Charging station electrical equipment conversion efficiency adaptive calculation method and system based on load prediction

Abstract

The invention discloses a self-adaptive calculation method and system for the conversion efficiency of electrical equipment of a charging station based on load prediction, and the method comprises the steps: collecting load fluctuation condition data, current and voltage values and power factor information from the electrical equipment of the charging station in real time, and constructing an initial data set, obtaining a basic characteristic value of load change based on the initial data set; using a long short-term memory network model to predict a load fluctuation trend within a preset time, and determining an accuracy range of a load prediction result in combination with historical data and current and voltage values collected in real time; aiming at a load prediction result, constructing a calculation model of power factor deviation, and calculating a deviation value of a current power factor and an ideal value by utilizing a predicted load fluctuation trend and real-time monitoring data to obtain a specific numerical value of the power factor deviation; and if the specific value of the power factor deviation exceeds a preset threshold range, parameter adjustment is carried out, a targeted reactive power compensation instruction is generated, and the adjusted reactive power increment is determined.

Description

Technical Field

[0001] The present invention relates to the technical field of load prediction, and in particular to a method and system for adaptively calculating conversion efficiency of electrical equipment in a charging station based on load prediction. Background Art

[0002] In the field of modern energy management, charging stations are crucial infrastructure for the widespread adoption of electric vehicles. The operating efficiency of their electrical equipment is directly related to the effectiveness of energy utilization and the stability of the power grid, making research in this area particularly critical. With the rapid growth in the number of electric vehicles, optimizing the energy conversion efficiency of charging station equipment has become a focus of industry attention. However, many current charging station electrical equipment often uses fixed parameter settings during operation, failing to dynamically adjust to actual load changes. This results in energy waste and inefficient equipment operation. Especially under large load fluctuations, the system is prone to excessive reactive power or poor power factor.

[0003] Existing methods have obvious limitations when dealing with load changes. Most systems lack the ability to predict loads, making it difficult to achieve real-time optimization and adjustment of equipment parameters. This limitation may cause charging stations to overcompensate at low loads and undercompensate at high loads, thereby affecting overall operating efficiency. The more critical challenge lies in how to dynamically adjust the power factor to improve conversion efficiency under load fluctuations. Load changes directly cause the power factor to deviate from the optimal range, and the deviation in the power factor will further aggravate the generation of reactive power, increase the burden on the power grid, and even affect the life of the equipment. This causal relationship makes the dynamic correction of the power factor a technical problem that needs to be solved urgently, and it also places higher demands on the system's intelligent control capabilities.

[0004] Therefore, how to dynamically adjust the power factor of electrical equipment in charging stations based on load forecasting and calculate the optimal power factor value under different load conditions has become a key issue in improving conversion efficiency and energy utilization. Summary of the Invention

[0005] The present invention provides a method for adaptively calculating the conversion efficiency of electrical equipment in a charging station based on load prediction, which mainly includes:

[0006] Deploy a sensor network to collect real-time load fluctuation data, current and voltage values, and power factor information from charging station electrical equipment. Build an initial data set and pre-process the collected data to obtain basic characteristic values of load changes.

[0007] Based on the basic characteristic values of load changes, the long short-term memory network model is used to predict the load fluctuation trend within a preset time. The accuracy range of the load prediction results is determined by combining historical data and real-time current and voltage values.

[0008] Based on the load forecast results, a calculation model for power factor deviation is constructed. Using the predicted load fluctuation trend and real-time monitoring data, the deviation between the current power factor and the ideal value is calculated to obtain the specific value of the power factor deviation.

[0009] If the specific value of the power factor deviation exceeds the preset threshold range, the parameters are adjusted to generate targeted reactive power compensation instructions and determine the adjusted reactive power increase.

[0010] Optionally, the preprocessing process specifically includes:

[0011] Perform missing value filling and outlier detection on the initial data set, wherein the missing value filling method is specifically:

[0012] Setting a dataset segmentation length and a value missing rate, segmenting the initial dataset based on the segmentation length, traversing all segmented initial datasets, and using the initial datasets with a missing rate less than a preset value missing rate as sample data to generate a sample dataset;

[0013] Building a filling model based on a deep learning model, and training the filling model using the sample dataset, MSE, and SIV losses to obtain an optimized filling model;

[0014] Filling missing value data in the initial data set using the optimized filling model;

[0015] The method for detecting outliers is specifically as follows:

[0016] CNTS is used to detect outliers in the data after missing value filling, and outliers are filled until no outliers appear.

