A method, system, device and storage medium for predicting line loss in an active distribution station area

Abstract

The present application discloses a method, system, device and storage medium for predicting line loss in an active distribution substation area, and relates to the field of line loss detection technology. The method comprises: decomposing the historical line loss data of the active distribution substation area into the line loss data of each electric meter in the active distribution substation area; performing correlation analysis on the line loss influencing factors of the active distribution substation area and the electric meter line loss data, and obtaining the weight of each influencing factor on the electric meter line loss data; using multiple classifiers and a bidirectional long short-term memory network to predict the electric meter line loss, and superimposing and reconstructing to obtain the line loss prediction result of the active distribution substation area. The present application realizes the accurate prediction of line loss in active distribution substation areas with a high proportion of DG access.

Description

Technical Field

[0001] The present application relates to the technical field of line loss detection, and in particular to a method, system, device and storage medium for predicting line loss in an active distribution station area. Background Art

[0002] The Distribution Station Area (DSA) covers the power supply network from 10kV transformers to user power terminal equipment, and is a key infrastructure for low-voltage power supply. With the continuous development of new power systems, the proportion of distributed energy (DG) mainly based on photovoltaics in DSA has gradually increased. However, affected by various factors such as solar radiation intensity and temperature, the output power of DG has greater randomness and volatility, which makes the line loss influencing mechanism of DSA containing a high proportion of DG more complicated. The DSA line loss characteristics and their changing trends lack obvious rules, making it difficult to accurately predict DSA line losses. Therefore, conducting research on DSA line loss prediction under the influence of multiple factors is of great significance for the reasonable planning of DG access and the effective reduction of DSA line losses.

[0003] At present, a lot of research has been done on line loss prediction, which can be mainly divided into two categories: traditional line loss prediction technology and line loss prediction technology based on deep learning. In traditional line loss prediction technology, physical-based technologies such as the root mean square current method, the average current method and the equivalent resistance method are usually used to calculate DSA line loss. Among them, the root mean square current method calculates the power loss generated by the calculator based on the root mean square current flowing through the 0.4kV line in the DSA. This method ignores the differences in load power factors, resulting in low accuracy in line loss prediction. As a derivative method of the root mean square current method, the average current method estimates the power loss by calculating the average current in the DSA line. However, this method ignores the DSA voltage fluctuation factor, which affects the accuracy of line loss prediction. The equivalent resistance method calculates the line loss by constructing a virtual equivalent resistance of the simulated transformer and the DSA line. Because the assumptions of this method deviate from the actual operating conditions of the power grid, the accuracy of line loss prediction is low. In summary, the traditional line loss prediction technology has a small demand for collected data and is convenient for calculation, but the DSA network structure is complex and the load is diversified, resulting in low accuracy of traditional line loss prediction technology.

[0004] Line loss prediction technology based on deep learning refers to the prediction of DSA line loss using deep learning methods such as recurrent neural network (RNN), K-Means clustering and back propagation (BP) neural network, random forest, etc. Line loss prediction technology based on deep learning determines the influence weight of the selected collected data on the DSA line loss calculation through the hierarchical analysis method, and forms a data set to train the deep learning network to determine the optimal input parameters to improve the accuracy of DSA line loss prediction. Compared with traditional line loss prediction technology, line loss prediction technology based on deep learning has higher calculation accuracy and faster calculation speed. However, in the above studies, the impact of DG access on DSA line loss was not considered. With the high proportion of DG access to DSA, the traditional passive DSA has become a two-way active distribution station area (Active Compensation of Distribution Station Area, ADSA). The line loss research based on passive DSA is not well applicable to ADSA. Summary of the invention

[0005] In response to the problems of complex line loss influencing mechanism and low prediction accuracy caused by a high proportion of DG access to DSA, the present application provides a method, system, device and storage medium for predicting line loss in an active distribution station area. The present application decomposes the historical line loss data of the active distribution station area into the line loss data of each meter, and analyzes the weights of each influencing factor of the line loss data of each meter. Based on this, a multi-classifier and a bidirectional long short-term memory model are used to predict the meter line loss, and the line loss prediction results of the active distribution station area are obtained by superposition and reconstruction, thereby achieving accurate prediction of the line loss in the active distribution station area with a high proportion of DG access.

