Power load prediction method and device, electronic equipment and storage medium

Abstract

The invention discloses a power load prediction method and device, electronic equipment and a storage medium. The method comprises the following steps: determining first data; and performing power load prediction through a first model according to the first data to obtain second data. According to the method, the second data is predicted through the first model according to the first data, time sequence change features in the first data are considered, nonlinear change features in the first data are also considered, the adaptive limitation of a traditional neural network in a complex time sequence fluctuation scene can be effectively overcome, the prediction accuracy is improved, and meanwhile the prediction efficiency is improved. The recurrent neural network layer enables the model to have a better time-frequency localization characteristic, and can ensure that the influence of interference is reduced when the first model performs feature extraction on the first data, thereby improving the anti-interference capability of power load prediction.

Description

Technical Field

[0001] The present invention relates to the field of electric power technology, and in particular to a method, device, electronic equipment and storage medium for predicting electric load. Background Art

[0002] With the rapid development of smart grids, power load forecasting plays a vital role in power system planning, operation, and optimization. Traditional power load forecasting methods, such as time series analysis and regression analysis, have problems such as insufficient capture of nonlinear characteristics and incomplete consideration of external factors, resulting in inaccurate power load forecasts. Summary of the Invention

[0003] The present invention provides a power load forecasting method, device, electronic equipment and storage medium to solve the problem of inaccurate power load forecasting.

[0004] According to one aspect of the present invention, there is provided a method for power load forecasting, comprising:

[0005] Determine first data; the first data is determined based on meteorological change data, power load change data, and power consumption characteristic data of the target area within a first time period; the power consumption characteristic data is used to characterize the power consumption of equipment in the target area and changes in power load resources;

[0006] The power load is predicted by a first model based on the first data to obtain second data; the first model is configured with a fuzzy neural network layer and a recursive neural network layer; the fuzzy neural network layer is used to obtain nonlinear feature information in the first data; the recursive neural network layer is used to obtain dynamic feature information of the first data changing over time; the fuzzy neural network layer and the recursive neural network layer in the first model respectively predict the first data, and the obtained feature information is weighted in the output layer; the second data is used to characterize the power load changes in the target area within a first time period.

[0007] According to another aspect of the present invention, there is provided a power load forecasting device, comprising:

[0008] a first data determination module for determining first data; the first data is determined based on meteorological change data, power load change data, and power consumption characteristic data of the target area within a first time period; the power consumption characteristic data is used to characterize the power consumption of equipment in the target area and changes in power load resources;

[0009] A second data determination module is used to predict the power load based on the first data through the first model to obtain second data; the first model is configured with a fuzzy neural network layer and a recursive neural network layer; the fuzzy neural network layer is used to obtain nonlinear feature information in the first data; the recursive neural network layer is used to obtain dynamic feature information of the first data changing over time; the fuzzy neural network layer and the recursive neural network layer in the first model respectively predict the first data, and the obtained feature information is weighted in the output layer; the second data is used to characterize the power load changes in the target area within the first time period.

[0010] According to another aspect of the present invention, an electronic device is provided, comprising:

[0011] at least one processor; and

[0012] a memory communicatively connected to the at least one processor; wherein,

[0013] The memory stores a computer program that can be executed by the at least one processor. The computer program is executed by the at least one processor so that the at least one processor can execute the power load forecasting method described in any embodiment of the present invention.

[0014] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the power load forecasting method according to any embodiment of the present invention when executed.

[0015] The technical solution of the embodiment of the present invention determines first data; performs power load forecasting through a first model based on the first data to obtain second data. The fuzzy neural network layer in the first model can capture the nonlinear characteristics in the first data; the recursive neural network layer can capture the time series characteristics in the first data, and can enable the first model to capture different change characteristics in the first data. The combination of different change characteristics can improve the accuracy of the model prediction, so that the obtained second data is more in line with the actual power load changes. This method predicts the second data based on the first data through the first model, which not only takes into account the time series change characteristics in the first data, but also takes into account the nonlinear change characteristics in the first data. It can effectively overcome the adaptability limitations of traditional neural networks in complex time series fluctuation scenarios, improve the accuracy of the prediction, and at the same time, the recursive neural network layer enables the model to have better time-frequency localization characteristics, which can ensure that the first model reduces the impact of interference when extracting features from the first data, thereby improving the anti-interference ability of power load forecasting.

[0016] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 A flow chart of a method for power load forecasting provided by an embodiment of the present invention;

[0019] Figure 2 A comparison chart of prediction results of a first model provided by an embodiment of the present invention;

[0020] Figure 3 A schematic structural diagram of a first model provided by an embodiment of the present invention;

[0021] Figure 4 A schematic diagram of a first model fitness change provided by an embodiment of the present invention;

[0022] Figure 5 A schematic diagram of a first model loss function provided by an embodiment of the present invention;

[0023] Figure 6 A schematic diagram of a feature quantization score provided by an embodiment of the present invention;

[0024] Figure 7 A schematic structural diagram of a power load forecasting device provided by an embodiment of the present invention;

[0025] Figure 8 A schematic diagram of the structure of an electronic device for implementing the power load forecasting method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0027] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] Figure 1 This is a flow chart of a power load forecasting method provided by an embodiment of the present invention. This embodiment is applicable to situations where power load forecasting is performed. The method can be executed by a power load forecasting device. The power load forecasting device can be implemented in the form of hardware and / or software. The power load forecasting device can be configured in any electronic device with network communication function. Figure 1 As shown, the method includes:

[0029] S110. Determine first data; the first data is determined based on meteorological change data, power load change data and power consumption characteristic data of the target area within the first time period; the power consumption characteristic data is used to characterize the power consumption of equipment in the target area and the changes in power load resources.

