Day-ahead prediction method and device for load cluster of medium and low voltage distribution network

Abstract

The invention relates to the technical field of power system operation, in particular to a day-ahead prediction method and device for a load cluster of a medium and low voltage distribution network, and the method comprises the steps: recognizing at least one load of which the similarity meets a preset condition, so as to generate at least one clustering cluster of a medium and low voltage transformer; performing day-ahead prediction on each cluster of the at least one cluster in a preset prediction mode to obtain an aggregation load of each cluster; and extracting at least one strong correlation feature by using the aggregated load of each cluster, and obtaining a final prediction model for day-ahead prediction of the load cluster of the medium and low voltage distribution network according to the at least one strong correlation feature. Therefore, the problem that load fluctuation and intermittency caused by distributed power supply access cannot be effectively processed due to the fact that the influence of important factors such as load user behaviors and characteristic diversity on load prediction is not fully considered in a load prediction mode in the related technology is solved; and refined modeling cannot be carried out for different load characteristics.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system operation, and in particular relates to a medium and low voltage distribution network load cluster day-ahead prediction method and device. BACKGROUND

[0002] The diversification and dynamics of loads bring many uncertainties to power grid operation and planning. Since the load is not only affected by meteorological conditions, economic activities and social behavior patterns, but also closely related to the generation characteristics of distributed resources and demand response behavior. Under this complex background, accurately predicting load demand, especially considering the dynamic changes of distributed resources and flexible loads, has become one of the core problems of intelligent regulation of power grid; with the decrease of voltage level, the randomness of load distribution increases significantly, and the load curve in medium and low voltage distribution network is difficult to accurately predict.

[0003] In related technologies, most researches predict the load of distribution network through the way of independent optimization of clustering algorithm and load prediction algorithm, wherein the clustering algorithm is used to classify the historical electricity load data, so as to infer the future load change trend; and then the load prediction algorithm is used to predict and analyze the load.

[0004] However, the way of separate optimization of clustering algorithm and load prediction algorithm in related technologies fails to fully consider the prediction target, which affects the overall performance of the model. Due to the diversity of load user behavior and characteristics, the load mode of different medium and low voltage distribution networks is significantly different, and a single model is difficult to adapt to the diversified load characteristics, further exacerbating the instability of the prediction result; at the same time, the cumulative error of random load will be significantly amplified in the aggregation process of medium and low voltage distribution network load prediction, which weakens the reliability of prediction; the high penetration rate of distributed photovoltaic and other new energy in distribution network further aggravates the volatility and intermittency of load curve, increases the prediction complexity, and the existing model has limited ability in processing non-stationary time series.

[0005] In summary, the way of predicting load in related technologies fails to fully consider the influence of important factors such as the diversity of load user behavior and characteristics on load prediction, which further leads to problems such as being unable to effectively handle the load volatility and intermittency caused by the access of distributed power supply, being unable to fine-grained modeling for different load characteristics, and the like, which need to be improved. SUMMARY

[0006] The present application provides a medium and low voltage distribution network load cluster day-ahead prediction method and device to solve the problem that the way of predicting load in related technologies fails to fully consider the influence of important factors such as the diversity of load user behavior and characteristics on load prediction, which further leads to problems such as being unable to effectively handle the load volatility and intermittency caused by the access of distributed power supply, being unable to fine-grained modeling for different load characteristics, and the like.

[0007] The first aspect embodiment of the present application provides a medium and low voltage distribution network load cluster day-ahead prediction method, comprising the following steps: identifying at least one load with similarity meeting a preset condition to generate at least one clustering cluster of medium and low voltage transformers; performing day-ahead prediction on each clustering cluster of the at least one clustering cluster in a preset prediction mode to obtain an aggregated load of each cluster; and extracting at least one strong correlation feature by using the aggregated load of each cluster, and obtaining a final prediction model for medium and low voltage distribution network load cluster day-ahead prediction according to the at least one strong correlation feature.

[0008] Through the above technical solution, the embodiment of the present application can generate at least one clustering cluster of low voltage transformers by introducing load cluster clustering; can obtain an aggregated load of each cluster and extract at least one strong correlation feature by introducing a gradient boosting algorithm, and combining load cluster clustering and the gradient boosting algorithm, the accuracy and stability of load prediction are significantly improved. This method not only effectively handles the load volatility and intermittency caused by the access of distributed power sources, but also fine-tunes the modeling for different load characteristics, providing a solid data support for the safe and stable operation of the power grid. In addition, through the feature selection and error evaluation steps, the prediction model is further optimized, ensuring the reliability and practicality of the prediction results, which is of great significance to the development of smart grid.

[0009] Optionally, in an embodiment of the present application, the identification of at least one load with similarity meeting a preset condition to generate a clustering cluster comprises: generating a representative load curve for each district; generating a load curve matrix according to the representative load curve; and grouping the medium and low voltage transformers with similar load characteristics into the same cluster based on the load curve matrix to obtain the clustering cluster.

[0010] Through the above technical solution, the embodiment of the present application can first generate a representative load curve for each district as the input of the clustering algorithm, and then apply the clustering algorithm to determine the optimal number of clusters. This way, the medium and low voltage transformers with similar load characteristics can be grouped into the same cluster, providing more targeted data support for subsequent load prediction.

[0011] Optionally, in an embodiment of the present application, the day-ahead prediction on each clustering cluster of the at least one clustering cluster to obtain an aggregated load of each cluster comprises: decomposing a representative load curve of each clustering cluster based on trend and seasonal components and random fluctuations and outliers in the curve to obtain a stationary part and a residual part; and predicting the stationary part and the residual part respectively by using a corresponding prediction mode in the preset prediction mode to obtain the aggregated load of each cluster.

