A power time series data fusion method for power grid planning optimization

Abstract

The application relates to the technical field of data fusion, and discloses a power time series data fusion method for power grid planning optimization, which comprises the following steps: collecting real-time load rates of a power grid, wind power and photovoltaic power of all power stations, and meteorological data of all meteorological stations, determining corrected meteorological data of the power stations, and respectively establishing a wind power feature matrix and a photovoltaic feature matrix; constructing a dynamic adjustment adjacency matrix, determining a fusion power value according to root mean square errors of wind power and photovoltaic power of the power stations, real-time load rates of the power grid, wind power and photovoltaic power of all the power stations, determining a composite loss function according to differences between the wind power and differences between the photovoltaic power at adjacent collection time points; and obtaining a data fusion result according to the dynamic adjustment adjacency matrix, the composite loss function, the wind power feature matrix and the photovoltaic feature matrix. The application can improve the accuracy of power time series data fusion for power grid planning optimization.

Description

Technical Field

[0001] This application relates to the field of data fusion technology, specifically to a power time-series data fusion method for power grid planning optimization. Background Technology

[0002] Power time-series data fusion for power grid planning optimization integrates and processes power time-series data from different sources and time scales to improve the accuracy and optimization effectiveness of power grid planning. Data fusion improves data quality, reduces data redundancy, and provides more accurate and comprehensive basic data for power grid planning, thereby enhancing the scientific rigor and effectiveness of the plan. New energy power is characterized by volatility and intermittency; therefore, the supply of new energy power from different sources is unstable, requiring flexible adjustments on the load side and energy storage side using traditional thermal power. Thus, new energy power and thermal power have spatiotemporal complementarity, and new energy power from different sources also exhibits spatiotemporal complementarity.

[0003] Existing methods often overlook the complex spatiotemporal complementary relationship between wind power and photovoltaic power generation when fusing power time series data, and lack constraints on the upper limit of power, resulting in inaccurate power time series data fusion results for grid planning optimization. Summary of the Invention

[0004] This application provides a power time-series data fusion method for power grid planning optimization to address the problem of inaccurate power time-series data fusion results caused by neglecting the spatiotemporal complementary effect between wind power generation and photovoltaic power generation. The specific technical solution adopted is as follows:

[0005] One embodiment of this application provides a power time-series data fusion method for power grid planning optimization, the method comprising the following steps:

[0006] The real-time load rate of the power grid at different collection times, as well as the wind power and photovoltaic power of all power plants and the meteorological data of all meteorological stations, are collected. Based on the distance between the power plants and all meteorological stations and the meteorological data of all meteorological stations, the corrected meteorological data of the power plants are determined. Combining the wind power, photovoltaic power and location of all power plants at all collection times, the wind power characteristic matrix and photovoltaic characteristic matrix of the power plants are established respectively.

[0007] Based on the wind power and photovoltaic power of all power plants, a dynamically adjusted adjacency matrix is ​​constructed. Based on the root mean square error of the wind power and photovoltaic power of the power plants, the real-time load rate of the grid, and the wind power and photovoltaic power of all power plants, the fused power value is determined. Based on the fused power value, the difference between wind power and photovoltaic power at all adjacent acquisition times, the composite loss function is determined.

[0008] The data fusion results are obtained by dynamically adjusting the adjacency matrix, composite loss function, wind power characteristic matrix, and photovoltaic characteristic matrix of the power station.

[0009] Furthermore, the specific method for determining the corrected meteorological data for the power station based on the distance between the power station and all meteorological stations and the meteorological data from all meteorological stations includes:

[0010] Meteorological data includes wind speed and irradiance; any one of these meteorological data types can be recorded as the target meteorological data.

[0011] The sum of the distances between the power station and all weather stations is recorded as the cumulative distance between the power station and the weather stations. The ratio of the distance between the power station and the weather stations to the cumulative distance between the power station and the weather stations is recorded as the distance weight between the power station and the weather stations.

[0012] The distance weight between the power station and the meteorological station is used as the weight of the target meteorological data of the meteorological station. The target meteorological data of all meteorological stations are weighted and summed, and the result of the weighted sum is recorded as the corrected target meteorological data of the power station.

[0013] Furthermore, the wind power feature matrix is ​​specifically as follows:

[0014] The wind power feature matrix contains three columns. The wind power of all power stations at the same acquisition time is arranged into one column to obtain the wind power column at the same acquisition time. The wind power of all acquisition times is arranged into one column according to the order of acquisition time to obtain the first column of the wind power feature matrix.

