Load regulation and control method based on regional load gap prediction

Abstract

The invention provides a load regulation and control method based on regional load gap prediction, and belongs to the technical field of power grid dispatching. Comprising the following steps: acquiring historical data of a power grid node, and preprocessing the historical data; performing regional division on the target power grid according to the difference of the historical meteorological data of the power grid nodes; constructing an independent data set for each region according to historical data of the power grid nodes; constructing an independent load gap prediction model for each region, and completing model training by using a corresponding data set; predicting the load gap of the region according to the load gap prediction model of each region; and carrying out load regulation and control according to the load gaps of different areas. According to the method, through regional accurate load gap prediction, the trans-regional power, the standby power supply and the user side flexible load can be flexibly allocated according to the gap scale, the key load demand is preferentially guaranteed, and the resource utilization efficiency is optimized while the power grid stability is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of power grid dispatching, and particularly relates to a load regulation method based on regional load gap prediction. Background Art

[0002] Load regulation is one of the core tasks in the operation of the power system, aiming to achieve real-time balance between power grid supply and demand by dynamically adjusting power load distribution or power generation output of power sources, and ensuring the safety, reliability, and economy of power supply. The formulation of load regulation strategies depends on the prediction of load gaps. In recent years, the penetration rate of distributed power sources represented by photovoltaic and wind power in the power grid has increased rapidly. Their output highly depends on meteorological conditions and has obvious intermittency, volatility, and randomness. On the one hand, the spatio-temporal differences of distributed power sources exacerbate the uncertainty of regional load characteristics and power grid dynamic balance. On the other hand, extreme weather events may simultaneously trigger a sharp increase in load demand and a sharp drop in new energy output, resulting in the mutability of load gaps.

[0003] Traditional load gap prediction models are difficult to effectively integrate multi-dimensional data (such as distributed power source output, meteorological characteristics, geographical connection relationships), and lack the ability to capture characteristics at key time points (such as extreme weather periods), resulting in large errors in prediction results. Therefore, there is an urgent need for a method to improve the accuracy of load gap prediction to provide reliable support for precise load regulation. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a load regulation method based on regional load gap prediction, a load gap prediction method based on regional modeling, capable of comprehensively integrating multi-source heterogeneous data and incorporating spatio-temporal feature extraction and attention mechanism, to provide reliable support for load regulation.

[0005] To solve one or part or all of the above technical problems, the technical solution adopted by the present invention is: A load regulation method based on regional load gap prediction, comprising: obtaining historical data of power grid nodes and preprocessing the historical data; dividing the target power grid into regions according to the differences in historical meteorological data of power grid nodes; constructing an independent data set for each region according to the historical data of power grid nodes; constructing an independent load gap prediction model for each region and completing model training using the corresponding data set; predicting the load gap of each region according to the load gap prediction model of each region; and performing load regulation according to the load gaps of different regions.

[0006] Further, the method for regional division of the target power grid according to the differences in historical meteorological data of power grid nodes includes: obtaining the statistical characteristics of each meteorological index in the meteorological data, and constructing the meteorological data feature vector of each power grid node; constructing the meteorological difference degree matrix between power grid nodes according to the meteorological data feature vector; constructing the geographical distance matrix and connection relationship matrix between power grid nodes; constructing the comprehensive difference degree matrix according to the meteorological difference degree matrix, the geographical distance matrix and the connection relationship matrix; and performing regional division on the power grid nodes according to the comprehensive difference degree matrix.

[0007] Further, the method for performing regional division on power grid nodes according to the comprehensive difference degree matrix includes: (a) Initializing each power grid node as an independent region, defining the difference degree between two regions as the maximum value of the comprehensive difference degrees between all nodes in the two regions, and initializing the geographical distance threshold as the 90th percentile of the geographical distances between nodes; (b) Obtaining the two regions with the smallest difference degree among all regions as candidate regions; (c) If the maximum geographical distance between all nodes in the candidate regions does not exceed the geographical distance threshold and there is a transmission line connection between the regions, then merging the candidate regions into a new region; otherwise, selecting the two regions with the second largest difference degree from the remaining regions as candidate regions, and repeating step (c); (e) Repeating steps (b)-(e) until the number of regions reaches a preset value; if all candidate regions cannot be merged and the number of regions has not reached the preset value, and the current geographical distance threshold does not exceed 2 times the initial geographical distance threshold, then setting the geographical distance threshold to 1.2 times the current value, and repeating steps (b)-(e).

[0008] Further, the comprehensive difference degree matrix , where N is the number of power grid nodes, , where α, β, and γ are weights, , W is the meteorological difference degree matrix, G is the geographical distance matrix, and C is the connection relationship matrix.

