Power grid load prediction method and system based on artificial intelligence and subspace aggregation

Abstract

The invention discloses a power grid load prediction method and system based on artificial intelligence and subspace aggregation, and the method comprises the steps: firstly collecting original data related to power grid load fluctuation, carrying out the fuzzification of the original data, and forming a multi-dimensional fuzzy set of the original data; a fuzzy rule base is established according to the power grid load fluctuation and fuzzy rules between the influence factors of the power grid load fluctuation, and a fuzzy inference engine is designed through the fuzzy rule base; and dividing a multi-dimensional feature space of the multi-dimensional fuzzy set into a plurality of random subspaces, establishing a fuzzy logic submodel in each random subspace based on a fuzzy rule base and a fuzzy inference engine, and aggregating prediction results of all the random subspaces to generate a final power grid load prediction model. Through the fuzzy logic system and the random subspace division, the calculation time required by model training and prediction is reduced, and the strategy of random subspace division and multi-model integration is adopted, so that the calculation deviation caused by a single model is effectively reduced, and the stability and accuracy of the prediction model are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power grids, and in particular relates to a power grid load prediction method and system based on artificial intelligence and subspace aggregation. Background Art

[0002] The stable operation of the power system is crucial to the normal operation of the social economy. In modern smart grids, accurate prediction of load fluctuations is an important part of ensuring the stability of power supply. Traditional load fluctuation prediction methods rely on some single models or simple statistical methods. These methods have problems such as insufficient prediction accuracy and weak model generalization ability when dealing with complex nonlinear relationships or high-dimensional data, resulting in large fluctuations in prediction results in practical applications and unable to meet the needs of power grid operation for high-precision prediction. In addition, traditional models also lack effective methods in data feature space division and integration strategies, resulting in significant deficiencies in the stability and accuracy of prediction results when dealing with large-scale power grid load prediction tasks.

[0003] In recent years, with the rapid development of artificial intelligence and machine learning technologies, some advanced prediction models have been proposed, such as load prediction models based on deep learning. However, when faced with multi-dimensional data characteristics and complex load fluctuations, deep learning models still have the limitation of being unable to fully capture nonlinear characteristics. In addition, in terms of large-scale data processing, a large amount of computing resources are often required. Existing technologies make it difficult to optimize computing efficiency while ensuring high-precision model predictions.

[0004] Therefore, how to introduce more advanced algorithms in grid load fluctuation prediction, improve the stability and accuracy of the prediction model, and optimize the utilization efficiency of computing resources is still a technical problem that needs to be solved urgently. Summary of the invention

[0005] The purpose of the present invention is to provide a power grid load forecasting method and system based on artificial intelligence and subspace aggregation with high prediction efficiency and high prediction accuracy in order to solve the above problems existing in the prior art.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] In a first aspect, the present invention proposes a power grid load forecasting method based on artificial intelligence and subspace aggregation, comprising:

[0008] S1. Collecting raw data related to power grid load fluctuations and fuzzifying the raw data to form a multi-dimensional fuzzy set of the raw data;

[0009] S2. Based on the multidimensional fuzzy set of the original data, a fuzzy rule base is established according to the fuzzy rules between the power grid load fluctuation and its various influencing factors;

[0010] S3, designing a fuzzy reasoning engine using a fuzzy rule base, and establishing a fuzzy logic model based on the fuzzy rule base and the fuzzy reasoning engine;

[0011] S4, dividing the multidimensional feature space of the multidimensional fuzzy set into multiple random subspaces, and establishing a fuzzy logic submodel in each random subspace based on the fuzzy rule base and the fuzzy inference engine, and aggregating the prediction results of all random subspaces to generate the final power grid load prediction model;

[0012] S5. Input the real-time data related to the grid load fluctuation into the final grid load prediction model to perform real-time prediction of the grid load.

[0013] The S1 includes:

[0014] S11. Use the following formula to map various types of original data into fuzzy sets to obtain the fuzzy value of each original data:

[0015]

[0016]

[0017]

[0018] In the above formula, is the fuzzy value of the grid load power data, is the fuzzy value of weather data, is the fuzzy value of social and economic data, x is the data sample of various types of original data, μ P (x) is the membership function of the power grid load data, μ w (x) is the membership function of weather data, μ E (x) is the membership function of the socio-economic data, P(t) is the grid load power at time t, W(t) is the weather data at time t, and E(t) is the socio-economic data at time t;

[0019] S12. Combining the fuzzy values ​​of various types of original data to form a multidimensional fuzzy set:

[0020]

[0021] In the above formula, is a multidimensional fuzzy set.

[0022] The S2 includes:

[0023] S21. Based on the multidimensional fuzzy set of the original data, several fuzzy rules are defined to describe the nonlinear relationship between the power grid load fluctuation and its influencing factors. The form of each fuzzy rule is as follows:

[0024]

[0025] In the above formula, R i (t+1) is the i-th fuzzy rule R at time t+1 i The generated grid load prediction result, f is the fuzzy reasoning function, M is the total number of grid load power data, is the membership degree of the mth grid load power data, is the weight coefficient of the power grid load data, N is the total number of weather data, is the membership degree of the nth weather data, is the weight coefficient of weather data, L is the total number of social and economic data, is the membership degree of the lth socioeconomic data, is the weight coefficient of socioeconomic data;

[0026] S22, combining the fuzzy rules to form a fuzzy rule base:

[0027]

[0028] In the above formula, is the fuzzy rule base composed of all fuzzy rules, and K is the total number of rules in the fuzzy rule base.