[0012] By the technical solution, the embodiment of the present application can distinguish random fluctuation abnormal load from smooth load by decomposing the representative load curve of each clustering cluster to obtain a smooth part and a residual part, and perform fine modeling and prediction according to different load characteristics, thereby providing solid data support for the safe and stable operation of the power grid.

[0013] Optionally, in an embodiment of the present application, further comprising: in the prediction mode corresponding to the residual part, calculating the importance of each feature according to the degree used in the prediction process; and using Bayesian optimization search to optimize the hyperparameters of the gradient boosting algorithm in the prediction mode.

[0014] By the technical solution, the embodiment of the present application can use the Bayesian optimization search algorithm to find the nearly optimal combination of hyperparameters in a small number of iterations, and further improve the prediction performance of the gradient boosting algorithm by optimizing the hyperparameters.

[0015] Optionally, in an embodiment of the present application, the extracting at least one strongly correlated feature using the aggregate load of each cluster and obtaining a final prediction model for day-ahead forecasting of the medium / low voltage distribution network load cluster according to the at least one strongly correlated feature comprises: determining the importance of each feature based on the importance of each feature to obtain the at least one strongly correlated feature; and calculating the minimum prediction error and the optimal feature set of each cluster according to the at least one strongly correlated feature to obtain the final prediction model.

[0016] By the technical solution, the embodiment of the present application can use the gradient boosting algorithm to train and optimize each cluster respectively, identify the relatively unimportant features of each cluster according to the typical behavior and load characteristics of the customers connected to the medium / low voltage transformer, and use the features with low importance to further improve the prediction accuracy of each cluster to avoid the overfitting caused by the features with low importance. The feature selection process is applied to each cluster to improve the prediction of each cluster and thus improve the prediction of the aggregate load. The error evaluation step further optimizes the prediction model to ensure the reliability and practicability of the prediction result, which is of great significance to the development of smart grid.

[0017] Optionally, in an embodiment of the present application, the calculation formula of the importance is:

[0018]

[0019] wherein, I j represents the importance of feature j used by all M decision trees in the gradient boosting algorithm, M represents the number of decision trees used in the gradient boosting algorithm, T m represents the mth decision tree in the gradient boosting algorithm, and m is the serial number of the decision tree.

[0020] Through the technical solution, the embodiment of the application can construct at least one clustering cluster of low-voltage transformers by introducing load cluster clustering. j , T m and other parameters to construct a feature importance calculation model, calculate the importance of each feature using the constructed feature importance calculation model, and then filter out non-important features, effectively preventing model overfitting and improving prediction accuracy.

[0021] The second aspect embodiment of the application provides a medium and low voltage power distribution network load cluster day-ahead prediction device, comprising: an identification module for identifying at least one load with similar characteristics to generate at least one clustering cluster of medium and low voltage transformers; a first prediction module for day-ahead prediction of each clustering cluster of the at least one clustering cluster in a preset prediction mode to obtain the aggregated load of each cluster; a second prediction module for extracting at least one strongly correlated feature using the aggregated load of each cluster and obtaining a final prediction model for medium and low voltage power distribution network load cluster day-ahead prediction according to the at least one strongly correlated feature.

[0022] Through the technical solution, the embodiment of the application can generate at least one clustering cluster of low-voltage transformers by introducing load cluster clustering; can obtain the aggregated load of each cluster and extract at least one strongly correlated feature by introducing gradient boosting algorithm, and combine load cluster clustering and gradient boosting algorithm to significantly improve the accuracy and stability of load prediction. This method not only effectively handles the load volatility and intermittency caused by distributed power supply access, but also fine-tuned modeling for different load characteristics, providing a solid data support for the safe and stable operation of the power grid. In addition, through the feature selection and error evaluation steps, the prediction model is further optimized to ensure the reliability and practicality of the prediction results, which is of great significance to the development of smart grid.

[0023] Optionally, in an embodiment of the application, the identification module comprises: a first generation unit for generating a representative load curve for each area; a second generation unit for generating a load curve matrix according to the representative load curve; a grouping unit for grouping the medium and low voltage transformers with similar load characteristics into the same cluster based on the load curve matrix to obtain the clustering cluster.

[0024] Through the technical solution, the embodiment of the application can first generate a representative load curve for each area as the input of the clustering algorithm, and then apply the clustering algorithm to determine the optimal cluster number. This way, the medium and low voltage transformers with similar load characteristics can be grouped into the same cluster, providing more targeted data support for subsequent load prediction.

[0025] Optionally, in an embodiment of the present application, the first prediction module comprises: a decomposition unit configured to decompose the representative load curve of each cluster group based on trend and seasonal components and random fluctuations and outliers in the curve to obtain a stationary part and a residual part; and a prediction unit configured to predict the stationary part and the residual part respectively in a corresponding prediction mode in the preset prediction mode to obtain the aggregated load of each cluster.

[0026] Through the above technical solution, the embodiment of the present application can distinguish random fluctuation abnormal load from stationary load by decomposing the representative load curve of each cluster group to obtain a stationary part and a residual part, and perform fine modeling and prediction according to different load characteristics, thereby providing solid data support for the safe and stable operation of the power grid.

[0027] Optionally, in an embodiment of the present application, the first prediction module is further configured to: in the prediction mode corresponding to the residual part, calculate the importance of each feature according to the degree used in the prediction process; and use Bayesian optimization search to optimize the hyperparameters of the gradient boosting algorithm in the prediction mode.

[0028] Through the above technical solution, the embodiment of the present application can use the Bayesian optimization search algorithm to find the near-optimal combination of hyperparameters in a small number of iterations, and by optimizing the hyperparameters, the prediction performance of the gradient boosting algorithm can be further improved.