[0009] Further, the meteorological difference degree matrix , and the matrix element W(i,j) represents the meteorological difference degree between nodes i and j, , is the value of the b-th statistical index of the a-th meteorological index of node i, n is the number of meteorological indices, and m is the number of statistical indices.

[0010] Further, the meteorological data feature vector of node i is expressed as , represents the statistical characteristics of the n-th meteorological index of node i, ; The statistical features include mean, variance, maximum value, minimum value, and trend slope.

[0011] Furthermore, the geographical distance matrix , where the matrix element G(i,j) represents the Euclidean distance between the geographical locations of nodes i and j; the connection relationship matrix , where the matrix element C(i,j) represents whether there is a connection between nodes i and j through a transmission line.

[0012] Furthermore, the load gap prediction model adopts a hybrid model of LSTM+Attention, including an input layer, an embedding layer, at least two LSTM layers, an Attention layer, and an output layer connected in sequence.

[0013] Furthermore, the input data of the load gap prediction model is the installed capacity data of distributed power sources, the output data of traditional power sources, meteorological data, the annual day number of the sampling moment, and the rest day identifier, and the output data is the hourly load gap data for the next 72 hours.

[0014] Furthermore, the historical data includes the load data of the node, the output data of distributed power sources within the node, the installed capacity data of distributed power sources within the node, the output data of traditional power sources within the node, and the meteorological data of the node.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: Through multi-source data integration, the present invention constructs a comprehensive difference matrix by combining meteorological statistical features, geographical distances, and grid connection relationships, and divides the power grid area according to the comprehensive difference matrix to ensure the consistency of meteorological characteristics within the area, thereby laying a foundation for regional independent modeling and significantly improving the local adaptability of the prediction model.

[0016] The present invention adopts a hybrid model of LSTM+Attention. It not only captures long-term time series dependence relationships through the LSTM layer but also dynamically identifies the key impacts of extreme weather and special moments on the load gap through the attention mechanism, enhancing the model's ability to extract complex non-linear features and significantly improving the spatio-temporal resolution and accuracy of future load gap prediction.

[0017] Through accurate load gap prediction by region, at the level of control strategies, through a hierarchical control mechanism and dynamic threshold adjustment, it is possible to flexibly allocate cross-regional power, standby power sources, and user-side flexible loads according to the gap scale, prioritize the guarantee of key load demands, and optimize the resource utilization efficiency while ensuring the stability of the power grid.

[0018] Especially in the context of high penetration of new energy, the present invention effectively addresses the problem of sudden changes in load gaps caused by fluctuations in new energy output and extreme weather by integrating multi-dimensional data such as distributed power output, weather forecasts, and geographical connection relationships, significantly enhancing the dynamic balance ability and risk resistance ability of the power system in complex scenarios. Detailed implementation manners

[0019] To better understand the present invention, the content of the present invention will be further clearly elaborated below in conjunction with embodiments. However, the protection scope of the present invention is not limited to the following embodiments. In the following description, a large number of specific details are given to provide a more thorough understanding of the present invention. However, it is obvious to those skilled in the art that the present invention can be implemented without one or more of these details.

[0020] Embodiment 1: The purpose of this embodiment is to provide a load regulation method based on regional load gap prediction, including the following steps: Step S1, obtain historical data of grid nodes and preprocess the historical data.

[0021] The historical data includes load data of nodes, distributed power output data within nodes, distributed power installed capacity data within nodes, traditional power output data within nodes, and meteorological data of nodes. The distributed power installed capacity data is installed capacity data by type, such as distributed photovoltaic installed capacity data and distributed wind power installed capacity data.

[0022] Subsequently, perform outlier and missing value filling, resampling at the same sampling frequency and sampling time, and normalization processing on the obtained historical data. The sampling interval of the resampled data is 15 minutes.

[0023] Step S2, divide the target power grid into regions according to the differences in historical meteorological data of grid nodes.

[0024] Specifically, this step includes: S21. Obtain the statistical characteristics of each meteorological index in the meteorological data and construct the meteorological data feature vector of each grid node.

[0025] The statistical characteristics include m statistical indexes such as mean, variance, maximum value, minimum value, and trend slope. The meteorological data includes n meteorological indexes such as temperature, humidity, wind direction, wind speed, visibility distance, air pressure, solar irradiance, snow depth, precipitation, and cloud type.

[0026] The meteorological data feature vector of node i is expressed as , represents the statistical characteristic of the nth meteorological index of node i, , Represents the value of the m-th statistical index of the n-th meteorological index of node i.

