CN114330935B

CN114330935B - New energy power prediction method and system based on multiple combination strategies integrated learning

Info

Publication number: CN114330935B
Application number: CN202210234586.2A
Authority: CN
Inventors: 包涛; 李鹏; 姚森敬; 马溪原; 陈炎森; 陈元峰; 李卓环; 程凯; 周悦; 张子昊; 周长城
Original assignee: Southern Power Grid Digital Grid Research Institute Co Ltd
Current assignee: Southern Power Grid Digital Grid Research Institute Co Ltd
Priority date: 2022-03-10
Filing date: 2022-03-10
Publication date: 2022-07-29
Anticipated expiration: 2042-03-10
Also published as: CN114330935A

Abstract

The application relates to a new energy power prediction method and system based on multiple combined strategy ensemble learning, wherein the method comprises the following steps: acquiring sample new energy data according to the target prediction object; the sample new energy data comprises training new energy data and verification new energy data; training through the training new energy data to obtain a plurality of primary prediction models; respectively carrying out integrated processing on the plurality of primary prediction models through a plurality of preset integrated instructions to obtain a plurality of integrated prediction models; and according to the performance indexes of the integrated prediction models, determining a target prediction model from the integrated prediction models, and predicting the new energy power through the target prediction model. The method can improve the power prediction precision of the new energy station.

Description

New energy power prediction method and system based on multiple combined strategy integrated learning

Technical Field

The application relates to the technical field of electric power, in particular to a new energy power prediction method, a system, a device, computer equipment and a storage medium based on multiple combined strategy ensemble learning.

Background

With the development of society, the demand of human daily life and production for energy is increasing day by day, and the energy on earth is limited, so how to improve the utilization efficiency of energy, implementing renewable energy source substitution action is the current main research direction. The basis for consuming the high-proportion renewable energy is to correctly carry out power prediction on the generated power of the renewable energy, and the existing new energy power prediction method and prediction system basically adopt numerical weather forecast with large granularity for prediction, so that the prediction accuracy is not high, and the actual requirement of large-scale access of the new energy to a power system in the future cannot be met.

Disclosure of Invention

In view of the above, it is necessary to provide a new energy power prediction method, a system, an apparatus, a computer device, and a computer readable storage medium, for solving the technical problem of low prediction accuracy of the new energy power prediction method.

In a first aspect, the application provides a new energy power prediction method based on multiple combined strategy ensemble learning. The method comprises the following steps:

acquiring sample new energy data according to the target prediction object; the sample new energy data comprises training new energy data and verification new energy data;

Training through the training new energy data to obtain a plurality of primary prediction models;

respectively carrying out integrated processing on the plurality of primary prediction models through a plurality of preset integrated instructions to obtain a plurality of integrated prediction models;

and according to the performance indexes of the integrated prediction models, determining a target prediction model from the integrated prediction models, and predicting the new energy power through the target prediction model.

In one embodiment, the obtaining sample new energy data according to the target prediction object includes:

determining a weather forecast data type and a power type according to the target prediction object;

acquiring historical weather forecast data of the target prediction object according to the weather forecast data type, and acquiring historical power data of the target prediction object according to the power type;

and obtaining the sample new energy data based on the historical weather forecast data and the historical power data.

In one embodiment, the obtaining the sample new energy data based on the historical weather forecast data and the historical power data includes:

obtaining initial sample new energy data based on the historical weather forecast data and the historical power data;

Removing data meeting preset removing conditions in the initial sample new energy data, and/or carrying out zero setting on negative power data in the initial sample new energy data, and/or carrying out interpolation processing on missing data in the sample new energy data to obtain processed sample new energy data;

and carrying out standardization processing on the processed sample new energy data to obtain the sample new energy data.

In one embodiment, the training by the new energy data to obtain a plurality of primary prediction models includes:

according to the target prediction type, determining an input variable and an actual output variable of each training step length from the training new energy data; the input variables comprise historical weather forecast data of a current prediction period and historical power of a previous prediction period, and the actual output variables represent the historical power of the current prediction period;

inputting the input variables into a plurality of primary prediction models to be trained respectively to obtain the prediction results of the primary prediction models to be trained;

and training each primary prediction model to be trained based on the loss value between the prediction result and the actual output variable to obtain a plurality of primary prediction models.

In one embodiment, the performing, by a plurality of preset integrated instructions, integrated processing on the plurality of primary prediction models respectively to obtain a plurality of integrated prediction models includes:

acquiring mean coefficients of the plurality of primary prediction models; the average coefficient is used for carrying out average processing on the prediction results of the primary prediction models;

and obtaining a first integrated prediction model based on the mean coefficient and each primary prediction model.

In one embodiment, the performing, by a plurality of preset integrated instructions, integrated processing on the plurality of primary prediction models respectively to obtain a plurality of integrated prediction models further includes:

inputting input variables in the training new energy data into each primary prediction model to obtain a prediction result of each primary prediction model;

acquiring the weight of each primary prediction model based on the prediction accuracy of the prediction result of each primary prediction model; the prediction accuracy is obtained based on the prediction result and the actual output variable in the verification new energy data;

and obtaining a second integrated prediction model based on each primary prediction model and the weight corresponding to each primary prediction model.

inputting input variables in the verified new energy data into each primary prediction model to obtain a prediction result of each primary prediction model;

training a secondary prediction model to be trained according to the prediction result and the actual output variable in the new energy data to obtain a trained secondary prediction model;

and obtaining a third integrated prediction model based on each primary prediction model and the trained secondary prediction model.

In one embodiment, the determining a target prediction model from the plurality of integrated prediction models according to the performance index of each integrated prediction model includes:

obtaining the prediction accuracy of each integrated prediction model;

and determining the integrated prediction model with the highest accuracy from the plurality of integrated prediction models as a target prediction model.

