Low-Voltage Substation Area Load Forecasting Method Based on Data Mining

Abstract

The present invention discloses a low-voltage substation area load forecasting method based on data mining, which specifically relates to the technical field of load forecasting. It includes determining the load characteristic data of adjacent nodes in the low-voltage substation area network topology, determining the sensitivity of reactive power between adjacent nodes, obtaining the load characteristic information and topological attribute information of adjacent nodes in the low-voltage substation area network topology, evaluating the influence of reactive power between adjacent nodes in the network topology, collecting the reactive power generated by each node according to the historical records of reactive power generated by each node, and determining the direct influence of the reactive power generated by each node on load forecasting by constructing a regression model, obtaining the uncertainty information and influence information of the nodes, and determining the nodes with poor performance in low-voltage substation area load forecasting accuracy. The present invention realizes the precise positioning and reactive power optimization control of key nodes in the low-voltage substation area, reduces the inaccuracy of load forecasting that may be caused by reactive power, and identifies the nodes with low load forecasting accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of load forecasting, and more specifically, to a load forecasting method for low-voltage power distribution areas based on data mining. Background Art

[0002] In a low-voltage power distribution network, load forecasting is the basis for key decisions such as power dispatch, operation optimization, and reactive power compensation. However, the load characteristics of low-voltage power distribution areas have strong time-variability and uncertainty, especially the volatility of reactive power, which has an important impact on the accuracy of load forecasting.

[0003] Traditional load forecasting methods mainly rely on active power characteristics. However, since reactive power is closely related to factors such as voltage fluctuation, load characteristics, and network topology, relying solely on active power may not accurately describe the load change trend. In addition, there are differences in the reactive power sensitivity of different nodes in the low-voltage power distribution network, and the reactive power changes of adjacent nodes will affect each other, resulting in low load forecasting accuracy for some nodes.

[0004] To solve the above defects, a technical solution is provided now. Summary of the Invention

[0005] In order to overcome the above defects of the prior art, an embodiment of the present invention provides a load forecasting method for low-voltage power distribution areas based on data mining to solve the problems raised in the above background art.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A load forecasting method for low-voltage power distribution areas based on data mining specifically includes the following steps:

[0008] S1: According to the load characteristic data of adjacent nodes in the low-voltage power distribution area network topology, determine the sensitivity of reactive power between adjacent nodes, and obtain the load characteristic information and topological attribute information of adjacent nodes in the low-voltage power distribution area network topology.

[0009] S2: Comprehensively analyze the load characteristic information and topological attribute information of adjacent nodes in the low-voltage power distribution area network topology, and evaluate the influence of reactive power between adjacent nodes in the network topology.

[0010] S3: According to the historical records of reactive power generated by each node, collect the reactive power generated by each node, and by constructing a regression model, determine the direct influence of the reactive power generated by each node on load forecasting, and determine the uncertainty information and influence information of the node.

[0011] S4: Combine and analyze the uncertainty information and influence information of the node, construct a prediction evaluation model, and obtain the nodes with poor load forecasting accuracy performance in the low-voltage power distribution area.

[0012] Among them, the uncertainty information of the node is represented by the distribution entropy value change coefficient, and the influence information of the node is represented by the joint influence evaluation coefficient.

[0013] In a preferred embodiment, the load characteristic information of adjacent nodes in the low-voltage distribution network topology includes:

[0014] The load characteristic information of adjacent nodes in the low-voltage distribution network topology is represented by the dynamic reactive power stock sensitivity coefficient;

[0015] The acquisition logic of the dynamic reactive power stock sensitivity coefficient is as follows: Obtain the actual reactive power load and rated reactive power capacity of adjacent nodes in the low-voltage distribution network topology, determine the reactive power load rate of adjacent nodes, and introduce a time weight factor based on dynamic characteristics;

[0016] Solve the power flow equation by the Newton-Raphson method to obtain the voltage and phase angle under the steady state of the system. Calculate the sub-matrix of the Jacobian matrix at the steady-state operating point to determine the sensitivity of the node reactive power to the voltage of its adjacent nodes;

[0017] Traverse the maximum value of the reactive power-voltage sensitivity of adjacent nodes in the low-voltage distribution network topology and perform normalization processing;

[0018] The calculation formula for the reactive power load rate of the adjacent node is:

[0019] ;

[0020] ;