[0030] The target area is the area that can be covered by the electric energy provided by the power system.

[0031] The first time period is a time period adjacent to the time period for which power load forecasting is required, and is located before the time period for which power load forecasting is required.

[0032] The meteorological change data is used to represent data generated by meteorological changes in the target area during the first time period. Exemplarily, the meteorological change data includes at least one of the following: temperature, humidity, wind speed, rainfall, sunshine duration, and PM2.5.

[0033] The power usage characteristic data is used to characterize the different power usage patterns of devices within the target area during a first time period, as well as changes in the target area's power resources during the first time period. Exemplarily, the power usage characteristic data includes at least one of the following: GDP growth rate, power load resources, industrial production index, device types within the target area, and power usage pattern clustering results. The power usage pattern clustering results are the result of classifying the device's power usage behavior according to preset dimensions. The preset dimensions include at least one of the following: time series and load characteristics.

[0034] The first data is a data set obtained by matching meteorological change data, power load change data, and power consumption characteristic data of the target area within a first time period according to the timestamp of the power load change data.

[0035] The power load change data is used to represent the power load values ​​of the target area at different times during the first time period. The power load data is collected at a preset time interval. The preset time interval may be 1 hour.

[0036] Furthermore, the power load change data includes a timestamp. The timestamp identifier included in a power load data includes at least one of the following: year, season, month, day, time, day of the week, and whether the day is a working day or a holiday.

[0037] Specifically, to determine the first data, first obtain power load change data from the SCADA system (Supervisory Control and Data Acquisition); obtain corresponding data from the database of corresponding categories according to different types of data, that is, obtain meteorological change data within the first time period from the meteorological data network; obtain electricity consumption characteristic data within the first time period from the electricity trading website, and filter the obtained meteorological change data and electricity consumption characteristic data according to the timestamp of the power load change data to obtain the first data.

[0038] Furthermore, the obtained weather change data and electricity consumption characteristic data are filtered according to the timestamp of the power load change data to obtain the first data. The steps include: obtaining date data and time data from the power load change data, merging the date data and time data to serve as a time identifier. Based on the time identifier, data at the same time is filtered from the weather change data and electricity consumption characteristic data, and matched with the power load change data to obtain the first data.

[0039] Among them, SCADA (Supervisory Control and Data Acquisition) is a computer system used to monitor and control the power load generation process of the power grid in real time.

[0040] S120. Power load is predicted through the first model according to the first data to obtain second data; the first model is configured with a fuzzy neural network layer and a recursive neural network layer; the fuzzy neural network layer is used to obtain nonlinear characteristic information in the first data; the recursive neural network layer is used to obtain dynamic characteristic information of the first data changing over time; the fuzzy neural network layer and the recursive neural network layer in the first model respectively predict the first data, and the obtained characteristic information is weighted in the output layer; the second data is used to characterize the power load changes in the target area within the first time period;.

[0041] The fuzzy neural network layer consists of a membership function layer and a fuzzy rule layer. The membership function layer is the perception layer of the fuzzy system, responsible for converting the first data into membership values ​​of fuzzy sets. The fuzzy rule layer performs logical reasoning based on the output of the membership function layer to obtain nonlinear feature information from the first data.

[0042] The recursive neural network layer is composed of a single recursive wavelet neuron, which combines the time-frequency analysis capability of wavelet transform with the memory characteristics of recursive structure, thereby being able to obtain dynamic feature information of the first data that changes over time.

[0043] The first model is further configured with an input layer, an output layer, and an improved particle swarm model. The input layer is used to input the first data; the output layer is used to perform weighted output of the output results of the fuzzy neural network layer and the recursive neural network layer; and the improved particle swarm model is used to modify the parameters of the fuzzy neural network layer, the recursive neural network layer, the input layer, and the output layer.

[0044] Specifically, the first data is input into the first model, the membership function layer in the fuzzy neural network layer converts the first data into the membership value of the fuzzy set, and the fuzzy rule layer performs logical reasoning based on the output of the membership function layer to obtain the first prediction value; the recursive neural network layer captures the dynamic feature information of the first data that changes with time and generates a second prediction value, and the first prediction value and the second prediction value are weighted in the output layer to obtain the second data.

[0045] Furthermore, if Figure 2 The prediction results of the first and third models are shown in Figure 2. It can be seen that the prediction value of the first model is closer to the actual value of the power load change data. The third model consists of a fuzzy neural network layer, a recursive neural network layer, and a particle swarm model.

[0046] Furthermore, if Figure 3As shown in Figure 1, the specific structure of the first model is as follows: the first and fifth layers are the input and output layers, respectively. The first layer is primarily responsible for receiving and processing the first data; the fifth layer is the total output of all connections between the third and fourth layers of the first model, that is, a weighted output of the third and fourth layers. The fuzzy neural network layer consists of the second and third layers, with the second layer being the membership function layer and the third layer being the fuzzy rule layer. The fourth layer is the recursive neural network layer.