[0006] In the first aspect, the present application is implemented through the following technical solutions:

[0007] A method for predicting line loss in an active distribution station area, the method comprising:

[0008] Decomposing the historical line loss data of the active distribution substation area into the line loss data of each electric meter in the active distribution substation area;

[0009] Conduct correlation analysis on the influencing factors of line loss in active distribution substation area and meter line loss data, and obtain the weight of each influencing factor on meter line loss data;

[0010] Based on the line loss data of each meter in the active distribution station area and its influencing factors and weights, multiple classifiers and a bidirectional long short-term memory network are used to predict the meter line loss, and the line loss prediction result of the active distribution station area is obtained by superimposing and reconstructing.

[0011] In some implementations, decomposing the historical line loss data of the active distribution station area into the line loss data of each electric meter in the active distribution station area specifically includes:

[0012] Collecting and repairing the historical data of electric meters in the active distribution station area, and calculating the historical line loss data of the active distribution station area based on the repaired historical data of electric meters;

[0013] Calculate the line length from each electric meter to the power supply point in the active distribution station area;

[0014] The line loss data of each electric meter in the active distribution station area is obtained by calculating the line length from each electric meter to the power supply point in the active distribution station area and the power drawn by the electric meter.

[0015] In some implementations, the collecting and abnormal repairing of the historical data of electric meters in the active distribution station area, and calculating the historical line loss data of the active distribution station area based on the repaired historical data of electric meters, specifically includes:

[0016] Normalize the original meter data sequence;

[0017] The normalized electric meter data sequence is used as the input of the recurrent neural network, and the sequence output after the recurrent neural network repair is the repaired electric meter data sequence;

[0018] The active distribution station area is calculated by the following formula j Line loss rate at a certain moment for:

[0019]

[0020] Where: , They are respectively the total list of active distribution stations j The amount of electricity offline and online at each moment; For the user's meter in the active distribution station area j The sum of the off-grid electricity at each moment; For the active distribution station area j The sum of DG online power consumption at each moment.

[0021] In some embodiments, the length of the line from the electric meter to the power supply point is L DM The calculation formula is:

[0022]

[0023] Where: t BG Timestamp for the signal emitted by the power point;t END The time when the electric meter receives the signal transmitted by the power source; v DM is the speed at which the signal propagates.

[0024] In some implementations, the process of calculating the line loss data of each electric meter in the active distribution station area includes:

[0025] The initial meter line loss rate is calculated based on the line length from the meter to the power source and the power drawn by the meter.

[0026] Calculate the active distribution station area j The difference between the direct line loss of the point and all the meters under it forms the error factor;

[0027] Based on the initial electric meter line loss rate and the error factor, a final line loss rate of each electric meter is calculated.

[0028] In some implementations, the correlation analysis between the line loss influencing factors of the active distribution station area and the meter line loss data to obtain the weight of each influencing factor on the meter line loss data specifically includes:

[0029] The partial maximum information coefficient is used to analyze the influencing factors of meter line loss, and the Akaike information rule is used as the end condition for screening the weight set of influencing factors;

[0030] When the screening end value reaches the minimum, the optimal weight set of influencing factors is obtained, and a line loss weight matrix of electric meters in the active distribution station area is formed.