[0015] Arrange the corrected wind speeds of all power plants at the same acquisition time into a column to obtain the column of corrected wind speeds at the same acquisition time. Arrange the corrected wind speeds of all acquisition times into a column according to the order of acquisition time to obtain the second column of the wind power feature matrix.

[0016] Arrange the locations of all power plants into a column, obtain the location column, and repeatedly fill the third column of the wind power feature matrix with the location column until the third column of the wind power feature matrix is ​​completely filled.

[0017] Furthermore, the photovoltaic feature matrix is ​​specifically as follows:

[0018] The photovoltaic feature matrix contains three columns. The photovoltaic power of all power plants at the same acquisition time is arranged into one column to obtain the photovoltaic power at the same acquisition time. The photovoltaic power of all acquisition times is arranged into one column according to the order of acquisition time to obtain the first column of the photovoltaic feature matrix.

[0019] Arrange the corrected irradiance of all power plants at the same acquisition time into a column to obtain the column of corrected irradiance at the same acquisition time. Arrange the corrected irradiance of all acquisition times into a column according to the order of acquisition time to obtain the second column of the photovoltaic feature matrix.

[0020] Arrange the locations of all power plants into a column, obtain the location column, and repeatedly fill the third column of the photovoltaic feature matrix with the location column until the third column of the photovoltaic feature matrix is ​​completely filled.

[0021] Furthermore, the specific details of dynamically adjusting the adjacency matrix are as follows:

[0022] Based on the total wind power output of two different power plants, determine the similarity of their wind power output; based on the total photovoltaic power output of two different power plants, determine the similarity of their photovoltaic power output.

[0023] Dynamically adjust the element values ​​corresponding to two different power plants in the adjacency matrix:

[0024]

[0025] in, This indicates that the power plants in the adjacency matrix are dynamically adjusted. and power station The corresponding element value; , , These represent the preset first influence weight parameter, second influence weight parameter, and third influence weight parameter, respectively, and the sum of the first influence weight parameter, second influence weight parameter, and third influence weight parameter is 1; Represents an exponential function with the natural constant as the base; Indicates power station and power station The distance between them; This indicates the preset attenuation parameter; Indicates power station and power station The similarity of wind power output; Indicates power station and power station The photovoltaic power similarity.

[0026] Furthermore, the formula for calculating the fusion power value is:

[0027]

[0028] in, Indicates the fusion power value; Indicates the weight of wind power output; Indicates the photovoltaic power weight; This represents the sum of wind power output from all power plants. This represents the sum of the photovoltaic power output of all power plants. This represents the weighting adjustment factor.

[0029] Furthermore, the specific methods for obtaining the wind power weight and the photovoltaic power weight are as follows:

[0030] The sum of the root mean square error of wind power and the number 1 is used as the denominator, and the number 1 is used as the numerator. This value is denoted as the wind power weight.

[0031] The sum of the root mean square error of photovoltaic power and the number 1 is used as the denominator, and the number 1 is used as the numerator. This fraction is denoted as the photovoltaic power weight.

[0032] Furthermore, the specific method for obtaining the weight adjustment factor is as follows:

[0033] The absolute value of the difference between the real-time load rate of the power grid and the preset economic security threshold of the power grid, plus the number 1, is used as the denominator, and the number 1 is used as the value of the fraction in the numerator, which is denoted as the weight adjustment factor.

[0034] Furthermore, the composite loss function is specifically as follows:

[0035]

[0036] In the formula, Represents the composite loss function; , , These represent the preset first loss weight parameter, second loss weight parameter, and third loss weight parameter, respectively, and the sum of the first loss weight parameter, second loss weight parameter, and third loss weight parameter is 1; This represents the mean square error of the ST-GCN model; Indicates the first The absolute power difference at each acquisition moment, specifically, will be the power difference at the first acquisition moment. The data collection time and the first The absolute value of the difference in wind power at each data collection time is denoted as the nth data collection time. The difference in adjacent wind power at the i-th acquisition time will be used to... The data collection time and the first The absolute value of the difference in photovoltaic power at the nth acquisition time is denoted as the nth acquisition time. The difference in photovoltaic power between adjacent data collection times will be used to determine the next data collection time. The sum of the differences in adjacent wind power and adjacent photovoltaic power at the i-th data collection time is denoted as the i-th data collection time. The absolute difference in power at each acquisition moment; This indicates the number of all data collection moments; This indicates the preset safe carrying capacity threshold of the power grid; This represents the maximum value function.

