An intelligent measurement and management system for distribution network line losses

Abstract

The invention discloses an intelligent measurement and management system for distribution network line loss, which relates to the field of intelligent measurement of distribution network line loss. The technical scheme thereof mainly includes: collecting multi-dimensional heterogeneous data of the distribution network within the current cycle time; obtaining a comprehensive data set, and performing cluster analysis on the comprehensive data set to obtain a feature vector set; constructing a neural network YC‑Net measurement model improved based on a genetic algorithm and training it, inputting the feature vector set into the trained YC‑Net measurement model to obtain a predicted line loss set of the distribution network within the current cycle time, the predicted line loss set consisting of predicted line loss values, and each predicted line loss value corresponds to sub-data in the feature vector; performing line loss rate difference calculation between each predicted line loss value in the predicted line loss set and the corresponding actual line loss value, and when the line loss rate difference exceeds a preset range, checking the power supply area corresponding to the power supply node, so as to realize intelligent measurement and management of the distribution network line loss.

Description

Technical Field

[0001] The present invention relates to the field of intelligent measurement of line losses in distribution networks, and more specifically, to an intelligent measurement and management system for line losses in distribution networks. Background Art

[0002] The line loss of the distribution network refers to the sum of the power losses generated in the entire process from the power plant sending electricity, through transmission, transformation, distribution and supply to customers. The loss generated in the power transmission process is called "line loss". This loss will lead to energy waste and reduce the efficiency of power transmission. With the continuous increase in user electricity demand, the distribution load continues to rise, and the line loss of the distribution network remains high, which has increased the cost investment and safety risks of the distribution network to a certain extent. In addition, it is difficult to measure and process current and voltage signals, and it is difficult to conduct in-depth resource mining. Complex wiring methods, long line radius, and obvious voltage fluctuations are prone to excessive line loss, which affects the economic benefits of power transmission and distribution. Due to the complex interlacing of lines, it is difficult to quantify the line loss of the distribution network. How to quantify and evaluate the line loss of the distribution network and accurately analyze the abnormal line loss nodes are the key points to be solved by the present invention. Therefore, it is necessary to design the intelligent measurement and management function of the distribution network line loss. Summary of the invention

[0003] In view of the problems existing in the prior art, the purpose of the present invention is to provide an intelligent measurement and management system for distribution network line losses, so as to realize intelligent measurement and management of distribution network line losses.

[0004] To achieve the above object, the present invention provides the following technical solution: a distribution network line loss intelligent measurement and management system, the distribution network line loss intelligent measurement and management system comprising:

[0005] Data acquisition module: used to collect multi-dimensional heterogeneous data of the distribution network within the current cycle time;

[0006] The data processing module is used to comprehensively analyze the multi-dimensional heterogeneous data, obtain a comprehensive data set, and perform cluster analysis on the comprehensive data set to obtain a feature vector cluster;

[0007] The analysis and management module is used to construct and train the neural network YC-Net calculation model based on the genetic algorithm, collect and input the feature vectors into the trained YC-Net calculation model, and obtain the predicted line loss set of the distribution network in the current cycle time. The predicted line loss set consists of predicted line loss values, and each predicted line loss value corresponds to the sub-data in the feature vector;

[0008] Adjustment module: within the current cycle, the line loss rate difference is calculated between each predicted line loss value in the predicted line loss set and the corresponding actual line loss value. When the line loss rate difference exceeds the preset range, the coordinates of the power supply node are extracted, and the power supply area corresponding to the power supply node is checked. When the line loss rate difference does not exceed the preset range, the line loss of the distribution network is normal.

[0009] The modules are connected to each other via wired and / or wireless means.

[0010] Preferably, the multi-dimensional heterogeneous data includes power grid topology, voltage and current data information and environmental impact coefficient;

[0011] The grid topology includes distribution lines, power supply nodes, and the power supply areas corresponding to the power supply nodes. The distribution lines, power supply areas, and power supply nodes are all obtained through the official website of the grid center or the information database of the power supply enterprise;

[0012] Voltage and current data information is obtained through power and voltage sensors;

[0013] The environmental impact coefficient is affected by temperature data, humidity data, wind speed data and distribution equipment function data. The temperature data, humidity data and wind speed data are obtained based on the regional weather information disclosed by the weather forecast. The distribution equipment function data is represented by the average depreciation rate of the power supply equipment in the power supply area.