[0031] In some embodiments, the use of multiple classifiers and bidirectional long short-term memory networks to predict the meter line loss, and superimposing and reconstructing to obtain the line loss prediction result of the active distribution station area, specifically includes:

[0032] Inputting the line loss data of each electric meter in the active distribution station area into a multi-classifier;

[0033] According to the line loss data of each electric meter in the active distribution station area, a reference classifier of each electric meter is constructed respectively, and each reference classifier is trained separately based on a bidirectional long short-term memory network to obtain an electric meter line loss model suitable for the corresponding electric meter;

[0034] In the bidirectional long short-term memory network, the input meter line loss data and its influencing factors and weights are combined. The forward network is first used to extract the meter line loss data features, and then the reverse network is used to feed forward to the same output layer of the meter line loss model to form a hidden state. The correlation between the hidden state and the query value is calculated through the scoring function to obtain the score, and then the normalization function is used to normalize the score.

[0035] The normalized value is mapped to the output layer through the fully connected layer to obtain the line loss prediction result of the active distribution station area.

[0036] In a second aspect, the present application proposes a line loss prediction system for an active distribution station area, the prediction system comprising:

[0037] A decomposition module, the decomposition module is used to decompose the historical line loss data of the active distribution station area into the line loss data of each electric meter in the active distribution station area;

[0038] A correlation analysis module, which is used to perform correlation analysis on the line loss influencing factors of the active distribution station area and the meter line loss data, and obtain the weight of each influencing factor on the meter line loss data;

[0039] And, a prediction module, which predicts the meter line loss based on the line loss data of each meter in the active distribution station area and its influencing factors and weights, uses multiple classifiers and a bidirectional long short-term memory network to predict the meter line loss, and superimposes and reconstructs to obtain the line loss prediction result of the active distribution station area.

[0040] In a third aspect, the present application proposes a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0041] In a fourth aspect, the present application proposes a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0042] The present application proposes a method, system, device and storage medium for predicting line loss in an active distribution area. First, the historical line loss data of the active distribution area is decomposed into the line loss data of each meter in the area, and the data is used as the basis for subsequent prediction of line loss in the active area. In this process, the collected meter data is repaired for abnormalities, thereby ensuring the reliability and accuracy of the historical line loss data. Then, the correlation analysis of the influencing factors of the line loss in the active area and the meter line loss data is performed to obtain the weight of each influencing factor on the meter line loss data; on this basis, multiple classifiers and bidirectional long short-term memory networks are used to predict the meter line loss, and the line loss prediction results of the active distribution area are obtained by superimposing and reconstructing, which realizes the reliable and effective prediction of the line loss in the active distribution area, and provides technical support for the power supply security of the active distribution area. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The drawings described herein are used to provide a further understanding of the embodiments of the present application, constitute a part of the present application, and do not constitute a limitation on the embodiments of the present application. In the drawings:

[0044] Figure 1A flow chart of the prediction method provided in the embodiment of the present application;

[0045] Figure 2 A structural diagram of a prediction model provided in an embodiment of the present application;

[0046] Figure 3 This is a functional block diagram of the prediction system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0047] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with examples and drawings. The illustrative implementation scheme of the present application and its description are only used to explain the present application and are not intended to limit the present application.

[0048] Embodiment: The existing DSA line loss prediction technology does not consider the impact of high-proportion DG access on DSA line loss, and the prediction accuracy is low. In view of this, this embodiment proposes a line loss prediction method for active distribution station area.

[0049] like Figure 1 As shown, the method proposed in this embodiment includes the following steps:

[0050] Step 100, decomposing the historical line loss data of the active distribution station area into the line loss data of each electric meter in the active distribution station area.

[0051] The decomposition process of historical line loss data in active distribution area includes the following steps:

[0052] Step 101, collect and repair the historical data of the electric meters in ADSA, and calculate and obtain the ADSA historical line loss data.