[0037] Furthermore, the specific method for obtaining the data fusion result based on dynamically adjusting the adjacency matrix, composite loss function, wind power feature matrix, and photovoltaic feature matrix of the power station includes:

[0038] The composite loss function is used as the loss function of the ST-GCN model, and the dynamically adjusted adjacency matrix is ​​used as the dynamic adjacency matrix in the ST-GCN model. The wind power feature matrix and photovoltaic feature matrix of the power station are input into the ST-GCN model to obtain the fusion result.

[0039] The beneficial effects of this application are:

[0040] This application first considers that meteorological data directly affects the wind power output of wind turbines. However, there is a significant distance between meteorological stations and wind turbine power plants. To improve the accuracy of the power fusion value obtained after fusing power time-series data, precise mapping is performed based on the meteorological data of each meteorological station to determine the accurate meteorological data for each power plant's location, i.e., determining the corrected meteorological data for the power plant. Combining the wind power, photovoltaic power, and location of all power plants at all data collection times, wind power characteristic matrices and photovoltaic characteristic matrices for each power plant are established. Due to the high proportion of wind and photovoltaic power integrated into the grid, wind power and photovoltaic power exhibit dual fluctuation characteristics of minute-level meteorological abrupt changes and seasonal energy structure shifts. To capture the correlation between wind power and photovoltaic power of different power plants, as well as the impact of different power plant locations on their wind power and photovoltaic power, a dynamically adjusted adjacency matrix is ​​constructed. Furthermore, the wind power and photovoltaic power of the power plants are adjusted accordingly. The influence weight of the photovoltaic power of the power plant on the fused power value is adjusted so that when one power data is affected, the fused power value is more inclined to be determined based on the other power data, thus improving the accuracy of the fusion result. At the same time, the real-time load rate of the power grid is combined in the process of determining the fused power value, so that the real-time state of the power grid can be embedded into the data fusion process. The real-time carrying capacity of the power grid is considered in the data fusion process to avoid the problem of instantaneous overload of the power grid line caused by the drastic fluctuation of wind power and photovoltaic power due to weather. Then, a composite loss function is determined to balance the prediction accuracy, power grid security constraints and wind-solar synergy effect. Finally, the data fusion result is obtained by dynamically adjusting the adjacency matrix, composite loss function, wind power feature matrix and photovoltaic feature matrix of the power plant. This solves the problem of inaccurate power time series data fusion results in power grid planning optimization caused by ignoring the spatiotemporal complementary effect between wind power generation and photovoltaic power generation, thus improving the accuracy of power time series data fusion in power grid planning optimization. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 A schematic flowchart of a power time-series data fusion method for power grid planning optimization provided in one embodiment of this application;

[0043] Figure 2 This is a flowchart illustrating the process of acquiring corrected meteorological data according to one embodiment of this application. Detailed Implementation

[0044] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0045] Please see Figure 1 The diagram illustrates a flowchart of a power time-series data fusion method for power grid planning optimization provided in one embodiment of this application. The method includes the following steps:

[0046] Step S001: Collect the real-time load rate of the power grid at different collection times, as well as the wind power and photovoltaic power of all power plants and the meteorological data of all meteorological stations. Based on the distance between the power plants and all meteorological stations and the meteorological data of all meteorological stations, determine the corrected meteorological data of the power plants. Combine the wind power, photovoltaic power and location of all power plants at all collection times to establish the wind power characteristic matrix and photovoltaic characteristic matrix of the power plants respectively.

[0047] Wind power and solar power output were collected from different power plants, and wind speed and irradiance were collected from different weather stations. Both wind speed and irradiance were recorded as meteorological data. The location of each weather station was collected using the GPS global positioning system. The real-time load rate of the power grid was collected using smart meters.

[0048] Preferably, in one embodiment of this application, when collecting real-time load rate, wind power, photovoltaic power, and meteorological data, this embodiment collects real-time load rate, wind power, photovoltaic power, and meteorological data every 30 seconds, for a total of 100 collection points. In practical applications, as other implementation methods, implementers can determine the sampling frequency and number of samples for real-time load rate, wind power, photovoltaic power, and meteorological data according to actual conditions; this application does not impose any special restrictions.