In a second aspect, the application further provides a new energy power prediction system based on multiple combined strategy integrated learning. The system comprises data acquisition equipment and a terminal; wherein the content of the first and second substances,

The data acquisition equipment is used for acquiring sample new energy data of the target prediction object; the sample new energy data comprises training new energy data and verification new energy data;

the terminal is used for acquiring the sample new energy data from the data acquisition equipment, obtaining a plurality of primary prediction models through training of the training new energy data, performing integrated processing on the plurality of primary prediction models respectively through a plurality of preset integrated instructions to obtain a plurality of integrated prediction models, determining a target prediction model from the plurality of integrated prediction models according to performance indexes of the integrated prediction models, and predicting new energy power through the target prediction model.

In a third aspect, the application further provides a new energy power prediction device based on multiple combined strategy integrated learning. The device comprises:

the acquisition module is used for acquiring sample new energy data according to the target prediction object; the sample new energy data comprises training new energy data and verification new energy data;

the training module is used for obtaining a plurality of primary prediction models through training of the new energy training data;

the integration module is used for respectively carrying out integration processing on the plurality of primary prediction models through a plurality of preset integration instructions to obtain a plurality of integrated prediction models;

And the determining module is used for determining a target prediction model from the plurality of integrated prediction models according to the performance indexes of the integrated prediction models and predicting the new energy power through the target prediction model.

In a fourth aspect, the present application further provides a computer device. The computer device comprises a memory storing a computer program and a processor implementing the following steps when executing the computer program:

In a fifth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of:

According to the new energy power prediction method, the system, the device, the computer equipment and the storage medium based on the multiple combined strategy integrated learning, after sample new energy data are obtained according to a target prediction object, a plurality of primary prediction models are obtained through training of new energy data in the sample new energy data, the plurality of primary prediction models are respectively subjected to integrated processing through multiple preset integrated instructions to obtain a plurality of integrated prediction models, finally, according to performance indexes of the integrated prediction models, a target prediction model is determined from the plurality of integrated prediction models, and new energy power prediction is carried out through the target prediction model. According to the method, a plurality of primary prediction models are trained firstly, each primary prediction model fully utilizes the learning capability of a neural network, the advantages of better establishing the fluctuation characteristic of a time sequence and the like are achieved, and the power fluctuation characteristic of a target prediction object can be well predicted. And then, integrating the primary prediction models by adopting various integration methods, and selecting a target prediction model from the obtained integrated prediction models, so that the model for predicting the power of the target prediction object is an optimal prediction model, and the power prediction precision of the new energy station is improved.

Drawings

FIG. 1 is a schematic flow chart illustrating a new energy power prediction method based on multi-combination strategy ensemble learning according to an embodiment;

FIG. 2 is a schematic flowchart of a sample new energy data acquisition step in one embodiment;

FIG. 3 is a diagram of a data prediction pre-processing module in one embodiment;

FIG. 4 is a diagram of an ensemble learning structure in one embodiment;

FIG. 5 is a schematic diagram of a new energy power prediction system based on multiple combined strategy ensemble learning in one embodiment;

FIG. 6 is a block diagram of a new energy power prediction device based on multiple integrated strategies for learning in one embodiment;

FIG. 7 is a diagram illustrating an internal structure of a computer device according to an embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

It should be noted that the terms "first," "second," and "third," etc. in the description and claims of this application and the above-described drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the application described herein are capable of operation in sequences other than those illustrated or described herein.

The following describes the various primary predictive models involved in the present application, which may include a bidirectional long-short term memory network, a circular gate unit model, and a bayesian network, wherein:

the bidirectional long and short term memory Network (BLSTM) is formed by combining a forward LSTM and a backward LSTM, the LSTM is one of Recurrent Neural Networks (RNN), unit calculation of the bidirectional Neural Network is communicated with one direction, a hidden layer of the bidirectional Neural Network needs to store two values, one value participates in forward calculation, and the other value participates in backward calculation.

The recurrent gate unit model (GRU) is also a kind of Recurrent Neural Network (RNN), and is proposed to solve the problems of long-term memory and gradient in back propagation, like the long-term memory network (LSTM). There are only two gates in the GRU model: an update gate and a reset gate, respectively, the update gate being used to control the extent to which the state information at the previous time is brought into the current state, the larger the value of the update gate, the more the state information at the previous time is brought in. Reset gates control how much information was written to the current candidate set from the previous state. The GRU combines the forgetting gate and the input gate into a single update gate, and also combines the cell state C and the hidden state h, so that the model is simpler than the standard LSTM model, and the mathematical expression is as follows:

Wherein the gate control signalz _t The range of (1) is 0 to 1. The closer the gating signal is to 1, the more data is represented to be memorized; and closer to 0 represents more "forgetting". Compared with the LSTM, the GRU has less 'gating' inside, has less parameters than the LSTM, but can achieve the equivalent functions of the LSTM, so the GRU can use less computing resources and time cost of a computer, and the method has obvious effect on adapting to a large-scale training set and a new energy power prediction scene sensitive to the prediction efficiency.

The bayesian neural network is different from a general neural network in that a weight parameter is a random variable rather than a certain value. BNN is modeled as follows:

assuming that the neural network parameters are W, p (W) is a priori distribution of the parameters, given observed data D = X, Y, where X is the input data and Y is the label data. BNN is expected to give the following distribution, i.e. predicted values:

since W is a random variable, the predictor is also a random variable. Wherein:

wherein P (W | D) is the a posteriori distribution, P (D | W) is the likelihood function, and P (D) is the edge likelihood. From the first equation, it can be seen that the core of probabilistic modeling and prediction of data using BNN lies in making efficient approximate posterior inferences, while variational inference VI or sampling is a very suitable approach. P (W | D) is evaluated by sampling the posterior distribution P (W | D), if sampled, and f (X | W) is calculated per sample, where f is the neural network. The output is a distribution, rather than a value, so that the uncertainty of the prediction can be estimated.