[0014] Environmental impact coefficient acquisition includes:

[0015]

[0016] Among them, α and β are weight coefficients, k1, k2 and k3 are influencing factors, Qw is the temperature data, Sd is the humidity data, and Fs is the wind speed data. represents the functional data of the power distribution equipment, mj represents the area of ​​the power supply area, and exp represents the natural exponential with e as the base.

[0017] Preferably, the comprehensive analysis of multi-dimensional heterogeneous data includes:

[0018] Constructing a comprehensive data set: constructing a two-dimensional data model according to the power grid topology; mapping the environmental impact coefficient to the corresponding power supply node in the two-dimensional space data model after normalization processing; adding a time tag to the preprocessed voltage and current data information, and mapping the voltage and current data information after adding the time tag to the corresponding power supply node in the two-dimensional space data model, storing the environmental impact coefficient and the voltage and current data information as attribute fields of the two-dimensional space model to form a comprehensive data set, which is composed of sub-data, each of which includes the voltage and current data information, the time tag and the environmental impact coefficient at the corresponding power supply node;

[0019] The specific method of adding the time tag is as follows: obtaining the periodic time period collected by the power and voltage sensors, binary encoding the periodic time period, obtaining a string of binary characters, and associating the binary characters with the voltage and current data information within the periodic time period;

[0020] Ways to build a two-dimensional data model include:

[0021] According to the river network topology, the power grid topology structure is obtained, including distribution lines, power supply nodes and the power supply areas corresponding to the power supply nodes. The distribution lines are taken as linear elements, the power supply nodes are converted into point elements, and the power supply areas are taken as surface elements. The power supply nodes and distribution lines are converted into a unified coordinate system to obtain the plane coordinates of each power supply node and distribution line.

[0022] Preferably, the preprocessing of the voltage and current data information includes:

[0023] During the cycle time, the voltage and current data information at each power supply node is obtained, and the voltage and current data information is moved and averaged;

[0024] Moving averages include:

[0025] Step S21: Select a window size of M, and set the original signal data point in the voltage and current data information to x[n], where n=0, 1, ..., N-1;

[0026] Step S22: Calculate the moving average value, average the values ​​of consecutive data points within the window size M to obtain the moving average value, and the moving average value is calculated using the moving average processing formula.

[0027] Moving average processing formula:

[0028]

[0029] represents the moving average x[n] at position n pj From Location arrive The average value of M data points at the position nk, where x[nk] represents the value of the original signal data point at the position nk, k represents the moving scale, f1 represents the smoothing coefficient, and N represents the number of original signal data points in the voltage and current data information;

[0030] Step S23: applying moving average to replace the value of the original data point with the calculated moving average value;

[0031] Step S24: repeating steps S22 to S23, applying moving average processing to all original signal data points in the voltage and current data information in sequence until all original data points are smoothed;

[0032] Step S25: repeat step S24, apply moving average processing to the voltage and current data information at each power supply node in turn, and obtain smoothed voltage and current data information.

[0033] Preferably, the smoothing coefficient f1 is obtained by:

[0034] Get the values ​​of continuous data points within the window size M, and extract the data point xi with the largest value within the window size M max The data point xi with the smallest sum min , then set the window size M except for the data point xi max and data point xi min The data points outside are xi p , and p = 0, 1, ..., M-2, η and is the weight coefficient, then

[0035]

[0036] Preferably, cluster analysis of the comprehensive data set comprises:

[0037] Using the improved k-means algorithm, the voltage and current data information at each power supply node is used as the feature data to perform cluster analysis on the sub-data of the comprehensive data set;

[0038] Assign data points to the nearest cluster center: randomly initialize the cluster center, calculate the Euclidean distance between the voltage and current data information in each sub-data and the cluster center, and assign them to the cluster center with the nearest distance. All sub-data filled in each cluster center are regarded as a cluster;

[0039] Update cluster center: introduce cost function I, set cost threshold a, if I≥a, it means cost function I is greater than or equal to cost threshold a, then you need to update cluster center, calculate the mean of all sub-data in each cluster, take this mean as the new cluster center, and distribute sub-data, if I<a, it means cost function I is less than cost threshold a, the lower cost function I is, the closer to cluster center, then you do not need to recalculate cluster center, cost function I calculation formula is:

[0040]

[0041] Where D represents the number of cluster centers, d represents the index of the cluster center, R represents the number of sub-data in the comprehensive dataset, r represents the index of the sub-data, and ||Vr-Ld|| represents the Euclidean distance between the r-th sub-data and the d-th cluster center;

[0042] Construct feature vector collection: take all sub-data of each cluster as a feature vector, and combine all feature vectors into a feature vector collection.