[0029] The fuzzy logic model in S3 includes:

[0030]

[0031]

[0032]

[0033] In the above formula, y i (t+1) is the i-th fuzzy rule R at time t+1 i The intermediate prediction result obtained by inference, f i is the fuzzy reasoning function based on the fuzzy rule base, is a multidimensional fuzzy set, is the fuzzy output result at time t+1, K is the total number of rules in the fuzzy rule base, is the i-th fuzzy rule R i When entering data The applicability of the above equation is given in Figure 2. Y(t+1) is the load prediction value determined at time t+1, X is the multidimensional feature space of the multidimensional fuzzy set, The fuzzy output result at time t+1 is the input data The membership function at .

[0034] The S4 includes:

[0035] S41. Use the following formula to distribute the multidimensional feature space of the multidimensional fuzzy set to each random subspace:

[0036]

[0037] In the above formula, X n is the nth random subspace, q n is the number of features contained in the nth random subspace, X n，j is the jth input feature assigned to the nth random subspace, X is the multidimensional feature space of the multidimensional fuzzy set;

[0038] S42, in each random subspace, based on the fuzzy rule base and fuzzy reasoning engine, establish several fuzzy logic sub-models M n，m , m=1,2,...,M n , M n，m Represents the mth fuzzy logic submodel in the nth random subspace, and generates several independent prediction results in each random subspace by independently training each fuzzy logic submodel is the prediction result of the mth fuzzy logic sub-model in the nth random subspace at time t+1;

[0039] S43, integrating the prediction results of all fuzzy logic sub-models in the nth random subspace, and obtaining the prediction results of each random subspace using the following formula:

[0040]

[0041] In the above formula, is the prediction result of the nth random subspace at time t+1, M n are all fuzzy logic submodels in the nth random subspace, α n，m is the weight coefficient of the mth fuzzy logic sub-model;

[0042] S45. Aggregate the prediction results of all random subspaces to generate the final power grid load prediction model:

[0043]

[0044] In the above formula, is the predicted value of the grid load at time t+1, β n The contribution weight of each random subspace to the final prediction result.

[0045] In a second aspect, the present invention proposes a power grid load prediction system based on artificial intelligence and subspace aggregation, including a multidimensional fuzzy set forming module, a fuzzy rule base establishing module, a fuzzy logic model establishing module, a power grid load prediction model generating module, and a power grid load real-time prediction module;

[0046] The multidimensional fuzzy set forming module is used to collect raw data related to power grid load fluctuations and perform fuzzification processing on the raw data to form a multidimensional fuzzy set of the raw data;

[0047] The fuzzy rule base building module is used to build a fuzzy rule base based on the multidimensional fuzzy set of the original data and the fuzzy rules between the power grid load fluctuation and its various influencing factors;

[0048] The fuzzy logic model building module is used to design a fuzzy reasoning engine using a fuzzy rule base, and to build a fuzzy logic model based on the fuzzy rule base and the fuzzy reasoning engine;

[0049] The power grid load prediction model generation module is used to divide the multidimensional feature space of the multidimensional fuzzy set into multiple random subspaces, and establish a fuzzy logic submodel in each random subspace based on the fuzzy rule base and the fuzzy inference engine, and aggregate the prediction results of all random subspaces to generate the final power grid load prediction model;

[0050] The grid load real-time prediction module is used to input real-time data related to grid load fluctuation into a final grid load prediction model to perform real-time prediction of the grid load.

[0051] The multidimensional fuzzy set forming module includes a fuzzy value calculation unit and a multidimensional fuzzy set combining unit;

[0052] The fuzzy value calculation unit is used to map various types of original data into fuzzy sets using the following formula to obtain the fuzzy value of each original data:

[0053]

[0054]

[0055]

[0056] In the above formula, is the fuzzy value of the grid load power data, is the fuzzy value of weather data, is the fuzzy value of social and economic data, x is the data sample of various types of original data, μ P (x) is the membership function of the power grid load data, μ W (x) is the membership function of weather data, μ E(x) is the membership function of the socio-economic data, P(t) is the grid load power at time t, W(t) is the weather data at time t, and E(t) is the socio-economic data at time t;

[0057] The multidimensional fuzzy set combination unit is used to combine the fuzzy values ​​of various types of original data to form a multidimensional fuzzy set:

[0058]

[0059] In the above formula, is a multidimensional fuzzy set.

[0060] The fuzzy rule base building module includes a fuzzy rule definition unit and a fuzzy rule base forming unit;

[0061] The fuzzy rule definition unit is used to define a number of fuzzy rules for describing the nonlinear relationship between the power grid load fluctuation and its influencing factors based on the multidimensional fuzzy set of the original data. The form of each fuzzy rule is as follows:

[0062]

[0063] In the above formula, R i (t+1) is the i-th fuzzy rule R at time t+1 i The generated grid load prediction result, f is the fuzzy reasoning function, M is the total number of grid load power data, is the membership degree of the mth grid load power data, is the weight coefficient of the power grid load data, N is the total number of weather data, is the membership degree of the nth weather data, is the weight coefficient of weather data, L is the total number of social and economic data, is the membership degree of the lth socioeconomic data, is the weight coefficient of socioeconomic data;

[0064] The fuzzy rule base forming unit is used to combine the fuzzy rules to form a fuzzy rule base:

[0065]

[0066] In the above formula, is the fuzzy rule base composed of all fuzzy rules, and K is the total number of rules in the fuzzy rule base.

[0067] The fuzzy logic model in the fuzzy logic model building module includes:

[0068]

[0069]

[0070]

[0071] In the above formula, y i (t+1) is the i-th fuzzy rule R at time t+1 i The intermediate prediction result obtained by inference, f i is the fuzzy reasoning function based on the fuzzy rule base, is a multidimensional fuzzy set, is the fuzzy output result at time t+1, K is the total number of rules in the fuzzy rule base, is the i-th fuzzy rule R i When entering data The applicability of the above equation is given in Figure 2. Y(t+1) is the load prediction value determined at time t+1, X is the multidimensional feature space of the multidimensional fuzzy set, The fuzzy output result at time t+1 is the input data The membership function at .