[0017] Optionally, the process of using the long short-term memory network model to predict the load fluctuation trend within a preset time period specifically includes:

[0018] Calculating the mean and variance of the basic eigenvalues;

[0019] The mean and variance are input into the LSTM neural network model for calculation to obtain an initial prediction value of the preset dimension;

[0020] The initial prediction value of the preset dimension is optimized using a loss function and an activation function to obtain a load fluctuation trend prediction value.

[0021] Optionally, the parameter adjustment process is specifically as follows:

[0022] The initial prediction value of the preset dimension is optimized using a loss function and an activation function to obtain a prediction result, the difference between the prediction result and the actual value is calculated, and the difference is corrected using the ARIMA method to obtain a corrected load fluctuation trend prediction value.

[0023] The present invention also discloses a system for adaptively calculating conversion efficiency of electrical equipment in a charging station based on load prediction, the system comprising:

[0024] The data acquisition module is used to use the deployed sensor network to collect real-time load fluctuation data, current and voltage values, and power factor information from the electrical equipment in the charging station, construct an initial data set, and pre-process the collected data to obtain the basic characteristic values of load changes;

[0025] The load forecasting module is used to predict the load fluctuation trend within a preset time based on the basic characteristic values of load changes and apply the long short-term memory network model. It combines historical data and real-time current and voltage values to determine the accuracy range of the load forecast results.

[0026] The deviation calculation module is used to build a power factor deviation calculation model based on the load forecast results. It uses the predicted load fluctuation trend and real-time monitoring data to calculate the deviation between the current power factor and the ideal value, and obtain the specific value of the power factor deviation;

[0027] The deviation compensation module is used to adjust parameters if the specific value of the power factor deviation exceeds a preset threshold range, generate targeted reactive power compensation instructions, and determine the adjusted reactive power increase.

[0028] Optionally, the preprocessing process specifically includes:

[0029] Perform missing value filling and outlier detection on the initial data set, wherein the missing value filling method is specifically:

[0030] Setting a dataset segmentation length and a value missing rate, segmenting the initial dataset based on the segmentation length, traversing all segmented initial datasets, and using the initial datasets with a missing rate less than a preset value missing rate as sample data to generate a sample dataset;

[0031] Building a filling model based on a deep learning model, and training the filling model using the sample dataset, MSE, and SIV losses to obtain an optimized filling model;

[0032] Filling missing value data in the initial data set using the optimized filling model;

[0033] The method for detecting outliers is specifically as follows:

[0034] CNTS is used to detect outliers in the data after missing value filling, and outliers are filled until no outliers appear.

[0035] Optionally, the process of using the long short-term memory network model to predict the load fluctuation trend within a preset time period specifically includes:

[0036] Calculating the mean and variance of the basic eigenvalues;

[0037] The mean and variance are input into the LSTM neural network model for calculation to obtain an initial prediction value of the preset dimension;

[0038] The initial prediction value of the preset dimension is optimized using a loss function and an activation function to obtain a load fluctuation trend prediction value.

[0039] Optionally, the parameter adjustment process is specifically as follows:

[0040] The initial prediction value of the preset dimension is optimized using a loss function and an activation function to obtain a prediction result, the difference between the prediction result and the actual value is calculated, and the difference is corrected using the ARIMA method to obtain a corrected load fluctuation trend prediction value.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] This method uses a long-short-term memory network model combined with fundamental characteristic values of load variation to predict load fluctuation trends and determine the accuracy range of the prediction results, enabling more accurate estimation of the load conditions of charging station electrical equipment over different time periods. Compared to traditional prediction methods, this model better captures the time series characteristics of the load, providing more accurate load baseline data for conversion efficiency calculations, making conversion efficiency assessments more closely aligned with actual operating conditions.