[0049] To avoid missing data affecting subsequent analysis, missing real-time load factor, wind power, photovoltaic power, and meteorological data are filled in. This embodiment selects median imputation to fill in the missing real-time load factor, wind power, photovoltaic power, and meteorological data. In practical applications, as other implementation methods, while achieving the goal of filling in missing data, implementers can use other existing methods such as median imputation, mode imputation, random sampling imputation, K-nearest neighbor imputation, and regression imputation to fill in the missing data; this application does not impose any special limitations.

[0050] It should be noted that, for ease of calculation, all wind power, photovoltaic power, and meteorological data involved in the calculation in this embodiment have undergone data preprocessing to eliminate the influence of dimensions. This embodiment uses the Z-Score standard normalization method to perform dimensionless processing on wind power, photovoltaic power, and meteorological data respectively. In practical applications, implementers can use other methods such as the existing technology of maximum-minimum normalization for dimensionless processing, which are not limited here.

[0051] Meteorological data directly affects the wind power output of wind turbines. Wind power plants are usually located in relatively open and spacious areas, while meteorological stations are often far away from wind power plants. In order to improve the accuracy of the power fusion value obtained after the fusion of power time series data, it is necessary to accurately map the meteorological data of each meteorological station to determine the accurate meteorological data of each power plant location.

[0052] The cumulative distance between the power station and all meteorological stations is recorded as the cumulative distance between the power station and the meteorological stations. The ratio of the distance between the power station and the meteorological stations to the cumulative distance between the power station and the meteorological stations is recorded as the distance weight between the power station and the meteorological stations. Any type of meteorological data is recorded as the target meteorological data. The distance weight between the power station and the meteorological stations is used as the weight of the target meteorological data of the meteorological stations. The target meteorological data of all meteorological stations are weighted and summed. The result of the weighted sum is recorded as the corrected target meteorological data of the power station.

[0053] Corrected meteorological data for any type of meteorological data can be obtained using the same method. The flowchart for obtaining corrected meteorological data is as follows: Figure 2 As shown.

[0054] It is understandable that meteorological data includes wind speed and irradiance. Therefore, the target meteorological data for power plant correction includes two types of data: wind speed and irradiance. Thus, the corrected wind speed and corrected irradiance for each power plant can be obtained separately. The corrected wind speed and corrected irradiance for each power plant are the accurate wind speed and irradiance at the power plant location.

[0055] Based on the corrected wind speed, corrected irradiance, wind power, photovoltaic power, and location of all power stations at all data acquisition times, establish the wind power characteristic matrix and photovoltaic characteristic matrix of the power stations.

[0056] Preferably, as an embodiment of this application, the wind power feature matrix comprises three columns: the wind power of all power stations at the same acquisition time is arranged into one column to obtain the wind power column at the same acquisition time; the wind power at all acquisition times is arranged into one column according to the order of acquisition time to obtain the first column of the wind power feature matrix; the corrected wind speed of all power stations at the same acquisition time is arranged into one column to obtain the corrected wind speed column at the same acquisition time; the corrected wind speed at all acquisition times is arranged into one column according to the order of acquisition time to obtain the second column of the wind power feature matrix; the location of all power stations is arranged into one column to obtain the location column; and the location column is filled sequentially in the third column of the wind power feature matrix until there are no empty positions in the third column of the wind power feature matrix.

[0057] Preferably, as an embodiment of this application, the photovoltaic feature matrix comprises three columns: the photovoltaic power of all power plants at the same acquisition time is arranged into one column to obtain the photovoltaic power column at the same acquisition time; the photovoltaic power at all acquisition times is arranged into one column according to the order of acquisition time to obtain the first column of the photovoltaic feature matrix; the corrected irradiance of all power plants at the same acquisition time is arranged into one column to obtain the corrected irradiance column at the same acquisition time; the corrected irradiance at all acquisition times is arranged into one column according to the order of acquisition time to obtain the second column of the photovoltaic feature matrix; the location of all power plants is arranged into one column to obtain the location column; and the location column is filled sequentially in the third column of the photovoltaic feature matrix until there are no empty positions in the third column of the photovoltaic feature matrix.

[0058] In this model, the same row of the wind power feature matrix and the photovoltaic feature matrix of the power station corresponds to the same power station, and the power stations corresponding to each data in each column are arranged in the same order.

[0059] Thus, the real-time load rate of the power grid at different collection times, as well as the wind power, photovoltaic power, wind power characteristic matrix and photovoltaic characteristic matrix of all power plants, and meteorological data and corrected meteorological data of all meteorological stations are obtained.