[0021] The calculation formula for the sensitivity of the node reactive power to the voltage of its adjacent node is:

[0022] ;

[0023] The calculation formula for the dynamic reactive power stock sensitivity coefficient is:

[0024] ;

[0025] In the formula, i and j are the numbers of adjacent nodes in the low-voltage distribution network topology, is the reactive power load rate of node i, is the reactive power load rate of node j, is the time weight factor, is the reactive power load of node i at time t, is the reactive power load stock of node i at time t, is the reactive power load of node j at time t, is the reactive power load stock of node j at time t, is the reactive power-voltage sensitivity between node i and node j, J is the Jacobian matrix, representing the partial derivative matrix of the power flow equation, is the sensitivity of the reactive power of node j to the voltage of adjacent node i, is the dynamic reactive power stock sensitivity coefficient, is the maximum value of the reactive power-voltage sensitivities of all node pairs in the system.

[0026] In a preferred embodiment, the topological attribute information of adjacent nodes in the low-voltage substation network topology includes:

[0027] The topological attribute information of adjacent nodes in the low-voltage substation network topology is represented by the reactive power coupling response coefficient;

[0028] The acquisition logic of the reactive power coupling response coefficient is: by determining the reactive power disturbance amount of node i and the peak value of the reactive power response of node j, determine the reactive power floating sensitivity coefficient;

[0029] By analyzing the waveform time stamp, determine the time from the occurrence of the reactive power disturbance of node i to the peak value of the reactive power response of node j, determine the time delay attenuation, and introduce the equivalent impedance to reflect the influence of the physical characteristics of the line on the reactive power transmission;

[0030] The calculation formula of the reactive power floating sensitivity coefficient is:

[0031] ;

[0032] The calculation formula of the equivalent impedance is:

[0033] ;

[0034] The calculation formula of the reactive power coupling response coefficient is:

[0035] ;

[0036] In the formula, is the reactive power floating sensitivity coefficient, is the peak value of the reactive power response of node j, is the reactive power disturbance amount of node i, is the equivalent impedance between node i and node j, is the resistance of the connecting line between node i and node j, is the reactance of the connecting line between node i and node j, is the response time delay, is the reactive power coupling response coefficient.

[0037] In a preferred embodiment, evaluating the influence of reactive power between adjacent nodes in the network topology includes:

[0038] Comprehensively analyze the load characteristic information and topological attribute information of adjacent nodes in the low-voltage substation network topology, construct a node evaluation model based on the BP neural network algorithm, generate a node evaluation coefficient, and the calculation formula of the node evaluation coefficient is:

[0039] ;

[0040] In the formula, is the node evaluation coefficient, , are the proportionality coefficients of the dynamic reactive power stock sensitivity coefficient and the reactive power coupling response coefficient, , are both greater than 0;

[0041] Set the node evaluation coefficient threshold, obtain the node evaluation coefficients of each adjacent node in the low-voltage substation network topology, compare the node evaluation coefficients of each adjacent node in the low-voltage substation network topology with the node evaluation coefficient threshold. If the node evaluation coefficient is greater than the node evaluation coefficient threshold, mark the adjacent node as a key influencing node. If the node evaluation coefficient is less than the node evaluation coefficient threshold, do not mark it.

[0042] In a preferred embodiment, the uncertainty information of the node includes:

[0043] The acquisition logic of the distribution entropy value change coefficient is as follows: Obtain the historical time series data of the reactive power of each node, set the time granularity to divide the unit time period, extract the reactive power of each unit time period according to the time granularity to form a sub-dataset;

[0044] Estimate the probability density function of the reactive power data of each unit time period using the Gaussian kernel function, and determine the distribution entropy value change coefficient through entropy value calculation;

[0045] The calculation formula of the probability density function of the reactive power of different nodes in each unit time period is:

[0046] ;

[0047] The calculation formula of the distribution entropy value change coefficient is:

[0048] ;

[0049] In the formula, is the probability density function of the kth unit time period, k is the number of different unit time periods, is the unit time period The total number of samples within, n is the sample index, Q is the specified value of reactive power, is the unit time period The reactive power within, is the variation coefficient of the distribution entropy value, and h is the bandwidth.