[0047] Furthermore, the membership function layer in the fuzzy neural network layer of the first model uses the Gaussian membership function as the activation function, and the membership function layer also contains the input membership function of the model. The output of each node in the membership function layer is expressed as:

[0048]

[0049] Where: c ij and σ ij They are expressed as the central parameter and width coefficient of the membership function respectively; A ij is a fuzzy subset.

[0050] Furthermore, each node in the fuzzy rule layer of the fuzzy neural network layer represents a fuzzy rule, R j is the rule of the jth node; the number of nodes is R1 to The total number of rules, the output of each node in this layer is expressed as follows:

[0051]

[0052] Among them, ∏ is the minimum operation of AND or; N r is the number of nodes in the current layer.

[0053] The fuzzy rules are:

[0054]

[0055] Among them, x i is the i-th input variable, where i=1:N in ; A ij is a fuzzy subset; j is the node output; w j is the connection weight; ψ j is the wavelet function.

[0056] Furthermore, if Figure 3 As shown, the first layer of the first model is the input layer, and its neurons transmit to the fuzzy neural network layer.

[0057] Furthermore, the recursive neural network layer is constructed by using the wavelet activation function in the neurons of the single hidden layer, and the Gaussian wavelet function is selected as the activation function. Calculated by the following formula:

[0058]

[0059] Considering discrete time k:

[0060]

[0061] Among them, t ij is the translation coefficient of the wavelet, d ij is the expansion coefficient of the wavelet; θ ij is the gain coefficient from the feedback loop; the subscript ij represents the i-th input and its corresponding j-th rule; Input a memory term into the recursive neural network layer to record data for a preset time period, wherein the preset time period is a historical time period at the current moment.

[0062] Furthermore, in the product layer located in the recurrent neural network layer, the processing result of each wavelet function product is as follows:

[0063]

[0064] Furthermore, the corresponding output of the recurrent neural network layer is:

[0065] υ j (k) = w j ·ψ j ,j=1:N r ;

[0066] Among them, w j is the corresponding connection weight value between the product layer of the fourth layer and the output layer of the fifth layer.

[0067] Furthermore, the product of the output nodes between the fourth layer and the third layer is calculated as follows:

[0068]

[0069] in,

[0070] Furthermore, the overall output of the output layer of the first model is the total contribution from each rule, and the total output is expressed as follows:

[0071]

[0072] Furthermore, the first model also incorporates an improved particle swarm optimization algorithm. The improved particle swarm optimization algorithm (MPSO) achieves efficient coordination between global search and local optimization through a two-layer nested structure, and combines it with the gradient descent algorithm (GDA) to significantly improve model initialization efficiency and parameter optimization accuracy. This allows the first model to effectively overcome the randomness flaws of a single algorithm while alleviating the problem of particle distribution convergence, achieving the optimization goal with a small population and a low number of iterations, while also balancing computational efficiency and solution quality.

[0073] Furthermore, the step of updating the parameters of the fuzzy neural network layer and the recursive neural network layer using the improved particle swarm model includes steps A1-A3:

[0074] Step A1: Initialization. Set the particle swarm size (psize) and the final number of optimization iterations (Maxgen); randomly initialize the velocity and position of each particle in the swarm, and set the initial number of iterations to 1.

[0075] Step A2: Calculate the fitness value of each particle in the group and pass fitness i To express, where i = 1: psize, and find the optimal previous position of the i-th particle Then select the best previous position gbest among all particles m , where the optimal previous position is the position where the fitness value meets the preset requirements among the positions experienced by all particles from the initial iteration to the current moment.

[0076] Among them, the fitness value can be expressed by the following formula:

[0077]

[0078] Among them, K is the number of parameters in the negative gradient direction of the gradient function; for y l is the output of the second model; The expectation of the second model; Θ is the parameter of the fuzzy neural network layer and the recursive neural network layer in the second model, and its value is Θ=(c ij ,σ ij ,w j ,t ij ,d ij ,θ ij ); The second model is a model that has not been parameter trained and has the same structure as the first model.

[0079] Step A3: Update the position and velocity of each particle. Randomly sort the third data and record the new third data order as Let l = 1. Calculate the fitness value of each particle and through Infitness i , to represent the fitness of the i-th particle. And search for the most preferred previous position of the i-th particle Select the most preferred previous position Ingbest among all particles m , and update the velocity and position of the particles through the update model. After the update is completed, let l = l + 1. If l < K + 1, return to recalculate the fitness value of each particle

[0080] Among them, the fitness value of the particle can be expressed by the following formula:

[0081]

[0082] Among them, the update model can be expressed by the following formula:

[0083]

[0084] Furthermore, the particle position and velocity are updated by the following formula:

[0085]

[0086] Among them, c1 and c2 are acceleration coefficients of position constants; r1 and r2 are two independent random numbers uniformly distributed between [0, 1]; ω is the inertia weight factor; is the next new position based on the previous position , is the next new velocity based on the previous velocity .

[0087] Furthermore, along the d-th dimension of the i-th particle at the m-th iteration, is the optimal position; along the d-th dimension at the m-th iteration, is the most preferred previous position of all particles in the group.

[0088] Step A4. Let m = m + 1. Calculate the fitness value fitness possessed by the i-th particle i , search for the most preferred previous position of the i-th Then select the most preferred previous position gbest among all particles m . If m > Maxgen, pause and select this gbest m , otherwise return to step A3.