[0043] This invention deploys a sensor network to collect key data such as load fluctuations, current and voltage values, and power factor from charging station electrical equipment in real time. This data set is then constructed and preprocessed to obtain basic characteristic values of load variation, enabling real-time monitoring of the charging station's operating status. Based on this real-time data, the impact of load variations on conversion efficiency can be promptly identified. Combined with a power factor deviation calculation model and load forecast results, the operating parameters of the electrical equipment can be dynamically adjusted to ensure consistent operation within a high conversion efficiency range, enhancing the charging station's adaptability to varying load conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in 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 paying any creative work.

[0045] Figure 1 This is a method step diagram of a method for adaptively calculating conversion efficiency of electrical equipment in a charging station based on load prediction according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] Example 1

[0048] An adaptive calculation method for conversion efficiency of electrical equipment in charging stations based on load prediction, such as Figure 1 As shown, the method includes:

[0049] By deploying a sensor network, we collect real-time load fluctuation data, current and voltage values, and power factor information from the electrical equipment in the charging station. We construct an initial data set and pre-process the collected data to obtain the basic characteristic values of load changes.

[0050] The preprocessing process specifically includes: filling missing values and detecting outliers on the initial data set, wherein the method of filling missing values is specifically: setting the data set segmentation length and the value missing rate, segmenting the initial data set based on the segmentation length, traversing all segmented initial data sets, and using the initial data sets with a missing rate less than the preset value missing rate as sample data to generate a sample data set; constructing a filling model based on a deep learning model, using the sample data set, MSE and SIV losses to train the filling model to obtain an optimized filling model; and using the optimized filling model to fill the missing value data in the initial data set.

[0051] The specific method of outlier detection is: use CNTS to detect outliers in the data after missing value filling, and fill the outliers until no outliers appear.

[0052] According to the basic characteristic values of load changes, the long short-term memory network model is applied to predict the load fluctuation trend within a preset time. The accuracy range of the load prediction results is determined by combining historical data and real-time collected current and voltage values.

[0053] The process of using the long short-term memory network model to predict the load fluctuation trend within a preset time specifically includes: calculating the mean and variance of the basic eigenvalues; inputting the mean and variance into the LSTM neural network model for calculation to obtain the initial prediction value of the preset dimension; using the loss function and activation function to optimize the initial prediction value of the preset dimension to obtain the load fluctuation trend prediction value.

[0054] The LSTM neural network model consists of two parts: forward propagation and backward propagation. The forward propagation content is as follows:

[0055] The input gate is:

[0056]

[0057] The forget gate is:

[0058]

[0059] The neurons are:

[0060]

[0061] Output gate:

[0062]

[0063] Neuron output:

[0064]

[0065] The LSTM in this embodiment uses BPTT to solve parameters, and the back propagation process is as follows:

[0066]

[0067] Neuron output:

[0068]

[0069] Output gate;

[0070]

[0071] Neuron status:

[0072]

[0073] Neurons:

[0074]

[0075] Output gate:

[0076]

[0077] Input Gate:

[0078]

[0079] The reconstructed sequence and its mean and variance are used as the input of the LSTM neural network, that is, the input dimension is 62. The output layer dimension is 1, and the loss function is L 2 Error function. The activation function is ReLU, and batch gradient descent is used for learning, ReLU(x) = max(0,x).

[0080] Based on the load forecast results, a calculation model for power factor deviation is constructed. The predicted load fluctuation trend and real-time monitoring data are used to calculate the deviation between the current power factor and the ideal value, and the specific value of the power factor deviation is obtained.

[0081] The load prediction module obtains prediction results, analyzes load fluctuation trends, and determines characteristic data of the fluctuation trend. Based on the characteristic data of the fluctuation trend and combined with monitoring data collected by the real-time monitoring system, the two types of data are fused to form a comprehensive data set. The comprehensive data set is input into a pre-established power factor calculation model to obtain the current power factor value. The current power factor value is compared with the preset ideal value to determine the difference between the two. If the difference exceeds the preset threshold range, the specific deviation value is calculated using the deviation calculation formula: Deviation value = current power factor value - ideal value, where the current power factor value represents the power factor calculated in real time and the ideal value represents the preset standard value. Based on the deviation value and combined with the characteristic data of the fluctuation trend, the support vector machine model is used to predict subsequent power factor fluctuations and obtain a predicted deviation trend. Based on the predicted deviation trend, the parameters of the calculation model are adjusted to determine the optimized power factor deviation value.