[0050] In a preferred embodiment, the influence information of the node includes:

[0051] The acquisition logic of the joint influence evaluation coefficient is as follows: different nodes are divided into different types according to the magnitude of the node evaluation coefficient, and the node evaluation coefficients of the nodes adjacent to the node within a unit time period are obtained;

[0052] Determine the types of the adjacent nodes of the node within a unit time period, determine the accuracy deviation of the load prediction of the low-voltage substation area node by different types of adjacent nodes through historical records, and based on the accuracy deviation of the load prediction of the low-voltage substation area node by different types of adjacent nodes, perform accuracy scoring on different types of adjacent nodes;

[0053] Construct a regression model according to the scores of different types of adjacent nodes, the number of different types of adjacent nodes, and the duration ratio of different types of adjacent nodes, and generate a joint influence evaluation coefficient;

[0054] The calculation formula of the joint influence evaluation coefficient is:

[0055] ;

[0056] In the formula, is the joint influence evaluation coefficient, 1, 2, 3,..., m are the numbers of different types of nodes, , , ..., are the scores of different types of adjacent nodes, , , ..., are the numbers of different types of adjacent nodes, , , ..., are the duration ratios of different types of adjacent nodes, , , ..., are the weights of different types of adjacent nodes, and e is a natural number.

[0057] In a preferred embodiment, determining the nodes with poor accuracy performance in the load prediction of the low-voltage substation area includes:

[0058] Through comprehensive analysis of the uncertainty information and influence information of the node, perform weighted calculation on the distribution entropy value change coefficient and the joint influence evaluation coefficient, construct a prediction evaluation model, and generate a prediction evaluation coefficient. The calculation formula of the prediction evaluation coefficient is:

[0059] ;

[0060] Wherein, is the prediction evaluation coefficient, , are respectively the change coefficient of distribution entropy value and the proportional coefficient of the combined influence evaluation coefficient, , are both greater than 0;

[0061] Set the prediction evaluation coefficient threshold, obtain the prediction evaluation coefficients of nodes in different time periods, compare the prediction evaluation coefficients of nodes in different time periods with the prediction evaluation coefficient threshold. If the prediction evaluation coefficient is greater than the prediction evaluation coefficient threshold, a warning signal is generated. If the node still has continuous warnings and the reactive power fluctuates greatly, no warning signal is generated.

[0062] The technical effects and advantages of the present invention:

[0063] 1. The present invention collects real-time reactive power load, voltage, and current data of adjacent nodes in the low-voltage substation area through devices such as smart meters and micro PMUs, and through the physical-data fusion analysis of the dynamic reactive power stock sensitivity coefficient and the reactive power coupling response coefficient, combined with the intelligent evaluation of the BP neural network, realizes the accurate positioning and reactive power optimization control of key nodes in the low-voltage substation area, and reduces the inaccuracy of load prediction that may be caused by reactive power.

[0064] 2. The present invention analyzes the historical reactive power records of each node, quantifies the change and fluctuation of reactive power, and through the mutual influence between the reactive powers of each node, quantifies the influence of reactive power on the load accuracy, and identifies the nodes with low load prediction accuracy, which helps to take targeted optimization measures after identifying the nodes with low prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] For the convenience of those skilled in the art to understand, the present invention will be further described below with reference to the accompanying drawings;

[0066] Figure 1 is a schematic flow chart of the low-voltage substation area load prediction method based on data mining of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0067] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0068] Embodiment 1

[0069] Figure 1 This is a schematic flowchart of the low-voltage substation area load forecasting method based on data mining according to the present invention, which specifically includes the following steps:

[0070] S1: According to the load characteristic data of adjacent nodes in the low-voltage substation area network topology, determine the sensitivity of reactive power between adjacent nodes, and obtain the load characteristic information and topological attribute information of adjacent nodes in the low-voltage substation area network topology;

[0071] S2: Comprehensively analyze the load characteristic information and topological attribute information of adjacent nodes in the low-voltage substation area network topology, and evaluate the influence of reactive power between adjacent nodes in the network topology;

[0072] S3: According to the historical records of reactive power generated by each node, collect the reactive power generated by each node, and by constructing a regression model, determine the direct influence of the reactive power generated by each node on load forecasting, and obtain the uncertainty information and influence information of the node;

[0073] S4: Combine and analyze the uncertainty information and influence information of the node, construct a prediction evaluation model, and determine the nodes with poor load forecasting accuracy performance in the low-voltage substation area.