[0072] The power grid load prediction model generation module includes a subspace allocation unit, a submodel prediction result training generation unit, a subspace prediction result integration unit, and a power grid load prediction model aggregation unit;

[0073] The subspace allocation unit is used to allocate the multidimensional feature space of the multidimensional fuzzy set to each random subspace using the following formula:

[0074]

[0075] In the above formula, X n is the nth random subspace, q n is the number of features contained in the nth random subspace, X n，j is the jth input feature assigned to the nth random subspace, X is the multidimensional feature space of the multidimensional fuzzy set;

[0076] The sub-model prediction result training generation unit is used to establish several fuzzy logic sub-models M in each random sub-space based on the fuzzy rule base and the fuzzy reasoning engine. n，m , m=1,2,...,M n , M n，m Represents the mth fuzzy logic submodel in the nth random subspace, and generates several independent prediction results in each random subspace by independently training each fuzzy logic submodel is the prediction result of the mth fuzzy logic sub-model in the nth random subspace at time t+1;

[0077] The subspace prediction result integration unit is used to integrate the prediction results of all fuzzy logic sub-models in the nth random subspace, and obtain the prediction results of each random subspace using the following formula:

[0078]

[0079] In the above formula, is the prediction result of the nth random subspace at time t+1, M n are all fuzzy logic submodels in the nth random subspace, α n，m is the weight coefficient of the mth fuzzy logic sub-model;

[0080] The power grid load prediction model aggregation unit is used to aggregate the prediction results of all random subspaces to generate a final power grid load prediction model:

[0081]

[0082] In the above formula, is the predicted value of the grid load at time t+1, β n The contribution weight of each random subspace to the final prediction result.

[0083] Compared with the prior art, the present invention has the following beneficial effects:

[0084] The present invention proposes a power grid load prediction method and system based on artificial intelligence and subspace aggregation. The method first collects original data related to power grid load fluctuations, fuzzifies the original data, and forms a multidimensional fuzzy set of the original data; based on the multidimensional fuzzy set of the original data, a fuzzy rule base is established according to the fuzzy rules between the power grid load fluctuations and its various influencing factors; then a fuzzy reasoning engine is designed using the fuzzy rule base, and a fuzzy logic model is established based on the fuzzy rule base and the fuzzy reasoning engine; finally, the multidimensional feature space of the multidimensional fuzzy set is divided into multiple random subspaces, a fuzzy logic submodel is established in each random subspace based on the fuzzy rule base and the fuzzy reasoning engine, and the prediction results of all random subspaces are aggregated to generate a final power grid load prediction model; real-time data related to power grid load fluctuations are input into the final power grid load prediction model to perform real-time prediction of the power grid load. On the one hand, this method reduces the computing time required for model training and prediction under the condition of limited computing resources through the fuzzy logic system and random subspace partitioning process, thereby improving the computational efficiency of the prediction model. Even in emergency situations, the model can still complete the prediction calculation in a short time, providing timely support for the dispatching decision of the power grid; on the other hand, this method divides the multidimensional feature space of the multidimensional fuzzy set into multiple random subspaces, and establishes a fuzzy logic sub-model in each random subspace based on the fuzzy rule base and fuzzy inference engine. The strategy of random subspace partitioning and multi-model integration is adopted to effectively reduce the computational deviation caused by a single model, and significantly improve the stability and accuracy of the prediction model by integrating the prediction results of different random subspaces. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 The figure is an overall flow chart of the method of the present invention.

[0086] Figure 2 The structure diagram of the system of the present invention is shown in FIG. DETAILED DESCRIPTION

[0087] The present invention is further described in detail below in conjunction with specific implementations and drawings.

[0088] The present invention proposes a power grid load prediction method and system based on artificial intelligence and subspace aggregation. The power grid system automatically collects original data related to power grid load fluctuations, and aggregates the prediction results of all random subspaces through a fuzzy logic system and a random subspace partitioning process to generate a final power grid load prediction model. Finally, the prediction performance of the power grid load fluctuation prediction model is evaluated, and the fuzzy rules and random subspace partitioning strategies are dynamically adjusted according to the performance evaluation results to optimize the power grid load fluctuation prediction model. In practical applications, fast and accurate load prediction is provided for the power grid system, providing strong technical support for the stable operation of the power grid.

[0089] Embodiment 1:

[0090] This embodiment takes the power grid system of a large industrial park in a coastal area as the research object. The power grid of the industrial park serves multiple high-energy-consuming manufacturing enterprises, and the production activities of the enterprises have obvious seasonal fluctuation characteristics. On July 25, 2024, the power grid system of the industrial park faced a sudden power load fluctuation problem. Between 2 and 4 a.m. that day, the power grid monitoring system recorded a series of abnormal load fluctuations, and there was a significant deviation between the load prediction model and the actual load data. The traditional load prediction model failed to accurately predict the fluctuations, causing the power grid dispatching center to face the risk of unstable power supply. In order to respond to this emergency, the power grid dispatching center quickly launched an emergency analysis program based on the power grid load prediction method of the present invention.