[0089] Furthermore, in order to improve the convergence performance of the particle swarm algorithm, a linearly decreasing inertia weight is adopted in the velocity update formula, and its expression is:

[0090] ω(k)=ω start -(ω start -ω end )*m / Maxgen;

[0091] Among them, ω start is the initial inertia weight; ω end is the inertia weight when the iteration reaches the maximum number

[0092] Furthermore, the inertia weight is taken as ω start =0.9,ω end =0.4. During the iteration process, the inertia weight decreases linearly from 0.9 to 0.4 as the iteration proceeds. The larger ω in the early stage of the iteration enables the algorithm to maintain good global exploration performance, while the smaller ω in the later stages of the iteration process is beneficial to the algorithm for more accurate local development, taking into account the particle's global exploration and local optimization capabilities, and improving the convergence performance and accuracy of the evolution process.

[0093] For example, Figure 4 As shown in the figure, it can be seen that the improved particle swarm algorithm begins to converge after 40 iterations, while the particle swarm algorithm begins to converge after 60 iterations.

[0094] Optionally, at least one method for determining the first model includes steps B1-B4:

[0095] Step B1, determining third data; the third data is data obtained by feature screening of the original data.

[0096] Specifically, the original data is preprocessed, and the feature importance of the preprocessed data is evaluated according to the CatBoost algorithm, and feature screening is performed based on the scoring results to obtain the third data.

[0097] Among them, preprocessing includes missing value incompleteness and outlier processing.

[0098] Among them, the CatBoost algorithm integrates decision trees, gradient boosting and category feature processing calculation methods, which can automatically process different types of data and perform feature importance scoring.

[0099] Step B2: input the third data into the second model to perform power load forecasting to obtain fourth data; the second model is a model that has not been parameter trained and has the same structure as the first model.

[0100] Among them, the second model consists of a fuzzy neural network layer, a recursive neural network layer and an improved particle swarm model, but the parameters are randomly generated and have not been trained and corrected.

[0101] Specifically, the third data is input into the second model, the membership function layer in the fuzzy neural network layer converts the first data into the membership value of the fuzzy set, and the fuzzy rule layer performs logical reasoning based on the output of the membership function layer to obtain the third prediction value; the recursive neural network layer captures the dynamic feature information of the first data that changes with time and generates a fourth prediction value, and the third prediction value and the fourth prediction value are weighted in the output layer to obtain the fourth data.

[0102] Step B3: perform an error analysis based on the fourth data and the third data, and determine a first correction parameter based on the error analysis result; the first correction parameter is used to correct the first parameter combination; the first parameter combination includes all parameters in the second model; the error analysis is used to evaluate the ability of the parameters in the second model to capture the data change patterns in the third data.

[0103] The error analysis is used to compare the fourth data with the third data, and determine whether the second model meets the requirements based on the comparison result.

[0104] Specifically, an error parameter is determined based on the fourth data and the third data, and the error parameter is compared with a first preset error parameter. If the error parameter is greater than the first preset error parameter, it is considered that the second model parameters do not meet the requirements and need to be corrected, and a first correction parameter is generated based on the error parameter. If the error parameter is less than the first preset error parameter, it is considered that the second model parameters meet the requirements and no parameter correction is required.

[0105] The error parameters can be determined by using the root mean square error and the mean absolute error.

[0106] Furthermore, generating the first correction parameter according to the error parameter may be: generating the first correction parameter according to the error parameter and the product of the current parameter of the second model and the learning rate.

[0107] Step B4: correct the second model according to the first correction parameter, and predict the third data according to the corrected second model until the obtained first error parameter meets the first preset error parameter, and then use the second model that meets the first preset error parameter as the first model.

[0108] Specifically, the second model is corrected according to the first correction parameter, and the third data is input into the corrected second model for training to obtain new prediction data. A new error parameter is determined based on the new prediction data and the third data, and it is determined whether the new error parameter meets the first preset error parameter. If so, the second model is used as the first model. If not, a new first correction parameter is determined, and the second model is corrected according to the new first correction parameter. Prediction is performed on the third data until the obtained first error parameter meets the first preset error parameter. The second model that meets the first preset error parameter is then used as the first model.

[0109] Optionally, after the second model is corrected according to the first correction parameter, steps C1 to C4 are included:

[0110] Step C1: Initialize the parameters of the recursive neural network layer to obtain a second parameter group.

[0111] Specifically, the parameters of the recursive neural network layer in the modified second model are initialized to obtain a second parameter group.

[0112] Step C2: input the third data into the second model to perform power load forecasting using the second parameter group to obtain fifth data.

[0113] Specifically, the third data is input into the second model after the parameters of the recursive neural network layer are initialized, the membership function layer in the fuzzy neural network layer converts the third data into the membership value of the fuzzy set, and the fuzzy rule layer performs logical reasoning based on the output of the membership function layer to obtain the fifth prediction value; the recursive neural network layer captures the dynamic feature information of the third data that changes with time and generates a sixth prediction value, and the fifth prediction value and the sixth prediction value are weighted in the output layer to obtain the fifth data.

[0114] Step C3: performing error analysis based on the fifth data and the third data, and generating a second correction parameter based on the analysis result.

[0115] Specifically, the fifth data and the third data are input into the loss function to perform an error calculation to obtain a second error parameter. If the second error parameter does not meet the second preset parameter error, the gradient value of each parameter in the second parameter is determined based on the fifth data and the third data, and a second correction parameter is generated based on the obtained gradient value of each parameter, the weight learning rate, and the current parameter value. If the preset requirements are met, the second model is used as the first model.