[0074] In the network topology of the low-voltage substation area, it is connected by cables and wires to form a tree-like or radial structure, and electric energy flows from the transformer to each user device. Among them, the flow of reactive power will directly affect the voltage level in the power grid, and further affect the power quality of users. By considering the influence of reactive power between adjacent nodes in the network topology of the low-voltage substation area, the change trend of the load can be more accurately reflected, and by reasonably configuring and dynamically controlling reactive power compensation devices, the reactive power distribution in the low-voltage substation area can be significantly improved, thereby reducing the mutual influence of reactive power between nodes and providing more accurate data support for load forecasting.

[0075] By collecting the load characteristic information and topological attribute information of adjacent nodes in the low-voltage substation area network topology, evaluating the influence of reactive power between adjacent nodes in the network topology, representing the load characteristic information of adjacent nodes in the low-voltage substation area network topology through the dynamic reactive power stock sensitivity coefficient, and representing the topological attribute information of adjacent nodes in the low-voltage substation area network topology through the reactive power coupling response coefficient.

[0076] Among them, the advantage of the dynamic reactive power stock sensitivity coefficient is that it can accurately capture the actual changes in the reactive power demand of each node by quantifying the reactive power load of each node, avoiding the deviation under simple assumptions, and by calculating the sensitivity between nodes, it is determined that some nodes have a greater impact on the change of reactive power, identifying pairs of nodes with strong electrical coupling. For these nodes, adjusting the reactive power is particularly important for the overall stability of the system.

[0077] The acquisition logic of the dynamic reactive power stock sensitivity coefficient is as follows: Obtain the actual reactive power load and rated reactive power capacity of adjacent nodes in the low-voltage substation network topology, determine the reactive power load rate of adjacent nodes, and introduce a time weight factor based on dynamic characteristics;

[0078] Solve the power flow equation by the Newton-Raphson method to obtain the voltage and phase angle under the steady state of the system. Calculate the sub-matrix of the Jacobian matrix at the steady-state operating point to determine the sensitivity of the node reactive power to the voltage of its adjacent nodes;

[0079] Traverse the maximum value of the reactive power-voltage sensitivity of adjacent nodes in the low-voltage substation network topology and perform normalization processing;

[0080] The calculation formula for the reactive power load rate of the adjacent nodes is:

[0081] ;

[0082] ;

[0083] The calculation formula for the sensitivity of the node reactive power to the voltage of its adjacent nodes is:

[0084] ;

[0085] The calculation formula for the dynamic reactive power stock sensitivity coefficient is:

[0086] ;

[0087] In the formula, i and j are the numbers of adjacent nodes in the low-voltage substation network topology, is the reactive power load rate of node i, is the reactive power load rate of node j, is the time weight factor, is the reactive power load of node i at time t, is the reactive power load stock of node i at time t, is the reactive power load of node j at time t, is the reactive power load stock of node j at time t, is the reactive power-voltage sensitivity between node i and node j, J is the Jacobian matrix, representing the partial derivative matrix of the power flow equation, is the sensitivity of the reactive power of node j to the voltage of adjacent node i, is the dynamic reactive power stock sensitivity coefficient, is the maximum value of the reactive power-voltage sensitivity of all node pairs in the system.

[0088] It should be noted that The dynamic weight reflecting the load fluctuation or urgency, for example, it increases during peak hours and decreases during low load hours. The larger the dynamic reactive power stock sensitivity coefficient, the more necessary it is to perform coordinated control on nodes in real-time scheduling. By deploying dynamic reactive power compensation devices, the prediction accuracy of the load can be improved.

[0089] The advantages of the reactive power coupling response coefficient are as follows:

[0090] In low-voltage lines, due to the large resistance, the reactive power flow is not only affected by the voltage amplitude difference but also related to the phase angle difference. Therefore, the reactive power coupling response coefficient can consider the influence of the resistance.

[0091] The load fluctuation in the low-voltage area may lead to frequent reactive power changes. In practical applications, through statistical methods or machine learning models, the reactive power coupling response coefficient is extracted from a large amount of real-time data. By combining data-driven methods and hierarchical control strategies, the reactive power coupling response coefficient can be effectively utilized to optimize reactive power management and improve voltage stability.

[0092] The acquisition logic of the reactive power coupling response coefficient is as follows: By determining the reactive power disturbance amount of node i and the peak value of the reactive power response of node j, the reactive power floating sensitivity coefficient is determined.