[0082] If the specific value of the power factor deviation exceeds the preset threshold range, the parameters are adjusted to generate targeted reactive power compensation instructions and determine the adjusted reactive power increase.

[0083] The parameter adjustment process is specifically as follows: using the loss function and the activation function to optimize the initial prediction value of the preset dimension to obtain the prediction result, calculating the difference between the prediction result and the actual value, and using the ARIMA method to perform residual correction on the difference to obtain the corrected load fluctuation trend prediction value.

[0084] The ARIMA model is expressed as:

[0085]

[0086] Φ(B)=1-φ11B-…-φ p Bp

[0087] Θ(B)=1-θ1B-…-θ q B q

[0088] AIC=-2ln(L)+2k

[0089] BIC=-2ln(L)+ln(n)×k

[0090] After inputting the data, the unit root test (Augmented Dickey-Fuller, ADF) is first performed to determine the stationarity of the sequence. If it is stationary, the next step of modeling is carried out. If it is not stationary, the difference operation is performed until it is stationary before modeling analysis. The ARIMA (p, d, q) model is constructed through the autocorrelation function (ACF) and partial autocorrelation function (PACF) of the sequence. The Durbin Watson test (DW) is used to test whether the residuals meet the normal distribution, and then the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) values are used to determine the optimal model.

[0091] in, is the difference, d is the number of differences, Φ(B) is the p-order autoregressive coefficient polynomial, Θ(B) is the q-order moving average coefficient polynomial, L is the maximum likelihood number, n is the number of data, and k is the number of model variables.

[0092] Example 2

[0093] A system for adaptively calculating conversion efficiency of electrical equipment in a charging station based on load prediction, the system comprising:

[0094] The data acquisition module uses a deployed sensor network to collect real-time load fluctuation data, current and voltage values, and power factor information from the charging station's electrical equipment. It then constructs an initial data set and preprocesses the collected data to obtain basic characteristic values of load variation.

[0095] The preprocessing process specifically includes: filling missing values and detecting outliers on the initial data set, wherein the method of filling missing values is specifically: setting the data set segmentation length and the value missing rate, segmenting the initial data set based on the segmentation length, traversing all segmented initial data sets, and using the initial data sets with a missing rate less than the preset value missing rate as sample data to generate a sample data set; constructing a filling model based on a deep learning model, using the sample data set, MSE and SIV losses to train the filling model to obtain an optimized filling model; and using the optimized filling model to fill the missing value data in the initial data set.

[0096] The specific method of outlier detection is: use CNTS to detect outliers in the data after missing value filling, and fill the outliers until no outliers appear.

[0097] The load forecasting module uses a long-short-term memory network model to predict load fluctuation trends within a preset timeframe based on the fundamental characteristic values of load variation. It then combines historical data with real-time current and voltage data to determine the accuracy of the load forecast results.

[0098] The process of using the long short-term memory network model to predict the load fluctuation trend within a preset time specifically includes: calculating the mean and variance of the basic eigenvalues; inputting the mean and variance into the LSTM neural network model for calculation to obtain the initial prediction value of the preset dimension; using the loss function and activation function to optimize the initial prediction value of the preset dimension to obtain the load fluctuation trend prediction value.

[0099] The LSTM neural network model consists of two parts: forward propagation and backward propagation. The forward propagation content is as follows:

[0100] The input gate is:

[0101]

[0102] The forget gate is:

[0103]

[0104] The neurons are:

[0105]

[0106] Output gate:

[0107]

[0108] Neuron output:

[0109]

[0110] The LSTM in this embodiment uses BPTT to solve parameters, and the back propagation process is as follows:

[0111]

[0112] Neuron output:

[0113]

[0114] Output gate;

[0115]

[0116] Neuron status:

[0117]

[0118] Neurons:

[0119]

[0120] Output gate:

[0121]

[0122] Input Gate:

[0123]

[0124] The reconstructed sequence and its mean and variance are used as the input of the LSTM neural network, that is, the input dimension is 62. The output layer dimension is 1, and the loss function is L 2 Error function. The activation function is ReLU, and batch gradient descent is used for learning, ReLU(x) = max(0,x).