[0029] Optionally, in an embodiment of the present application, the second prediction module comprises: a determination unit configured to determine the importance of each feature based on the importance of each feature to obtain at least one strongly correlated feature; and a calculation unit configured to calculate the minimum prediction error and the optimal feature set of each cluster according to the at least one strongly correlated feature to obtain the final prediction model.

[0030] Through the above technical solution, the embodiment of the present application can use the gradient boosting algorithm to train and optimize each cluster respectively, identify the relatively unimportant features of each cluster according to the typical behavior and load characteristics of the customers connected to the medium / low voltage transformer, and use the features with low importance to further improve the prediction accuracy of each cluster to avoid the situation that the features with low importance may cause overfitting. The feature selection process is applied to each cluster to improve the prediction of each cluster and thus improve the prediction of the aggregated load. The error evaluation step further optimizes the prediction model to ensure the reliability and practicality of the prediction results, which is of great significance to the development of smart grids.

[0031] Optionally, in an embodiment of the present application, the calculation formula of the importance is:

[0032]

[0033] wherein, I j represents the importance of feature j used by all M decision trees in the gradient boosting algorithm, M represents the number of decision trees used in the gradient boosting algorithm, T m represents the mth decision tree in the gradient boosting algorithm, m is the decision tree serial number.

[0034] Through the above technical solutions, the embodiments of the present application can construct a feature importance calculation model through I j , T m and other parameters, calculate the importance of each feature using the constructed feature importance calculation model, and then filter out non-important features, effectively preventing model overfitting and improving prediction accuracy.

[0035] The third aspect of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the medium and low voltage power distribution network load cluster day-ahead prediction method as described in the above embodiments.

[0036] The fourth aspect of the present application provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the medium and low voltage power distribution network load cluster day-ahead prediction method as described above.

[0037] The fifth aspect of the present application provides a computer program product, which stores a computer program, and the program is executed by a processor to implement the medium and low voltage power distribution network load cluster day-ahead prediction method as described above.

[0038] The embodiments of the present application can generate at least one cluster cluster of low-voltage transformers by introducing load cluster clustering; can obtain the aggregated load of each cluster by introducing the gradient boosting algorithm and extract at least one strongly correlated feature, combine load cluster clustering and gradient boosting algorithm, and significantly improve the accuracy and stability of load prediction. This method not only effectively handles the load volatility and intermittency caused by the access of distributed power sources, but also fine-tuned modeling for different load characteristics, providing a solid data support for the safe and stable operation of the power grid. In addition, through the feature selection and error evaluation steps, the prediction model is further optimized, ensuring the reliability and practicality of the prediction results, which is of great significance to the development of smart grid. Thus, the present application solves the problem that the related art does not fully consider the influence of important factors such as the diversity of load user behavior and characteristics on load prediction, which leads to the inability to effectively handle the load volatility and intermittency caused by the access of distributed power sources, and the inability to fine-tune modeling for different load characteristics.

[0039] Additional aspects and advantages of the application will be set forth in part in the description which follows, and in part will become apparent to those skilled in the art upon examination of the following and / or can be learned by practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0040] The above and / or additional aspects and advantages of the application will become apparent and be well understood from the following description, taken in conjunction with the drawings, wherein:

[0041] Figure 1 A flow chart of a medium and low voltage power distribution network load cluster day-ahead forecasting method according to an embodiment of the application;

[0042] Figure 2 A flow chart of a medium and low voltage power distribution network load cluster day-ahead forecasting method according to an embodiment of the application;

[0043] Figure 3 An example diagram of a medium and low voltage power distribution network load cluster day-ahead forecasting device according to an embodiment of the application;

[0044] Figure 4 A structural diagram of an electronic device according to an embodiment of the application. DETAILED DESCRIPTION

[0045] Embodiments of the application are described in detail below with reference to the attached drawing figures, wherein the same or like component have the same or similar designations. The embodiments described below are merely examples used to explain the application and should not be considered limiting of the application.

[0046] A method and device for medium and low voltage distribution network load cluster day-ahead prediction are described below with reference to the accompanying drawings. In view of the fact that the related art does not fully consider the influence of important factors such as the diversity of load user behavior and characteristics on load prediction, and thus cannot effectively handle the load volatility and intermittency caused by the access of distributed power sources, and cannot perform fine modeling for different load characteristics, the present application provides a method for medium and low voltage distribution network load cluster day-ahead prediction. In this method, at least one cluster group of low voltage transformers can be generated by introducing load cluster clustering. The aggregated load of each cluster can be obtained by introducing a gradient boosting algorithm, and at least one strongly correlated feature can be extracted. The combination of load cluster clustering and gradient boosting algorithm significantly improves the accuracy and stability of load prediction. This method not only effectively handles the load volatility and intermittency caused by the access of distributed power sources, but also performs fine modeling for different load characteristics, providing solid data support for the safe and stable operation of the power grid. In addition, through the feature selection and error evaluation steps, the prediction model is further optimized to ensure the reliability and practicality of the prediction results, which is of great significance to the development of smart grids. Thus, the problem of related art that the way of predicting load does not fully consider the influence of important factors such as the diversity of load user behavior and characteristics on load prediction, and thus cannot effectively handle the load volatility and intermittency caused by the access of distributed power sources, and cannot perform fine modeling for different load characteristics, etc. is solved.

[0047] Specifically, Figure 1 A flowchart of a method for medium and low voltage distribution network load cluster day-ahead prediction is provided in the present application.