[0116] The second preset error parameter may be the same as or different from the first preset error parameter.

[0117] Step C4: correct the second model according to the second correction parameter.

[0118] Specifically, the corresponding parameters in the second model are updated and corrected according to the second correction parameter, and the third data is predicted according to the corrected second model until the obtained second error parameter meets the second preset error parameter, then the prediction is stopped and the second model is used as the first model.

[0119] Furthermore, if Figure 5 As shown in Figure 3, after 300 iterations, the loss function of the second model tends to be stable.

[0120] Optionally, performing error analysis based on the fifth data and the third data, and generating a second correction parameter based on the analysis result, includes steps D1-D3:

[0121] Step D1: Determine a second error parameter through a loss function based on the fifth data and the third data; the second error parameter is used to characterize the prediction accuracy of the second model.

[0122] Specifically, the fifth data and the third data are input into the loss function to perform error calculation to obtain a second error parameter.

[0123] Among them, the loss function is expressed as:

[0124]

[0125] Among them, y d (k) is the expectation of the second model at time k; y(k) is the power load at time k in the fifth data.

[0126] Step D2: Determine the gradient value of each parameter in the second parameter group according to the second error parameter.

[0127] Specifically, if the error parameter does not satisfy the second preset parameter error, the gradient value of each parameter in the second parameter is determined according to the fifth data and the third data.

[0128] Step D3: Determine a second correction parameter according to the gradient value of each parameter.

[0129] Specifically, the second correction parameter is generated according to the obtained gradient value of each parameter, the weight learning rate, and the current parameter value.

[0130] Furthermore, the second correction parameter can be updated by the following formula:

[0131]

[0132] Among them, w j (k) is the connection weight between the product layer of the fourth layer and the output layer of the fifth layer at time k; t ij (k) is the wavelet translation coefficient at time k; d ij (k) is the wavelet expansion coefficient at time k; θ ij (k) is the gain coefficient of the self-feedback loop at time k; the subscript ij is the jth rule corresponding to the i-th input; γ = [γ w ,γ t ,γ d ,γ θ ] is the diagonal matrix of weight learning rate, that is, 0<γ<1.

[0133] Furthermore, according to the chain rule definition of calculus, the partial derivatives are as follows:

[0134]

[0135] Among them, y d (k) is the expectation of the second model at time k; y(k) is the power load at time k in the fifth data; υ j (k) is the output of the fourth layer; ψ j (z) is the result of the product of each wavelet function in the fourth product layer; μ j (x) is the product of the output nodes between the fourth layer and the third layer; N r is the number of fuzzy rules; d ij is the wavelet expansion coefficient at time k, and the Gaussian wavelet function is used as the wavelet mother function in the above formula.

[0136] Optionally, determining the third data includes steps E1-E5:

[0137] Step E1: Determine sixth data, which is the power load change data of the target area in the third time period; the third time period is before the first time period.

[0138] Specifically, the power load change data of the target area in the third time period is obtained from the SCADA system.

[0139] Among them, SCADA (Supervisory Control and Data Acquisition) is a computer system used to monitor and control the power load generation process of the power grid in real time.

[0140] Step E2: Match the sixth data with the meteorological change data and the electricity consumption characteristic data in the third time period according to the timestamp to obtain the seventh data.

[0141] Specifically, the date data and time data in the sixth data are obtained, and the date data and time data are merged to obtain a timestamp. According to the obtained timestamp, the data at the corresponding time is obtained from the meteorological change data and the power consumption characteristic data, and the data is merged with the power load change data to obtain the seventh data.

[0142] Furthermore, before data fusion, the acquired power load change data, meteorological change data and electricity consumption characteristic data need to be cleaned.

[0143] Furthermore, data cleaning was performed using PySpark. Specifically, the following steps were performed: Outliers, missing values, and duplicate data were processed. Power load values ​​outside the acceptable range (e.g., negative values ​​or exceeding installed capacity) were considered abnormal and replaced with the mean of the preceding and following values. Missing values ​​in the meteorological data were marked as NaN and retained for subsequent interpolation.

[0144] Among them, the abnormal value is the mutation point of the load value; the missing value is the missing meteorological change data; and the repeated data is the repeatedly sampled power load value.

[0145] Among them, PySpark is an integrated model framework for processing and analyzing data.

[0146] Step E3: performing missing value filling, data quantization and normalization processing on the seventh data to obtain the eighth data; data quantization is to convert the data described in text into numerical values; and normalization processing is used to eliminate differences between different data types.

[0147] The data quantization is to encode the data amount of the text description in the seventh data to convert the text data into a numerical value.

[0148] Furthermore, One-Hot coding is used to encode the power load change data and power consumption characteristic data.

[0149] Among them, One-Hot encoding is a technique for converting categorical variables into binary vectors, which is used to map the category labels of the text category data in the seventh data to a binary vector with a vector length equal to the total number of categories; each category corresponds to a position (index) in the vector, the position value is 1, and the remaining positions are 0. For example, taking the week as an example, the encoding of Monday is [1, 0, 0, 0, 0, 0]; the encoding of Tuesday is [0, 1, 0, 0, 0, 0]; the encoding of Wednesday is [0, 0, 1, 0, 0, 0]; the encoding of Thursday is [0, 0, 0, 1, 0, 0, 0]; the encoding of Friday is [0, 0, 0, 0, 1, 0, 0]; the encoding of Saturday is [0, 0, 0, 0, 0, 1, 0]; and the encoding of Seventh Day is [0, 0, 0, 0, 0, 0, 1].