In one embodiment, as shown in fig. 1, a new energy power prediction method based on multiple combined policy ensemble learning is provided, and this embodiment is illustrated by applying the method to a terminal, and it is to be understood that the method may also be applied to a server, and may also be applied to a system including a terminal and a server, and is implemented by interaction between the terminal and the server. The terminal can be but not limited to various personal computers, notebook computers, smart phones, tablet computers, internet of things equipment and portable wearable equipment, and the internet of things equipment can be smart sound boxes, smart televisions, smart air conditioners, smart vehicle-mounted equipment and the like. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, and the like. The server may be implemented as a stand-alone server or as a server cluster comprised of multiple servers. In this embodiment, the method includes the steps of:

in step S110, sample new energy data is obtained according to the target prediction object; the sample new energy data comprises training new energy data and verification new energy data.

The prediction objects can comprise wind power plants and photovoltaic power plants.

Wherein the sample new energy data may include historical weather forecast data and historical power data.

In specific implementation, the demands of the prediction data corresponding to different prediction objects are different, so that a target prediction object needs to be determined in advance, a corresponding input feature quantity is constructed according to the target prediction object, and the sample new energy data is obtained based on the input feature quantity. More specifically, the corresponding weather forecast data type and power type may be determined according to the target prediction object, historical weather forecast data and historical power data of the target prediction object are respectively collected according to the weather forecast data type and the power type, and the historical weather forecast data and the historical power data are used to form sample new energy data of the target prediction object.

Further, after the sample new energy data is obtained, the sample new energy data can be divided into training new energy data used for model training and verification new energy data used for verifying the model effect, so that an optimal integrated prediction model can be selected subsequently.

In step S120, a plurality of primary prediction models are obtained by training new energy data.

The primary prediction model may include three models, i.e., a Bi-directional Long Short-Term Memory network (BLSTM), a Gate-round Unit (GRU) and a Bayesian Neural Network (BNN).

In the specific implementation, before each primary prediction model is trained, besides the input characteristic quantity of the model, an input variable and an output variable of each training step are required to be constructed, then the training functions of the primary prediction models such as the bidirectional long-short term memory network, the cyclic gate unit model and the Bayesian neural network are called, and the primary prediction models are trained respectively to obtain a plurality of trained primary prediction models.

More specifically, training steps of different prediction types are different, so before training each primary prediction model, a target prediction type needs to be determined, and an input variable and an actual output variable of each training step are determined from new energy source training data according to the target prediction type, wherein the input variable comprises historical weather forecast data of a current prediction period and historical power of a previous prediction period, and the actual output variable represents the historical power of the current prediction period. Respectively inputting the input variables into a plurality of primary prediction models to be trained to obtain the prediction result of each primary prediction model to be trained; and training each primary prediction model to be trained based on the loss value between the prediction result and the actual output variable to obtain a plurality of primary prediction models.

In step S130, a plurality of primary prediction models are respectively integrated through a plurality of preset integration instructions, so as to obtain a plurality of integrated prediction models.

In specific implementation, the preset multiple integrated instructions may be an integrated instruction based on an averaging method, an integrated instruction based on a voting method, and an integrated instruction based on a Stacking strategy, and the trained primary prediction models, such as a bidirectional long-short term memory network, a loop gate unit model, a bayesian neural network, and the like, are integrated by the integrated instructions to obtain multiple integrated prediction models.

In step S140, a target prediction model is determined from the plurality of integrated prediction models according to the performance index of each integrated prediction model, and new energy power prediction is performed through the target prediction model.

The performance index may be an accuracy of a prediction result of the integrated prediction model, and it can be understood that in some other scenarios, the performance index may also be a prediction rate, which may be specifically determined according to actual needs.

The relation of the prediction accuracy can be expressed as:

wherein, the first and the second end of the pipe are connected with each other,r ₁ the rate of accuracy of the prediction is represented,nrepresents the number of samples of the test set,C _k to representkTotal capacity of boot-up of a time slot.

In specific implementation, in order to improve the accuracy of power prediction of new energy in the power system, the accuracy of each integrated prediction model can be obtained by taking a performance index as the accuracy of a prediction result, the integrated prediction model with the highest prediction accuracy in the multiple integrated prediction models is determined as a target prediction model, and power prediction is performed on the new energy through the target prediction model.

More specifically, the input variables in the verified new energy data may be input into each integrated prediction model to obtain the prediction power of each integrated prediction model, the prediction accuracy of each integrated prediction model may be obtained based on the comparison result between the prediction power and the actual output power, and the integrated prediction model with the highest prediction accuracy may be further determined from the plurality of integrated prediction models to serve as the target prediction model.

According to the new energy power prediction method based on the multiple combined strategies integrated learning, after sample new energy data are obtained according to a target prediction object, a plurality of primary prediction models are obtained through training of new energy data in the sample new energy data, the plurality of primary prediction models are respectively subjected to integrated processing through preset multiple integrated instructions to obtain a plurality of integrated prediction models, and finally the target prediction model is determined from the plurality of integrated prediction models according to performance indexes of the integrated prediction models. According to the method, a plurality of primary prediction models are trained firstly, each primary prediction model fully utilizes the learning capacity of a neural network, and the method has the advantages of being capable of well establishing the fluctuation characteristic of a time sequence and the like, and can well predict the power fluctuation characteristic of a target prediction object. And then, integrating the primary prediction models by adopting various integration methods, and selecting a target prediction model from the obtained integrated prediction models, so that the model for predicting the power of the target prediction object is an optimal prediction model, and the power prediction precision of the new energy station is improved.

In an exemplary embodiment, the step S110 may be implemented by:

step S110a, determining weather forecast data type and power type according to the target prediction object;

step S110b, acquiring historical weather forecast data of the target prediction object according to the weather forecast data type, and acquiring historical power data of the target prediction object according to the power type;

and step S110c, obtaining sample new energy data based on the historical weather forecast data and the historical power data.

In specific implementation, after the target prediction object is determined, historical weather forecast data and historical power data can be acquired according to the weather forecast data type and the power type corresponding to the target prediction object, so that sample new energy data can be obtained.