[0053] In order to promote the construction of new power grid services such as smart energy use in user households, DG access, and vehicle-grid interaction, a new generation of smart meters has been built. This type of meter collects user current, voltage, electricity and other data at a frequency of 15 minutes, providing strong data support for ADSA historical line loss calculation. However, during the meter data collection process, due to factors such as low-voltage high-speed carrier transmission channel (High Power Line Communication, HPLC) and meter clock deviation, the electricity data used to calculate ADSA historical line loss data has errors and errors, and the data collection success rate of the new generation of smart meters is less than 97%. Therefore, it is necessary to repair the collected 15-minute frequency meter electricity data to eliminate the impact of errors and abnormal data on the decomposition of ADSA historical line loss data. In view of this, this embodiment uses a recurrent neural network (RNN) to repair the historical collection data. The recurrent neural network consists of an input layer, a hidden layer, and an output layer, wherein the input layer integrates the current input state and the state of the previous moment to generate a new state for capturing long-term dependencies in the sequence; the hidden layer determines whether the historical state is lost and determines the historical state value to be retained at the current time. The specific repair process is as follows:

[0054] First, the original meter data sequence is normalized. Q NORM for:

[0055] (1)

[0056] In the formula, Q ORI is the original meter data sequence; Q AVG is the mean value of the electric meter data; δ is the standard deviation.

[0057] Then, the normalized meter data series Q NORM As the input of the recurrent neural network, the sequence output by the recurrent neural network repair is the repaired meter data sequence Q OUT Specific treatment of repaired meter data The previous data sequence is used for RNN forward training, and the repaired meter data is used The subsequent data sequence is trained with RNN reverse training, and the RNN hyperparameters are optimized to improve the training accuracy. The trained RNN is used to train the meter data to be repaired. Perform forward and reverse bidirectional prediction, and perform weighted averaging of the forward and reverse prediction sequences to obtain the final repair data .

[0058] (2)

[0059] In the formula, For the RNN network j Forward prediction data; For the RNN network j Reverse prediction data; k is the length of the meter data sequence.

[0060] ADSA line loss is composed of down-grid line loss and up-grid line loss, where down-grid line loss is the line loss between the 10kV distribution transformer and the user's meter, and up-grid line loss is the line loss caused by DG up-grid. j Line loss rate at a certain moment for:

[0061] (3)

[0062] Where: , They are respectively j The amount of electricity offline and online at each moment; For the user meter in ADSA j The sum of the off-grid electricity at each moment; For ADSA j The sum of DG online power consumption at each moment.

[0063] The line loss data of a high proportion of DG connected to ADSA is highly nonlinear and random. Therefore, this embodiment decomposes the ADSA line loss into calculating the line loss of each ADSA meter. First, it is necessary to calculate the line length from the meter to the ADSA mid-grid power supply point (i.e., distribution transformer) and the distributed power supply (photovoltaic, energy storage, etc.) device to the meter. Then, the ADSA meter line loss is calculated based on the power provided by the power supply point to the single meter and the line length from the power supply to the meter.

[0064] Step 102, calculating the line length from each electric meter in the ADSA to the power source point.

[0065] When calculating the line length from the power source point to the meter, a signal with a timestamp is transmitted from the power source to the meter, so that the line length is calculated according to the time point. L DM for:

[0066] (4)

[0067] Where: t BG Timestamp for the signal emitted by the power point; t ENDThe time when the electric meter receives the signal transmitted by the power source; v DM is the speed at which the signal propagates.

[0068] Step 103, calculating the ADSA meter line loss according to the line length from each meter to the power source point and the power drawn by the meter, thereby completing the decomposition of the ADSA line loss data.

[0069] First, calculate the initial meter line loss rate based on the line length and the power drawn by the meter q LOSS ( t )for:

[0070] (5)

[0071] Where: For the i The power supplied to the meter by a power supply, For the i A power supply provides a constant voltage to the meter. For the i The length of the line from the power source to the meter; is the unit resistivity of the line (Ω / meter); n The number of power supplies.

[0072] Then, calculate ADSA j The difference between the direct line loss of the meter at the moment and all the meters under it forms the error factor. for:

[0073] (6)

[0074] Where: For the k Line loss rate of each electric meter; m For the j The number of all electricity meters at this moment.