[0150] Specifically, the Lagrange interpolation method is used to complete missing values, quantize data, and normalize the seventh data to obtain the eighth data.

[0151] Furthermore, when performing missing value filling, each feature field needs to be interpolated independently to avoid multivariate coupling errors.

[0152] Missing values ​​are filled using Lagrange interpolation, which uses known meteorological data to construct a polynomial and then estimates the missing values ​​based on the polynomial's changing trend. For example, if the power load value at a certain moment is missing, the known power load values ​​at several moments before and after are used to interpolate the value.

[0153] Furthermore, the process of missing value completion is to first establish a basis function to convert the vertical coordinate y of the first n points of the missing value i With the basis function f i(x) and then multiplied and accumulated in sequence to obtain the Lagrange polynomial function Then input the missing value coordinates into the Lagrange polynomial function for calculation.

[0154] Among them, the basis function can be expressed as:

[0155]

[0156] Among them, x i is a point among the n points before the missing value, x j For any other point. The basis function is at x i The value is 1 at the point where the digit is located, and 0 at other points.

[0157] Normalization involves standardizing (Z-Score) continuous features in the seventh data, such as temperature and power load values, to a uniform numerical range to eliminate the impact of dimensional differences on the model. Dimensional differences refer to significant differences in the units or numerical ranges of different features in the seventh data.

[0158] Furthermore, data normalization can be performed using the following formula:

[0159]

[0160] Among them, x' is the eighth data, x is the seventh data, is the mean of the seventh data, and σ is the standard deviation of the seventh data.

[0161] Step E4: Score the seventh data based on the feature importance according to the relationship between the feature and the load to obtain a scoring result.

[0162] Specifically, the seventh data set is shuffled to obtain the ninth data set used for the current tree construction. During CatBoost training, the categorical features of each data type in the ninth data set are dynamically converted to target encoding using the built-in encoding mechanism to obtain the tenth data set. During the symmetric tree construction process, the split gain value of each feature in the tenth data set is calculated and accumulated as the basis for feature importance. The feature importance score is determined based on the total gain contribution or usage frequency of each feature in all global splits using the built-in statistical logic of the get_feature_importance() function.

[0163] Among them, get_feature_importance() is used to evaluate the importance of each feature in the seventh data.

[0164] Step E5: Perform feature selection on the eighth data according to the scoring result to obtain third data.

[0165] Specifically, the eighth data is sorted according to the scoring result, and a preset number W of features are selected in sequence to obtain the third data.

[0166] Furthermore, if Figure 6 The following table shows the scoring results of the features in the third data. For example, the score at time T exceeds 5000; the scores for date and temperature at time T are around 3500. The load score at time T the previous day is 4000; and the scores for the remaining features are below 1000.

[0167] The preset number W is determined using the elbow rule. The elbow rule is a method for determining the optimal number of clusters in cluster analysis. The elbow rule calculates the clustering metric for different W values, observes the curve of the metric as it changes with W, and finds the inflection point in the curve that resembles an elbow. The W value corresponding to this inflection point is the optimal number of clusters.

[0168] Optionally, the seventh data is scored according to the relationship between the feature and the load to obtain a scoring result, including steps F1-F3:

[0169] Step F1: Arrange the seventh data in random order to obtain ninth data for current tree construction.

[0170] The shuffling is a process of randomly rearranging the seventh data sequence.

[0171] Specifically, the seventh data is rearranged in a random order to obtain ninth data used for constructing the current tree.

[0172] For example, the seventh data T={(x m ,y m )} m＝1Ln , its characteristics Perform random processing to obtain the ninth data for current tree construction

[0173]

[0174] Step F2: Replace the different categorical features in the ninth data with numerical features of the corresponding types to obtain the tenth data.

[0175] Specifically, Catboost automatically replaces each different categorical feature in the ninth data with a numerical feature of the corresponding type.

[0176] Furthermore, the numerical features are expressed as:

[0177]

[0178] Where p is the added prior term, a is the weight coefficient greater than 0, is the σ of the mth sample in the ninth data j , σ p Features, is the σth j The tag value where the data type is located.

[0179] Step F3: During the process of constructing the symmetric tree, the first splitting gain value of each feature in the tenth data is calculated in real time, and the feature importance score result is determined according to the first splitting gain value.

[0180] The first splitting gain value is a splitting gain value obtained by calculating the splitting gain value of each feature in the tenth data in real time during the process of constructing the symmetric tree, and accumulating all the splitting gain values.

[0181] Specifically, during the construction of the symmetric tree, the split gain value of each feature in the tenth data is calculated in real time, and all split gain values ​​are accumulated as the first split gain value. Based on the total gain contribution or usage frequency of each feature in all global splits, the feature importance score is determined by the built-in statistical logic of the get_feature_importance() function.

[0182] The technical solution of this embodiment determines the first data; performs power load forecasting through the first model based on the first data to obtain the second data. The fuzzy neural network layer in the first model can capture the nonlinear characteristics in the first data; the recursive neural network layer can capture the time series characteristics in the first data, and can enable the first model to capture different change characteristics in the first data. The combination of different change characteristics can improve the accuracy of the model prediction, so that the obtained second data is more in line with the actual power load changes. This method predicts the second data based on the first data through the first model, which not only takes into account the time series change characteristics in the first data, but also takes into account the nonlinear change characteristics in the first data. It can effectively overcome the adaptability limitations of traditional neural networks in complex time series fluctuation scenarios, improve the accuracy of the prediction, and at the same time, the recursive neural network layer enables the model to have better time-frequency localization characteristics, which can ensure that the first model reduces the impact of interference when extracting features from the first data, thereby improving the anti-interference ability of power load forecasting.