More specifically, for a predicted object wind farm, the weather forecast data types may include: wind speed 170m high, wind speed 100m high, wind direction 170m high, wind direction 100m high, air pressure, surface air pressure, humidity and temperature, the power type of which may be wind power. For a predicted subject photovoltaic plant, the weather forecast data types may include: temperature, cloud cover, short wave radiation, long wave radiation, earth surface pressure, large scale precipitation, convection precipitation and humidity, and the power type can be photovoltaic power generation power.

In this embodiment, the weather forecast data type and the power type of the target prediction object are determined first, so that historical weather forecast data and historical power data corresponding to the target prediction object are obtained according to the weather forecast data type and the power type, sample new energy data is obtained, and subsequent training of a prediction model is facilitated.

Further, in an exemplary embodiment, as shown in fig. 2, the step S110c may be specifically implemented by the following steps:

step S210, obtaining initial sample new energy data based on historical weather forecast data and historical power data;

step S220, removing data meeting preset removing conditions in the initial sample new energy data, and/or carrying out zero setting on negative power data in the initial sample new energy data, and/or carrying out interpolation processing on missing data in the sample new energy data to obtain processed sample new energy data;

and step S230, carrying out standardization processing on the processed sample new energy data to obtain the sample new energy data.

The preset rejection condition may include that the data is error data and the data is in a power-limited period.

In the specific implementation, when most of the sample data is incomplete and inconsistent, data mining cannot be directly performed, or the mining result is not satisfactory. The data quality determines the upper limit of new energy power prediction, and in the power prediction practice of a wind power plant or a photovoltaic power station, missing values, repeated values and the like may exist in historical power data and historical weather forecast data, so that in order to improve the quality of data mining, data preprocessing is needed before initial sample new energy data such as the historical power data and the historical weather forecast data are used, and the accuracy of power prediction is improved.

More specifically, the preprocessing of the new energy power prediction data mainly comprises the following steps: the method comprises the following specific steps of power-limited time period rejection, negative power data zero setting, wrong data rejection, missing value processing and data standardization, as shown in fig. 3, the method is a schematic diagram of a data prediction preprocessing module, and each preprocessing comprises the following specific steps:

and eliminating the power limiting time period. Due to the intermittency and fluctuation of the power of the wind power plant and the photovoltaic power station, the quality of the supplied electric energy is poor; on the other hand, because the power grid is a system with real-time supply and demand balance, the situations of wind abandoning and power limiting and light abandoning and power limiting are inevitable. The method has the obvious characteristic in the sample set, namely under the numerical weather forecast conditions suitable for power generation of a wind power plant or a photovoltaic power station and corresponding to wind speed or irradiance in a time period, the output power of the new energy source station suddenly drops in a longer time period, the time period is judged to be the power limiting time period, or the sample set directly provides information of the power limiting time period, and data in the time period are removed.

The negative power data is zeroed. The new energy power generation power is a value greater than or equal to 0, and if the power data is less than 0, the power data in the time period can be set to zero.

And (4) removing error data. The new energy power generation data may have the problems of time interval errors, numerical values exceeding installed capacity, empty power numerical values or error fields in a certain time interval and the like, the historical weather forecast data may have the problems that the time interval errors, the numerical values are empty, the fields are wrong, or infinite or infinitesimal power values or the like obviously do not meet the actual conditions, and the power data and the weather forecast data in the corresponding time interval need to be removed at the same time.

And (4) processing missing values. Since the resolutions of the power data and the numerical weather forecast data are both 15 minutes, and the distances of the sample attributes are clear, the missing values are interpolated by using the average value of the effective values of the corresponding attributes by using a mean interpolation method.

And (4) standardizing the data. Because the influence when different attributes such as numerical weather forecast or historical power have different magnitudes needs to be eliminated: the difference of the orders of magnitude leads to the property with larger orders of magnitude to occupy the dominant position; ② the difference of order of magnitude will lead to the slow convergence speed of the neural network iteration. Therefore, the samples were normalized using z-score, i.e., the data was normalized based on the mean and standard deviation of the raw data. The original value x is normalized to x' using z-score, i.e. new data = (original data-mean)/standard deviation.

In this embodiment, the quality of the sample new energy data can be improved by preprocessing the initial sample new energy data, so that the prediction accuracy of the prediction model for predicting the new energy power, which is obtained by training according to the sample new energy data, can be improved.

In an exemplary embodiment, in the step S120, the training of the new energy data to obtain a plurality of primary prediction models may specifically be implemented by the following steps:

Step S120a, determining an input variable and an actual output variable of each training step from the training new energy data according to the target prediction type; the input variables comprise historical weather forecast data of a current prediction period and historical power of a previous prediction period, and the actual output variables represent the historical power of the current prediction period;

step S120b, respectively inputting input variables into a plurality of primary prediction models to be trained to obtain the prediction results of the primary prediction models to be trained;

and step S120c, training each primary prediction model to be trained based on the loss value between the prediction result and the actual output variable to obtain a plurality of primary prediction models.

The prediction types may include short-term prediction and ultra-short-term prediction, among others.

In specific implementation, the prediction types are different, and the corresponding prediction parameters are also different, so that the input variable and the actual output variable of the training step length are also different, and therefore, before training each primary prediction model, the target prediction type needs to be determined. According to the target prediction type, determining an input variable and an actual output variable of each training step length from the new energy source training data, then respectively inputting the input variables into a plurality of primary prediction models to be trained to obtain the prediction result of each primary prediction model to be trained, and training each primary prediction model to be trained on the basis of the loss value between the prediction result and the actual output variable to obtain a plurality of primary prediction models.

More specifically, the prediction parameters may include a prediction time period, a time resolution, and a rolling prediction period, wherein the time resolution refers to a minimum time interval between two adjacent telemetric observations made in the same region.

For example, for the short-term prediction type, the prediction time period can be 0-72 hours in the future, the time resolution is not less than 15 minutes, and the rolling prediction period is 24 hours; for the ultra-short-term prediction type, the prediction time period is 0-4 hours in the future, the time resolution is 15 minutes, and the rolling prediction period is 15 minutes.

