Power grid load forecasting method and system based on time series correlation analysis

Abstract

The present application relates to the field of data processing technology, and specifically to a method and system for predicting power grid load based on time series correlation analysis. The method comprises: collecting the total power of electrical equipment of all residents in a community at each moment, the temperature of the area where the community is located at each moment, and the power grid load of the power supply station to which the community belongs at each moment; obtaining each periodic item of the total power at each moment, and each periodic item of the temperature at each moment; setting weights for the objective function in the XGBoost algorithm, correcting the objective function and regularization parameters in the XGBoost algorithm, and predicting the power grid load at each moment using the XGBoost algorithm; determining the influence weights of each periodic item of the total power and the temperature at each moment, reconstructing the data of the total power and the temperature at each moment, and using the reconstructed data in combination with a time series prediction model to obtain a prediction result of the power grid load. This improves the prediction accuracy of the power grid load.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a power grid load prediction method and system based on time series correlation analysis. Background Art

[0002] Grid load forecasting uses historical power load data and other relevant information, along with mathematical models and algorithms, to predict grid power demand at a specific moment or over a period of time. Its purpose is to provide data support for power system scheduling, planning, and operations, ensuring the stability and security of power supply. Accurate load forecasting helps power stations rationally schedule power production and transmission, avoiding overinvestment or insufficient power supply. It also optimizes grid resource allocation and reduces operating costs.

[0003] However, residents' electricity consumption behavior is highly random and uncertain. The time and amount of electricity consumption of many households may be affected by factors such as living habits, weather changes, and holiday arrangements, which makes load forecasting complicated. In addition, periodic factors such as seasonal changes and holidays may also make load forecasting more difficult, resulting in low accuracy of power grid load forecasting. Summary of the Invention

[0004] In order to solve the above technical problems, the purpose of this application is to provide a power grid load forecasting method and system based on time series correlation analysis. The technical solutions adopted are as follows:

[0005] In a first aspect, an embodiment of the present application provides a power grid load forecasting method based on time series correlation analysis, the method comprising the following steps:

[0006] Collect the total power consumption of all residents' electrical equipment at all times, the temperature of the area where the community is located at all times, and the grid load of the power supply station to which the community belongs at all times;

[0007] Performing time series decomposition on the total power and the temperature at each moment to obtain each periodic item of the total power and each periodic item of the temperature at each moment;

[0008] All periodic items of the total power and the temperature at each moment are used as training data for the XGBoost algorithm. Based on the electricity consumption patterns of community residents at different time periods during the day, weights are set for the objective function in the XGBoost algorithm. The objective function and regularization parameters in the XGBoost algorithm are modified, and the grid load at each moment is predicted using the XGBoost algorithm.

[0009] Determine the influence weights of each periodic item of the total power and the temperature at each moment based on the difference in objective function values ​​before and after the generation of each decision tree during the XGBoost algorithm training process, and the importance change trend of the corresponding periodic item during the iteration process;

[0010] Based on the impact weight, data of the total power and the temperature at each moment is reconstructed, and the reconstructed total power and the temperature are combined with the grid load to obtain a grid load prediction result using a time series prediction model.

[0011] In one embodiment, setting a weight for the objective function in the XGBoost algorithm includes:

[0012] The weight of the time periods corresponding to the morning and evening peaks in a day is set to w1, the weight of the time period corresponding to the daytime on weekends is set to w2, and the weight of the remaining time periods is set to w3, where w1>w2>w3, and w1+w2+w3=1.

[0013] In one embodiment, the objective function in the modified XGBoost algorithm includes:

[0014] The expression of the objective function L in the XGBoost algorithm is: ; Where N is the number of samples, is the weight of the time period to which time t belongs, is the true value of the grid load at time t, is the predicted value of the grid load at time t.

[0015] In one embodiment, the expression of the regularization parameter is:

[0016] Where, is the regularization parameter after correction at time t, is the initial regularization parameter.

[0017] In one embodiment, determining the influence weights of the periodic items of the total power and the temperature at each moment includes:

[0018] The difference between the objective function values ​​before and after the generation of each decision tree during the XGBoost algorithm training process is calculated, the first weight and the second weight are preset, and the performance evaluation index of each decision tree is determined in combination with the importance change trend to obtain the influence weights of each periodic item of the total power and the temperature at each moment.