[0060] Step S002: Based on the wind power and photovoltaic power of all power plants, construct a dynamically adjusted adjacency matrix. Based on the root mean square error of the wind power and photovoltaic power of the power plants, the real-time load rate of the grid, and the wind power and photovoltaic power of all power plants, determine the fused power value. Based on the fused power value, the difference between wind power and photovoltaic power at all adjacent acquisition times, determine the composite loss function.

[0061] Wind speed is subject to high-frequency pulsating components due to terrain disturbances, and the power response of wind turbines lags behind wind speed changes. Irradiance is affected by cloud movement, resulting in minute-level power drops, and the power curve shows a strong periodic correlation with the solar altitude angle. Therefore, in grid planning with a high proportion of wind and solar power integrated, wind and solar power exhibit dual fluctuation characteristics of minute-level meteorological abrupt changes and seasonal energy structure shifts.

[0062] The ST-GCN (Spatial-Temporal Graph Convolutional Networks) model is a deep learning model that combines the GCN (Graph Convolutional Network) and TCN (Temporal Convolutional Network) to integrate spatial and temporal information processing capabilities. The ST-GCN model captures spatial dependencies between nodes in a graph structure through graph convolutional layers, and simultaneously captures temporal dependencies in time-series data through temporal convolutional layers, effectively capturing complex relationships in spatiotemporal data. Therefore, this application uses the ST-GCN model to achieve power time-series data fusion for power grid planning optimization.

[0063] Furthermore, wind power and photovoltaic power exhibit dynamic changes over time and geographical correlations in space. Wind speed fluctuates on a minute-by-minute scale due to turbulence, and adjacent wind farms share similar meteorological conditions such as wind speed and cloud cover. Therefore, the wind power and photovoltaic power of adjacent wind farms show a significant correlation. Photovoltaic power plant clusters are affected by the day-night cycle, with high photovoltaic power during the day and zero at night. Influenced by topography and sunlight distribution, the irradiance distribution in different areas is uniform, and the power fluctuates synchronously.

[0064] To capture the correlation between wind power and photovoltaic power of different power plants, and the impact of the location of different power plants on their wind power and photovoltaic power, a dynamically adjusted adjacency matrix is ​​constructed.

[0065] All wind power outputs of the power plants are arranged in chronological order of their acquisition times to obtain the wind power output sequence. The Pearson correlation coefficient between the wind power output sequences of two different power plants is recorded as the wind power output similarity between the two power plants. Similarly, all photovoltaic power outputs of the power plants are arranged in chronological order of their acquisition times to obtain the photovoltaic power output sequence. The Pearson correlation coefficient between the photovoltaic power output sequences of two different power plants is recorded as the photovoltaic power output similarity between the two power plants.

[0066] Preferably, as an embodiment of this application, the element values ​​corresponding to two different power plants in the adjacency matrix are dynamically adjusted as follows:

[0067]

[0068] in, This indicates that the power plants in the adjacency matrix are dynamically adjusted. and power station The corresponding element value; , , These represent the preset first influence weight parameter, second influence weight parameter, and third influence weight parameter, respectively. The sum of the first influence weight parameter, second influence weight parameter, and third influence weight parameter is 1, and all of them are preset parameter values. In this embodiment, the values ​​of the first influence weight parameter, second influence weight parameter, and third influence weight parameter are all... ; Represents an exponential function with the natural constant as the base; Indicates power station and power station The distance between them; This indicates the preset attenuation parameter; in this embodiment, the attenuation parameter is set to 15. Indicates power station and power station The similarity of wind power output; Indicates power station and power station The photovoltaic power similarity.

[0069] The ST-GCN model can obtain the root mean square error of wind power and photovoltaic power of a power plant, respectively.

[0070] The calculation of the root mean square error (RMSE) of the ST-GCN model is a well-known technique and will not be elaborated further. Specifically, the ST-GCN model can obtain the predicted wind power and photovoltaic power of power plants. The sum of the absolute values ​​of the differences between the predicted wind power and the actual wind power of all power plants is the RMSE of the wind power of the power plant; the sum of the absolute values ​​of the differences between the predicted photovoltaic power and the actual photovoltaic power of all power plants is the RMSE of the photovoltaic power of the power plant.

[0071] The combined power value is determined based on the root mean square error of the wind and solar power of the power plants, the real-time load factor of the grid, and the wind and solar power of all power plants.