Further, the input and output of the training steps can be determined based on the prediction parameters, for example, in the ultra-short term prediction, the rolling prediction period is 15 minutes, i.e. once every 15 minutes, and the prediction time points are t0, t1 and t2 from the time t0, then there are t2-t1= t1-t0=15 minutes, the prediction time period is 0-4 hours in the future, the time resolution is 15 minutes, i.e. every 15 minutes corresponds to one prediction time point in 4 hours, then there are 16 time points in the prediction time period, so the input and output of each training step are as shown in table 1 below:

TABLE 1 input and output of training step size for ultra-short term prediction

In short-term prediction, the rolling prediction period is 24 hours, i.e. prediction is performed once every 24 hours, starting from time t0, the prediction time points are t0, t96 and t192, then there are t192-t96= t96-t0=24 hours, the prediction time period is 0-72 hours in the future, the time resolution is 15 minutes, i.e. every 15 minutes corresponds to one prediction time point in 72 hours, and then there are 288 time points in the prediction time period, so the input and output of each training step are shown in table 2 below:

TABLE 2 input and output of training step size for short term prediction

As can be seen from tables 1 and 2, when each primary prediction model is trained by training new energy data, the input variable of each training step is formed by the historical power of the previous rolling prediction period and the historical weather forecast data in the current prediction period, and the output is the prediction power in the current prediction period. Therefore, when each primary prediction model is trained, input variables and actual output variables of each training are determined from new energy source training data according to a target prediction type, then the input variables are sequentially input into each primary prediction model according to a training step length, an obtained prediction result and actual output variables are input into a training function (or called loss function), a loss value between the prediction result and the actual output variables is obtained, and each primary prediction model to be trained is trained through the loss value to obtain each trained primary prediction model.

In this embodiment, the input variable and the actual output variable of each training step are determined by the target prediction type, and each of the primary prediction models to be trained is trained based on the input variable and the output variable to obtain the plurality of primary prediction models.

In an exemplary embodiment, the step S130 includes: acquiring mean coefficients of a plurality of primary prediction models; the average coefficient is used for carrying out average processing on the prediction results of the primary prediction models; and obtaining a first integrated prediction model based on the mean coefficient and each trained primary prediction model.

In a specific implementation, the first integrated prediction model may be an averaging-based integrated instruction. The averaging method is to average the outputs of a plurality of primary prediction models to obtain a final prediction output, in this embodiment, an arithmetic averaging method is used, that is, the final prediction is the average of the prediction results of each primary prediction model, and the relation of the first integrated prediction model can be expressed as:

wherein the content of the first and second substances,H(x) Represents the prediction output of the first integrated prediction model, 1-TIt is possible to express the coefficient of the mean value,hirepresenting the primary predictive model.

In another embodiment, if each primary predictive model has a weightwThen, the relationship of the first integrated prediction model can be expressed as:

wherein the content of the first and second substances,wirepresenting a primary prediction modelhiIs usually given a weight of

。

In this embodiment, the first integrated prediction model is obtained through the mean coefficient and each trained primary prediction model, and the integrated processing of each trained primary prediction model based on the averaging method is realized.

In an exemplary embodiment, the step S130 further includes: inputting input variables in the training new energy data into each primary prediction model to obtain a prediction result of each primary prediction model; acquiring the weight of each primary prediction model based on the prediction accuracy of the prediction result of each primary prediction model; the prediction accuracy is obtained based on the prediction result and the output variable in the verification new energy data; and obtaining a second integrated prediction model based on the primary prediction models and the weights corresponding to the primary prediction models.

The weight and the prediction accuracy rate form a positive correlation relationship, namely the higher the prediction accuracy rate is, the larger the weight is, the lower the prediction accuracy rate is, and the smaller the weight is.

In a specific implementation, the second integrated predictive model may be an integrated instruction based on a voting method. In this embodiment, the voting method used is a weighted voting method, that is, the number of classification votes of each primary prediction model needs to be multiplied by a weight, and finally the weighted votes of each category are summed, and the category corresponding to the largest value is the final category.

More specifically, after the prediction result of each primary prediction model is obtained, the prediction result is compared with the actual output variable in the training new energy data, if the comparison result is within the error range, the prediction is judged to be accurate, otherwise, the prediction is judged to be inaccurate, the prediction accuracy of each primary prediction model is obtained according to the comparison result of each sample data, the prediction accuracy of each primary prediction model is ranked, the weight of each primary prediction model is determined according to the ranking result, and further, the corresponding primary prediction model is weighted and summed through each weight to obtain a second integrated prediction model.

For example, the accuracy of the BLSTM, GRU and BNN primary prediction models is ranked, the proportions of 0.5, 0.3 and 0.2 are respectively given according to the ranking, the results output by the BLSTM, GRU and BNN algorithms are voted, and finally the power prediction result based on the voting method is obtained.

In the embodiment, the prediction accuracy of each stage of prediction model is determined according to the prediction result of each primary prediction model, the weight of each primary prediction model is obtained based on the accuracy, the second integrated prediction model is further obtained based on the weight of each primary prediction model, and the integrated processing of each trained primary prediction model based on a weighted voting method is realized.

In an exemplary embodiment, the step S130 further includes: inputting input variables in the verified new energy data into each primary prediction model to obtain a prediction result of each primary prediction model; training the secondary prediction model to be trained through the prediction result and the actual output variable in the new energy data to obtain a trained secondary prediction model; and obtaining a third integrated prediction model based on each trained primary prediction model and each trained secondary prediction model.

In a specific implementation, the third integrated prediction model may be an integrated instruction based on a Stacking policy. And taking the prediction result of each primary prediction model obtained by training on the verification new energy data as input, taking the actual output variable in the verification new energy data as output, and retraining a secondary prediction model to obtain a final prediction result, thereby constructing an integrated prediction model based on a Stacking strategy. For the verification of new energy data, firstly, each primary prediction model is used for predicting once to obtain an input sample of a secondary prediction model, and then, the secondary prediction model is used for predicting once to obtain a final prediction result. The specific idea is as follows:

A plurality of primary predictive models are trained from an initial data set, and then a new data set is "generated" for training a secondary predictive model. In this new data set generated, the output of the primary predictive model is treated as a sample input feature for the new data set, while the label (i.e., the actual output variable) of the original data set is still treated as a sample label. That is, assuming that there are M primary prediction models, then for a sample (x; y) in an original data set, there are M outputs { h1(x), h2(x),.., hm (x) } from the M primary prediction models, { h1(x), h2(x),. hm, (x); y } as a sample of the new data, so the output of a primary prediction model is used as a feature of the corresponding sample in the new data set, and it is marked as a marker of that sample in the original data.