[0091] like Figure 1 As shown, the power grid load forecasting method based on artificial intelligence and subspace aggregation is carried out in the following steps:

[0092] 1. The power grid system automatically collects raw data related to power grid load fluctuations between 2 a.m. and 4 a.m., and fuzzifies the collected data to form a multi-dimensional fuzzy set of raw data;

[0093] The raw data related to the grid load fluctuation include grid load power data: 1750MW at 2:00 a.m., 1790MW at 2:30 a.m., 1820MW at 3:00 a.m., 1900MW at 3:30 a.m., and 1950MW at 4:00 a.m.; local weather data: temperature 32°C, humidity 85%, wind speed 10m / s; socio-economic data: industrial production index 110, residential electricity consumption 850MW;

[0094] The collected raw data is preprocessed by data cleaning, outlier removal, missing value filling and other preprocessing operations, and the preprocessed data is fuzzified to convert the continuous data set into a fuzzy set;

[0095] The membership function corresponding to the original data related to the power grid load fluctuation is set. The membership function is used to map the original data into a fuzzy set. Each type of data is fuzzified according to the preset membership function. The following formula is used to map various types of original data into fuzzy sets to obtain the fuzzy value of each original data:

[0096]

[0097]

[0098]

[0099] In the above formula, is the fuzzy value of the grid load power data, is the fuzzy value of weather data, is the fuzzy value of social and economic data, x is the data sample of various types of original data, μ P (x) is the membership function of the power grid load data, μ W (x) is the membership function of weather data, μ E (x) is the membership function of social and economic data, P(t) is the power grid load power at time t, W(t) is the weather data at time t, including temperature data T(t), humidity data H(t) and wind speed data V(t) at time t, E(t) is the social and economic data at time t, including industrial production index data I(t) and residential electricity consumption data R(t) at time t;

[0100] The fuzzy values ​​of various types of original data are combined to form a multidimensional fuzzy set:

[0101]

[0102] In the above formula, is a multidimensional fuzzy set.

[0103] 2. Based on the multidimensional fuzzy set of the original data, a fuzzy rule base is established according to the fuzzy rules between the power grid load fluctuation and its various influencing factors;

[0104] Based on the statistical analysis results of historical power grid load data and expert experience, the correlation between power grid load fluctuation and its influencing factors is analyzed. The influencing factors include power grid load power factor, weather factor and socio-economic factor. Several fuzzy rules are defined to describe the nonlinear relationship between power grid load fluctuation and its influencing factors. The form of each fuzzy rule is as follows:

[0105]

[0106] In the above formula, R i (t+1) is the i-th fuzzy rule R at time t+1 i The generated grid load prediction result, f is the fuzzy reasoning function, M is the total number of grid load power data, is the membership degree of the mth grid load power data, is the weight coefficient of the power grid load data, N is the total number of weather data, is the membership degree of the nth weather data, is the weight coefficient of weather data, L is the total number of social and economic data, is the membership degree of the lth socioeconomic data, is the weight coefficient of social and economic data; the weight coefficient of each data type reflects the relative importance of different factors to grid load fluctuation;

[0107] For each fuzzy rule, the following formula is used to define its corresponding membership function, which is used to describe the applicability of the fuzzy rule under different input conditions:

[0108]

[0109] In the above formula, is the i-th fuzzy rule R i When entering data The applicability of α m , β n , γ1 are weighted indices of grid load power data, weather data, and socio-economic data, respectively;

[0110] Combine the fuzzy rules to form a fuzzy rule base:

[0111]

[0112] In the above formula, is the fuzzy rule base composed of all fuzzy rules, and K is the total number of rules in the fuzzy rule base.

[0113] 3. Design a fuzzy reasoning engine using the fuzzy rule base, and establish a fuzzy logic model based on the fuzzy rule base and the fuzzy reasoning engine;

[0114] The fuzzy logic model includes the intermediate prediction results of the power grid load, the fuzzy output results and the determined load prediction value;

[0115] A fuzzy reasoning engine is designed based on the fuzzy rule base to reason on the multidimensional fuzzy set of the original data and obtain the intermediate prediction results of the power grid load:

[0116]

[0117] In the above formula, y i (t+1) is the i-th fuzzy rule R at time t+1 i The intermediate prediction result obtained by inference, f i is the fuzzy reasoning function based on the fuzzy rule base, is a multidimensional fuzzy set;

[0118] The weighted average method is used to perform fuzzy aggregation on the intermediate prediction results of the power grid load under each fuzzy rule to generate the final fuzzy output result:

[0119]

[0120] In the above formula, is the fuzzy output result at time t+1, K is the total number of rules in the fuzzy rule base, is the i-th fuzzy rule R i When entering data the extent of applicability of the following;

[0121] The weighted integral method is used to defuzzify the final fuzzy output result after fuzzy aggregation and convert it into a definite load prediction value:

[0122]

[0123] In the above formula, Y(t+1) is the load prediction value determined at time t+1, X is the multidimensional feature space of the multidimensional fuzzy set, The fuzzy output result at time t+1 is the input data The membership function at .

[0124] 4. Divide the multidimensional feature space of the multidimensional fuzzy set into multiple random subspaces, and establish a fuzzy logic submodel in each random subspace based on the fuzzy rule base and fuzzy inference engine, and aggregate the prediction results of all random subspaces to generate the final power grid load prediction model;

[0125] The following formula is used to distribute the multidimensional feature space of the multidimensional fuzzy set into various random subspaces, each of which contains several features:

[0126]

[0127] In the above formula, X n is the nth random subspace, q n is the number of features contained in the nth random subspace, X n，j is the jth input feature assigned to the nth random subspace, X is the multidimensional feature space of the multidimensional fuzzy set, x = {x1, x2, ..., x p};

[0128] The load prediction value of each random subspace is calculated using the following formula:

[0129]

[0130] In the above formula, Y n (t+1) is the load prediction value generated by the nth random subspace at time t+1, f n is the inference prediction function of the nth random subspace, which performs inference calculation based on the rules in the fuzzy rule base, w n，j is the corresponding input feature X in the nth random subspace n，j The weight of For input feature X n，j Membership degree at time t;