[0183] Figure 7 This is a schematic diagram of the structure of a power load prediction device provided by an embodiment of the present invention. This embodiment is applicable to the case of predicting power load. The power load prediction device can be implemented in the form of hardware and / or software. The power load prediction device can be configured in any electronic device with network communication function. Figure 7 As shown, the device includes: a first data determination module 210 and a second data determination module 220, wherein:

[0184] First data determination module 210: used to determine first data; the first data is determined based on meteorological change data, power load change data, and power consumption characteristic data of the target area during the first time period; the power consumption characteristic data is used to characterize the power consumption of equipment in the target area and the changes in power load resources;

[0185] The second data determination module 220 is used to predict the power load according to the first data through the first model to obtain the second data; the first model is configured with a fuzzy neural network layer and a recursive neural network layer; the fuzzy neural network layer is used to obtain nonlinear feature information in the first data; the recursive neural network layer is used to obtain dynamic feature information of the first data that changes with time; the fuzzy neural network layer and the recursive neural network layer in the first model predict the first data respectively, and the obtained feature information is weighted in the output layer; the second data is used to characterize the power load changes in the target area within the first time period;.

[0186] Optionally, the second data determination module 220 includes:

[0187] A third data determining unit is used to determine third data; the third data is data obtained by feature screening of the original data;

[0188] a fourth data determining unit, configured to input the third data into a second model for power load forecasting to obtain fourth data; the second model being a model that has not undergone parameter training and has the same structure as the first model;

[0189] a first correction parameter determination unit, configured to perform an error analysis based on the fourth data and the third data, and determine a first correction parameter based on the error analysis result; the first correction parameter is used to correct the first parameter combination; the first parameter combination includes all parameters in the second model; the error analysis is used to evaluate the ability of the parameters in the second model to capture the data variation pattern in the third data;

[0190] The correction unit is used to correct the second model according to the first correction parameter and predict the third data according to the corrected second model until the obtained first error parameter meets the first preset error parameter, and the second model that meets the first preset error parameter is used as the first model.

[0191] Optionally, the second data determination module 220 includes:

[0192] A second parameter group determining unit is configured to initialize the parameters of the recursive neural network layer to obtain a second parameter group;

[0193] A fifth data determining unit is configured to input the third data into the second model and perform power load forecasting using the second parameter group to obtain fifth data;

[0194] A second correction parameter determination unit is configured to perform error analysis based on the fifth data and the third data, and generate a second correction parameter based on the analysis result;

[0195] The correction unit is used to correct the second model according to the second correction parameter.

[0196] Optionally, the second correction parameter determination unit includes:

[0197] An error parameter determination subunit is configured to determine a second error parameter based on the fifth data and the third data through a loss function; the second error parameter is used to characterize the prediction accuracy of the second model;

[0198] A parameter gradient value determination subunit: configured to determine a gradient value of each parameter in the second parameter group according to the second error parameter;

[0199] The second correction parameter determination subunit is configured to determine the second correction parameter according to the gradient value of each parameter.

[0200] Optionally, the third data determination unit includes:

[0201] A sixth data determination subunit is configured to determine sixth data, where the sixth data is the power load change data of the target area in a third time period; the third time period is before the first time period;

[0202] a seventh data determination subunit configured to match the sixth data with the meteorological change data and the electricity consumption characteristic data within the third time period according to the timestamp to obtain the seventh data;

[0203] The eighth data determination subunit is used to perform missing value filling, data quantification and normalization processing on the seventh data to obtain the eighth data; data quantization is to convert the data described in text into numerical values; normalization processing is used to eliminate the differences between different data types;

[0204] Scoring result determination subunit: used to score the seventh data according to the relationship between the feature and the load, and obtain the scoring result;

[0205] The third data determination subunit is used to perform feature selection on the eighth data according to the scoring result to obtain the third data.

[0206] Optionally, the scoring result determination subunit is specifically used to:

[0207] Arrange the seventh data in random order to obtain the ninth data used for current tree construction;

[0208] Replace the different categorical features in the ninth data with the numerical features of the corresponding types to obtain the tenth data;

[0209] In the process of constructing the symmetric tree, the first splitting gain value of each feature in the tenth data is calculated in real time, and the feature importance score result is determined according to the first splitting gain value.

[0210] The power load prediction device provided in the embodiment of the present invention can execute the power load prediction method provided in any embodiment of the present invention mentioned above, and has the corresponding functions and beneficial effects of executing the power load prediction method. For detailed process, please refer to the relevant operations of the power load prediction method in the above embodiment.

[0211] Figure 8 Schematic diagram of the structure of an electronic device for implementing the power load forecasting method of an embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are only examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0212] like Figure 8 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11. The memory stores a computer program that can be executed by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. Various programs and data required for the operation of the electronic device 10 can also be stored in the RAM 13. The processor 11, ROM 12, and RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0213] Multiple components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0214] The processor 11 may be any general-purpose and / or specialized processing component with processing and computing capabilities. Examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any other suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the power load forecasting method.