[0019] In one embodiment, the performance evaluation index is expressed as:

[0020] Where, is the performance evaluation index when the i-th decision tree is generated, To preset the first weight, is the difference between the objective function values ​​before and after the i-th decision tree is generated, To preset the second weight, sigmoid() is the sigmoid function, To preset the first parameter, is the generation sequence number of the decision tree, where .

[0021] In one embodiment, the expression for the influence weight of each periodic item of the total power and the temperature at each moment is:

[0022] Where, is the influence weight of the k-th period item, is the performance evaluation index when the i-th decision tree is generated, T is the total number of decision trees, is a binary function with a value of 0 or 1. When the feature selected by the splitting node of the i-th decision tree is the k-th period item, ,otherwise, .

[0023] In one embodiment, reconstructing data of the total power and the temperature at each moment includes:

[0024] For the total power at each moment, weighted sum all periodic terms of the total power by the influence weight of each periodic term, and add the sum to the trend term and residual term after time series decomposition to obtain the total power after data reconstruction;

[0025] Accordingly, the temperature after data reconstruction is obtained.

[0026] In one embodiment, obtaining the prediction result of the power grid load includes:

[0027] The total power and temperature reconstructed at each moment are used as inputs of the time series prediction model, and the grid load of the power supply station belonging to the cell at each moment is used as a label to train the time series prediction model. The trained time series prediction model is used to obtain the prediction results of the grid load.

[0028] In the second aspect, an embodiment of the present application also provides a power grid load forecasting system based on time series correlation analysis, including a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor implements the steps of any one of the above methods when executing the computer program.

[0029] This application has at least the following beneficial effects:

[0030] This application decomposes the total power and temperature of the electrical equipment of all residents in the community at each moment in time series, obtains the periodic terms of the total power at each moment, and the periodic terms of the temperature at each moment, thereby improving the analytical ability of load-driven influencing factors, clarifying the independent contribution of residents' electricity consumption patterns and climate impacts, and enhancing the interpretability of features; further, setting weights for the objective function in the XGBoost algorithm, correcting the objective function and regularization parameters in the XGBoost algorithm, and differentially weighting the objective function according to the electricity intensity in different time periods, thereby improving the adaptability of the model to residents' behavior, avoiding the problems of overfitting in low-peak periods and underfitting in peak periods caused by traditional equal weights, and strengthening the high-regularity period by reducing the weights for low-regularity period. The prediction dominance of the relevant time period is suppressed, the interference of noise is suppressed, and the regularization parameter and the objective function are modified in a coordinated manner, which reduces the overfitting risk of the model and enhances the ability to capture nonlinear relationships; determining the influence weights of each periodic item of the total power and the temperature at each moment is helpful to identify the key influencing factors and realize the real-time calibration of the feature importance; based on the influence weights, the total power and the temperature at each moment are reconstructed, the signal-to-noise ratio of the data reconstruction is optimized, the quality of the model input data is improved, and redundant feature interference is avoided; using the reconstructed total power and the temperature, combined with the power grid load, a time series prediction model is used to obtain the prediction result of the power grid load, thereby improving the accuracy of the power grid load prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0032] Figure 1 A flowchart of the steps of a power grid load forecasting method based on time series correlation analysis provided in one embodiment of the present application;

[0033] Figure 2 Reconstruct a flow chart for the data of total power and temperature of electrical equipment. DETAILED DESCRIPTION

[0034] In order to further illustrate the technical means and effects adopted by this application to achieve the predetermined invention objectives, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation methods, structures, features, and effects of the power grid load forecasting method and system based on time series correlation analysis proposed in this application. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable form.

[0035] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0036] The specific scheme of the power grid load forecasting method and system based on time series correlation analysis provided by this application is described in detail below with reference to the accompanying drawings.

[0037] See also Figure 1 , which shows a flowchart of a method for grid load forecasting based on time series correlation analysis provided by an embodiment of the present application, the method comprising the following steps:

[0038] S1 collects the total power consumption of all residents' electrical equipment at all times, the temperature of the area where the community is located at all times, and the grid load of the power supply station to which the community belongs at all times.