[0075] Finally, by repeatedly calculating formula (6), until the error factor converges to the minimum value, it is taken as the minimum error factor , the final line loss rate of each meter is calculated based on the minimum error factor:

[0076] (7)

[0077] Step 200: performing correlation analysis on the influencing factors of the line loss in the active distribution station area and the meter line loss data to obtain the weight of each influencing factor on the meter line loss data.

[0078] The key factors affecting ADSA line loss include power supply radius, 10kV distribution transformer capacity, DG output power, load factor, power factor, voltage fluctuation and daily temperature.

[0079] Among them, power supply radius: As the power supply radius increases, the line impedance increases accordingly, resulting in increased line loss.

[0080] 10kV distribution transformer capacity: This capacity directly affects the current of the 0.4kV line. The larger the capacity, the smaller the current passing through the transformer, thus reducing line losses.

[0081] DG output power: When the DG output power is lower than the ADSA load power, the generated power is completely consumed inside the ADSA, and the line loss is low. On the contrary, if the DG output power exceeds the load power, part of the power needs to be connected to the grid through the 10kV distribution transformer, and the line loss increases.

[0082] Load factor: Within the rated power range of the transformer, an increase in the load factor means an increase in the utilization rate of electrical equipment, which helps to reduce line losses.

[0083] Power factor: A lower power factor will increase reactive power loss, thereby increasing line loss rate.

[0084] Voltage fluctuation: An increase in voltage helps reduce line losses.

[0085] Daily temperature: The increase in daily temperature will cause the impedance of the 0.4kV line to increase, thereby increasing line losses.

[0086] This embodiment uses the Partial Maximum Information Coefficient (PMIC) to analyze the factors affecting the line loss of ADSA meters. PMIC is an algorithm that measures the relationship between two variables. The algorithm combines partial mutual information and maximum information coefficient to mine features that have a greater impact on the results and eliminate the error caused by coupled information. It is suitable for processing high-dimensional data and complex relationship data. The specific analysis process is as follows:

[0087] In PIMC, the influencing factor of ADSA line loss is assumed to be z , No. i The analytical signal of the mode (referring to the ith frequency component of the meter line loss data after Hilbert transform) is the meter line loss variable , the output variable (the output after PIMC processing under the i-th frequency component) is .like and z There is a relevant coupling relationship between them, then and Therefore, this embodiment adopts the expected condition To eliminate Include information.

[0088] (8)

[0089] Where: λ i For the i Output variables The marginal probability density function of ; z i For the i Output variables Line loss influencing factors.

[0090] This embodiment uses the Akaike Information Criterion (AIC) as the end condition for ADSA line loss PMIC screening. Screening end value o T for:

[0091] (9)

[0092] In the formula, b i The line loss variable of the ADSA meter has been selected calculate No. i residuals; h is the number of selected ADSA meter line loss variables; q is the number of residuals; as the iterations continue, o T When the value reaches the minimum, the optimal weight set of influencing factors of ADSA is screened and the ADSA meter line loss weight matrix is ​​formed. w MAT .

[0093] (10)

[0094] Where: The meter line loss weight matrix size is m * a ;in m is the number of line loss sequences (equal to the number of meters); a is the number of influencing factors.

[0095] In the process of analyzing the factors affecting the line loss of ADSA meters, the correlation analysis is performed based on the influencing factors such as DG output power and temperature and the line loss of ADSA meters to obtain the weight of each influencing factor on the line loss data of the meter. On this basis, the weight matrix affecting the line loss of ADSA meters is calculated.

[0096] Step 300, using multiple classifiers and bidirectional long short-term memory networks to predict the meter line loss, and superimposing and reconstructing to obtain the line loss prediction result of the active distribution station area.

[0097] In order to improve the ADSA line loss prediction accuracy, this embodiment first uses multiple classifiers to predict the ADSA meter line loss to improve the prediction accuracy of each meter line loss. On this basis, the ADSA line loss prediction results are superimposed and reconstructed to form an overall ADSA line loss prediction value.