[0043] Preferably, the YC-Net calculation model construction includes:

[0044] Three parallel paths are used, and the three paths are set to 1×1, 3×3, and 5×5 convolution kernels respectively. The 1×1 convolution kernel is set before the 5×5 and 3×3 convolution kernels;

[0045] The number of Inception modules is set to 3, the number of Inception convolution kernels is set to 512, and two fully connected layers are added above the softmax output layer. The fully connected layers use ReLU as the activation function, and the number of neurons in the fully connected layers is set to 2048 and 1024 respectively;

[0046] The YC-Net measurement model defines the loss function as the Huber Loss loss function, which calculates the loss difference between the actual line loss value of each power supply node in the distribution network and the corresponding predicted line loss value.

[0047] Preferably, the hidden layer node setting of the YC-Net calculation model includes:

[0048] Constrain the hidden layer nodes and set the hidden layer node value constraints as:

[0049]

[0050] Among them, Nc represents the number of hidden layer nodes, No represents the number of output nodes, Ni represents the number of input nodes, and m represents the adjustment constant.

[0051] Preferably, after the hidden layer node values ​​of the YC-Net calculation model are determined, the index of the line loss prediction accuracy of the model is evaluated according to the fitness function, and the YC-Net calculation model with the best line loss prediction capability when the fitness function is larger is selected.

[0052] The fitness function is expressed as:

[0053]

[0054] Among them, ff represents the fitness function, which is used to measure the quality of the model, w represents the index of the iteration number, that is, the current iteration number, W is the iteration number, b is the index of the hidden layer node, which is used to traverse each hidden layer node. is the theoretical line loss value of the bth node at the wth iteration. The theoretical line loss value is the expected loss obtained based on the actual distribution network using the traditional power calculation method. is the calculated line loss value of the bth node at the wth iteration. The calculated line loss value is the line loss value predicted by inputting the feature vector aggregation into the trained neural network model.

[0055] Preferably, the YC-Net calculation model training process includes:

[0056] S11, collecting characteristic vectors of the distribution network within the historical period time t as a sample set, and the sample set has sufficient samples;

[0057] S11, input each feature vector of each feature vector set in the sample set into the YC-Net measurement model in turn for forward propagation;

[0058] S12. Obtain the predicted line loss value output by the YC-Net measurement model, calculate the loss value using the Huber Loss function, and update the weight and bias of the model through the back propagation algorithm;

[0059] S13. Repeat S12 and S13 until the loss function value of the YC-Net calculation model no longer changes, stop training, and obtain the trained YC-Net calculation model.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] In the present invention, the present invention realizes smooth and high-quality input of signal data through moving average preprocessing of voltage and current data information, improves the effect of subsequent data processing and model training as well as retrieval and response speed, provides an accurate data basis for subsequent analysis and optimization, and realizes rapid positioning of power supply nodes; through the constraint restrictions of hidden layer nodes and the selection of the maximum value of fitness function, the learning efficiency of the YC-Net calculation model is accelerated, the efficient retrieval processing of the model is significantly improved, and new kinetic energy is injected into the agility and flexibility of data processing. The line loss prediction based on the YC-Net calculation model implemented by this system deeply mines and utilizes the power resources of the distribution network; the adaptive optimization mechanism continuously improves the overall efficiency of data processing, and multiple modules cooperate with each other to realize comprehensive data collection, processing and analysis, provide real-time and accurate data support, and the learning model helps managers better understand the line loss situation of the power system operation and make corresponding decision recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 The present invention proposes a structural schematic diagram of a distribution network line loss intelligent measurement and management system;

[0063] Figure 2 It is a schematic diagram of the method applied to the intelligent measurement and management system of distribution network line losses in the present invention. DETAILED DESCRIPTION

[0064] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0065] Embodiment 1

[0066] Reference Figure 1 and Figure 2 Embodiment 1 further illustrates a distribution network line loss intelligent measurement and management system proposed by the present invention.