[0093] By analyzing the waveform timestamp, determine the time from the occurrence of the reactive power disturbance at node i to the peak value of the reactive power response at node j, determine the time delay attenuation, and introduce the equivalent impedance to reflect the influence of the physical characteristics of the line on the reactive power transmission.

[0094] The calculation formula for the reactive power floating sensitivity coefficient is:

[0095] ;

[0096] The calculation formula for the equivalent impedance is:

[0097] ;

[0098] The calculation formula for the reactive power coupling response coefficient is:

[0099] ;

[0100] In the formula, is the reactive power floating sensitivity coefficient, is the peak value of the reactive power response of node j, is the reactive power disturbance amount of node i, is the equivalent impedance between node i and node j, is the resistance of the connecting line between node i and node j, is the reactance of the connecting line between node i and node j, is the response time delay, is the reactive power coupling response coefficient.

[0101] It should be noted that for data acquisition, simulation experiments can be carried out through high-precision smart meters, micro PMUs, or programmable load simulators. The larger the reactive power coupling response coefficient, the stronger the reactive power influence of node i on node j, and the easier the influence is to transmit. By deploying dynamic reactive power compensation devices at key nodes, voltage fluctuations can be effectively suppressed and the coupling effect can be weakened, thereby improving the accuracy of the load prediction model.

[0102] Comprehensively analyze the load characteristic information and topological attribute information of adjacent nodes in the low-voltage distribution network topology, and construct a node evaluation model based on the bp neural network algorithm to generate a node evaluation coefficient. The calculation formula of the node evaluation coefficient is:

[0103] ;

[0104] In the formula, is the node evaluation coefficient, , are the proportionality coefficients of the dynamic reactive power stock sensitivity coefficient and the reactive power coupling response coefficient, , are both greater than 0.

[0105] Set the node evaluation coefficient threshold, obtain the node evaluation coefficients of each adjacent node in the low-voltage distribution network topology, and compare the node evaluation coefficients of each adjacent node in the low-voltage distribution network topology with the node evaluation coefficient threshold. If the node evaluation coefficient is greater than the node evaluation coefficient threshold, mark the adjacent node as a key influence node and deploy a dynamic reactive power compensation device at the key node. If the node evaluation coefficient is less than the node evaluation coefficient threshold, do not mark it.

[0106] In this embodiment, real-time reactive power load, voltage, and current data of adjacent nodes in the low-voltage distribution network are collected through devices such as smart meters and micro PMUs, and through physical-data fusion analysis of the dynamic reactive power stock sensitivity coefficient and the reactive power coupling response coefficient, combined with intelligent evaluation of the BP neural network, accurate positioning and reactive power optimization control of key nodes in the low-voltage distribution network are achieved, reducing the possible inaccuracy of reactive power on load prediction.

[0107] Embodiment 2

[0108] According to the network topology of the low-voltage power distribution area, identify the nodes in the network topology of the low-voltage power distribution area that affect the accuracy of load forecasting due to the reactive power of the nodes. Based on the historical records of reactive power generation at each node, collect the reactive power generated at each node, determine the uncertainty information of the nodes, represent the uncertainty information of the nodes through the distribution entropy change coefficient, determine the direct impact of the reactive power generated at each node on load forecasting by constructing a regression model, determine the impact information of the nodes, and represent the impact information of the nodes through the joint impact evaluation coefficient.

[0109] The advantages of the distribution entropy change coefficient are as follows:

[0110] The distribution entropy change coefficient can identify the uncertainty of nodes, help identify the nodes with unstable reactive power generation during different electricity consumption periods, and thus improve the accuracy of load forecasting during different electricity consumption periods.

[0111] The acquisition logic of the distribution entropy change coefficient is as follows: Obtain the historical time series data of the reactive power of each node, set the time granularity to divide the unit time period, extract the reactive power of each unit time period according to the time granularity to form a sub-dataset;

[0112] Use the Gaussian kernel function to estimate the probability density function for the reactive power data of each unit time period, and determine the distribution entropy change coefficient through entropy calculation;

[0113] The calculation formula for the probability density function of the reactive power of different nodes in each unit time period is:

[0114] ;

[0115] The calculation formula for the distribution entropy change coefficient is:

[0116] ;

[0117] In the formula, is the probability density function of the k-th unit time period, k is the number of different unit time periods, is the total number of samples within the unit time period , n is the sample index, Q is the specified value of reactive power, is the reactive power within the unit time period , is the distribution entropy change coefficient, h is the bandwidth.