[0098] Multi-classifier is an ensemble learning method that completes the task of ADSA line loss prediction by building and combining multiple learners. Multi-classifier first builds a set of individual ADSA meter line loss learners to learn the ADSA meter line loss, and then uses an ensemble strategy to combine the learners to form a prediction result.

[0099] This embodiment uses a bidirectional long short-term memory network (Bi-directional Long Short-TemmMemory, BLSTM) as a multi-classifier learner for ADSA meter line loss prediction. Among them, BLSTM is an optimized recurrent neural network, which consists of two long short-term memory networks (Long Short-TemmMemory, LSTM) in opposite directions. The forward LSTM is responsible for forward propagation and the reverse LSTM is responsible for backward propagation. BLSTM considers the information of the previous and subsequent moments in the processing time series at the same time, which can solve the long-term context dependency of the time series, thereby improving the prediction accuracy.

[0100] This embodiment uses BLSTM to predict ADSA line loss. First, the ADSA meter line loss data is input, and the BLSTM model of each meter is obtained through multi-classifier training. Then, the input ADSA meter line loss data and the influencing factors and their weights are combined to extract positive and negative features, and the extracted features are output. On this basis, the attention mechanism model is used to score each meter prediction model, increase the weight of the effective features of the ADSA meter prediction model, and improve the prediction accuracy. Finally, the ADSA line loss prediction result is obtained through full connection layer mapping.

[0101] Specific as Figure 2 As shown in the figure, the structure of using BLSTM for ADSA line loss prediction includes six parts: input layer, multi-classifier layer, bidirectional LSTM layer, attention mechanism, fully connected layer and output layer.

[0102] Among them, the input layer: ADSA meter line loss data q 1, q 2,… q n The loss is input into the multi-classifier.

[0103] Multiple classifiers: First, multiple benchmark classifiers are constructed based on the ADSA meter line loss data to make the best prediction for each ADSA meter data. Then, each line loss benchmark classifier is trained separately based on BLSTM to ensure that each BLSTM is suitable for the corresponding meter line loss prediction; finally, the weight of each benchmark classifier is determined.

[0104] Bidirectional LSTM layer: In the bidirectional BLSTM, ADSA meter line loss data, influencing factors and their weights are input, and the forward network is used to extract the features of the ADSA meter line loss data. Then, the reverse network is used to feed forward to the same output layer of the ADSA meter line loss model to form a hidden state. h 1, h 2,… h n . And calculate the input and query values ​​through the scoring function σ Then, the score is normalized using the normalization function, and the normalized value is f =( f 1, f 2,… f n ), No. i Normalized value f i for:

[0105] (11)

[0106] Where: exp is the normalized function softmax; r is the scoring function; h i is the output of BLSTM i A hidden state.

[0107] Fully connected layer: In the fully connected layer of the ADSA line loss prediction BLSTM, the output value of the attention mechanism layer is connected and mapped to the output result.

[0108] Output layer: Converts the data of the fully connected layer into ADSA line loss prediction results.

[0109] Through multi-classifier training of different BLSTM models, secondly, combined with the input ADSA meter line loss data, forward and reverse feature privileges are performed, and the extracted features are output. On this basis, the attention mechanism model is used to score each meter prediction model, increase the weight of the effective features of the ADSA meter prediction model, and improve the prediction accuracy. Finally, the ADSA meter line loss prediction result is output through the fully connected layer.

[0110] The ADSA line loss prediction technology proposed in this embodiment first decomposes the ADSA historical line loss data into the line loss data of each meter as the basis for ADSA line loss prediction, and then performs correlation analysis on the DG output power, temperature and other data with the meter line loss data to obtain the weight of each influencing factor on the meter line loss data; on this basis, multiple classifiers and BLSTM are used to predict the meter line loss, and the ADSA line loss prediction results are superimposed and reconstructed, thereby realizing reliable and effective prediction of ADSA line loss.