[0067] The line loss of the distribution network refers to the sum of the power losses generated during the entire process from the power plant sending electricity to the power transmission, transformation, distribution and supply to customers. The loss generated during the power transmission process is called "line loss". This loss will lead to energy waste and reduce the efficiency of power transmission. As the demand for electricity from users continues to increase, the distribution load continues to rise, and the line loss of the distribution network remains high, which to a certain extent increases the cost investment and safety risks of the distribution network. Complex wiring methods, long line radius, and obvious voltage fluctuations are prone to excessive line loss, affecting the economic benefits of power transmission and distribution. How to quantify and analyze the causes of distribution network line loss and accurately analyze abnormal line loss nodes are the key points to be solved by the present invention.

[0068] The calculation of the actual total loss of the power grid is affected by operating parameters, equipment parameters and other factors. From the current line loss composition, it is mainly divided into fixed loss, variable loss and other losses. Fixed loss does not change with load changes, such as the iron loss of the transformer; variable loss changes according to load changes. This loss is proportional to the square of the current, that is, the greater the current, the greater the loss. In the distribution network, variable loss is an important consideration because it directly affects the operating efficiency and power supply quality of the power grid. To reduce the overall loss, management and optimization are required from all aspects. The online monitoring system has a complete framework, including modules such as data acquisition, data processing, data analysis, visual interface, mechanism adjustment, etc. These modules work together to achieve comprehensive data collection, processing and analysis, provide real-time and accurate data support, and the learning model helps managers better understand the line loss of the power system and make corresponding decision recommendations. In addition, the system also autonomously integrates a modern intelligent database, which can not only process and analyze a large amount of data efficiently and accurately, but also realize more intelligent and automated power transmission and supply, which makes the entire system more in line with the development needs of modern power grids and injects new vitality into the sustainable development of the power industry.

[0069] In addition, the system also includes a database, which is used to store system operation data and is provided with a temporary storage area for interface data.

[0070] Data acquisition module: used to collect multi-dimensional heterogeneous data of the distribution network within the current cycle time;

[0071] Multi-dimensional heterogeneous data includes power grid topology, voltage and current data information and environmental impact coefficients;

[0072] The power grid topology includes distribution lines, power supply nodes and the power supply areas corresponding to the power supply nodes. The distribution lines, power supply areas and power supply nodes are all obtained through the official website of the power grid center or the information database of the power supply enterprise. The power supply node represents the power equipment, the distribution line represents the connection between the power equipment, and the power supply area represents the effective power supply range that the power supply node can provide.

[0073] The voltage and current data information is obtained through power and voltage sensors. The present invention collects the voltage and current data information at each power supply node, that is, various sensors are installed on the power equipment of the distribution network, such as current sensors, voltage sensors, power sensors, etc. These sensors will continuously monitor the changes in parameters such as current, voltage and power;

[0074] The environmental impact coefficient is affected by temperature data, humidity data, wind speed data and distribution equipment function data. The temperature data, humidity data and wind speed data are obtained based on the regional weather information disclosed by the weather forecast. The distribution equipment function data is represented by the average depreciation rate of the power supply equipment in the power supply area. The average depreciation rate is averaged by the depreciation rates of various power supply equipment counted by management personnel. The depreciation rate is obtained by the ratio of the actual usage time of the power supply equipment to the standard usage time.

[0075] The collected data will be transmitted to the data processing module through the communication network. This usually involves network communication technologies such as Ethernet, Wi-Fi, LoRa, etc. to ensure reliable data transmission.

[0076] Environmental impact coefficient acquisition includes:

[0077]

[0078] Among them, α and β are weight coefficients, k1, k2 and k3 are influencing factors, Qw is the temperature data, Sd is the humidity data, and Fs is the wind speed data. represents the functional data of the power distribution equipment, mj represents the area of ​​the power supply area, and exp represents the natural exponential with e as the base;

[0079] The data processing module is used to comprehensively analyze the multi-dimensional heterogeneous data, obtain a comprehensive data set, and perform cluster analysis on the comprehensive data set to obtain a feature vector cluster;

[0080] Comprehensive analysis of multi-dimensional heterogeneous data includes:

