characteristic value sample data is replaced by the corresponding load history data, so that the corresponding +. >

Class load history data:

；

wherein,,

indicate->

Class load history data from->

And obtaining sample data of the class load characteristic index.

To the above classification

And category->

Similarity calculation was performed to calculate +.>

Is +.>

Is prepared from the following components in proportion:

；

wherein,,

representation->

All load history data belonging to +.>

Ratio of->

Representing at the same time->

And

load history data quantity,/->

Representation->

Is a total number of load history data of the vehicle. />

Calculate all

And a similarity table is obtained, as shown in the following Table 1, table 1 below intuitively shows +.>

And (3) with

Is a similarity of (3).

Table 1 similarity table

If the influence factor data a is highly correlated with the load history data, then

And->

Will be in one-to-one correspondence and highly similar. Based on this basic knowledge, the calculation of the correlation is specifically as follows:

extracting the maximum value of each column from the table

All columns +.>

Form a vector

. The least significant element in the vector is used as influencing factor data +.>

The correlation with the load history data is as follows:

；

and calculating the correlation degree of each influence factor data with discrete variable characteristics and the load history data through the steps.

After screening and reserving to obtain training factor data, taking the training factor data as independent variables and the load history data as dependent variables, and constructing a data set of the training data. Wherein the data set includes a training set, a validation set, and a test set.

In one example, as shown in fig. 2, the power load prediction model training method of another embodiment further includes step S300 and step S301:

s300, testing the accuracy of the power load prediction model according to training factor data and load history data;

s301, adjusting screening processing according to the precision.

The screening process is based on the set threshold, and therefore, the size of the set threshold can be adjusted according to the accuracy. The method comprises the following steps: if the accuracy is relatively high, the set threshold is lowered, otherwise, the set threshold is raised to execute adjustment processing, for example:

1) And training to obtain an electric load prediction model which is taken as a reference model without considering influence factor data with discrete variable characteristics.

2) Determining an initial set threshold value according to experience;

3) And training to obtain a new power load prediction model by taking the influence factor data with discrete variable characteristics, the correlation of which is larger than a set threshold value, as a part of model input.

4) If the prediction precision of the power load prediction model is improved, the set threshold value is reduced; if the prediction accuracy of the power load prediction model is reduced, the set threshold is raised.

5) Repeating the steps 3) and 4) until the set threshold value is not changed any more, obtaining a set value of the final set threshold value, and obtaining an optimal power load prediction model.

In one embodiment, the power load prediction model is based on a neural network. As a preferred implementation mode, the LSTNet network is selected to establish a power load prediction model, so that a better prediction effect can be obtained in a (ultra) short-term power load prediction scene.

Wherein, since the power load data is time series data, the power load prediction can be regarded as a time series modeling problem. Compared with the traditional neural network, the Long Short-Term Memory (LSTM) network has special Memory capacity and gate structure, so that the time sequence and nonlinearity of the power load data can be considered at the same time, and the Long-Term sequence information of the load can be better learned. Therefore, LSTM has greater accuracy in predicting future load demands. The LSTNet network can learn the correlation between multiple variables (influencing factor data) well and extract highly nonlinear long-short-term and linear features in the data.

Fig. 3 is a diagram of LSTNet model, as shown in fig. 3, the LSTNet model built based on the LSTNet network is composed of a nonlinear part composed of a CNN convolution layer, an RNN loop layer and a jump layer, and a linear part composed of an autoregressive linear layer. The LSTNet model simultaneously utilizes the advantages of CNN and RNN, the CNN can extract short-term local dependency relationship between power load data, the RNN and the jump layer can capture long-term dependency relationship, and the optimization process is simplified based on periodicity of power load. Finally, a traditional autoregressive linear model is added, so that the nonlinear deep learning model has stronger robustness on the power load time sequence. The final LSTNet prediction result, i.e. the power load prediction result, is obtained by superimposing the results of the nonlinear and linear portions.

In the process of constructing the power load prediction model by adopting the LSTNet network, the influence factor data with small correlation is used as the input of the model to participate in training, so that the prediction accuracy of the model is reduced, and the convergence time of training is prolonged. Therefore, the prediction accuracy can be effectively improved by screening training factor data.

In order to better explain the effects of the embodiments of the present disclosure, the embodiments of the present disclosure are explained below in one specific application of the embodiment.