[0072]

[0073]

[0074]

[0075]

[0076] in, Indicates the fusion power value; Indicates the weight of wind power output; Indicates the photovoltaic power weight; The root mean square error of wind power output; The root mean square error represents the photovoltaic power output. Indicates the weighting adjustment factor; This indicates the real-time load factor of the power grid. This represents the economic security threshold of the power grid, specifically 85% of the grid's maximum load. This represents the sum of wind power output from all power plants. This represents the sum of the photovoltaic power output of all power plants.

[0077] Based on the root mean square error (RMSE) of the wind power and photovoltaic (PV) power of the power plants, the weights of wind power and PV power can be adjusted to influence the combined power value based on the sum of wind power from all power plants and the sum of PV power from all power plants. When wind power is significantly affected by complex meteorological factors such as wind speed and turbulence, the RMSE of wind power is larger, and the weight of wind power is smaller. In this case, the sum of PV power from all power plants has a larger influence on the combined power value, thus improving the accuracy of the combined result. Conversely, when PV power is significantly affected by cloud cover, the RMSE of PV power is larger, and the weight of PV power is smaller. In this case, the sum of wind power from all power plants has a larger influence on the combined power value, thus improving the accuracy of the combined result.

[0078] The weighting adjustment factor determined based on the real-time load rate of the power grid can embed the real-time status of the power grid into the data fusion process. During the data fusion process, the real-time carrying capacity of the power grid is taken into account, avoiding the problem of instantaneous overload of the power grid lines that may be caused by the drastic fluctuations in wind power and photovoltaic power due to weather conditions.

[0079] The closer the real-time load factor of the power grid is to the economic safety threshold of the power grid, and the lower the real-time load factor is than the economic safety threshold, the higher the penetration rate of new energy sources. Meanwhile, when the real-time load factor of the power grid is greater than the economic safety threshold, the problem of power grid overload can be automatically avoided.

[0080] By dynamically adjusting the weights of the fused power value based on the root mean square error of the wind and solar power of the power plant and the real-time load rate of the grid, and by determining the composite loss function, the prediction accuracy, grid security constraints and wind-solar synergy can be balanced. At the same time, it can adapt to sudden weather changes such as sudden wind speed drops and cloud cover. The smaller the root mean square error of the ST-GCN model, the more the fused power value should dominate the fusion result, and the higher the reliability of the fused value obtained by the ST-GCN model.

[0081] Furthermore, by jointly optimizing the three objectives of prediction accuracy, power smoothness, and grid security, a composite loss function is constructed to ensure the accuracy of the fusion results while meeting the constraints of actual engineering.

[0082] Specifically, the composite loss function is determined based on the fused power value, the differences in wind power and photovoltaic power between all adjacent acquisition times. .

[0083] Preferably, as an embodiment of this application, the composite loss function Specifically:

[0084]

[0085] In the formula, Represents the composite loss function; , , These represent the preset first loss weight parameter, second loss weight parameter, and third loss weight parameter, respectively. The sum of the first loss weight parameter, second loss weight parameter, and third loss weight parameter is 1, and all of them are preset parameter values. In this embodiment, the values ​​of the first loss weight parameter, second loss weight parameter, and third loss weight parameter are all... ; This represents the mean square error of the ST-GCN model; Indicates the first The absolute power difference at each acquisition moment, specifically, will be the power difference at the first acquisition moment. The data collection time and the first The absolute value of the difference in wind power at each data collection time is denoted as the nth data collection time. The difference in adjacent wind power at the i-th acquisition time will be used to... The data collection time and the first The absolute value of the difference in photovoltaic power at the nth acquisition time is denoted as the nth acquisition time. The difference in photovoltaic power between adjacent data collection times will be used to determine the next data collection time. The sum of the differences in adjacent wind power and adjacent photovoltaic power at the i-th data collection time is denoted as the i-th data collection time. The absolute difference in power at each acquisition moment; This indicates the number of all data collection moments; This represents the safe carrying capacity threshold of the power grid, specifically set at 90% of the grid's maximum load. Indicates the fusion power value; This represents the maximum value function. Specifically, the maximum value function takes the maximum value among the comma-separated values ​​within the parentheses.

[0086] The calculation of the mean square error of the ST-GCN model is a well-known technique and will not be elaborated further. Specifically, the ST-GCN model can obtain the predicted wind power and photovoltaic power of the power plant. The square of the difference between the predicted wind power and the actual wind power is denoted as the first square, and the square of the difference between the predicted photovoltaic power and the actual photovoltaic power is denoted as the second square. The sum of the first square and the second square is the mean square error of the ST-GCN model.