[0048] As Figure 1 shown, the method for medium and low voltage distribution network load cluster day-ahead prediction includes the following steps:

[0049] In step S101, at least one load with similarity satisfying a preset condition is identified to generate at least one cluster group of medium and low voltage transformers.

[0050] The cluster group can be understood as a category set formed by grouping load curves with similar characteristics through a clustering algorithm, and the preset condition can be set by a person skilled in the art according to the actual scene.

[0051] Specifically, the power load of a city distribution network is equal to the sum of all medium and low voltage distribution transformer loads. Considering that these distribution transformer loads have certain similarity, in order to more accurately predict these loads, the present application can first identify and cluster loads with similarity. The higher the similarity of the load patterns in the same cluster group, the higher the accuracy of load prediction.

[0052] The embodiment of the application can construct a representative clustering cluster by clustering similar loads, make the regularity of similar loads stronger, realize the transformation of a large number of users to a small number of representative clusters, reduce the amount of training data, and make the model easier to learn, so as to improve the load prediction accuracy.

[0053] In step S102, day-ahead prediction is performed on each clustering cluster of the at least one clustering cluster in a preset prediction mode, to obtain the aggregated load of each clustering.

[0054] It can be understood that day-ahead prediction refers to predicting a variable (such as power load, photovoltaic / wind power output, electricity price, etc.) in the next 24 hours (the next day) at the current time, which is usually used for dispatching plan, market transaction and resource optimization of a power system.

[0055] In actual implementation, the embodiment of the application can use a gradient boosting algorithm to perform day-ahead prediction on the aggregated load of each clustering using the gradient boosting algorithm; the algorithm first decomposes the load curve into a stationary part and a residual part, and then performs prediction respectively. Of course, it should be understood by those skilled in the art that other algorithms can also be used to predict the clustering cluster, and a suitable day-ahead prediction algorithm can be selected according to the actual application scenario.

[0056] After generating the at least one clustering cluster of the medium and low voltage transformer, the embodiment of the application can calculate the aggregated load of each clustering by using the gradient boosting algorithm, and then decompose the aggregated load curve of each clustering into a stationary part and a residual part, wherein the residual part can represent random fluctuations and outliers in the load curve, which not only realizes data fusion of the clustering algorithm and the gradient boosting algorithm, but also fully considers the diversity of load user behavior and characteristics.

[0057] In step S103, at least one strongly correlated feature is extracted using the aggregated load of each clustering, and a final prediction model for day-ahead prediction of the medium and low voltage distribution network load cluster is obtained according to the at least one strongly correlated feature.

[0058] In machine learning and data analysis, a strongly correlated feature can be understood as a variable that has a high statistical correlation with a target variable (prediction object) or other features.

[0059] As an implementable way, the embodiment of the application can explain the correlation between the application features and the predicted load based on the gradient boosting algorithm according to the relative importance of the calculated features, and then extract the strongly correlated features.

[0060] The embodiments of the present application can eliminate low importance features by extracting at least one strongly correlated feature of the aggregated load of each cluster, thereby reducing the probability that low importance features may cause overfitting, improving the prediction of each cluster, and improving the prediction of the aggregated load.

[0061] Optionally, in an embodiment of the present application, identifying at least one load with similar characteristics satisfying a preset condition to generate a cluster group comprises: generating a representative load curve for each area; generating a load curve matrix according to the representative load curve; grouping medium and low voltage transformers with similar load characteristics into the same cluster based on the load curve matrix to obtain the cluster group.

[0062] The related explanation of the cluster group has been described in detail above and will not be repeated here.

[0063] In a specific embodiment of the present application, the power load of the urban power distribution network can be set to be equal to the sum of all N medium / low voltage distribution transformer loads, and the specific implementation steps of the load clustering are as follows:

[0064] (1) First, a representative load curve is generated for each area as the input of the clustering algorithm; let matrix L = [l1, …, ln] ∈ R N ]∈R P×N is the load of the medium / low voltage transformer, where P is the total number of data points recorded within a certain time, and N is the number of medium / low voltage transformers. Each vector is the data of the nth low voltage transformer. Next, the representative load curve (RLP) of each transformer is generated by averaging the load data points of the transformer within a day, which can be represented as where m is the number of averaged data points. In order to standardize the load level difference between different transformers, the RLP is standardized to obtain the standardized RLP matrix where are the mean and standard deviation of , respectively.

[0065] After processing the dataset and converting it into a matrix X, the next step is to apply a clustering algorithm to determine the optimal number of clusters. The core objective of the clustering algorithm is to group the medium and low voltage transformers with similar load characteristics into the same cluster based on the pattern similarity of all the load curves (i.e. representative load profiles, RLPs). By trying different numbers of clusters (K values) and utilizing clustering quality assessment indices (such as CDI (Clustering Dispersion Indicator), MDI (Mean Dispersion Index), DBI (Davies-Bouldin Index), and CHI (Calinski-Harabasz Index / Variance Ratio Criterion)) to evaluate the quality of each clustering solution, the optimal number of clusters can be determined. The clustering algorithm assigns each load curve in the dataset to a unique cluster based on the similarity of their load variation patterns throughout the day.

[0066] CDI can be defined as the ratio of the intra-cluster distance of RLPs in the same cluster k to the distance between the k clusters represented by the centroid set C:

[0067]

[0068] where, is the intra-centroid distance, which reflects the separation degree of different clusters; Ω k is the Kth cluster plan. MDI can calculate the distance ratio between the cluster with the maximum dispersion and the nearest centroid of the K cluster:

[0069]

[0070] where, is the dispersion of the kth cluster, i.e. the distance between data points, c i and c j represent two different centroids, respectively.