[0039] This embodiment installs a smart meter at the distribution box of the community to collect the total power consumption of the electrical equipment of all residents in the community at all times. Secondly, a temperature sensor is used to collect the temperature of the area where the community is located at all times, and a power load monitoring system is used to collect the grid load of the power supply station to which the community belongs at all times.

[0040] It should be noted that the total power, temperature and grid load of the electrical equipment are all collected synchronously, and the collection time interval is set to 1s. The implementer can set it according to the actual situation, and this embodiment does not impose any restrictions on this.

[0041] S2, performing time series decomposition on the total power and the temperature at each moment, and obtaining each periodic item of the total power and each periodic item of the temperature at each moment.

[0042] Residential electricity consumption often exhibits a bimodal pattern on a daily basis, with daily electricity consumption concentrated in the morning and evening peak periods. During these periods, household members and activities are typically larger than at other times, leading to a significant increase in electricity consumption. On a weekly basis, a "stay-at-home economy" effect often emerges, with significant peaks in the morning and evening on weekdays and lower daytime electricity loads. During weekends, increased daytime appliance usage can lead to daytime electricity loads approaching evening peak levels. On a seasonal basis, electricity consumption is prone to a "temperature-sensitive load" effect, with air conditioning loads dominating in the summer, extending the evening peak period. In winter, heating equipment usage increases, while in spring and autumn, the load on these appliances decreases.

[0043] Based on the above analysis, residential electricity consumption typically exhibits three cyclical patterns: daily, weekly, and seasonal. Therefore, this embodiment uses the total power and temperature of electrical equipment collected at each moment as the output of the MSTL (Multiple Seasonal-Trend Decomposition Using Loess) time series decomposition algorithm. This output includes the trend term, residual term, and cycle terms for the total power of electrical equipment at each moment, as well as the trend term, residual term, and cycle terms for the temperature at each moment. The MSTL decomposition algorithm is well-known, and the specific process is not detailed here.

[0044] S3, using all periodic items of the total power and the temperature at each moment as training data for the XGBoost algorithm, setting weights for the objective function in the XGBoost algorithm based on the electricity consumption patterns of community residents in different time periods of the day, correcting the objective function and regularization parameters in the XGBoost algorithm, and using the XGBoost algorithm to predict the grid load at each moment.

[0045] XGBoost is an ensemble learning algorithm based on the gradient boosting decision tree (GBDT). It optimizes prediction performance by iteratively constructing a decision tree and weightedly combining the results of weak learners. Its core idea is the efficient implementation of the gradient boosting framework, which combines regularization, parallel computing, distributed computing, and sparse-aware optimization technology, thereby greatly improving the computational efficiency and generalization ability of the algorithm. In this embodiment, the grid load and each periodic item are trained using XGBoost to obtain the characteristic importance of each periodic item to the grid load. Among them, when using the XGBoost algorithm for training, each periodic item of the total power of the electrical equipment at each moment and each periodic item of the temperature at each moment are respectively used as input, and the grid load at each moment is used as a label to predict the grid load at each moment.

[0046] It should be noted that when the periodic items of the total power of electrical equipment at each moment and the periodic items of the temperature at each moment are used as inputs to the XGBoost algorithm, this embodiment uses the maximum-minimum normalization method to normalize them, eliminating the interference of dimensional differences on the feature importance evaluation, and ensuring that the algorithm fairly evaluates the feature importance.

[0047] Furthermore, this embodiment sets weights for the objective function in the XGBoost algorithm based on the electricity consumption patterns of community residents in different time periods of the day. The peak electricity consumption period has a greater impact on the load forecast results, and the load fluctuations within a day mainly occur during the peak electricity consumption period. Therefore, by setting adaptive weights for time points in different time periods, the feature splitting contribution of the peak period is amplified, thereby improving the sensitivity of the prediction model to data changes during the peak period.

[0048] Based on the above analysis, this embodiment sets the weight of the time periods corresponding to the morning and evening peaks as w1, the weight of the time period corresponding to the weekend daytime as w2, and the weight of the remaining time periods as w3, where w1>w2>w3, and w1+w2+w3=1. In this embodiment, the morning peak corresponds to 7:00-9:00, the evening peak corresponds to 18:00-22:00, and the weekend daytime corresponds to 10:00-18:00. Implementers can adjust the time periods corresponding to the morning peak, evening peak, and weekend daytime according to seasonal changes. For example, the evening peak in summer can be extended from 22:00 to 24:00.