[0215] In some embodiments, the power load forecasting method may be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the power load forecasting method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to execute the power load forecasting method in any other suitable manner (e.g., via firmware).

[0216] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system comprising at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0217] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0218] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0219] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0220] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0221] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0222] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0223] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A method for predicting power load, characterized in that: include: determining first data; The first data is determined based on meteorological change data, power load change data, and power consumption characteristic data of the target area within a first time period; The power consumption characteristic data is used to characterize the power consumption of equipment in the target area and the changes in power load resources; Performing power load forecasting using a first model based on the first data to obtain second data; The first model is configured with a fuzzy neural network layer and a recursive neural network layer; The fuzzy neural network layer is used to obtain nonlinear feature information in the first data; The recursive neural network layer is used to obtain dynamic characteristic information of the first data changing over time; the fuzzy neural network layer and the recursive neural network layer in the first model respectively predict the first data, and the obtained characteristic information is weighted in the output layer; the second data is used to characterize the power load changes in the target area within the first time period.

2. The method according to claim 1, characterized in that At least one determination method of the first model includes: Determine third data; the third data is data obtained by feature screening the original data; Inputting the third data into a second model for power load forecasting to obtain fourth data; the second model is a model that has not been parameter trained and has the same structure as the first model; performing an error analysis based on the fourth data and the third data, and determining a first correction parameter based on the error analysis result; the first correction parameter is used to correct a first parameter combination; the first parameter combination includes all parameters in the second model; and the error analysis is used to evaluate the ability of the parameters in the second model to capture data variation patterns in the third data; The second model is corrected according to the first correction parameter, and the third data is predicted according to the corrected second model until the obtained first error parameter meets the first preset error parameter, and the second model that meets the first preset error parameter is used as the first model.

3. The method according to claim 2, characterized in that After the second model is corrected according to the first correction parameter, the method includes: Initializing the parameters of the recurrent neural network layer to obtain a second parameter set; Inputting the third data into the second model to perform power load forecasting using the second parameter group to obtain fifth data; performing an error analysis based on the fifth data and the third data, and generating a second correction parameter based on the analysis result; The second model is corrected according to the second correction parameter.

4. The method according to claim 3, characterized in that The performing error analysis based on the fifth data and the third data and generating a second correction parameter based on the analysis result includes: Determining a second error parameter using a loss function based on the fifth data and the third data; the second error parameter is used to characterize the prediction accuracy of the second model; determining a gradient value of each parameter in a second parameter group according to the second error parameter; A second correction parameter is determined according to the gradient value of each parameter.

5. The method according to claim 2, characterized in that The determining of the third data includes: Determining sixth data, where the sixth data is power load change data of the target area in a third time period; the third time period is before the first time period; Matching the sixth data with the meteorological change data and the electricity consumption characteristic data within the third time period according to the timestamp to obtain seventh data; Performing missing value filling, data quantification, and normalization processing on the seventh data to obtain eighth data; the data quantization is to convert the data described in text into numerical values; the normalization processing is used to eliminate differences between different data types; Score the seventh data according to the relationship between the feature and the load to obtain the score result; Feature selection is performed on the eighth data according to the scoring result to obtain third data.

6. The method according to claim 5, characterized in that The seventh data is scored according to the relationship between the feature and the load to obtain a scoring result, including: Arrange the seventh data in random order to obtain the ninth data used for current tree construction; Replacing different category features in the ninth data with numerical features of corresponding types to obtain tenth data; In the process of constructing the symmetric tree, the first splitting gain value of each feature in the tenth data is calculated in real time, and the feature importance score result is determined according to the first splitting gain value.

7. A power load forecasting device, characterized in that: include: A first data determining module, configured to determine first data; The first data is determined based on meteorological change data, power load change data, and power consumption characteristic data of the target area within a first time period; the power consumption characteristic data is used to characterize the power consumption of equipment in the target area and changes in power load resources; a second data determination module, configured to perform power load forecasting using a first model based on the first data to obtain second data; The first model is configured with a fuzzy neural network layer and a recursive neural network layer; the fuzzy neural network layer is used to obtain nonlinear feature information in the first data; The recursive neural network layer is used to obtain dynamic characteristic information of the first data changing over time; the fuzzy neural network layer and the recursive neural network layer in the first model respectively predict the first data, and the obtained characteristic information is weighted in the output layer; the second data is used to characterize the power load changes in the target area within the first time period.

8. The device according to claim 7, characterized in that The second data determination module includes: A third data determining unit, configured to determine third data; the third data being data obtained by feature screening the original data; a fourth data determining unit, configured to input the third data into a second model for power load forecasting to obtain fourth data; the second model is a model that has not undergone parameter training and has the same structure as the first model; a first correction parameter determination unit, configured to perform an error analysis based on the fourth data and the third data, and determine a first correction parameter based on the error analysis result; the first correction parameter is used to correct a first parameter combination; the first parameter combination includes all parameters in the second model; and the error analysis is used to evaluate the ability of the parameters in the second model to capture data variation patterns in the third data; The correction unit is used to correct the second model according to the first correction parameter and predict the third data according to the corrected second model until the obtained first error parameter meets the first preset error parameter, and the second model that meets the first preset error parameter is used as the first model.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so as to enable the at least one processor to perform the power load forecasting method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the power load forecasting method according to any one of claims 1 to 6 when executed.

Images