[0071] DBI can quantify the maximum similarity between K clusters, which are divided by the interval of the set:

[0072]

[0073] where, represents the average distance between the centroid of the kth cluster set and the data points within the kth cluster set, represents the average distance between the centroid of the ith cluster set and the data points within the ith cluster set, c k represents the centroid of the kth cluster set.

[0074] CHI can compute a ratio that compares the dispersion within cluster k to the dispersion of the centroid representing each cluster:

[0075]

[0076] where Tr(g) is the trace operator, S W is the intra-set scatter matrix, S B is the inter-set scatter matrix. It is defined as:

[0077]

[0078] where the variable is the average of all RLPs in the set X, x represents a single data point in the dataset X, denotes an exponential variable. This exponential penalty uses a large number of clusters K, using the second term of the CHI index.

[0079] Through the technical solutions described above, the embodiments of the present application can first generate a representative load curve for each substation area as the input of the clustering algorithm, and then apply the clustering algorithm to determine the optimal number of clusters. This way, medium and low voltage transformers with similar load characteristics can be grouped into the same cluster, providing more targeted data support for subsequent load forecasting.

[0080] Optionally, in an embodiment of the present application, the day-ahead forecasting is performed on each cluster of the at least one cluster cluster to obtain the aggregated load of each cluster, including: based on the trend and seasonal components and random fluctuations and outliers in the curve, decomposing the representative load curve of each cluster cluster to obtain a stationary part and a residual part; and using a corresponding prediction mode in the preset prediction mode to respectively predict the stationary part and the residual part to obtain the aggregated load of each cluster.

[0081] As an implementable way, the embodiments of the present application can perform load forecasting based on the gradient boosting algorithm. For each cluster, the gradient boosting algorithm is used to perform day-ahead forecasting on its aggregated load. The algorithm first decomposes the load curve into a stationary part and a residual part, and then performs prediction respectively.

[0082] Load curve decomposition can decompose the aggregated load curve of each cluster into a stationary part and a residual part. The stationary part represents the trend and seasonal components in the load curve, while the residual part represents the random fluctuations and outliers in the load curve. The stationary part can be predicted using an autoregressive algorithm, because the autoregressive model is good at capturing the trend and seasonal components in time series data. The residual part can be predicted using the gradient boosting algorithm, because the gradient boosting algorithm can handle nonlinear data and complex patterns.

[0083] Gradient boosting is an ensemble learning method that combines the predictions of multiple decision trees to improve overall prediction accuracy. In each iteration, the algorithm updates the model parameters based on the prediction errors of the current model to minimize the prediction errors. In gradient boosting, the importance of each feature can be calculated by analyzing the extent to which it is used in the prediction process. This feature importance information can be used in subsequent feature selection steps.

[0084] Through the above technical solutions, the embodiments of the present application can distinguish random fluctuation abnormal load from smooth load by decomposing the representative load curve of each cluster group to obtain the smooth part and the residual part, and perform fine modeling and prediction for different load characteristics, thereby providing solid data support for the safe and stable operation of the power grid.

[0085] Optionally, in an embodiment of the present application, it further includes: in the prediction mode corresponding to the residual part, calculating the importance of each feature according to the extent to which it is used in the prediction process; and using Bayesian optimization search to optimize the hyperparameters of the gradient boosting algorithm in the prediction mode.

[0086] Bayesian optimization can be understood as a hyperparameter optimization method based on a probability model, which is specifically used for black box function optimization (target function calculation is expensive, has no analytical expression, and may contain noise). It is widely used in the fields of machine learning hyperparameter tuning, automated experimental design, and engineering optimization.

[0087] In actual execution, the embodiments of the present application can use Bayesian optimization search to optimize the hyperparameters (such as learning rate, number of trees, and tree depth) of the gradient boosting algorithm. It can find the global optimal solution with as few evaluations as possible by constructing a probability proxy model (such as Gaussian process) of the target function and balancing exploration using a sampling function.

[0088] Bayesian optimization is an efficient global optimization algorithm, and the embodiments of the present application can use the Bayesian optimization search algorithm to find a near-optimal combination of hyperparameters in a small number of iterations. By optimizing the hyperparameters, the prediction performance of the gradient boosting algorithm can be further improved.

[0089] Optionally, in an embodiment of the present application, at least one strongly correlated feature is extracted using the aggregated load of each cluster, and a final prediction model for day-ahead prediction of the medium and low voltage distribution network load cluster is obtained according to the at least one strongly correlated feature, including: determining the importance of each feature based on the importance of each feature to obtain at least one strongly correlated feature; and calculating the minimum prediction error and the optimal feature set of each cluster based on the at least one strongly correlated feature to obtain the final prediction model.

[0090] The meaning of strongly correlated features has been explained in detail above and will not be repeated here.

[0091] In one embodiment of the present application, strong correlation feature extraction and prediction error evaluation can be achieved as follows:

[0092] (1) Strong correlation feature extraction: Feature selection is a key step to further improve the prediction accuracy. By removing unimportant features, the complexity of the model can be reduced and overfitting can be prevented. Based on the gradient boosting algorithm, the correlation between the application features and the prediction load can be explained according to the calculated relative importance of the features. The importance of feature j is as follows:

[0093]

[0094] where T represents the decision tree, t is the non-terminal node related to the terminal node J, v t is the split variable related to feature j, is the square improvement of DT after splitting with feature j.

[0095] The importance of all features j used in the gradient boosting algorithm can be calculated as follows:

[0096]

[0097] where I j represents the importance of feature j used in the gradient boosting algorithm, M represents the number of decision trees used in the gradient boosting algorithm, T m represents the mth decision tree in the gradient boosting algorithm, and m is the decision tree number.