[0125] The deviation calculation module is used to build a power factor deviation calculation model based on load forecast results. Using the predicted load fluctuation trend and real-time monitoring data, it calculates the deviation between the current power factor and the ideal value and obtains the specific value of the power factor deviation.

[0126] The load prediction module obtains prediction results, analyzes load fluctuation trends, and determines characteristic data of the fluctuation trend. Based on the characteristic data of the fluctuation trend and combined with monitoring data collected by the real-time monitoring system, the two types of data are fused to form a comprehensive data set. The comprehensive data set is input into a pre-established power factor calculation model to obtain the current power factor value. The current power factor value is compared with the preset ideal value to determine the difference between the two. If the difference exceeds the preset threshold range, the specific deviation value is calculated using the deviation calculation formula: Deviation value = current power factor value - ideal value, where the current power factor value represents the power factor calculated in real time and the ideal value represents the preset standard value. Based on the deviation value and combined with the characteristic data of the fluctuation trend, the support vector machine model is used to predict subsequent power factor fluctuations and obtain a predicted deviation trend. Based on the predicted deviation trend, the parameters of the calculation model are adjusted to determine the optimized power factor deviation value.

[0127] The deviation compensation module is used to adjust parameters if the specific value of the power factor deviation exceeds the preset threshold range, generate targeted reactive power compensation instructions, and determine the adjusted reactive power increase.

[0128] The parameter adjustment process is specifically as follows: using the loss function and the activation function to optimize the initial prediction value of the preset dimension to obtain the prediction result, calculating the difference between the prediction result and the actual value, and using the ARIMA method to perform residual correction on the difference to obtain the corrected load fluctuation trend prediction value.

[0129] The ARIMA model is expressed as:

[0130]

[0131] Φ(B)=1-φ11B-…-φ p B p

[0132] Θ(B)=1-θ1B-…-θ q B q

[0133] AIC=-2ln(L)+2k

[0134] BIC=-2ln(L)+ln(n)×k

[0135] After inputting the data, the unit root test (Augmented Dickey-Fuller, ADF) is first performed to determine the stationarity of the sequence. If it is stationary, the next step of modeling is carried out. If it is not stationary, the difference operation is performed until it is stationary before modeling analysis. The ARIMA (p, d, q) model is constructed through the autocorrelation function (ACF) and partial autocorrelation function (PACF) of the sequence. The Durbin Watson test (DW) is used to test whether the residuals meet the normal distribution, and then the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) values are used to determine the optimal model.

[0136] in, is the difference, d is the number of differences, Φ(B) is the p-order autoregressive coefficient polynomial, Θ(B) is the q-order moving average coefficient polynomial, L is the maximum likelihood number, n is the number of data, and k is the number of model variables.

[0137] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A method for adaptively calculating conversion efficiency of electrical equipment in charging stations based on load prediction, characterized in that: The method comprises: Deploy a sensor network to collect real-time load fluctuation data, current and voltage values, and power factor information from charging station electrical equipment. Build an initial data set and pre-process the collected data to obtain basic characteristic values of load changes. Based on the basic characteristic values of load changes, the long short-term memory network model is used to predict the load fluctuation trend within a preset time. The accuracy range of the load prediction results is determined by combining historical data and real-time current and voltage values. Based on the load forecast results, a calculation model for power factor deviation is constructed. Using the predicted load fluctuation trend and real-time monitoring data, the deviation between the current power factor and the ideal value is calculated to obtain the specific value of the power factor deviation. If the specific value of the power factor deviation exceeds the preset threshold range, the parameters are adjusted to generate targeted reactive power compensation instructions and determine the adjusted reactive power increase.