The photovoltaic power plant used in this example was located in australia and experiments were performed with actual power load data for the photovoltaic power plant 2021, 10 months 1 to 2022, 9 months 15. The method provided by the invention is used for calculating the correlation degree of each influence factor with discrete variable characteristics and the power load. By constructing a comparison experiment, taking 1) discrete influence factors with low correlation degree with the power load as a part of model input; 2) Discrete influencing factors with low correlation to the electrical load are not part of the model input, and an LSTNet model is trained and predicted for each case. The evaluation result of the discrete influence factors and the power load correlation degree and the prediction results of the two LSTNet models are described, so that the effectiveness of the technical scheme of the embodiment is verified.

In this example, MAE (Mean Absolute Error, average absolute error) and RMSE (Root Mean Square Error ) were used as error evaluation indices. MAE and RMSE were calculated as follows:

；/>

；

wherein,,

representing predicted data points, ++>

Indicate->

Normalized actual load value of each predicted point, +.>

Indicate->

The normalized model of each predicted point predicts the load value.

And step 1, data acquisition and abnormal value processing. Fig. 4 is a graph of the abnormal value processing results, and as shown in fig. 4, the actual power load data of 2021, 10 months, 1 to 2022, 9 months, 15 are divided into a training set (60%), a verification set (20%) and a test set (20%). The present example collects historical data of power load, temperature, somatosensory temperature, humidity, precipitation, wind speed, light intensity, holidays, weather type, season from day 10 of 2021 to day 9 of 2022. Normalizing the original data, adopting an isolated forest anomaly detection algorithm to detect the anomalies of the original load data, and reducing the influence of the anomalies on load prediction.

And 2, calculating the similarity of the discrete variables. Among the load influencing factors, there are holidays, weather types and seasons with discrete characteristics. The correlation degree between the 3 influencing factors and the power load is calculated by adopting the method provided by the invention. For example, it is calculated whether the holiday is correlated with the power load:

the load data is divided into 2 different classes according to whether the holiday is 2 different values (holiday, non-holiday), and the set of the classes is expressed as

；

Based on 6 characteristic indexes (daily maximum load, daily minimum load, daily average load, daily load rate, daily peak-valley difference and daily peak Gu Chalv) of the load, the classification of the load data is completed by adopting a K-means clustering algorithm. Into 2 classes, the set of these classes being expressed as

；

And carrying out similarity calculation on the 2 classification results. Separately calculate

Load data of (2) belong to->

Obtaining a similarity table, as shown in table 2;

TABLE 2 similarity of holidays

And calculating the correlation degree of seasons and power loads. Extracting the maximum value of each column from the table to form a vector

The minimum element in the vector is +.>

As to whether holidays are correlated with electric load, i.e. +.>

。

Similarly, the calculated weather type, season and power load similarity are shown in tables 3 and 4, and the calculated weather type, season and power load correlations are respectively

，

。

Table 3 seasonal similarity table

Table 4 weather type similarity table

It can be seen from the above table whether the 3 discrete variables holidays, weather types and seasons are less correlated with the power load. Model training is performed on the model without adding the three discrete variables and with adding the three discrete variables respectively, and a comparison experiment is performed on the two models.

Case 1: whether 3 discrete variables of holidays, weather types and seasons are not used as a prediction model input by the model, and the model input is as follows: power load, temperature, somatosensory temperature, humidity, precipitation, wind speed, visibility, and illumination intensity.

Case 2: whether 3 discrete variables of holidays, weather types and seasons are taken as a prediction model input by the model, and the model input is as follows: power load, temperature, somatosensory temperature, humidity, precipitation, wind speed, visibility, illumination intensity, holidays, weather type and season.

And 3, training an LSTNet model. And (3) training an LSTNet short-term load prediction model by the two inputs in the step (2). The input of the model is historical data (comprising historical load data and influencing factors) of 96 hours in the past, and the output is a load prediction result of 24 hours in the future. The evaluation index of the two models is shown in table 5 below.

TABLE 5 weather type similarity Table

As can be seen from table 5, the addition of the model of whether or not holidays, weather types, and seasons increased 8.39%, 7.99%, 2.76% in MAE and 3.79%, 4.00%, 0.95% in RMSE, i.e., the addition of the influence factors of whether or not holidays, weather types, and seasons as inputs did not have any improvement effect on the performance of the system, but rather reduced the prediction accuracy of the model.