[0118] It should be noted that k is determined by the time resolution, the time resolution is determined by the staff in the professional field, and n represents traversing the unit time period For the nth data point within, the greater the variation coefficient of the distribution entropy value, the greater the change in the reactive power at the node within the unit time period, and the more irregular it may be. Therefore, in load forecasting, the error usually increases with the increase in the forecasting duration, and the complex change in reactive power will exacerbate the error accumulation, resulting in a decrease in the accuracy of long-term forecasting.

[0119] The advantages of the combined influence evaluation coefficient are as follows:

[0120] Through the combined influence evaluation coefficient, the influence of different types of adjacent nodes on the load forecasting accuracy of the target node can be quantified. Through node type division, it can dynamically adapt to different electricity consumption patterns, avoid using a fixed model for evaluation, and improve the adaptability to complex load changes;

[0121] By analyzing the deviation of different types of adjacent nodes from the load forecasting through historical records, a data-driven accuracy scoring system is established, reducing the dependence on human experience. Combining multi-time period data analysis can identify long-term influence patterns, which helps to more accurately evaluate the difficulty of load forecasting for nodes.

[0122] The acquisition logic of the combined influence evaluation coefficient is as follows: Different nodes are divided into different types according to the magnitude of the node evaluation coefficient, and the node evaluation coefficients of the nodes adjacent to the node within the unit time period are obtained;

[0123] Determine the types of adjacent nodes of the node within the unit time period, determine the accuracy deviation of different types of adjacent nodes from the load forecasting of the low-voltage substation area nodes through historical records, and based on the accuracy deviation of different types of adjacent nodes from the load forecasting of the low-voltage substation area nodes, perform accuracy scoring on different types of adjacent nodes;

[0124] According to the scores of different types of adjacent nodes, the numbers of different types of adjacent nodes, and the proportion of the duration of different types of adjacent nodes, a regression model is constructed to generate the combined influence evaluation coefficient;

[0125] The calculation formula of the combined influence evaluation coefficient is:

[0126] ;

[0127] In the formula, is the combined influence evaluation coefficient, 1, 2, 3, ……, m are the numbers of different types of nodes, , , ……, are the scores of different types of adjacent nodes, , , ……, are the numbers of different types of adjacent nodes, , , ……, is the ratio of the duration of different types of adjacent nodes, , , ……, are the weights of adjacent nodes of different types, and e is a natural number.

[0128] It should be noted that there may be multiple nodes adjacent to a node, that is, users using the same distribution box. Based on the electricity consumption of users in a unit time period, there may be different patterns of electricity consumption changes in adjacent nodes. Therefore, nodes may be divided into different node types within a unit time period. The larger the combined influence evaluation coefficient, the stronger the interference of the node load prediction by adjacent nodes, and the more attention and compensation are required, and the lower the prediction accuracy.

[0129] Through comprehensive analysis of the uncertainty information and influence information of nodes, the change coefficient of distribution entropy value and the combined influence evaluation coefficient are weighted and calculated to construct a prediction evaluation model and generate a prediction evaluation coefficient. The calculation formula of the prediction evaluation coefficient is:

[0130] ;

[0131] In the formula, is the prediction evaluation coefficient, , are the proportionality coefficients of the change coefficient of distribution entropy value and the combined influence evaluation coefficient respectively, , are both greater than 0.

[0132] Set the threshold of the prediction evaluation coefficient, obtain the prediction evaluation coefficients of nodes in different time periods, and compare the prediction evaluation coefficients of nodes in different time periods with the threshold of the prediction evaluation coefficient. If the prediction evaluation coefficient is greater than the threshold of the prediction evaluation coefficient, a warning signal is generated to notify relevant staff to optimize the load prediction algorithm for the warning node and time period. If the node still has continuous warnings and large reactive power fluctuations, the grid operation strategy is appropriately adjusted, such as increasing reactive power compensation equipment or adjusting the topological structure. If the prediction evaluation coefficient is less than the threshold of the prediction evaluation coefficient, no warning signal is generated.

[0133] In this embodiment, by analyzing the historical reactive power records of each node, the change and fluctuation of reactive power are quantified, and through the mutual influence between the reactive powers of each node, the influence of reactive power on the load accuracy is quantified, and the nodes with low load prediction accuracy are identified, which helps to take targeted optimization measures after identifying the nodes with low prediction accuracy.