In this embodiment, a secondary prediction model is trained according to the prediction result of each primary prediction model and the actual output variable in the new energy data, and a third integrated model is obtained from each primary prediction model and each secondary prediction model, so that the integrated processing of a plurality of primary prediction models based on a Stacking strategy is realized.

In one embodiment, to facilitate understanding of the embodiments of the present application by those skilled in the art, the following description, taken in conjunction with the specific examples of the drawings, includes the following steps:

(1) And preprocessing the historical power data and the historical weather forecast data of the new energy. In the practice of power prediction of a wind power plant or a photovoltaic power station, missing values, repeated values and the like may exist in historical power data and numerical weather forecast data, and in order to improve the quality of data mining, data preprocessing is required before the data is used, so that the accuracy of power prediction is improved. The data preprocessing can be realized by means of power-limited time period elimination, negative power data zero setting, wrong data elimination, missing value processing, data standardization and the like.

(2) And (5) building a primary prediction model. And (3) building primary prediction models such as a bidirectional long-short term memory network, a cycle gate unit model and a Bayesian network.

(3) And building an integrated prediction model based on a plurality of combination strategies. Referring to fig. 4, which is a schematic diagram of an integrated learning structure, a plurality of primary prediction models are trained, and then an integrated prediction model with a stronger learning ability is formed through a certain combination strategy. The method specifically comprises the following steps: building an integrated prediction model based on an averaging method and a strategy, building an integrated prediction model based on a voting method, and building an integrated prediction model based on a Stacking strategy.

(4) And cutting the sample set. Before power prediction is carried out, an algorithm model is trained by using a proper training set, and the effect of the algorithm is evaluated by using a verification set, so that the optimal integrated learning strategy is selected, and the optimal integrated prediction model is finally output. The method comprises the following steps of carrying out 15-minute-resolution historical power and 15-minute-resolution numerical weather forecast according to the following steps of: and 4, segmenting the training set and the verification set.

(5) A primary prediction model is trained. And respectively constructing input characteristic quantities for training the primary prediction model according to different prediction objects, constructing input and output of each training step length, calling training functions of BLSTM, GRU and BNN, and finally outputting the trained three types of primary prediction models.

(6) Training a secondary prediction model based on a packing strategy.

(7) And selecting an optimal combination strategy for ensemble learning based on the verified new energy data. In the process of verifying the new energy data, firstly, the characteristic quantity same as that of the training new energy data is used as input, and three primary prediction models, namely BLSTM, GRU and BNN, are used for power prediction to obtain a power prediction result. And then, respectively adopting an averaging method, a voting method and a Stacking strategy to carry out power prediction, and selecting an optimal strategy for integrated learning according to the accuracy of the final prediction result. The method comprises the following specific steps:

a. Power prediction based on averaging

In the process of verifying the new energy data, the results output by BLSTM, GRU and BNN algorithms are arithmetically averaged to obtain a power prediction result based on an averaging method.

b. Power prediction based on voting method

In the training of new energy data, the accuracy rates of three types of weak learners including BLSTM, GRU and BNN are ranked, the proportions of 0.5, 0.3 and 0.2 are respectively given according to the ranking, in the verification of the new energy data, the results output by BLSTM, GRU and BNN algorithms are voted, and finally, the power prediction result based on the voting method is obtained.

c. Power prediction based on Stacking strategy

And performing power prediction on the BLSTM, GRU and BNN primary prediction models, and taking the obtained power prediction result as the input of a secondary prediction model based on a Stacking strategy, thereby obtaining the power prediction result based on the Stacking strategy.

In the embodiment, a new energy power Short-Term and ultra-Short-Term prediction method based on a Bi-directional Long Short-Term Memory network (BLSTM), a Gate Recovery Unit (GRU) and a Bayesian Neural Network (BNN) is provided, Short-Term and ultra-Short-Term power generation power of a wind farm and a photovoltaic power station is predicted, the Short-Term prediction time period is 0-72 hours in the future, and the time resolution is 15 minutes; the ultra-short term prediction period is 0 to 4 hours into the future, with a time resolution of 15 minutes. The three neural network algorithms fully utilize the learning capability of the neural network, have the advantages of being capable of well establishing the time sequence fluctuation characteristic and the like, and can well predict the power fluctuation characteristic of the wind power plant and the photovoltaic power station. The method is characterized in that integrated learning is carried out on three types of individual learners by an averaging method, a voting method and a Stacking method to obtain three types of integrated learning models, an optimal integrated learning model is selected from the three types of integrated learning models, short-term and ultra-short-term power of a wind power plant and a photovoltaic power plant is predicted, and power prediction accuracy of a new energy plant is greatly improved.

It should be understood that, although the steps in the flowcharts related to the embodiments described above are shown in sequence as indicated by the arrows, the steps are not necessarily performed in sequence as indicated by the arrows. The steps are not limited to being performed in the exact order illustrated and, unless explicitly stated herein, may be performed in other orders. Moreover, at least a part of the steps in the flowcharts related to the embodiments described above may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, and the execution order of the steps or stages is not necessarily sequential, but may be performed alternately or alternately with other steps or at least a part of the steps or stages in other steps.

It can be understood that, although various power prediction methods are put into practical operation at the present stage, the generalization capability of the prediction model is low, and the prediction model cannot adapt to all new energy stations, or different algorithm models have respective prediction advantages for different new energy stations, and no better method is available for comprehensively utilizing the algorithm model with better performance, so that the actual requirement of future large-scale access of new energy to the power system cannot be met.