[0081] Constructing a comprehensive data set: constructing a two-dimensional data model according to the power grid topology; mapping the environmental impact coefficient to the corresponding power supply node in the two-dimensional space data model after normalization processing; adding a time tag to the preprocessed voltage and current data information, and mapping the voltage and current data information after adding the time tag to the corresponding power supply node in the two-dimensional space data model, storing the environmental impact coefficient and the voltage and current data information as attribute fields of the two-dimensional space model to form a comprehensive data set, which is composed of sub-data, each of which includes the voltage and current data information, the time tag and the environmental impact coefficient at the corresponding power supply node;

[0082] The specific method of adding the time tag is as follows: obtaining the periodic time period collected by the power and voltage sensors, binary encoding the periodic time period, obtaining a string of binary characters, and associating the binary characters with the voltage and current data information within the periodic time period;

[0083] Ways to build a two-dimensional data model include:

[0084] The power grid topology structure is obtained according to the river network topology structure, including distribution lines, power supply nodes and power supply areas corresponding to the power supply nodes. The distribution lines are taken as linear elements, the power supply nodes are converted into point elements, and the power supply areas are taken as surface elements. The power supply nodes and distribution lines are converted into a unified coordinate system to obtain the plane coordinates of each power supply node and distribution line.

[0085] Preprocessing of voltage and current data information includes:

[0086] When current and voltage are collected, a waveform signal is usually collected rather than a single value within a cycle time. The changes of current and voltage are usually continuous, so the current and voltage data collected over time are presented as continuous waveforms. In the actual data collection of the present invention, a sampling device (such as a sensor) is used to sample the current and voltage at a certain frequency to obtain the waveform signals of the current and voltage. These sampled data can be discrete data points or continuous analog signals, depending on the sampling frequency and data transmission method.

[0087] The collected current waveform signal can provide more comprehensive and detailed information only after preprocessing, including the amplitude, frequency, waveform characteristics, etc. of current and voltage, making these data more prominent. Therefore, in the intelligent measurement and management system of distribution network line losses, the collection and preprocessing of waveform signals is a crucial link.

[0088] During the cycle time, voltage and current data information at each power supply node is obtained, and the voltage and current data information are pre-processed in a manner including moving average.

[0089] Moving averages include:

[0090] Step S21: Select the window size as M. The window size determines the number of data points for averaging the signal and affects the degree of smoothing. Set the original signal data points in the voltage and current data information as x[n], where n=0, 1, ..., N-1.

[0091] Step S22: Calculate the moving average value, average the values ​​of consecutive data points within the window size M to obtain the moving average value, and the moving average value is calculated using the moving average processing formula.

[0092] Moving average processing formula:

[0093]

[0094] represents the moving average x[n] at position n pj From Location arrive The average value of M data points at the position nk, where x[nk] represents the value of the original signal data point at the position nk, k represents the moving scale, f1 represents the smoothing coefficient, and N represents the number of original signal data points in the voltage and current data information;

[0095] Step S23: Apply moving average, and replace the value of the original data point with the calculated moving average to achieve signal smoothing. This can remove high-frequency noise and rapid changes in the signal, making the signal smoother;

[0096] Step S24: Repeat steps S22 to S23, and apply moving average processing to all original signal data points in the voltage and current data information in sequence until all original data points are smoothed.

[0097] Step S25: repeat step S24, apply moving average processing to the voltage and current data information at each power supply node in turn, and obtain smoothed voltage and current data information.

[0098] At the beginning and end of the signal, due to the limitation of window size, it is impossible to calculate the complete moving average, so boundary processing is required. Common methods include filling with zero values ​​or smoothing with data points before and after the boundary points.

[0099] The smoothing coefficient f1 is obtained by: obtaining the values ​​of continuous data points within the window size M, extracting the data point xi with the largest value within the window size M maxThe data point xi with the smallest sum min , then set the window size M except for the data point xi max and data point xi min The data points outside are xi p , and p = 0, 1, ..., M-2, η and is the weight coefficient, then

[0100]

[0101] The present invention sets M=3, that is, selects the window size as 3, and sets the original signal data points in the voltage and current data information as,

[0102] [50,51,48,49,300,52,47,45,53,48,49,48,47,48,200,52,48,49,46,45],

[0103] Use moving average to process, calculate the moving average and apply the moving average to get the processed signal as,

[0104] [50.0,49.67,49.67,132.67,133.67,49.67,48.0,48.33,48.67,50.0,48.33,48.0,47.67,48.33,100.0,100.0,49.67,47.67,46.33,46.0],

[0105] Cluster analysis of the comprehensive dataset includes:

[0106] Using the improved k-means algorithm, the voltage and current data information at each power supply node is used as the feature data to perform cluster analysis on the sub-data in the comprehensive data set.