[0131] Based on the input features of each random subspace and its corresponding fuzzy rule base, the load prediction value of each random subspace is trained. During the training process, the error function L is minimized. n Optimize the weight parameter W corresponding to the input feature in each random subspace n，j , get more accurate load prediction values ​​of each random subspace;

[0132] In each random subspace, several fuzzy logic submodels M are established based on the fuzzy rule base and fuzzy reasoning engine. n，m , m=1,2,...,M n , M n，m represents the mth fuzzy logic submodel in the nth random subspace and minimizes the error function L n Each fuzzy logic sub-model is trained independently, generating several independent prediction results in each random subspace is the prediction result of the mth fuzzy logic sub-model in the nth random subspace at time t+1;

[0133] The minimized error function L n for:

[0134]

[0135] In the above formula, T is the total length of the time series, Y n，m (t+1) is the actual value of the mth fuzzy logic submodel in the nth random subspace at time t+1;

[0136] The prediction results of all fuzzy logic sub-models in the nth random subspace are integrated through the random subspace aggregation algorithm, and the prediction results of each random subspace are obtained using the following formula:

[0137]

[0138] In the above formula, is the prediction result of the nth random subspace at time t+1, M n are all fuzzy logic submodels in the nth random subspace, α n，m is the weight coefficient of the mth fuzzy logic sub-model, satisfying

[0139] Aggregate the prediction results of all random subspaces to generate the final power grid load prediction model:

[0140]

[0141] In the above formula, is the predicted value of the grid load at time t+1, β n The contribution weight of each random subspace to the final prediction result satisfies

[0142] 5. Evaluate the prediction performance of the grid load fluctuation prediction model through the validation data set, namely historical grid load power data, weather data and socio-economic data, including:

[0143] The validation dataset V = {V1, V2, ..., V q}, input into the generated power grid load prediction model, and calculate the power grid load prediction value at time t+1 The root mean square error RMSE, mean absolute percentage error MAPE and R 2 , quantify the prediction performance of the grid load fluctuation prediction model:

[0144]

[0145]

[0146]

[0147] In the above formula, q is the total number of samples in the validation data set, Y true，j (t+1) is the actual load value of the jth sample in the validation data set at time t+1, is the predicted value of grid load fluctuation at time t+1, is the average value of the actual load values ​​of all samples in the validation data set; where R 2 It reflects the ability of the power grid load fluctuation prediction model to explain data variation. The closer the value is to 1, the better the performance of the power grid load fluctuation prediction model.

[0148] 6. Dynamically adjust the fuzzy rules and random subspace partitioning strategy of the original data related to power grid load fluctuations according to the performance evaluation results to optimize the power grid load fluctuation prediction model;

[0149] By fine-tuning the membership function in the fuzzy rule base, adjusting the division range of the fuzzy set, or reallocating the input feature variables in the multidimensional fuzzy set feature space, and optimizing the random subspace division strategy, the final power grid load forecasting model is optimized until the performance index of the power grid load forecasting model reaches the preset optimization target, thus forming the optimal power grid load forecasting model;

[0150] By introducing a dynamic adjustment mechanism to evaluate the prediction performance of the model in real time, dynamically optimizing the fuzzy rule base and random subspace partitioning strategy, it is possible to continuously optimize and adjust its own prediction model, quickly respond to changes in real-time data in the power grid system, and ensure that the model adapts to the complex fluctuations of the power grid load. When faced with real-time data, it can respond quickly and provide more accurate load prediction results, greatly improving the adaptability and stability of the prediction model, which is especially suitable for complex and changeable power grid operating environments.

[0151] 7. Input the real-time data related to the grid load fluctuation into the optimized grid load prediction model to perform real-time prediction of the grid load;

[0152] The actual grid load data collected between 2 a.m. and 4 a.m. is input into the grid load forecasting model to quickly generate the load forecast results for the next hour: the model predicts that the grid system load will reach 2000MW at 4:30 a.m. and may further climb to 2050MW at 5:00 a.m.

[0153] In order to verify the effect of the present invention, the dispatch center made a detailed comparison between the actual load data from 2:00 to 4:00 a.m. on July 25, 2024, the predicted data by the traditional method and the predicted data by the present invention, as shown in Table 1:

[0154] Table 1 Actual load data, traditional method prediction data and the prediction data of the present invention

[0155] time Actual load Traditional prediction method The present invention predicts 2:00 1750MW 1700MW 1745MW 2:30 1790MW 1720MW 1785MW 3:00 1820MW 1750MW 1815MW 3:30 1900MW 1800MW 1890MW 4:00 1950MW 1850MW 1945MW ;

[0156] According to the prediction data in the above table, the prediction performance of the traditional prediction model and the prediction model of the present invention is evaluated. The evaluation results show that in terms of prediction accuracy, the root mean square error of the traditional method is 75MW, the average absolute percentage error is 4.8%, while the root mean square error of the method of the present invention is only 25MW, the average absolute percentage error is 1.3%. In addition, the R 2 The value reached 0.97, which is better than the 0.85 of the traditional method. Compared with the traditional prediction method, the prediction method of the present invention shows higher prediction accuracy, and the prediction model can better explain the variability of power grid load fluctuations.