[0098] The characteristics initially applied by each cluster are the same time, weather, and distributed photovoltaic related characteristics. However, the gradient boosting algorithm is trained and optimized separately for each cluster.

[0099] The feature selection process starts with calculating the feature importance of each cluster. Subsequently, the algorithm removes the feature with the lowest calculated importance from the feature selection set, and predicts two days of time until the smallest prediction error is calculated, and the optimal feature set of each cluster is determined. This set of optimal features is used to predict a week of consecutive five days. The feature selection process is repeated once a week to consider the changing correlation between features and loads over a longer period of time due to external factors such as changes in weather conditions.

[0100] (2) Prediction error evaluation: Error evaluation is a key step to verify the performance of the model, by calculating the prediction error and using statistical significance test to judge the effectiveness of the improvement.

[0101] Error calculation: Normalized Root Mean Square Error (NRMSE), Mean Absolute Percentage Error (MAPE), and Pearson Correlation Coefficient (PCC) are used to evaluate the accuracy of the predictions. These indicators measure the absolute, relative, and linear correlation between predicted and true values, respectively. For each cluster and aggregated load, the values of the above three indicators are calculated. Then, by comparing the error indicator values under different cluster numbers, the impact of clustering on prediction performance is evaluated.

[0102] Normalized Root Mean Square Error (NRMSE) measures the normalized degree of absolute prediction error relative to the true value range, and the smaller the value, the more accurate the prediction. The calculation formula is as follows:

[0103]

[0104] Where: y i is the true value, is the predicted value, y represents the substation load value, and n is the sample size.

[0105] Range(y) = max(y) - min(y) is the true value range.

[0106] Mean Absolute Percentage Error (MAPE) measures the relative value of the prediction error, reflecting the deviation of the predicted value from the true value, and the smaller the value, the more accurate the prediction. The calculation formula is as follows:

[0107]

[0108] Pearson Correlation Coefficient (PCC) measures the linear correlation between predicted and true values, and the closer the value to 1, the stronger the linear relationship and the more reliable the prediction. The calculation formula is as follows:

[0109]

[0110] Where: is the mean of the true value of the substation load, is the mean of the predicted value of the substation load.

[0111] Statistical significance test: 95% confidence interval is used to determine whether the difference between the prediction error under the no clustering condition and the prediction error under the optimal clustering number K is statistically significant. If the difference value falls outside the confidence interval, it is considered that the improvement is significant.

[0112] The embodiments of the present application can train and optimize each cluster respectively by using the gradient boosting algorithm, identify the relatively unimportant features of each cluster according to the typical behavior and load characteristics of the customers connected to the medium / low voltage transformer, and use the low importance features to further improve the prediction accuracy of each cluster to avoid the overfitting caused by the low importance features. The feature selection process is applied to each cluster to improve the prediction of each cluster and thus improve the prediction of the aggregated load. The error evaluation step further optimizes the prediction model to ensure the reliability and practicability of the prediction results, which is of great significance to the development of smart grid.

[0113] Optionally, in an embodiment of the present application, the importance calculation formula is:

[0114]

[0115] wherein I j represents the importance of feature j used by all M decision trees in the gradient boosting algorithm, M represents the number of decision trees used in the gradient boosting algorithm, T m represents the mth decision tree in the gradient boosting algorithm, and m is the decision tree serial number.

[0116] The embodiments of the present application can construct a feature importance calculation model by using the parameters I j , T m , calculate the importance of each feature by using the constructed feature importance calculation model, filter out the unimportant features, effectively prevent the model from overfitting, and improve the prediction accuracy.

[0117] In summary, as shown in FIG. 1, the medium / low voltage distribution network load cluster day-ahead prediction method according to the embodiments of the present application can be summarized as being mainly realized by the following key steps, which are as follows: Figure 2

[0118] Step S202: Medium / low voltage distribution network load cluster aggregation

[0119] Step S203: Load prediction based on gradient boosting algorithm

[0120] Step S204: Strong correlation feature extraction and prediction error evaluation

[0121] ​The low-voltage power distribution network load cluster day-ahead prediction method provided by the embodiment of the application can generate at least one clustering cluster of low-voltage transformers by introducing load cluster clustering; the aggregated load of each cluster can be obtained by introducing a gradient boosting algorithm, and at least one strongly correlated feature can be extracted, thereby significantly improving the accuracy and stability of load prediction by combining load cluster clustering and the gradient boosting algorithm. The method can not only effectively handle the load volatility and intermittency caused by the access of distributed power sources, but also can fine-tune the modeling for different load characteristics, thereby providing a solid data support for the safe and stable operation of the power grid. In addition, the prediction model is further optimized through the feature selection and error evaluation steps, thereby ensuring the reliability and practicality of the prediction results, which is of great significance for promoting the development of smart grids. Thus, the problem that the related art fails to consider the influence of important factors such as the diversity of load user behaviors and characteristics on load prediction, thereby failing to effectively handle the load volatility and intermittency caused by the access of distributed power sources and failing to fine-tune the modeling for different load characteristics is solved.

[0122] Secondly, refer to the attached Figure 3 The low-voltage power distribution network load cluster day-ahead prediction device provided by the embodiment of the application is described.

[0123] Figure 3 The low-voltage power distribution network load cluster day-ahead prediction device provided by the embodiment of the application is described.

[0124] As Figure 3 shown, the low-voltage power distribution network load cluster day-ahead prediction device 10 comprises an identification module 100, a first prediction module 200, and a second prediction module 300.

[0125] The identification module 100 is configured to identify at least one load with a similarity satisfying a preset condition, to generate at least one clustering cluster of low-voltage transformers.