[0027] S22. Construct a meteorological difference matrix between power grid nodes based on the meteorological data eigenvector.

[0028] Meteorological difference matrix , where N is the number of power grid nodes, and the matrix element W(i, j) represents the meteorological difference between nodes i and j, which is represented by the Euclidean distance of the meteorological data eigenvector between the nodes. .

[0029] S23. Construct a geographical distance matrix and a connection relationship matrix between power grid nodes.

[0030] Geographical distance matrix , and the matrix element G(i, j) represents the Euclidean distance between the geographical locations of nodes i and j, which can be calculated according to the longitude and latitude coordinates of the nodes.

[0031] Connection relationship matrix , and the matrix element C(i, j) represents whether there is a connection through a transmission line between nodes i and j, which is represented by 0 or 1.

[0032] S24. Construct a comprehensive difference matrix based on the meteorological difference matrix, the geographical distance matrix, and the connection relationship matrix.

[0033] Comprehensive difference matrix , and the matrix element S(i, j) represents the comprehensive difference between nodes i and j. , where α is the weight of the meteorological difference, β is the weight of the geographical distance, and γ is the weight of the connection relationship. .

[0034] S25. Divide the power grid nodes into regions according to the comprehensive difference matrix.

[0035] (a) Initialize each power grid node as an independent region, define the difference between two regions as the maximum value of the comprehensive differences between all nodes in the two regions; initialize the geographical distance threshold as the 90th percentile of the geographical distances between nodes. (b) Obtain the two regions with the smallest difference among all regions as candidate regions. (c) If the maximum geographical distance between all nodes in the candidate regions does not exceed the geographical distance threshold and there is a transmission line connection between the regions, then merge the candidate regions into a new region; otherwise, select the two regions with the second largest difference from the remaining regions as candidate regions, and repeat step (c); the geographical distance between nodes can be directly obtained through the geographical distance matrix, and the connection relationship between regions can be directly obtained through the connection relationship matrix. (e) Repeat steps (b)-(e) until the number of regions reaches a preset value; if all candidate regions cannot be merged, the number of regions has not reached the preset value, and the current geographical distance threshold does not exceed 2 times the initial geographical distance threshold, then set the geographical distance threshold to 1.2 times the current value and repeat steps (b)-(e).

[0036] Step S3: Construct an independent data set for each region based on the historical data of power grid nodes.

[0037] Construct the data set for the region based on the itemized mean of the historical data of all power grid nodes in the region. The data set uses the same sampling frequency as the resampled historical data. The data set includes load gap data, distributed power generation installed capacity data, traditional power generation output data, meteorological data, the annual day number of the sampling moment, and the rest day flag of the sampling moment. The load gap is the difference between the load data and the distributed power generation output data and the traditional power generation output data.

[0038] Step S4: Construct an independent load gap prediction model for each region and complete model training using the corresponding data set.

[0039] The load gap prediction model uses a hybrid model of LSTM+Attention, including an input layer, an embedding layer, at least two LSTM layers, an Attention layer, and an output layer connected in sequence. The input layer receives input data; the embedding layer converts the input data into a vector and outputs it to the LSTM layer; the LSTM layer can gradually extract deep time series features and output the hidden state at each time step; the Attention layer can automatically identify the relevance of each hidden state to the target task, identify key time points, and assign higher weights to key time points, and generate a global context vector through weighted summation; the output layer outputs the prediction result according to the global context vector.

[0040] The input data of the load gap prediction model are distributed power generation installed capacity data, traditional power generation output data, meteorological data, the annual day number of the sampling moment, and the rest day flag, and the output data are the hourly load gap data for the next 72 hours.

[0041] Step S5: Predict the load gap of each region according to the load gap prediction model of each region.

[0042] First, obtain the hourly meteorological forecast data for the next 72 hours for each region and resample it at the sampling frequency of the data set by interpolation; the traditional power generation output data at each sampling moment uses the rated output of the traditional power source; then input the resampled data of the next-day meteorological forecast data and the traditional power generation output data into the load gap prediction model of each region to obtain the hourly load gap data for the next 72 hours.

[0043] Step S6, performing load regulation according to the load gaps in different areas.

[0044] Based on the predicted load gap in each region, a hierarchical control approach is adopted. Surplus electricity is allocated through cross-regional support of the power grid or the activation of backup power sources according to the size of the gap. Priority is given to key loads such as hospitals and transportation. Short-term rotation shutdowns are adopted for non-critical areas according to changes in the load gap at different times. Industrial users suspend the operation of some high-energy-consuming equipment as per the agreement, and the load interruption compensation plan is implemented in conjunction with contracted industrial and commercial users. Flexible loads such as air conditioning and energy storage are regulated through smart devices to carry out load control while minimizing the impact on users.