[0087] The wind and solar power output of power plants is greatly affected by weather changes, and high-precision prediction can reduce grid dispatching errors. In the composite loss function, the mean square error is used to avoid the impact of extreme errors on the grid. At the same time, the smoothing constraint can reduce the frequency regulation pressure on the grid caused by the fluctuation of wind and solar power. The difference between wind and solar power at adjacent acquisition times is more sensitive to small fluctuations, encouraging the composite loss function to transition smoothly. The grid carrying capacity is the most critical constraint, which can help the model approach the safety boundary while strictly limiting out-of-bounds behavior.

[0088] Thus, the dynamically adjusted adjacency matrix and composite loss function are obtained.

[0089] Step S003: Obtain the data fusion result based on the dynamically adjusted adjacency matrix, composite loss function, wind power characteristic matrix and photovoltaic characteristic matrix of the power station.

[0090] The composite loss function is used as the loss function of the ST-GCN model, and the dynamically adjusted adjacency matrix is ​​used as the dynamic adjacency matrix in the ST-GCN model. The wind power feature matrix and photovoltaic feature matrix of the power station are input into the ST-GCN model. Based on the dynamically adjusted adjacency matrix and the loss function, the ST-GCN model can output the fusion result.

[0091] The use of the ST-GCN model to obtain data fusion results is a well-known technique and will not be elaborated further.

[0092] This completes the fusion of power time-series data for power grid planning and optimization.

[0093] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A method for fusing power time-series data for power grid planning and optimization, characterized in that, The method includes the following steps: The real-time load rate of the power grid at different collection times, as well as the wind power and photovoltaic power of all power plants and the meteorological data of all meteorological stations, are collected. Based on the distance between the power plants and all meteorological stations and the meteorological data of all meteorological stations, the corrected meteorological data of the power plants are determined. Combining the wind power, photovoltaic power and location of all power plants at all collection times, the wind power characteristic matrix and photovoltaic characteristic matrix of the power plants are established respectively. Based on the wind power and photovoltaic power of all power plants, a dynamically adjusted adjacency matrix is ​​constructed. Based on the root mean square error of the wind power and photovoltaic power of the power plants, the real-time load rate of the grid, and the wind power and photovoltaic power of all power plants, the fused power value is determined. Based on the fused power value, the difference between wind power and photovoltaic power at all adjacent acquisition times, the composite loss function is determined. The data fusion results are obtained by dynamically adjusting the adjacency matrix, composite loss function, wind power characteristic matrix and photovoltaic characteristic matrix of the power station; The specific details of dynamically adjusting the adjacency matrix are as follows: Based on the total wind power output of two different power plants, determine the similarity of their wind power output; based on the total photovoltaic power output of two different power plants, determine the similarity of their photovoltaic power output. Dynamically adjust the element values ​​corresponding to two different power plants in the adjacency matrix: in, This indicates that the power plants in the adjacency matrix are dynamically adjusted. and power station The corresponding element value; , , These represent the preset first influence weight parameter, second influence weight parameter, and third influence weight parameter, respectively, and the sum of the first influence weight parameter, second influence weight parameter, and third influence weight parameter is 1; Represents an exponential function with the natural constant as its base; Indicates power station and power station The distance between them; This indicates the preset attenuation parameter; Indicates power station and power station wind power similarity; Indicates power station and power station The photovoltaic power similarity.

2. The power time-series data fusion method for power grid planning optimization according to claim 1, characterized in that, The method for determining the corrected meteorological data for the power station based on the distance between the power station and all meteorological stations and the meteorological data from all meteorological stations includes: Meteorological data includes wind speed and irradiance; any one of these meteorological data types can be recorded as the target meteorological data. The sum of the distances between the power station and all weather stations is recorded as the cumulative distance between the power station and the weather stations. The ratio of the distance between the power station and the weather stations to the cumulative distance between the power station and the weather stations is recorded as the distance weight between the power station and the weather stations. The distance weight between the power station and the meteorological station is used as the weight of the target meteorological data of the meteorological station. The target meteorological data of all meteorological stations are weighted and summed, and the result of the weighted sum is recorded as the corrected target meteorological data of the power station.