[0157] Previously, the traditional forecasting model predicted that the load at 4:30 would be 1900MW and the load at 5:00 would be 1920MW. The actual load was far beyond the forecast range, causing the power grid dispatching center to be passive in the initial response. The method of the present invention provides a more accurate load forecast through a multi-model integration strategy. In the subsequent operation, the power grid dispatching center dispatched the backup power supply in advance at 4:00 according to the forecast results of the present invention, and completed the access of the backup power supply at 4:15, ensuring a stable power supply from 4:30 to 5:00 and avoiding production interruptions caused by insufficient power supply. Subsequent analysis showed that if the traditional load forecasting model continued to be used, the power grid system would likely face the risk of increased power supply pressure around 5:00;

[0158] The successful handling of this incident proved the feasibility and effectiveness of the power grid load forecasting method of the present invention in practical applications. It not only far exceeds the traditional methods in forecasting accuracy, but also provides strong technical support in emergency handling and resource scheduling, laying a solid foundation for the stable operation of the smart grid.

[0159] Embodiment 2:

[0160] like Figure 2 As shown, the power grid load prediction system based on artificial intelligence and subspace aggregation includes a multi-dimensional fuzzy set forming module, a fuzzy rule base establishing module, a fuzzy logic model establishing module, a power grid load prediction model generating module, and a power grid load real-time prediction module;

[0161] The system includes a multi-dimensional fuzzy set forming module, a fuzzy rule base establishing module, a fuzzy logic model establishing module, a power grid load prediction model generating module, and a power grid load real-time prediction module;

[0162] The multidimensional fuzzy set forming module is used to collect raw data related to power grid load fluctuations and perform fuzzification processing on the raw data to form a multidimensional fuzzy set of the raw data;

[0163] The fuzzy rule base building module is used to build a fuzzy rule base based on the multidimensional fuzzy set of the original data and the fuzzy rules between the power grid load fluctuation and its various influencing factors;

[0164] The fuzzy logic model building module is used to design a fuzzy reasoning engine using a fuzzy rule base, and to build a fuzzy logic model based on the fuzzy rule base and the fuzzy reasoning engine;

[0165] The power grid load prediction model generation module is used to divide the multidimensional feature space of the multidimensional fuzzy set into multiple random subspaces, and establish a fuzzy logic submodel in each random subspace based on the fuzzy rule base and the fuzzy inference engine, and aggregate the prediction results of all random subspaces to generate the final power grid load prediction model;

[0166] The grid load real-time prediction module is used to input real-time data related to grid load fluctuation into a final grid load prediction model to perform real-time prediction of the grid load.

[0167] The multidimensional fuzzy set forming module includes a fuzzy value calculation unit and a multidimensional fuzzy set combining unit;

[0168] The fuzzy value calculation unit is used to map various types of original data into fuzzy sets using the following formula to obtain the fuzzy value of each original data:

[0169]

[0170]

[0171]

[0172] In the above formula, is the fuzzy value of the grid load power data, is the fuzzy value of weather data, is the fuzzy value of social and economic data, x is the data sample of various types of original data, μ P (x) is the membership function of the power grid load data, μ W (x) is the membership function of weather data, μ E (x) is the membership function of the socio-economic data, P(t) is the grid load power at time t, W(t) is the weather data at time t, and E(t) is the socio-economic data at time t;

[0173] The multidimensional fuzzy set combination unit is used to combine the fuzzy values ​​of various types of original data to form a multidimensional fuzzy set:

[0174]

[0175] In the above formula, is a multidimensional fuzzy set.

[0176] The fuzzy rule base building module includes a fuzzy rule definition unit and a fuzzy rule base forming unit;

[0177] The fuzzy rule definition unit is used to define a number of fuzzy rules for describing the nonlinear relationship between the power grid load fluctuation and its influencing factors based on the multidimensional fuzzy set of the original data. The form of each fuzzy rule is as follows:

[0178]

[0179] In the above formula, R i (t+1) is the i-th fuzzy rule R at time t+1 i The generated grid load prediction result, f is the fuzzy reasoning function, M is the total number of grid load power data, is the membership degree of the mth grid load power data, is the weight coefficient of the power grid load data, N is the total number of weather data, is the membership degree of the nth weather data, is the weight coefficient of weather data, L is the total number of social and economic data, is the membership degree of the lth socioeconomic data, is the weight coefficient of socioeconomic data;

[0180] The fuzzy rule base forming unit is used to combine the fuzzy rules to form a fuzzy rule base:

[0181]

[0182] In the above formula, is the fuzzy rule base composed of all fuzzy rules, and K is the total number of rules in the fuzzy rule base.

[0183] The fuzzy logic model in the fuzzy logic model building module includes:

[0184]

[0185]

[0186]

[0187] In the above formula, y i (t+1) is the i-th fuzzy rule R at time t+1 i The intermediate prediction result obtained by inference, f i is the fuzzy reasoning function based on the fuzzy rule base, is a multidimensional fuzzy set, is the fuzzy output result at time t+1, K is the total number of rules in the fuzzy rule base, is the i-th fuzzy rule R i When entering data The applicability of the above equation is given in Figure 2. Y(t+1) is the load prediction value determined at time t+1, X is the multidimensional feature space of the multidimensional fuzzy set, The fuzzy output result at time t+1 is the input data The membership function at .