2. The method for adaptively calculating conversion efficiency of electrical equipment in charging stations based on load prediction according to claim 1, characterized in that: The pretreatment process specifically includes: Perform missing value filling and outlier detection on the initial data set, wherein the missing value filling method is specifically: Setting a dataset segmentation length and a value missing rate, segmenting the initial dataset based on the segmentation length, traversing all segmented initial datasets, and using the initial datasets with a missing rate less than a preset value missing rate as sample data to generate a sample dataset; Building a filling model based on a deep learning model, and training the filling model using the sample dataset, MSE, and SIV losses to obtain an optimized filling model; Filling missing value data in the initial data set using the optimized filling model; The method for detecting outliers is specifically as follows: CNTS is used to detect outliers in the data after missing value filling, and outliers are filled until no outliers appear.

3. The method for adaptively calculating conversion efficiency of electrical equipment in charging stations based on load prediction according to claim 1, characterized in that: The process of using the long short-term memory network model to predict the load fluctuation trend within a preset time period specifically includes: Calculating the mean and variance of the basic eigenvalues; The mean and variance are input into the LSTM neural network model for calculation to obtain an initial prediction value of the preset dimension; The initial prediction value of the preset dimension is optimized using a loss function and an activation function to obtain a load fluctuation trend prediction value.

4. The method for adaptively calculating conversion efficiency of electrical equipment in charging stations based on load prediction according to claim 3, characterized in that: The parameter adjustment process is specifically as follows: The initial prediction value of the preset dimension is optimized using a loss function and an activation function to obtain a prediction result, the difference between the prediction result and the actual value is calculated, and the difference is corrected using the ARIMA method to obtain a corrected load fluctuation trend prediction value.

5. A system for adaptively calculating conversion efficiency of electrical equipment in a charging station based on load prediction, the system being used to implement the adaptive calculation method according to any one of claims 1 to 4, characterized in that: The system includes: The data acquisition module is used to use the deployed sensor network to collect real-time load fluctuation data, current and voltage values, and power factor information from the electrical equipment in the charging station, construct an initial data set, and pre-process the collected data to obtain the basic characteristic values of load changes; The load forecasting module is used to predict the load fluctuation trend within a preset time based on the basic characteristic values of load changes and apply the long short-term memory network model. It combines historical data and real-time current and voltage values to determine the accuracy range of the load forecast results. The deviation calculation module is used to build a power factor deviation calculation model based on the load forecast results. It uses the predicted load fluctuation trend and real-time monitoring data to calculate the deviation between the current power factor and the ideal value, and obtain the specific value of the power factor deviation; The deviation compensation module is used to adjust parameters if the specific value of the power factor deviation exceeds a preset threshold range, generate targeted reactive power compensation instructions, and determine the adjusted reactive power increase.

6. The adaptive calculation system for conversion efficiency of electrical equipment in charging stations based on load prediction according to claim 5, characterized in that: The pretreatment process specifically includes: Perform missing value filling and outlier detection on the initial data set, wherein the missing value filling method is specifically: Setting a dataset segmentation length and a value missing rate, segmenting the initial dataset based on the segmentation length, traversing all segmented initial datasets, and using the initial datasets with a missing rate less than a preset value missing rate as sample data to generate a sample dataset; Building a filling model based on a deep learning model, and training the filling model using the sample dataset, MSE, and SIV losses to obtain an optimized filling model; Filling missing value data in the initial data set using the optimized filling model; The method for detecting outliers is specifically as follows: CNTS is used to detect outliers in the data after missing value filling, and outliers are filled until no outliers appear.

7. The adaptive calculation system for conversion efficiency of electrical equipment in charging stations based on load prediction according to claim 5, characterized in that: The process of using the long short-term memory network model to predict the load fluctuation trend within a preset time period specifically includes: Calculating the mean and variance of the basic eigenvalues; The mean and variance are input into the LSTM neural network model for calculation to obtain an initial prediction value of the preset dimension; The initial prediction value of the preset dimension is optimized using a loss function and an activation function to obtain a load fluctuation trend prediction value.

8. The self-adaptive calculation system for charging station electrical equipment conversion efficiency based on load prediction according to claim 7, characterized in that: The parameter adjustment process is specifically as follows: The initial prediction value of the preset dimension is optimized using a loss function and an activation function to obtain a prediction result, the difference between the prediction result and the actual value is calculated, and the difference is corrected using the ARIMA method to obtain a corrected load fluctuation trend prediction value.

Images