In order to solve the problems, a new energy power prediction system based on multiple combination strategies integrated learning is also built on the basis of the new energy power prediction method. Referring to fig. 5, a schematic diagram of a new energy power prediction system based on multiple combined strategy ensemble learning, including a data acquisition device and a terminal, wherein,

the data acquisition equipment is used for acquiring sample new energy data of the target prediction object; the sample new energy data comprises training new energy data and verification new energy data.

Specifically, historical power and numerical weather forecast for a wind farm or photovoltaic power plant may be collected, for example, historical power at 15 minutes resolution and numerical weather forecast at 15 minutes resolution.

Wherein, the numerical weather forecast data requirement is as follows: (1) the numerical weather forecast content at least comprises air temperature, wind speed, wind direction, air pressure, irradiance, temperature, relative humidity and the like at different elevations of 0-100 meters away from the ground; (2) the numerical weather forecast data updating time period is not more than 12 hours, the time resolution is not less than 15 minutes, and the single forecast time is not less than 72 hours; (3) the space range of the numerical weather forecast covers the target station, and the prediction grid scale is not more than 3 kilometers multiplied by 3 kilometers.

The terminal is used for acquiring sample new energy data from the data acquisition equipment, obtaining a plurality of primary prediction models through training the new energy data, respectively carrying out integrated processing on the plurality of primary prediction models through a plurality of preset integrated instructions to obtain a plurality of integrated prediction models, determining a target prediction model from the plurality of integrated prediction models according to performance indexes of the integrated prediction models, and carrying out new energy power prediction through the target prediction model.

The primary prediction model can be generated based on three algorithms of BLSTM, GRU and BNN and used for predicting the power of the new energy. The integration instruction can comprise an averaging method, a voting method and a Stacking method.

The system can also comprise a database for storing relevant data such as the prediction result of the power prediction model.

In the new energy power prediction system with multiple integrated learning strategies provided by this embodiment, a primary prediction model for new energy power prediction is generated by using three algorithms, namely BLSTM, GRU and BNN, then three integrated prediction models for integrated learning are generated by using three types of strategies (averaging, voting and Stacking) for integrated learning, finally ranking comparison is performed according to the performances (performance is evaluated by the accuracy of power prediction) of the three integrated prediction models generated by the three types of methods, an optimal integrated learning strategy for short-term or ultra-short-term of a specific wind farm or photovoltaic power station is selected and applied to short-term/ultra-short-term power prediction of the wind farm/photovoltaic power station, and a power prediction result is output to a MySQL database. The system can realize the omnibearing, self-adaptive and high-precision power prediction of the new energy station.

Based on the same inventive concept, the embodiment of the application also provides a new energy power prediction device based on multi-combination strategy ensemble learning, which is used for realizing the new energy power prediction method based on multi-combination strategy ensemble learning. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the method, so that the specific limitations in one or more embodiments of the new energy power prediction device based on multi-combination strategy ensemble learning provided below can be referred to the limitations in the above description for the new energy power prediction method based on multi-combination strategy ensemble learning, and are not repeated herein.

In one embodiment, as shown in fig. 6, there is provided a new energy power prediction apparatus based on multiple combined strategy ensemble learning, including: an acquisition module 610, a training module 620, an integration module 630, and a determination module 640, wherein:

an obtaining module 610, configured to obtain sample new energy data according to a target prediction object; the sample new energy data comprises training new energy data and verification new energy data;

a training module 620, configured to obtain a plurality of primary prediction models through training new energy data;

An integration module 630, configured to perform integration processing on the multiple primary prediction models respectively through multiple preset integration instructions to obtain multiple integrated prediction models;

and the determining module 640 is configured to determine a target prediction model from the multiple integrated prediction models according to the performance index of each integrated prediction model, and perform new energy power prediction through the target prediction model.

In an embodiment, the obtaining module 610 is specifically configured to determine a weather forecast data type and a power type according to a target prediction object; acquiring historical weather forecast data of a target prediction object according to the weather forecast data type, and acquiring historical power data of the target prediction object according to the power type; and obtaining sample new energy data based on the historical weather forecast data and the historical power data.

In one embodiment, the device further comprises a preprocessing module, configured to obtain initial sample new energy data based on historical weather forecast data and historical power data; removing data meeting preset removing conditions in the initial sample new energy data, and/or carrying out zero setting on negative power data in the initial sample new energy data, and/or carrying out interpolation processing on missing data in the sample new energy data to obtain processed sample new energy data; and carrying out standardization processing on the processed sample new energy data to obtain the sample new energy data.

In an embodiment, the training module 620 is further configured to determine an input variable and an actual output variable of each training step from the training new energy data according to the target prediction type; the input variables comprise historical weather forecast data of a current prediction period and historical power of a previous prediction period, and the actual output variables represent the historical power of the current prediction period; respectively inputting input variables into a plurality of primary prediction models to be trained to obtain a prediction result of each primary prediction model to be trained; and training each primary prediction model to be trained based on the loss value between the prediction result and the actual output variable to obtain a plurality of primary prediction models.

In one embodiment, the integration module 630 is configured to obtain mean coefficients of a plurality of primary prediction models; the average coefficient is used for carrying out average processing on the prediction results of the primary prediction models; and obtaining a first integrated prediction model based on the mean coefficient and each primary prediction model.

In an embodiment, the integration module 630 is further configured to input variables in the training new energy data into each primary prediction model to obtain a prediction result of each primary prediction model; acquiring the weight of each primary prediction model based on the prediction accuracy of the prediction result of each primary prediction model; the prediction accuracy is obtained based on the prediction result and the actual output variable in the verified new energy data; and obtaining a second integrated prediction model based on the primary prediction models and the weights corresponding to the primary prediction models.

In an embodiment, the integration module 630 is further configured to input the input variables in the new energy data to each primary prediction model, so as to obtain a prediction result of each primary prediction model; training the secondary prediction model to be trained through the prediction result and the actual output variable in the new energy data to obtain a trained secondary prediction model; and obtaining a third integrated prediction model based on each primary prediction model and the trained secondary prediction model.