[0111] Based on the same technical concept as above, this embodiment also proposes a line loss prediction system for active distribution area, such as Figure 3 As shown, the system proposed in this embodiment includes:

[0112] A decomposition module is used to decompose the historical line loss data of the active distribution station area into the line loss data of each electric meter in the active distribution station area.

[0113] The correlation analysis module is used to perform correlation analysis on the line loss influencing factors of the active distribution station area and the meter line loss data to obtain the weight of each influencing factor on the meter line loss data. The correlation analysis process is specifically as described in the above step 200, and will not be repeated here.

[0114] And, a prediction module, which uses multiple classifiers and bidirectional long short-term memory networks to predict the meter line loss, and superimposes and reconstructs to obtain the active distribution area line loss prediction result. The line loss prediction process is specifically as described in the above step 300, and will not be repeated here.

[0115] Furthermore, the decomposition module also includes:

[0116] The data acquisition and processing unit is used to collect and repair the historical data of the electric meter in the ADSA, and calculate the ADSA historical line loss data. The specific process is as described in the above step 101, and will not be repeated here.

[0117] The first calculation unit is used to calculate the line length from each electric meter in the ADSA to the power supply point. The specific calculation method is as described in the above step 102, which will not be repeated here.

[0118] And, a second calculation unit, which calculates the ADSA meter line loss according to the line length from each meter to the power source point and the power drawn by the meter. The specific calculation process is as described in the above step 103, which will not be repeated here.

[0119] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0120] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0121] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0122] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0123] The specific implementation methods described above further illustrate the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only the specific implementation method of the present application and is not intended to limit the scope of protection of the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A method for predicting line loss in an active distribution area, characterized in that: The line loss prediction method comprises: Decomposing the historical line loss data of the active distribution substation area into the line loss data of each electric meter in the active distribution substation area; Conduct correlation analysis on the influencing factors of line loss in active distribution substation area and meter line loss data, and obtain the weight of each influencing factor on meter line loss data; Based on the line loss data of each meter in the active distribution substation area and its influencing factors and weights, a multi-classifier and a bidirectional long short-term memory network are used to predict the meter line loss, and the line loss prediction result of the active distribution substation area is obtained by superposition and reconstruction; the decomposition of the historical line loss data of the active distribution substation area into the line loss data of each meter in the active distribution substation area specifically includes: Collecting and repairing the historical data of electric meters in the active distribution station area, and calculating the historical line loss data of the active distribution station area based on the repaired historical data of electric meters; Calculate the line length from each electric meter to the power supply point in the active distribution station area; The line loss data of each meter in the active distribution substation area is calculated based on the line length from each meter in the active distribution substation area to the power supply point and the power drawn by the meter; the historical data of the meters in the active distribution substation area are collected and repaired abnormally, and the historical line loss data of the active distribution substation area is calculated based on the repaired historical data of the meters, specifically including: Normalize the original meter data sequence; The normalized electric meter data sequence is used as the input of the recurrent neural network, and the sequence output after the recurrent neural network repair is the repaired electric meter data sequence; The line loss rate of the active distribution station area at the jth moment is calculated by the following formula: for: Where: They are the off-grid power and on-grid power of the active distribution station area master meter at the jth moment respectively; It is the sum of the off-grid electricity of the user meters in the active distribution station area at the jth moment; is the sum of the DG online power at the jth moment in the active distribution station area; the line length L from the meter to the power supply point DM The calculation formula is: L DM =(t BG -t END )v DM Where: t BG The timestamp of the signal emitted by the power point; t END v is the time when the electric meter receives the signal transmitted by the power source; DM is the speed of signal propagation; the line loss data calculation process of each meter in the active distribution station area includes: The initial meter line loss rate is calculated based on the line length from the meter to the power source and the power drawn by the meter. Calculate the difference between the direct line loss of the jth point in the active distribution station area and all the electric meters under it to form an error factor; Based on the initial electric meter line loss rate and the error factor, a final line loss rate of each electric meter is calculated.