Two models were applied to the light storage system of the example from day 16 of 2022, 9, to day 17 of 2022, 9, and the load prediction results and actual loads of the two models are shown in fig. 5. Wherein, curve 1 represents an actual load value, curve 2 represents a load predicted value to which a model of whether holidays, weather types, and seasons are added, and curve 3 represents a load predicted value to which a model of whether holidays, weather types, and seasons are not added. The error results for both models are shown in fig. 6. As is apparent from fig. 5 and 6, the prediction accuracy of the model to which influence factors such as holidays, weather types, and seasons are added is lower.

According to the power load prediction model training method of any embodiment, the load historical data and the influence factor data are obtained, screening processing is carried out on the influence factor data according to the relativity of the load historical data and the influence factor data, training factor data are obtained, and finally the training factor data and the load historical data are used as training data to train the power load prediction model. Based on the above, the accuracy of the power load prediction can be improved by the correlation screening of the load history data and the influence factor data, and the method is particularly suitable for the power load prediction based on the influence factor data with discrete variable characteristics.

At least one embodiment of the present disclosure also provides a power load prediction model training device.

Fig. 7 is a block diagram of an electric power load prediction model training apparatus according to an embodiment, and as shown in fig. 7, the electric power load prediction model training apparatus according to an embodiment includes:

the data acquisition module 1000 is configured to acquire load history data and influence factor data; wherein the influencing factor data corresponds to the load history data;

the data screening module 1001 is configured to perform screening processing on the influence factor data according to the correlation between the load history data and the influence factor data, so as to obtain training factor data;

the data training module 1002 is configured to train the power load prediction model by using the training factor data and the load history data as training data.

According to the power load prediction model training device of any embodiment, the load history data and the influence factor data are obtained, screening processing is carried out on the influence factor data according to the relativity of the load history data and the influence factor data, training factor data are obtained, and finally the training factor data and the load history data are used as training data to train the power load prediction model. Based on the above, the accuracy of the power load prediction can be improved by the correlation screening of the load history data and the influence factor data, and the method is particularly suitable for the power load prediction based on the influence factor data with discrete variable characteristics.

Fig. 8 is a flowchart of a power load prediction method according to an embodiment, as shown in fig. 8, the power load prediction method according to an embodiment includes steps S400 and S401:

s400, acquiring influence factor data;

s401, inputting the influence factor data into the power load prediction model to obtain a power load prediction result output by the power load prediction model.

At least one embodiment of the present disclosure provides an electrical load prediction apparatus.

Fig. 9 is a block diagram of a power load predicting apparatus according to an embodiment, and as shown in fig. 9, the power load predicting apparatus according to an embodiment includes:

The data acquisition module 2000 is used for acquiring influence factor data;

the data output module 2001 is configured to input the influence factor data into the power load prediction model of any of the above embodiments, and obtain a power load prediction result output by the power load prediction model.

At least one embodiment of the present disclosure also provides a model training apparatus. Fig. 10 is a schematic block diagram of a model training apparatus provided in accordance with at least one embodiment of the present disclosure. For example, as shown in fig. 10, model training apparatus 20 may include one or more memories 200 and one or more processors 201. Memory 200 is used to non-transitory store computer-executable instructions; the processor 201 is configured to execute computer-executable instructions that, when executed by the processor 201, may cause the processor 201 to perform one or more steps in a power load prediction model training method according to any embodiment of the present disclosure.

For specific implementation and relevant explanation of each step of the power load prediction model training method, reference may be made to relevant content in the foregoing embodiment of the power load prediction model training method, which is not described herein. It should be noted that the components of model training apparatus 20 shown in fig. 10 are exemplary only and not limiting, and that model training apparatus 20 may have other components as desired for practical applications.

In one embodiment, the processor 201 and the memory 200 may communicate with each other directly or indirectly. For example, the processor 201 and the memory 200 may communicate over a network connection. The network may include a wireless network, a wired network, and/or any combination of wireless and wired networks, the disclosure is not limited in type and function of the network herein. For another example, processor 201 and memory 200 may also communicate via a bus connection. The bus may be a peripheral component interconnect standard (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. For example, the processor 201 and the memory 200 may be disposed at a remote data server (cloud) or a distributed energy system (local), or may be disposed at a client (e.g., a mobile device such as a mobile phone). For example, processor 201 may be a Central Processing Unit (CPU), tensor Processor (TPU), or graphics processor GPU, among other devices having data processing and/or instruction execution capabilities, and may control other components in model training apparatus 20 to perform desired functions. The Central Processing Unit (CPU) can be an X86 or ARM architecture, etc.