In an embodiment, the determining module 640 is further configured to obtain a prediction accuracy of each integrated prediction model; and determining the integrated prediction model with the highest accuracy from the plurality of integrated prediction models as a target prediction model.

The modules in the new energy power prediction device based on the multiple combined strategy ensemble learning can be wholly or partially realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as shown in fig. 7. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used for carrying out wired or wireless communication with an external terminal, and the wireless communication can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. The computer program is executed by a processor to implement a new energy power prediction method based on a plurality of integrated learning strategies. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.

It will be appreciated by those skilled in the art that the configuration shown in fig. 7 is a block diagram of only a portion of the configuration associated with the present application, and is not intended to limit the computing device to which the present application may be applied, and that a particular computing device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.

In an embodiment, a computer device is further provided, which includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above method embodiments when executing the computer program.

In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored which, when being executed by a processor, carries out the steps of the above-mentioned method embodiments.

It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) referred to in the present application are information and data authorized by the user or sufficiently authorized by each party.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, database, or other medium used in the embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high-density embedded nonvolatile Memory, resistive Random Access Memory (ReRAM), Magnetic Random Access Memory (MRAM), Ferroelectric Random Access Memory (FRAM), Phase Change Memory (PCM), graphene Memory, and the like. Volatile Memory can include Random Access Memory (RAM), external cache Memory, and the like. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM), among others.

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A new energy power prediction method based on multiple combined strategy ensemble learning is characterized by comprising the following steps:

acquiring sample new energy data according to the target prediction object; the sample new energy data comprises training new energy data and verification new energy data; the target prediction object comprises a wind power plant and a photovoltaic power station;

training through the training new energy data to obtain a plurality of primary prediction models; further comprising: determining corresponding prediction parameters according to the target prediction type, and determining an input variable and an actual output variable of each training step from the training new energy data according to the prediction parameters; inputting the input variables into a plurality of primary prediction models to be trained respectively to obtain the prediction results of the primary prediction models to be trained; training each primary prediction model to be trained based on the loss value between the prediction result and the actual output variable to obtain a plurality of primary prediction models; the input variables comprise historical weather forecast data of a current prediction period and historical power of a previous prediction period, and the actual output variables represent the historical power of the current prediction period; the target prediction type comprises short-term prediction and ultra-short-term prediction, the prediction parameters comprise a prediction time period, a time resolution and a rolling prediction period, and the time resolution is 15 minutes;

2. The method of claim 1, wherein obtaining sample new energy data based on the target predicted object comprises:

3. The method of claim 2, wherein the deriving the sample new energy data based on the historical weather forecast data and the historical power data comprises:

4. The method according to claim 1, wherein the integrating the plurality of primary prediction models by a plurality of preset integration instructions to obtain a plurality of integrated prediction models comprises:

5. The method according to claim 1, wherein the integrating the plurality of primary prediction models by a plurality of preset integrating instructions to obtain a plurality of integrated prediction models further comprises:

6. The method according to claim 1, wherein the integrating the plurality of primary prediction models by a plurality of preset integrating instructions to obtain a plurality of integrated prediction models further comprises:

7. A new energy power prediction system based on multiple combined strategy integrated learning is characterized by comprising data acquisition equipment and a terminal; wherein the content of the first and second substances,

the data acquisition equipment is used for acquiring sample new energy data of the target prediction object; the sample new energy data comprises training new energy data and verification new energy data; the target prediction object comprises a wind power plant and a photovoltaic power station;

the terminal is used for acquiring the sample new energy data from the data acquisition equipment, obtaining a plurality of primary prediction models through training of the training new energy data, performing integrated processing on the plurality of primary prediction models respectively through a plurality of preset integrated instructions to obtain a plurality of integrated prediction models, determining a target prediction model from the plurality of integrated prediction models according to performance indexes of the integrated prediction models, and performing new energy power prediction through the target prediction model; wherein the training process of the plurality of primary predictive models comprises: determining corresponding prediction parameters according to the target prediction type, and determining an input variable and an actual output variable of each training step from the training new energy data according to the prediction parameters; inputting the input variables into a plurality of primary prediction models to be trained respectively to obtain the prediction results of the primary prediction models to be trained; training each primary prediction model to be trained based on the loss value between the prediction result and the actual output variable to obtain a plurality of primary prediction models; the input variables comprise historical weather forecast data of a current prediction period and historical power of a previous prediction period, and the actual output variables represent the historical power of the current prediction period; the target prediction type comprises short-term prediction and ultra-short-term prediction, the prediction parameters comprise a prediction time period, a time resolution and a rolling prediction period, and the time resolution is 15 minutes.

8. A new energy power prediction apparatus based on multiple integrated learning strategies, the apparatus comprising:

the acquisition module is used for acquiring sample new energy data according to the target prediction object; the sample new energy data comprises training new energy data and verification new energy data; the target prediction object comprises a wind power plant and a photovoltaic power station;

the determining module is used for determining a target prediction model from the plurality of integrated prediction models according to the performance indexes of the integrated prediction models and predicting the new energy power through the target prediction model;

the training module is further used for determining corresponding prediction parameters according to the target prediction type, and determining an input variable and an actual output variable of each training step length from the training new energy data according to the prediction parameters; inputting the input variables into a plurality of primary prediction models to be trained respectively to obtain the prediction results of the primary prediction models to be trained; training each primary prediction model to be trained based on the loss value between the prediction result and the actual output variable to obtain a plurality of primary prediction models; the input variables comprise historical weather forecast data of a current prediction period and historical power of a previous prediction period, and the actual output variables represent the historical power of the current prediction period; the target prediction type comprises short-term prediction and ultra-short-term prediction, the prediction parameters comprise a prediction time period, a time resolution and a rolling prediction period, and the time resolution is 15 minutes.

9. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor implements the steps of the method of any one of claims 1 to 6 when executing the computer program.

10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 6.