[0045] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Other modifications or equivalent substitutions made to the technical solution of the present invention by ordinary technicians in the field should be included in the scope of the claims of the present invention as long as they do not depart from the spirit and scope of the technical solution of the present invention.

Claims

1. A load control method based on regional load gap prediction, characterized in that: include: Obtain historical data of power grid nodes and pre-process the historical data; The target power grid is divided into regions according to the differences in historical meteorological data of power grid nodes; Based on the historical data of the power grid nodes, an independent dataset is constructed for each region; Build an independent load gap prediction model for each region and use the corresponding data set to complete model training; Predict the load gap in each area according to the load gap prediction model of the area; Load regulation is carried out according to the load gap in different areas.

2. The load regulation method based on regional load gap prediction according to claim 1 is characterized in that: Methods for dividing the target power grid into regions based on the differences in historical meteorological data of power grid nodes include: Obtain the statistical characteristics of each meteorological indicator in the meteorological data and construct the meteorological data feature vector of each power grid node; Constructing a meteorological difference matrix between power grid nodes according to the meteorological data feature vector; Construct the geographical distance matrix and connection relationship matrix between power grid nodes; Constructing a comprehensive difference matrix according to the meteorological difference matrix, the geographic distance matrix and the connection relationship matrix; The power grid nodes are divided into regions according to the comprehensive difference matrix.

3. The load regulation method based on regional load gap prediction according to claim 2 is characterized in that: The method for dividing the grid nodes into regions according to the comprehensive difference matrix includes: (a) Each grid node is initialized as an independent region, the difference between two regions is defined as the maximum value of the comprehensive difference between all nodes in the two regions, and the initial geographical distance threshold is the 90th percentile of the geographical distance between nodes; (b) Obtain the two regions with the smallest difference among all regions as candidate regions; (c) If the maximum geographical distance between all nodes in the candidate area does not exceed the geographical distance threshold, and there is a transmission line connection between the areas, the candidate areas are merged into a new area; otherwise, two areas with the second largest difference are selected from the remaining areas as candidate areas, and step (c) is repeated; (e) Repeat (b)-(e) until the number of regions reaches the preset value; if all candidate regions cannot be merged and the number of regions does not reach the preset value, and the current geographic distance threshold does not exceed 2 times the initial geographic distance threshold, then set the geographic distance threshold to 1.2 times the current value and repeat (b)-(e).

4. The load regulation method based on regional load gap prediction according to claim 2 is characterized in that: The comprehensive difference matrix , N is the number of grid nodes, , where α, β, and γ are weights, , W is the meteorological difference matrix, G is the geographic distance matrix, and C is the connection relationship matrix.

5. The load regulation method based on regional load gap prediction according to claim 4 is characterized in that: The meteorological difference matrix , the matrix element W(i,j) represents the difference in weather conditions between nodes i and j, , is the value of the bth statistical indicator of the ath meteorological indicator at node i, n is the number of meteorological indicators, and m is the number of statistical indicators.

6. The load regulation method based on regional load gap prediction according to claim 5 is characterized in that: The meteorological data feature vector of node i is expressed as , represents the statistical characteristics of the nth meteorological index of node i, ; The statistical characteristics include mean, variance, maximum value, minimum value, and trend slope.

7. The load regulation method based on regional load gap prediction according to claim 4 is characterized in that: The geographic distance matrix , the matrix element G(i,j) represents the Euclidean distance between the geographical locations of nodes i and j; the connection relationship matrix , the matrix element C(i,j) indicates whether there is a connection between nodes i and j through a transmission line.

8. The load regulation method based on regional load gap prediction according to claim 1 is characterized in that: The load gap prediction model adopts a hybrid model of LSTM+Attention, including an input layer, an embedding layer, at least two LSTM layers, an Attention layer and an output layer connected in sequence.

9. The load regulation method based on regional load gap prediction according to claim 8 is characterized in that: The input data of the load gap prediction model are distributed power generation installed capacity data, traditional power generation output data, meteorological data, annual day number and rest day mark of the sampling time, and the output data are hourly load gap data for the next 72 hours.

10. The load regulation method based on regional load gap prediction according to claim 1 is characterized in that: The historical data includes the load data of the node, the output data of the distributed power source in the node, the installed capacity data of the distributed power source in the node, the output data of the traditional power source in the node and the meteorological data of the node.

Images