2. The method for predicting line loss in an active distribution area according to claim 1, characterized in that: The correlation analysis of the influencing factors of the active distribution station area line loss and the meter line loss data is performed to obtain the weight of each influencing factor on the meter line loss data, specifically including: The partial maximum information coefficient is used to analyze the influencing factors of meter line loss, and the Akaike information rule is used as the end condition for screening the weight set of influencing factors; When the screening end value reaches the minimum, the optimal weight set of influencing factors is obtained, and a line loss weight matrix of electric meters in the active distribution station area is formed.

3. The method for predicting line loss in an active distribution area according to claim 2, characterized in that: The method of using multiple classifiers and bidirectional long short-term memory networks to predict the line loss of electric meters and superimposing and reconstructing the line loss prediction results of the active distribution station area specifically includes: Inputting the line loss data of each electric meter in the active distribution station area into a multi-classifier; According to the line loss data of each electric meter in the active distribution station area, a reference classifier of each electric meter is constructed respectively, and each reference classifier is trained separately based on a bidirectional long short-term memory network to obtain an electric meter line loss model suitable for the corresponding electric meter; In the bidirectional long short-term memory network, the input meter line loss data and its influencing factors and weights are combined. The forward network is first used to extract the meter line loss data features, and then the reverse network is used to feed forward to the same output layer of the meter line loss model to form a hidden state. The correlation between the hidden state and the query value is calculated through the scoring function to obtain the score, and then the normalization function is used to normalize the score. The normalized value is mapped to the output layer through the fully connected layer to obtain the line loss prediction result of the active distribution station area.

4. An active distribution area line loss prediction system, characterized in that: The prediction system comprises: A decomposition module, the decomposition module is used to decompose the historical line loss data of the active distribution station area into the line loss data of each electric meter in the active distribution station area; A correlation analysis module, which is used to perform correlation analysis on the line loss influencing factors of the active distribution station area and the meter line loss data, and obtain the weight of each influencing factor on the meter line loss data; And, a prediction module, which uses a multi-classifier and a bidirectional long short-term memory network to predict the meter line loss based on the line loss data of each meter in the active distribution station area and its influencing factors and weights, and superimposes and reconstructs to obtain the line loss prediction result of the active distribution station area; The decomposition of the historical line loss data of the active distribution station area into the line loss data of each electric meter in the active distribution station area specifically includes: Collecting and repairing the historical data of electric meters in the active distribution station area, and calculating the historical line loss data of the active distribution station area based on the repaired historical data of electric meters; Calculate the line length from each electric meter to the power supply point in the active distribution station area; The line loss data of each meter in the active distribution substation area is calculated based on the line length from each meter in the active distribution substation area to the power supply point and the power drawn by the meter; the historical data of the meters in the active distribution substation area are collected and repaired abnormally, and the historical line loss data of the active distribution substation area is calculated based on the repaired historical data of the meters, specifically including: Normalize the original meter data sequence; The normalized electric meter data sequence is used as the input of the recurrent neural network, and the sequence output after the recurrent neural network repair is the repaired electric meter data sequence; The line loss rate of the active distribution station area at the jth moment is calculated by the following formula: for: Where: They are the off-grid power and on-grid power of the active distribution station area master meter at the jth moment respectively; It is the sum of the off-grid electricity of the user meters in the active distribution station area at the jth moment; is the sum of the DG online power at the jth moment in the active distribution station area; the line length L from the meter to the power supply point DM The calculation formula is: L DM =(t BG -t END )v DM Where: t BG The timestamp of the signal emitted by the power point; t END v is the time when the electric meter receives the signal transmitted by the power source; DM is the speed of signal propagation; the line loss data calculation process of each meter in the active distribution station area includes: The initial meter line loss rate is calculated based on the line length from the meter to the power source and the power drawn by the meter. Calculate the difference between the direct line loss of the jth point in the active distribution station area and all the electric meters under it to form an error factor; Based on the initial electric meter line loss rate and the error factor, a final line loss rate of each electric meter is calculated.

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 3 are implemented.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

Images