[0134] The above formulas are all dimensionless and only take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula that is closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0135] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer, or a data storage device such as a server or data center that contains one or more sets of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0136] It should be understood that in various embodiments of the present application, the magnitudes of the sequence numbers of the above processes do not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0137] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0138] In several embodiments provided by the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling, direct coupling, or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of the devices or units can be in electrical, mechanical, or other forms.

[0139] If the above functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs and other various media that can store program codes.

[0140] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present application can easily think of changes or substitutions, which should all be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A low voltage area load forecasting method based on data mining, characterized in that: The specific steps include: S1: According to the load characteristic data of adjacent nodes in the low-voltage area network topology, the sensitivity of reactive power between adjacent nodes is determined, and the load characteristic information and topological attribute information of adjacent nodes in the low-voltage area network topology are obtained; S2: Comprehensively analyze the load characteristic information and topology attribute information of adjacent nodes in the low-voltage area network topology to evaluate the impact of reactive power between adjacent nodes in the network topology; S3: According to the historical records of reactive power generated by each node, the reactive power generated by each node is collected, and by building a regression model, the direct impact of the reactive power generated by each node on the load forecast is determined to obtain the uncertainty information and impact information of the node; S4: Combine the uncertainty information and impact information of the node and build a prediction evaluation model to determine the nodes with poor load prediction accuracy in the low-voltage area; Among them, the uncertainty information of the node is represented by the distribution entropy value change coefficient, and the impact information of the node is represented by the joint impact assessment coefficient.

2. The low voltage area load forecasting method based on data mining according to claim 1 is characterized in that: Load characteristic information of adjacent nodes in the low-voltage area network topology, including: The load characteristic information of adjacent nodes in the low-voltage area network topology is represented by the dynamic reactive power stock sensitivity coefficient; The acquisition logic of the dynamic reactive stock sensitivity coefficient is: obtaining the actual reactive load and rated reactive capacity of the adjacent nodes in the low-voltage area network topology, determining the reactive load rate of the adjacent nodes, and introducing a time weight factor based on the dynamic characteristics; The power flow equation is solved by the Newton-Raphson method to obtain the voltage and phase angle of the system in steady state. The sub-matrix of the Jacobian matrix is ​​calculated at the steady-state operating point to determine the sensitivity of the node reactive power to the voltage of its adjacent nodes. Traverse the maximum reactive power-voltage sensitivity of adjacent nodes in the low-voltage area network topology and perform normalization processing; The calculation formula of the reactive load rate of the adjacent nodes is: ； ； The sensitivity calculation formula of the node reactive power to its adjacent node voltage is: ； The calculation formula of the dynamic reactive power stock sensitivity coefficient is: ； Where i, j are the numbers of adjacent nodes in the low voltage area network topology. is the node reactive load rate of node i, is the node reactive load rate of node j, is the time weight factor, is the reactive load of node i at time t, is the reactive load stock of node i at time t, is the reactive load of node j at time t, is the reactive load stock of node j at time t, is the reactive power-voltage sensitivity between node i and node j, J is the Jacobian matrix, which represents the partial derivative matrix of the power flow equation, is the sensitivity of the reactive power of node j to the voltage of the adjacent node i, is the dynamic reactive power stock sensitivity coefficient, It is the maximum value of reactive power-voltage sensitivity of all node pairs in the system.

3. The low voltage area load forecasting method based on data mining according to claim 2 is characterized in that: The topological attribute information of adjacent nodes in the low-voltage area network topology includes: The topological attribute information of the adjacent nodes of the low-voltage area network topology is represented by the reactive coupling response coefficient; The acquisition logic of the reactive coupling response coefficient is: by determining the reactive disturbance amount of node i and the reactive response peak value of node j, the reactive floating sensitivity coefficient is determined; By analyzing the waveform timestamp, the time from the occurrence of reactive disturbance at node i to the reactive response peak at node j is determined, the delay attenuation is determined, and the equivalent impedance is introduced to reflect the impact of the physical characteristics of the line on reactive power transmission; The calculation formula of reactive floating sensitivity coefficient is: ； The calculation formula for equivalent impedance is: ； The calculation formula of reactive coupling response coefficient is: ； In the formula, is the reactive floating sensitivity coefficient, is the peak value of the reactive power corresponding to node j, is the reactive disturbance of node i, is the equivalent impedance between node i and node j, is the resistance of the connection line between node i and node j, is the reactance of the line connecting node i and node j, To respond to the delay, is the reactive coupling response coefficient.