[0049] In this embodiment, w1=0.6, w2=0.3, and w3=0.1 are set. The implementer can adjust them according to the actual situation, and this embodiment does not impose any restrictions on this.

[0050] The load fluctuations in residential electricity consumption vary greatly during different periods, and the cost of prediction errors during critical periods is high. Therefore, higher loss weights are assigned to the morning and evening peaks, forcing the XGBoost algorithm to prioritize the prediction of critical periods. This allows the XGBoost algorithm to prioritize learning load changes during peak periods and seasonally sensitive periods. Based on this, this embodiment modifies the objective function and regularization parameters in the XGBoost algorithm based on the weights set for each time period, specifically:

[0051] The expression of the objective function L in the XGBoost algorithm is: ; Where N is the number of samples, is the weight of the time period to which time t belongs, is the true value of the grid load at time t, is the predicted value of the grid load at time t.

[0052] Regularization parameter in XGBoost algorithm Control the global regularization benchmark. This embodiment modifies the fixed regularization parameter in the original gain formula. Change to dynamic value , and time period weight Inversely proportional. The power grid load has significant time dependence, and the fluctuation patterns in different periods vary greatly. When the peak period fluctuates greatly, it is necessary to capture fine-grained changes. Reducing the regularization strength allows the model to select a finer-grained threshold when the node splits, capturing load mutations. When the trough period fluctuates less, it is necessary to suppress the model's overfitting to noise. Increasing the regularization strength inhibits the model's excessive splitting of the stable interval and prevents learning irrelevant noise. Therefore, the expression of the revised regularization parameter is:

[0053] Where, is the regularization parameter after correction at time t, is the initial regularization parameter.

[0054] S4. Determine the influence weights of each periodic item of the total power and the temperature at each moment based on the difference in objective function values ​​before and after the generation of each decision tree during the XGBoost algorithm training process, and the importance change trend of the corresponding periodic item during the iteration process.

[0055] During the training process of the XGBoost algorithm, each round of iteration of the XGboost algorithm generates a new decision tree, with the goal of reducing the prediction error of the previous round of decision trees. The construction of each decision tree is based on the prediction error of the previous round of decision trees, that is, the deviation between the true value and the predicted value is adjusted to improve the prediction accuracy of each subsequent round. Therefore, the present embodiment counts the features of the XGboost algorithm during each round of iteration as a node splitting during the training process, and counts the number of times each feature is selected when it is used as a splitting node. If a feature is used as a feature when a node splits, then the information gain of the left and right subtrees divided according to the feature is larger, the performance of the generated decision tree is more excellent, and the influence of the feature on the prediction accuracy is greater. It should be noted that, in the present embodiment, each cycle item represents a feature.

[0056] On the other hand, since the accuracy of newly generated decision trees gradually improves with the number of training rounds, the importance of the features selected when splitting nodes in these newly generated decision trees should be higher than the features selected when splitting nodes in the decision trees generated at the beginning of training. Furthermore, since the decomposed components of the total power, temperature, and grid load data collected under normal circumstances satisfy certain data distributions, such as the random distribution of periodic terms, trend terms, and residual terms, the probability of overfitting in the XGBoost algorithm increases with the number of training rounds. Therefore, the importance of the features selected when splitting nodes in the decision tree cannot be set solely by the generation order. As the generation order of the decision tree increases, the importance of the selected features should stabilize after reaching a maximum value as the number of iterations increases, rather than continuously increasing with each iteration.

[0057] Specifically, for any feature, in this embodiment, each periodic item input by the XGboost algorithm is regarded as a type of feature. Taking the kth feature as an example, that is, taking the kth periodic item as an example, for the kth feature, based on the number of times the kth feature is selected by the split node during the training process and the generation order of the decision tree where the split node is located, the influence weight of the kth feature on the prediction result is determined.