[0188] The power grid load prediction model generation module includes a subspace allocation unit, a submodel prediction result training generation unit, a subspace prediction result integration unit, and a power grid load prediction model aggregation unit;

[0189] The subspace allocation unit is used to allocate the multidimensional feature space of the multidimensional fuzzy set to each random subspace using the following formula:

[0190]

[0191] In the above formula, X n is the nth random subspace, q n is the number of features contained in the nth random subspace, X n，j is the jth input feature assigned to the nth random subspace, X is the multidimensional feature space of the multidimensional fuzzy set;

[0192] The sub-model prediction result training generation unit is used to establish several fuzzy logic sub-models M in each random sub-space based on the fuzzy rule base and the fuzzy reasoning engine. n，m , m=1,2,...,M n , M n，m Represents the mth fuzzy logic submodel in the nth random subspace, and generates several independent prediction results in each random subspace by independently training each fuzzy logic submodel is the prediction result of the mth fuzzy logic sub-model in the nth random subspace at time t+1;

[0193] The subspace prediction result integration unit is used to integrate the prediction results of all fuzzy logic sub-models in the nth random subspace, and obtain the prediction results of each random subspace using the following formula:

[0194]

[0195] In the above formula, is the prediction result of the nth random subspace at time t+1, M n are all fuzzy logic submodels in the nth random subspace, α n，m is the weight coefficient of the mth fuzzy logic sub-model;

[0196] The power grid load prediction model aggregation unit is used to aggregate the prediction results of all random subspaces to generate a final power grid load prediction model:

[0197]

[0198] In the above formula, is the predicted value of the grid load at time t+1, β n The contribution weight of each random subspace to the final prediction result.

Claims

1. A power grid load forecasting method based on artificial intelligence and subspace aggregation, characterized in that: include: S1. Collecting raw data related to power grid load fluctuations and fuzzifying the raw data to form a multi-dimensional fuzzy set of the raw data; S2. Based on the multidimensional fuzzy set of the original data, a fuzzy rule base is established according to the fuzzy rules between the power grid load fluctuation and its various influencing factors; S3, designing a fuzzy reasoning engine using a fuzzy rule base, and establishing a fuzzy logic model based on the fuzzy rule base and the fuzzy reasoning engine; S4, dividing the multidimensional feature space of the multidimensional fuzzy set into multiple random subspaces, and establishing a fuzzy logic submodel in each random subspace based on the fuzzy rule base and the fuzzy inference engine, and aggregating the prediction results of all random subspaces to generate the final power grid load prediction model; S5. Input the real-time data related to the grid load fluctuation into the final grid load prediction model to perform real-time prediction of the grid load.

2. The power grid load forecasting method based on artificial intelligence and subspace aggregation according to claim 1 is characterized in that: The S1 includes: S11. Use the following formula to map various types of original data into fuzzy sets to obtain the fuzzy value of each original data: In the above formula, is the fuzzy value of the grid load power data, is the fuzzy value of weather data, is the fuzzy value of social and economic data, X is the data sample of various types of original data, μ P (x) is the membership function of the power grid load data, μ W (x) is the membership function of weather data, μ E (x) is the membership function of the socio-economic data, P(t) is the grid load power at time t, W(t) is the weather data at time t, and E(t) is the socio-economic data at time t; S12. Combining the fuzzy values ​​of various types of original data to form a multidimensional fuzzy set: In the above formula, is a multidimensional fuzzy set.

3. The power grid load forecasting method based on artificial intelligence and subspace aggregation according to claim 1 is characterized in that: The S2 includes: S21. Based on the multidimensional fuzzy set of the original data, several fuzzy rules are defined to describe the nonlinear relationship between the power grid load fluctuation and its influencing factors. The form of each fuzzy rule is as follows: In the above formula, R i (t+1) is the i-th fuzzy rule R at time t+1 i The generated grid load prediction result, f is the fuzzy reasoning function, M is the total number of grid load power data, is the membership degree of the mth grid load power data, is the weight coefficient of the power grid load data, N is the total number of weather data, is the membership degree of the nth weather data, is the weight coefficient of weather data, L is the total number of social and economic data, is the membership degree of the lth socioeconomic data, is the weight coefficient of socioeconomic data; S22, combining the fuzzy rules to form a fuzzy rule base: In the above formula, is the fuzzy rule base composed of all fuzzy rules, and K is the total number of rules in the fuzzy rule base.

4. The power grid load forecasting method based on artificial intelligence and subspace aggregation according to claim 1 is characterized in that: The fuzzy logic model in S3 includes: In the above formula, y i (t+1) is the i-th fuzzy rule R at time t+1 i The intermediate prediction result obtained by inference, f i is the fuzzy reasoning function based on the fuzzy rule base, is a multidimensional fuzzy set, is the fuzzy output result at time t+1, K is the total number of rules in the fuzzy rule base, is the i-th fuzzy rule R i When entering data The applicability of the above equation is given in Figure 2. Y(t+1) is the load prediction value determined at time t+1, X is the multidimensional feature space of the multidimensional fuzzy set, The fuzzy output result at time t+1 is the input data The membership function at .

5. The power grid load forecasting method based on artificial intelligence and subspace aggregation according to claim 1 is characterized in that: The S4 includes: S41. Use the following formula to distribute the multidimensional feature space of the multidimensional fuzzy set to each random subspace: In the above formula, X n is the nth random subspace, q n is the number of features contained in the nth random subspace, X n，j is the jth input feature assigned to the nth random subspace, X is the multidimensional feature space of the multidimensional fuzzy set; S42, in each random subspace, based on the fuzzy rule base and fuzzy reasoning engine, establish several fuzzy logic sub-models M n，m , m=1,2,...,M n , M n，m Represents the mth fuzzy logic submodel in the nth random subspace, and generates several independent prediction results in each random subspace by independently training each fuzzy logic submodel is the prediction result of the mth fuzzy logic sub-model in the nth random subspace at time t+1; S43, integrating the prediction results of all fuzzy logic sub-models in the nth random subspace, and obtaining the prediction results of each random subspace using the following formula: In the above formula, is the prediction result of the nth random subspace at time t+1, M n are all fuzzy logic submodels in the nth random subspace, α n，m is the weight coefficient of the mth fuzzy logic sub-model; S45. Aggregate the prediction results of all random subspaces to generate the final power grid load prediction model: In the above formula, is the predicted value of the grid load at time t+1, β n The contribution weight of each random subspace to the final prediction result.