4. The low voltage area load forecasting method based on data mining according to claim 3 is characterized in that: Evaluate the impact of reactive power between adjacent nodes in the network topology, including: The load characteristic information and topological attribute information of adjacent nodes in the low-voltage area network topology are comprehensively analyzed, and a node evaluation model is constructed based on the BP neural network algorithm to generate a node evaluation coefficient. The calculation formula of the node evaluation coefficient is: ； In the formula, is the node evaluation coefficient, , is the proportional coefficient of the dynamic reactive stock sensitivity coefficient and reactive coupling response coefficient, , They are both greater than 0; Set the node evaluation coefficient threshold, obtain the node evaluation coefficient of each adjacent node in the low-voltage substation network topology, compare the node evaluation coefficient of each adjacent node in the low-voltage substation network topology with the node evaluation coefficient threshold, if the node evaluation coefficient is greater than the node evaluation coefficient threshold, then the adjacent node is marked as a key influencing node, if the node evaluation coefficient is less than the node evaluation coefficient threshold, then it is not marked.

5. The method for low voltage area load forecasting based on data mining according to claim 4 is characterized in that: Uncertainty information of the node, including: The acquisition logic of the distribution entropy value variation coefficient is as follows: acquiring the historical time series data of reactive power of each node, setting the time granularity to divide the unit time period, extracting the reactive power of each unit time period according to the time granularity, and forming a sub-data set; The probability density function is estimated using the Gaussian kernel function for the reactive power data of each unit time period, and the distribution entropy value variation coefficient is determined by entropy value calculation; The calculation formula of the probability density function of reactive power of different nodes in each unit time period is: ； The calculation formula of the distribution entropy value change coefficient is: ； In the formula, is the probability density function of the kth unit time period, k is the number of different unit time periods, Unit time period The total number of samples in, n is the sample index, Q is the specified value of reactive power, Unit time period The reactive power inside is the coefficient of change of distribution entropy, and h is the bandwidth.

6. The method for low voltage area load forecasting based on data mining according to claim 5 is characterized in that: Node impact information, including: The acquisition logic of the joint impact assessment coefficient is: different nodes are divided into different types according to the size of the node assessment coefficient, and the node assessment coefficient of the node adjacent to the node in a unit time period is obtained; Determine the types of adjacent nodes of a node within a unit time period, determine the accuracy deviation of different types of adjacent nodes on the load prediction of low-voltage area nodes through historical records, and perform accuracy scoring on different types of adjacent nodes based on the accuracy deviation of different types of adjacent nodes on the load prediction of low-voltage area nodes; According to the scores of different types of adjacent nodes, the number of different types of adjacent nodes, and the proportion of different types of duration of adjacent nodes, a regression model is constructed to generate a joint impact assessment coefficient; The calculation formula of the joint impact assessment coefficient is: ； In the formula, is the joint impact assessment coefficient, 1, 2, 3, ..., m are the numbers of different types of nodes, , , ……、 Scoring different types of neighboring nodes, , , ……、 is the number of adjacent nodes of different types, , , ……、 is the duration ratio of different types of adjacent nodes, , , ……、 are weights of adjacent nodes of different types, and e is a natural number.

7. The method for predicting low voltage load based on data mining according to claim 6 is characterized in that: Identify nodes with poor load forecasting accuracy in low-voltage areas, including: Through comprehensive analysis of the uncertainty information and impact information of the nodes, the distribution entropy value change coefficient and the joint impact assessment coefficient are weighted and calculated to build a prediction and assessment model and generate a prediction and assessment coefficient. The calculation formula of the prediction and assessment coefficient is: ； In the formula, is the prediction evaluation coefficient, , are the proportional coefficients of the distribution entropy change coefficient and the joint impact assessment coefficient, respectively. , They are both greater than 0; Set the prediction evaluation coefficient threshold, obtain the prediction evaluation coefficients of nodes in different time periods, compare the prediction evaluation coefficients of nodes in different time periods with the prediction evaluation coefficient threshold, and generate a warning signal if the prediction evaluation coefficient is greater than the prediction evaluation coefficient threshold; if the prediction evaluation coefficient is less than the prediction evaluation coefficient threshold, no warning signal is generated.

Images