[0107] Assign data points to the nearest cluster center: randomly initialize the cluster center, calculate the Euclidean distance between the voltage and current data information in each sub-data and the cluster center, and assign them to the cluster center with the nearest distance. All sub-data filled in each cluster center are regarded as a cluster;

[0108] Update cluster center: introduce cost function I, set cost threshold a, if I≥a, it means cost function I is greater than or equal to cost threshold a, then you need to update cluster center, calculate the mean of all sub-data in each cluster, take this mean as the new cluster center, and distribute sub-data, if I<a, it means cost function I is less than cost threshold a, the lower cost function I is, the closer to cluster center, then you do not need to recalculate cluster center, cost function I calculation formula is:

[0109]

[0110] Where D represents the number of cluster centers, d represents the index of the cluster center, R represents the number of sub-data in the comprehensive dataset, r represents the index of the sub-data, and ||Vr-Ld|| represents the Euclidean distance between the r-th sub-data and the d-th cluster center;

[0111] Construct feature vector collection: take all sub-data of each cluster as a feature vector, and combine all feature vectors into a feature vector collection.

[0112] The analysis and management module is used to construct and train the neural network YC-Net calculation model based on the genetic algorithm, collect and input the feature vectors into the trained YC-Net calculation model, and obtain the predicted line loss set of the distribution network in the current cycle time. The predicted line loss set consists of predicted line loss values, and each predicted line loss value corresponds to the sub-data in the feature vector;

[0113] Build the YC-Net calculation model:

[0114] Three parallel paths are used, and the three paths are set to 1×1, 3×3, and 5×5 convolution kernels respectively. The 1×1 convolution kernel is set before the 5×5 and 3×3 convolution kernels;

[0115] The number of Inception modules is set to 3, the number of Inception convolution kernels is set to 512, and two fully connected layers are added above the softmax output layer. The fully connected layers use ReLU as the activation function, and the number of neurons in the fully connected layers is set to 2048 and 1024 respectively;

[0116] The hidden layer nodes are constrained. Usually, the number of hidden layer nodes should not be too large to avoid overfitting the network to the training data, nor too small to ensure that the network can fully learn the characteristics of the data. The value constraints of the hidden layer nodes are set as follows:

[0117]

[0118] Among them, Nc represents the number of hidden layer nodes, No represents the number of output nodes, Ni represents the number of input nodes, and m represents the adjustment constant;

[0119] In summary, the function of this formula is to determine a suitable range of the number of hidden layer nodes Nc based on the number of output nodes No and the number of input nodes Ni. This can make the structure of the neural network reasonable, neither overfitting nor underfitting, thereby improving the generalization ability and prediction accuracy of the model.

[0120] After determining the value range of the number of hidden layer nodes, the model's line loss prediction accuracy is evaluated based on the fitness function. The larger the fitness function, the more accurate the model's prediction and the better the line loss parameter relationship is captured by the model. Therefore, in the optimization process of the genetic algorithm, individuals with larger fitness function values ​​are selected for further parameter adjustment and optimization. The purpose of the fitness function is to select the best individual, that is, the YC-Net measurement model with the best line loss prediction ability.

[0121] The fitness function is expressed as:

[0122]

[0123] Among them, ff represents the fitness function, which is used to measure the quality of the model, w represents the index of the iteration number, that is, the current iteration number, W is the iteration number, b is the index of the hidden layer node, which is used to traverse each hidden layer node. is the theoretical line loss value of the bth node at the wth iteration. The theoretical line loss value is the expected loss obtained based on the actual distribution network using the traditional power calculation method. is the calculated line loss value of the bth node at the wth iteration. The calculated line loss value is the line loss value predicted by inputting the feature vector aggregation into the trained neural network model.