3. The power time-series data fusion method for power grid planning optimization according to claim 2, characterized in that, The wind power feature matrix is ​​specifically as follows: The wind power feature matrix contains three columns. The wind power of all power stations at the same acquisition time is arranged into one column to obtain the wind power column at the same acquisition time. The wind power of all acquisition times is arranged into one column according to the order of acquisition time to obtain the first column of the wind power feature matrix. Arrange the corrected wind speeds of all power plants at the same acquisition time into a column to obtain the column of corrected wind speeds at the same acquisition time. Arrange the corrected wind speeds of all acquisition times into a column according to the order of acquisition time to obtain the second column of the wind power feature matrix. Arrange the locations of all power plants into a column, obtain the location column, and repeatedly fill the third column of the wind power feature matrix with the location column until the third column of the wind power feature matrix is ​​completely filled.

4. The power time-series data fusion method for power grid planning optimization according to claim 2, characterized in that, The photovoltaic feature matrix is ​​specifically as follows: The photovoltaic feature matrix contains three columns. The photovoltaic power of all power plants at the same acquisition time is arranged into one column to obtain the photovoltaic power at the same acquisition time. The photovoltaic power of all acquisition times is arranged into one column according to the order of acquisition time to obtain the first column of the photovoltaic feature matrix. Arrange the corrected irradiance of all power plants at the same acquisition time into a column to obtain the column of corrected irradiance at the same acquisition time. Arrange the corrected irradiance of all acquisition times into a column according to the order of acquisition time to obtain the second column of the photovoltaic feature matrix. Arrange the locations of all power plants into a column, obtain the location column, and repeatedly fill the third column of the photovoltaic feature matrix with the location column until the third column of the photovoltaic feature matrix is ​​completely filled.

5. The power time-series data fusion method for power grid planning optimization according to claim 1, characterized in that, The formula for calculating the fusion power value is: in, Indicates the fusion power value; Indicates the weight of wind power output; Indicates the photovoltaic power weight; This represents the sum of wind power output from all power plants. This represents the sum of the photovoltaic power output of all power plants. This represents the weighting adjustment factor.

6. The power time-series data fusion method for power grid planning optimization according to claim 5, characterized in that, The specific methods for obtaining the wind power weight and the photovoltaic power weight are as follows: The sum of the root mean square error of wind power and the number 1 is used as the denominator, and the number 1 is used as the numerator. This value is denoted as the wind power weight. The sum of the root mean square error of photovoltaic power and the number 1 is used as the denominator, and the number 1 is used as the numerator. This fraction is denoted as the photovoltaic power weight.

7. The power time-series data fusion method for power grid planning optimization according to claim 5, characterized in that, The specific method for obtaining the weight adjustment factor is as follows: The absolute value of the difference between the real-time load rate of the power grid and the preset economic security threshold of the power grid, plus the number 1, is used as the denominator, and the number 1 is used as the value of the fraction in the numerator, which is denoted as the weight adjustment factor.

8. A power time-series data fusion method for power grid planning optimization according to claim 5, characterized in that, The composite loss function is specifically as follows: In the formula, Represents the composite loss function; , , These represent the preset first loss weight parameter, second loss weight parameter, and third loss weight parameter, respectively, and the sum of the first loss weight parameter, second loss weight parameter, and third loss weight parameter is 1; This represents the mean square error of the ST-GCN model; Indicates the first The absolute power difference at each acquisition moment, specifically, will be the power difference at the first acquisition moment. The data collection time and the first The absolute value of the difference in wind power at each data collection time is denoted as the nth data collection time. The difference in adjacent wind power at the first acquisition time will be used to... The data collection time and the first The absolute value of the difference in photovoltaic power at the nth acquisition time is denoted as the nth acquisition time. The difference in photovoltaic power between adjacent data collection times will be used to determine the next data collection time. The sum of the differences in adjacent wind power and adjacent photovoltaic power at the i-th data collection time is denoted as the i-th data collection time. The absolute difference in power at each acquisition moment; This indicates the number of all data collection moments; This indicates the preset safe carrying capacity threshold of the power grid; This represents the maximum value function.

9. A power time-series data fusion method for power grid planning optimization according to claim 1, characterized in that, The specific method for obtaining data fusion results by dynamically adjusting the adjacency matrix, composite loss function, wind power characteristic matrix, and photovoltaic characteristic matrix of the power station includes: The composite loss function is used as the loss function of the ST-GCN model, and the dynamically adjusted adjacency matrix is ​​used as the dynamic adjacency matrix in the ST-GCN model. The wind power feature matrix and photovoltaic feature matrix of the power station are input into the ST-GCN model to obtain the fusion result.

Images