[0058] First, the performance of each decision tree is comprehensively evaluated based on the change of the objective function after each decision tree is generated and the decision tree generated in the previous round of training, as well as the order in which each decision tree is generated. The performance evaluation index of the i-th decision tree is expressed as , whose expression is:

[0059] Where, To preset the first weight, is the difference between the objective function values ​​before and after the i-th decision tree is generated, To preset the second weight, sigmoid() is the sigmoid function, To preset the first parameter, is the generation sequence number of the decision tree, where In this embodiment , , the implementer can set it according to the actual situation, and this embodiment does not limit it; The value range of is [0.1, 0.3]. .

[0060] Furthermore, the probability of the kth periodic item being selected as the node splitting feature in T decision trees is combined to determine the degree of influence of the kth periodic item on the prediction results when the collected temperature, total power of the electrical equipment, and the decomposed periodic items are used as features for training. Specifically, the influence weights of the total power of the electrical equipment collected at each moment and the periodic items of the temperature are determined, and the expression is:

[0061] Where, is the influence weight of the k-th period item, is the performance evaluation index when the i-th decision tree is generated, T is the total number of decision trees, is a binary function with a value of 0 or 1. When the feature selected by the splitting node of the i-th decision tree is the k-th period item, ,otherwise, .

[0062] S5, based on the impact weight, reconstructing the total power and the temperature at each moment, using the reconstructed total power and the temperature, combined with the grid load, and using a time series prediction model to obtain a grid load prediction result.

[0063] Assume that the total power Q of electrical equipment at each moment is decomposed into three periodic items: Q_day, Q_week, and Q_season. Similarly, assume that the temperature data T is decomposed into three periodic items: T_day, T_week, and T_season.

[0064] Based on the above analysis, the impact weight of each periodic item is calculated separately, and the impact weights of all periodic items are normalized using the maximum-minimum normalization method for subsequent data reconstruction. The specific data reconstruction process is as follows:

[0065]

[0066]

[0067] Where, is the total power of the electrical equipment after data reconstruction at time t, is the normalized value of the impact weight of the first period item of the total power decomposition of the electrical equipment at time t, is the first periodic item of the total power decomposition of the electrical equipment at time t, is the normalized value of the impact weight of the second period term of the total power decomposition of the electrical equipment at time t, is the second periodic term of the total power decomposition of the electrical equipment at time t, is the normalized value of the impact weight of the third period item of the total power decomposition of the electrical equipment at time t, is the third periodic item of the total power decomposition of the electrical equipment at time t, is the residual term of the total power decomposition of the electrical equipment at time t, is the trend item of the total power decomposition of electrical equipment at time t, is the temperature after data reconstruction at time t, is the normalized value of the influence weight of the first period term of temperature decomposition at time t, is the first periodic term of temperature decomposition at time t, is the normalized value of the influence weight of the second period term of temperature decomposition at time t, is the second periodic term of temperature decomposition at time t, is the normalized value of the influence weight of the third period term of temperature decomposition at time t, is the third periodic term of the temperature decomposition at time t, is the residual term of temperature decomposition at time t, is the trend item of temperature decomposition at time t. The data reconstruction flow chart of total power and temperature of electrical equipment is as follows: Figure 2 shown.

[0068] The purpose of data reconstruction is that in addition to periodic and trend characteristics, isolated noise points may appear during the temperature and power data collection process, or local fluctuations in the data may occur due to temporary behavior of a small number of community residents. Data reconstruction can reduce the interference of such isolated data points on the collected temperature and power data when predicting grid load, thereby improving the data quality of subsequent input into the load prediction model.

[0069] Finally, this embodiment uses the total power and temperature of electrical equipment reconstructed from data at each moment as input to an LSTM neural network model. The grid load of the cell's power supply station at each moment is used as a label to train the LSTM neural network model. This trained LSTM neural network model is then used to predict grid load at future moments. The LSTM neural network model is well-known technology, and the specific process is not detailed here. Implementers can choose other feasible time series prediction models, and this embodiment does not impose any restrictions on this.

[0070] Based on the same inventive concept as the above-mentioned method, an embodiment of the present application also provides a power grid load forecasting system based on time series correlation analysis, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-mentioned power grid load forecasting methods based on time series correlation analysis.