6. A power grid load forecasting system based on artificial intelligence and subspace aggregation, characterized in that: The system includes a multi-dimensional fuzzy set forming module, a fuzzy rule base establishing module, a fuzzy logic model establishing module, a power grid load prediction model generating module, and a power grid load real-time prediction module; The multidimensional fuzzy set forming module is used to collect raw data related to power grid load fluctuations and perform fuzzification processing on the raw data to form a multidimensional fuzzy set of the raw data; The fuzzy rule base building module is used to build a fuzzy rule base based on the multidimensional fuzzy set of the original data and the fuzzy rules between the power grid load fluctuation and its various influencing factors; The fuzzy logic model building module is used to design a fuzzy reasoning engine using a fuzzy rule base, and to build a fuzzy logic model based on the fuzzy rule base and the fuzzy reasoning engine; The power grid load prediction model generation module is used to divide the multidimensional feature space of the multidimensional fuzzy set into multiple random subspaces, and establish a fuzzy logic submodel in each random subspace based on the fuzzy rule base and the fuzzy inference engine, and aggregate the prediction results of all random subspaces to generate the final power grid load prediction model; The grid load real-time prediction module is used to input real-time data related to grid load fluctuation into a final grid load prediction model to perform real-time prediction of the grid load.

7. The power grid load forecasting system based on artificial intelligence and subspace aggregation according to claim 6 is characterized in that: The multidimensional fuzzy set forming module includes a fuzzy value calculation unit and a multidimensional fuzzy set combining unit; The fuzzy value calculation unit is used to map various types of original data into fuzzy sets using the following formula to obtain the fuzzy value of each original data: In the above formula, is the fuzzy value of the grid load power data, is the fuzzy value of weather data, is the fuzzy value of social and economic data, x is the data sample of various types of original data, μ P (x) is the membership function of the power grid load data, μ W (x) is the membership function of weather data, μ E (x) is the membership function of the socio-economic data, P(t) is the grid load power at time t, W(t) is the weather data at time t, and E(t) is the socio-economic data at time t; The multidimensional fuzzy set combination unit is used to combine the fuzzy values ​​of various types of original data to form a multidimensional fuzzy set: In the above formula, is a multidimensional fuzzy set.

8. The power grid load forecasting system based on artificial intelligence and subspace aggregation according to claim 6, characterized in that: The fuzzy rule base building module includes a fuzzy rule definition unit and a fuzzy rule base forming unit; The fuzzy rule definition unit is used to define a number of fuzzy rules for describing the nonlinear relationship between the power grid load fluctuation and its influencing factors based on the multidimensional fuzzy set of the original data. The form of each fuzzy rule is as follows: In the above formula, R i (t+1) is the i-th fuzzy rule R at time t+1 i The generated grid load prediction result, f is the fuzzy reasoning function, M is the total number of grid load power data, is the membership degree of the mth grid load power data, is the weight coefficient of the power grid load data, N is the total number of weather data, is the membership degree of the nth weather data, is the weight coefficient of weather data, L is the total number of social and economic data, is the membership degree of the lth socioeconomic data, is the weight coefficient of socioeconomic data; The fuzzy rule base forming unit is used to combine the fuzzy rules to form a fuzzy rule base: In the above formula, is the fuzzy rule base composed of all fuzzy rules, and K is the total number of rules in the fuzzy rule base.

9. The power grid load forecasting system based on artificial intelligence and subspace aggregation according to claim 6, characterized in that: The fuzzy logic model in the fuzzy logic model building module includes: In the above formula, y i (t+1) is the i-th fuzzy rule R at time t+1 i The intermediate prediction result obtained by reasoning, f is the fuzzy reasoning function based on the fuzzy rule base, is a multidimensional fuzzy set, is the fuzzy output result at time t+1, K is the total number of rules in the fuzzy rule base, is the i-th fuzzy rule R i When entering data The applicability of the above equation is given in Figure 2. Y(t+1) is the load prediction value determined at time t+1, X is the multidimensional feature space of the multidimensional fuzzy set, The fuzzy output result at time t+1 is the input data The membership function at .

10. The power grid load forecasting system based on artificial intelligence and subspace aggregation according to claim 6, characterized in that: The power grid load prediction model generation module includes a subspace allocation unit, a submodel prediction result training generation unit, a subspace prediction result integration unit, and a power grid load prediction model aggregation unit; The subspace allocation unit is used to allocate the multidimensional feature space of the multidimensional fuzzy set to each random subspace using the following formula: In the above formula, X n is the nth random subspace, q n is the number of features contained in the nth random subspace, X n，j is the jth input feature assigned to the nth random subspace, X is the multidimensional feature space of the multidimensional fuzzy set; The sub-model prediction result training generation unit is used to establish several fuzzy logic sub-models M in each random sub-space based on the fuzzy rule base and the fuzzy reasoning engine. n，m , m=1,2,...,M n , M n，m Represents the mth fuzzy logic submodel in the nth random subspace, and generates several independent prediction results in each random subspace by independently training each fuzzy logic submodel is the prediction result of the mth fuzzy logic sub-model in the nth random subspace at time t+1; The subspace prediction result integration unit is used to integrate the prediction results of all fuzzy logic sub-models in the nth random subspace, and obtain the prediction results of each random subspace using the following formula: In the above formula, is the prediction result of the nth random subspace at time t+1, M n are all fuzzy logic submodels in the nth random subspace, α n，m is the weight coefficient of the mth fuzzy logic sub-model; The power grid load prediction model aggregation unit is used to aggregate the prediction results of all random subspaces to generate a final power grid load prediction model: In the above formula, is the predicted value of the grid load at time t+1, β n The contribution weight of each random subspace to the final prediction result.

Images