[0126] The first prediction module 200 is configured to perform day-ahead prediction on each clustering cluster of the at least one clustering cluster in a preset prediction manner, to obtain the aggregated load of each cluster.

[0127] The second prediction module 300 is configured to extract at least one strongly correlated feature from the aggregated load of each cluster, and to obtain a final prediction model for low-voltage power distribution network load cluster day-ahead prediction according to the at least one strongly correlated feature.

[0128] Optionally, in an embodiment of the present application, the identifying module comprises a first generating unit, a second generating unit and a grouping unit; the first generating unit is configured to generate a representative load curve of each substation area; the second generating unit is configured to generate a load curve matrix according to the representative load curve; and the grouping unit is configured to group the medium and low voltage transformers with similar load characteristics into the same cluster based on the load curve matrix to obtain a cluster cluster.

[0129] Optionally, in an embodiment of the present application, the first prediction module 200 comprises a decomposition unit and a prediction unit; the decomposition unit is configured to decompose the representative load curve of each cluster cluster based on the trend and seasonal components and random fluctuations and outliers in the curve to obtain a stationary part and a residual part; and the prediction unit is configured to predict the stationary part and the residual part respectively in a corresponding prediction mode in the preset prediction mode to obtain the aggregated load of each cluster.

[0130] Optionally, in an embodiment of the present application, the first prediction module 200 is further configured to: in the prediction mode corresponding to the residual part, calculate the importance of each feature according to the degree used in the prediction process; and use Bayesian optimization search to optimize the hyperparameters of the gradient boosting algorithm in the prediction mode.

[0131] Optionally, in an embodiment of the present application, the second prediction module 300 comprises a determination unit and a calculation unit; the determination unit is configured to determine the importance of each feature based on the importance of each feature to obtain at least one strongly correlated feature; and the calculation unit is configured to calculate the minimum prediction error and the optimal feature set of each cluster according to the at least one strongly correlated feature to obtain a final prediction model.

[0132] Optionally, in an embodiment of the present application, the calculation formula of the importance is:

[0133]

[0134] wherein, I j represents the importance of feature j used by all M decision trees in the gradient boosting algorithm, M represents the number of decision trees used in the gradient boosting algorithm, T m represents the mth decision tree in the gradient boosting algorithm, m is the serial number of the decision tree.

[0135] It should be noted that the foregoing explanation and description of the medium and low voltage distribution network load cluster day-ahead prediction method embodiment also applies to the medium and low voltage distribution network load cluster day-ahead prediction device of this embodiment, which will not be described here.

[0136] The low-voltage distribution network load cluster day-ahead prediction device provided by the embodiment of the present application can generate at least one clustering cluster of low-voltage transformers by introducing load cluster clustering; the aggregated load of each cluster can be obtained by introducing a gradient boosting algorithm, and at least one strongly correlated feature can be extracted, thereby combining the load cluster clustering and the gradient boosting algorithm to significantly improve the accuracy and stability of load prediction. This method not only effectively handles the load volatility and intermittency caused by the access of distributed power sources, but also performs fine modeling for different load characteristics, thereby providing a solid data support for the safe and stable operation of the power grid. In addition, through the feature selection and error evaluation steps, the prediction model is further optimized, ensuring the reliability and practicality of the prediction results, which is of great significance for promoting the development of smart grids. Thus, the problem that the related art fails to consider the influence of important factors such as the diversity of load user behaviors and characteristics on load prediction, thereby failing to effectively handle the load volatility and intermittency caused by the access of distributed power sources, and failing to perform fine modeling for different load characteristics is solved.

[0137] Figure 4 The structure schematic diagram of the electronic device provided by the embodiment of the present application is provided. The electronic device can include:

[0138] The memory 401, the processor 402, and the computer program stored in the memory 401 and executable on the processor 402.

[0139] The processor 402 implements the low-voltage distribution network load cluster day-ahead prediction method provided in the above embodiments when executing the program.

[0140] Further, the electronic device further includes:

[0141] The communication interface 403 is used for communication between the memory 401 and the processor 402.

[0142] The memory 401 is used to store the computer program executable on the processor 402.

[0143] The memory 401 can include a high-speed RAM memory, and can also include a non-volatile memory such as at least one disk memory.

[0144] If the memory 401, the processor 402 and the communication interface 403 are implemented independently, the communication interface 403, the memory 401 and the processor 402 can be connected with each other through a bus and complete communication between each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For convenience of representation, Figure 4 Only one thick line is used to represent the bus in the figure, but it does not mean that there is only one bus or only one type of bus.

[0145] Optionally, in a specific implementation, if the memory 401, the processor 402 and the communication interface 403 are integrated on a chip, the memory 401, the processor 402 and the communication interface 403 can complete communication between each other through an internal interface.

[0146] The processor 402 can be a Central Processing Unit (CPU), or an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0147] The embodiments of the present application also provide a computer readable storage medium, having stored thereon a computer program, which, when executed by a processor, implements the above medium and low voltage power distribution network load cluster day-ahead forecasting method.

[0148] The embodiments of the present application also provide a computer program product, having stored thereon a computer program, which, when executed by a processor, implements the above medium and low voltage power distribution network load cluster day-ahead forecasting method.

[0149] In the description of the application, reference to "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" means that a particular feature, structure, material, or characteristic being described is included in at least one embodiment or example of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment or example. Furthermore, the described specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples. In addition, the usage of "N" means at least two, for example, two, three or the like, unless explicitly stated otherwise.

[0150] Furthermore, the terms "first", "second", or the like, are used merely as a designation of certain elements or features of the application, and do not imply or connote relative importance or a specific order of precedence. Thus, features defined with "first", "second", etc. can include at least one of the features, either explicitly or implicitly.