In one embodiment, memory 200 may comprise any combination of one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory. Volatile memory can include, for example, random Access Memory (RAM) and/or cache memory (cache) and the like. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, erasable programmable read-only memory (EPROM), portable compact disc read-only memory (CD-ROM), USB memory, flash memory, and the like. One or more computer-executable instructions may be stored on a computer-readable storage medium and the processor 201 may execute the computer-executable instructions to implement the various functions of the model training apparatus 20. Various applications and various data, as well as various data used and/or generated by the applications, etc., may also be stored in the memory 200.

It should be noted that, the model training device 20 may achieve similar technical effects as the foregoing power load prediction model training method, and the repetition is not repeated.

At least one embodiment of the present disclosure also provides a non-transitory computer-readable storage medium. FIG. 11 is a schematic diagram of a non-transitory computer-readable storage medium provided by at least one embodiment of the present disclosure. For example, as shown in FIG. 11, one or more computer-executable instructions 301 may be non-transitory stored on the non-transitory computer-readable storage medium 30. For example, the computer-executable instructions 301, when executed by a computer, may cause the computer to perform one or more steps in a power load prediction model training method according to any embodiment of the present disclosure.

In one embodiment, the non-transitory computer readable storage medium 30 may be used in the model training apparatus 20 described above, which may be, for example, the memory 200 in the model training apparatus 20.

In one embodiment, the description of the non-transitory computer readable storage medium 30 may refer to the description of the memory 200 in the embodiment of the model training apparatus 20, and the repetition is omitted.

It should be noted that the memory 200 stores different non-transitory computer-executable instructions that, when executed by the processor 201, may cause the processor 201 to perform one or more steps in a load prediction method according to any of the embodiments of the present disclosure, for example, the model training apparatus 20 corresponds as a load prediction apparatus.

For the purposes of this disclosure, the following points are also noted:

(1) The drawings of the embodiments of the present disclosure relate only to the structures to which the embodiments of the present disclosure relate, and reference may be made to the general design for other structures.

(2) In the drawings for describing embodiments of the present invention, thicknesses and dimensions of layers or structures are exaggerated for clarity. It will be understood that when an element such as a layer, film, region or substrate is referred to as being "on" or "under" another element, it can be "directly on" or "under" the other element or intervening elements may be present.

(3) The embodiments of the present disclosure and features in the embodiments may be combined with each other to arrive at a new embodiment without conflict. The above is only a specific embodiment of the present disclosure, but the protection scope of the present disclosure is not limited thereto, and the protection scope of the present disclosure should be subject to the protection scope of the claims

The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.

The foregoing examples represent only a few embodiments of the present application, which are described in more detail and are not thereby to be construed as limiting the scope of the claims. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application is to be determined by the claims appended hereto.

Claims

1. A method of training a power load predictive model, comprising the steps of:

Acquiring load history data and influence factor data; wherein the influence factor data corresponds to the load history data;

according to the relevance between the load historical data and the influence factor data, screening the influence factor data to obtain training factor data;

and taking the training factor data and the load history data as training data to train a power load prediction model.

2. The method of training a power load predictive model of claim 1, further comprising the steps of:

and adjusting the screening process according to the precision.

3. The method of claim 1, wherein the impact factor data comprises a numerical result of temperature, somatosensory temperature, humidity, precipitation, wind speed, light intensity, holiday, weather type, and/or season.

4. The power load prediction model training method of claim 1, wherein the load history data includes a load characteristic index over a set period.

5. The method of claim 4, wherein the load characteristic index comprises a maximum load, a minimum load, an average load, a load factor, a peak-to-valley difference, and/or a peak-to-valley difference rate.

6. The method for training a power load prediction model according to claim 2, wherein the process of performing a screening process on the influence factor data according to the correlation between the load history data and the influence factor data to obtain training factor data comprises the steps of:

7. The power load prediction model training method according to claim 6, wherein the process of adjusting the screening process according to the accuracy includes the steps of:

and if the precision is relatively improved, the set threshold is reduced, otherwise, the set threshold is improved.

8. The method of claim 6, wherein the process of calculating the correlation of the load history data and the influence factor data comprises the steps of:

9. The power load prediction model training method according to any one of claims 1 to 8, wherein the power load prediction model is built based on an LSTNet network.

10. A method of predicting electrical load, comprising the steps of:

acquiring influence factor data;

inputting the influence factor data into the power load prediction model according to any one of claims 1 to 9, and obtaining a power load prediction result output by the power load prediction model.