[0071] It should be noted that the order in which the embodiments of the present application are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0072] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

[0073] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A power grid load forecasting method based on time series correlation analysis, characterized in that: The method comprises the following steps: Collect the total power consumption of all residents' electrical equipment at all times, the temperature of the area where the community is located at all times, and the grid load of the power supply station to which the community belongs at all times; Performing time series decomposition on the total power and the temperature at each moment to obtain each periodic item of the total power and each periodic item of the temperature at each moment; All periodic items of the total power and the temperature at each moment are used as training data for the XGBoost algorithm. Based on the electricity consumption patterns of community residents at different time periods during the day, weights are set for the objective function in the XGBoost algorithm. The objective function and regularization parameters in the XGBoost algorithm are modified, and the grid load at each moment is predicted using the XGBoost algorithm. Determine the influence weights of each periodic item of the total power and the temperature at each moment based on the difference in objective function values ​​before and after the generation of each decision tree during the XGBoost algorithm training process, and the importance change trend of the corresponding periodic item during the iteration process; Based on the impact weight, data of the total power and the temperature at each moment is reconstructed, and the reconstructed total power and the temperature are combined with the grid load to obtain a grid load prediction result using a time series prediction model.

2. The power grid load forecasting method based on time series correlation analysis according to claim 1, characterized in that: Setting the weight of the objective function in the XGBoost algorithm includes: The weight of the time periods corresponding to the morning and evening peaks in a day is set to w1, the weight of the time period corresponding to the daytime on weekends is set to w2, and the weight of the remaining time periods is set to w3, where w1>w2>w3, and w1+w2+w3=1.

3. The power grid load forecasting method based on time series correlation analysis according to claim 1, characterized in that: The objective function in the modified XGBoost algorithm includes: The expression of the objective function L in the XGBoost algorithm is: ; Where N is the number of samples, is the weight of the time period to which time t belongs, is the true value of the grid load at time t, is the predicted value of the grid load at time t.

4. The power grid load forecasting method based on time series correlation analysis according to claim 3, characterized in that: The expression of the regularization parameter is: Where, is the regularization parameter after correction at time t, is the initial regularization parameter.

5. The power grid load forecasting method based on time series correlation analysis according to claim 1, characterized in that: Determining the influence weights of the periodic items of the total power and the temperature at each moment includes: The difference between the objective function values ​​before and after the generation of each decision tree during the XGBoost algorithm training process is calculated, the first weight and the second weight are preset, and the performance evaluation index of each decision tree is determined in combination with the importance change trend to obtain the influence weights of each periodic item of the total power and the temperature at each moment.

6. The method for power grid load forecasting based on time series correlation analysis according to claim 5, characterized in that: The expression of the performance evaluation index is: Where, is the performance evaluation index when the i-th decision tree is generated, To preset the first weight, is the difference between the objective function values ​​before and after the i-th decision tree is generated, To preset the second weight, sigmoid() is the sigmoid function, To preset the first parameter, is the generation sequence number of the decision tree, where .

7. The power grid load forecasting method based on time series correlation analysis according to claim 5, characterized in that: The expression for the influence weight of each periodic item of the total power and the temperature at each moment is: Where, is the influence weight of the k-th period item, is the performance evaluation index when the i-th decision tree is generated, T is the total number of decision trees, is a binary function with a value of 0 or 1. When the feature selected by the splitting node of the i-th decision tree is the k-th period item, ,otherwise, .

8. The method for power grid load forecasting based on time series correlation analysis according to claim 1, wherein: The data reconstruction of the total power and the temperature at each moment includes: For the total power at each moment, weighted sum all periodic terms of the total power by the influence weight of each periodic term, and add the sum to the trend term and residual term after time series decomposition to obtain the total power after data reconstruction; Accordingly, the temperature after data reconstruction is obtained.

9. The power grid load forecasting method based on time series correlation analysis according to claim 1, characterized in that: The obtaining of the prediction result of the power grid load includes: The total power and temperature reconstructed at each moment are used as inputs of the time series prediction model, and the grid load of the power supply station belonging to the cell at each moment is used as a label to train the time series prediction model. The trained time series prediction model is used to obtain the prediction results of the grid load.

10. A power grid load forecasting system based on time series correlation analysis, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

Images