[0151] Any process or method descriptions or blocks in flow charts or otherwise described herein represent embodiments of modules, segments, or portions of code which include one or more executable instructions for implementing specific logic functions or steps, and alternate implementations are possible. In some embodiments, the processes or methods described in flow charts or otherwise described herein are intended to be implemented by computer-executable instructions, such as program modules or functional processes, being executed by a computer. Generally, these processes or methods can be implemented as program modules or functional processes including routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. Moreover, those skilled in the art will appreciate that the functions or steps described in the processes or methods described herein can be performed by one or more of the components of the application, either alone or in combination with other components.

[0152] The logic and / or steps represented in the flowcharts and / or described herein, for example, can be considered as a sequence of executable instructions stored in a computer readable medium, which can be executed by an instruction execution system, apparatus or device, such as a computer-based system, a processor-based system, or other system that can fetch the instructions from the instruction execution system, apparatus or device and execute the instructions, or a combination of the above. For the purposes of this specification, a "computer readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus or device. The computer readable medium can be a computer readable storage medium or a computer readable signal medium. The computer readable storage medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a propagation medium. The computer readable signal medium can include, but is not limited to, a computer readable medium that facilitates transfer of the program from one place to another. A specific example of a computer readable medium is a non-transitory computer-readable storage medium. A specific example of a computer readable signal medium is a source or destination of the computer readable medium. Another specific example of a computer readable signal medium is a computer readable signal travelling through space. Thus, a computer readable medium can take many forms of hardware to carry out the program for use by or in connection with the instruction execution system, apparatus or device.

[0153] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, the N steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented in hardware and in another embodiment, the hardware can be implemented using any or a combination of the following technologies, which are each well known in the art: a discrete logic circuit(s) having logic gates for implementing logic functions upon an application of data signals, an application specific integrated circuit having appropriate combinational logic gates, a programmable gate array(s) (PGA), a field programmable gate array (FPGA), etc.

[0154] Those of skill in the art would understand that the steps of the methods carried out above can be carried out wholly or partly by a program instructing relevant hardware, and the program can be stored in a computer readable storage medium, and when executed, includes one or a combination of the steps of the method embodiments.

[0155] In addition, each of the functional units in the various embodiments of the present application can be integrated in one processing module, or each of the units can be physically present separately, or two or more units can be integrated in one module. The integrated module can be realized in the form of hardware or in the form of a software functional module. When the integrated module is realized in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer readable storage medium.

[0156] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A method for day-ahead load clustering of medium and low voltage distribution network, characterized in that, The method comprises the following steps: identifying at least one load with similarity meeting preset conditions to generate at least one cluster group of medium and low voltage transformers; performing day-ahead prediction on each cluster group of the at least one cluster group in a preset prediction mode to obtain aggregated load of each cluster; extracting at least one strongly correlated feature from the aggregated load of each cluster and obtaining a final prediction model for day-ahead prediction of the load cluster of the medium and low voltage distribution network according to the at least one strongly correlated feature.

2. The method of claim 1, wherein, The identification of at least one load with similarity meeting preset conditions to generate a cluster group comprises: generating a representative load curve of each area; generating a load curve matrix according to the representative load curve; grouping the medium and low voltage transformers with similar load characteristics into the same cluster based on the load curve matrix to obtain the cluster group.

3. The method of claim 2, wherein, The day-ahead prediction on each cluster group of the at least one cluster group to obtain the aggregated load of each cluster comprises: decomposing the representative load curve of each cluster group based on trend and seasonal components and random fluctuations and outliers in the curve to obtain a stationary part and a residual part; predicting the stationary part and the residual part respectively in a corresponding prediction mode in the preset prediction mode to obtain the aggregated load of each cluster.

4. The method of claim 3, wherein, Further comprising: in the corresponding prediction mode of the residual part, calculating the importance of each feature according to the degree used in the prediction process; using Bayesian optimization search to optimize the hyperparameters of the gradient boosting algorithm in the prediction mode.

5. The method of claim 4, wherein, The extraction of at least one strongly correlated feature from the aggregated load of each cluster and the obtaining of a final prediction model for day-ahead prediction of the load cluster of the medium and low voltage distribution network according to the at least one strongly correlated feature comprises: determining the importance of each feature based on the importance of each feature to obtain the at least one strongly correlated feature; calculating the minimum prediction error and the optimal feature set of each cluster according to the at least one strongly correlated feature to obtain the final prediction model.

6. The method of claim 5, wherein, The calculation formula of the importance is: where I j represents the importance of feature j used by all M decision trees in the gradient boosting algorithm, M represents the number of decision trees used in the gradient boosting algorithm, T m represents the mth decision tree in the gradient boosting algorithm, m is the decision tree serial number.

7. A medium and low voltage distribution network load cluster day-ahead forecasting device, characterized in that, comprising: an identification module configured to identify at least one load with similarity meeting preset conditions to generate at least one cluster group of medium and low voltage transformers; a first prediction module configured to perform day-ahead prediction on each cluster group of the at least one cluster group in a preset prediction mode to obtain aggregated load of each cluster; a second prediction module configured to extract at least one strongly correlated feature from the aggregated load of each cluster and obtain a final prediction model for day-ahead prediction of the load cluster of the medium and low voltage distribution network according to the at least one strongly correlated feature.

8. An electronic device, comprising: comprising: a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the method for day-ahead prediction of the load cluster of the medium and low voltage distribution network according to any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the method for day-ahead prediction of the load cluster of the medium and low voltage distribution network according to any one of claims 1-6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed for implementing the medium and low voltage power distribution network load cluster day-ahead forecasting method as claimed in any one of claims 1-6.

Images