[0124] The YC-Net calculation model defines the loss function as:

[0125]

[0126] Among them, y is the actual line loss value of each power supply node in the distribution network in the real state, δ is the control behavior parameter, which is used to control the behavior of the loss function. Output the predicted line loss value corresponding to the power supply node in the predicted line loss set for the YC-Net calculation model.

[0127] The YC-Net calculation model training process includes:

[0128] S11, collecting characteristic vectors of the distribution network within the historical period time t as a sample set, and the sample set has sufficient samples;

[0129] S11, input each feature vector of each feature vector set in the sample set into the YC-Net measurement model in turn for forward propagation;

[0130] S12. Obtain the predicted line loss value output by the YC-Net measurement model, calculate the loss value using the loss function, and update the weight and bias of the model through the back propagation algorithm;

[0131] S13. Repeat S12 and S13 until the loss function value of the YC-Net calculation model no longer changes, stop training, and obtain the trained YC-Net calculation model.

[0132] Adjustment module: In the current cycle, the line loss rate difference is calculated between each predicted line loss value in the predicted line loss set and the corresponding actual line loss value. When the line loss rate difference exceeds the preset range, the power supply node is extracted and the power supply area corresponding to the power supply node is checked. When the line loss rate difference does not exceed the preset range, the line loss of the distribution network is normal.

[0133] When the line loss rate difference exceeds the preset range, it is determined that there is abnormal distribution network line loss. The maintenance personnel should go to the site to inspect and handle it according to the power supply area corresponding to the power supply node. Investigate the cause of the abnormal line loss and analyze whether there is abnormal power consumption behavior. Carry out power saving guidance, line loss compensation control, optimize the distribution network line structure, and improve power quality for abnormal power consumption behavior.

[0134] The modules are connected to each other via wired and / or wireless means.

[0135] In this embodiment, the present invention realizes smooth and high-quality input of signal data through moving average preprocessing of voltage and current data information, improves the effect of subsequent data processing and model training as well as retrieval and response speed, provides an accurate data basis for subsequent analysis and optimization, and realizes rapid positioning of power supply nodes; through the constraint restrictions of hidden layer nodes and the selection of the maximum value of fitness function, the learning efficiency of the YC-Net calculation model is accelerated, the efficient retrieval processing of the model is significantly improved, and new kinetic energy is injected into the agility and flexibility of data processing. The line loss prediction based on the YC-Net calculation model implemented by this system deeply mines and utilizes the power resources of the distribution network; the adaptive optimization mechanism continuously improves the overall efficiency of data processing, and multiple modules cooperate with each other to realize comprehensive data collection, processing and analysis, provide real-time and accurate data support, and the learning model helps managers better understand the line loss situation of the power system operation and make corresponding decision recommendations.

[0136] Embodiment 2

[0137] Reference Figure 1 and Figure 2 , Example 2 further illustrates a distribution network line loss intelligent measurement and management system proposed by the present invention.

[0138] The method for intelligently measuring and managing line losses in a distribution network is applied to an intelligently measuring and managing system for line losses in a distribution network, and comprises the following steps:

[0139] Step S1: Collect multi-dimensional heterogeneous data of the distribution network within the current cycle time;

[0140] Step S2: performing moving average preprocessing on the voltage and current data information of each power supply node in the multi-dimensional heterogeneous data, and constructing the preprocessed multi-dimensional heterogeneous data into a comprehensive data set;

[0141] Step S3: Perform cluster analysis on the comprehensive data set to obtain feature vector clustering;

[0142] Step S4: construct a YC-Net calculation model, obtain a collection of characteristic vectors of the distribution network within the historical cycle time t, import them into the YC-Net calculation model for training, and obtain a trained YC-Net calculation model according to the Huber Loss function;

[0143] Step S5: import the feature vectors in the current cycle time into the trained YC-Net calculation model to obtain the predicted line loss set, calculate the line loss rate difference in sequence, extract the coordinates of the power supply nodes whose line loss rate difference exceeds the preset range, and check the power supply area corresponding to the power supply node.

[0144] The intelligent measurement and management system for distribution network line losses comprises: a data acquisition module for collecting multi-dimensional heterogeneous data of the distribution network within the current cycle time;

[0145] The data processing module is used to comprehensively analyze the multi-dimensional heterogeneous data, obtain a comprehensive data set, and perform cluster analysis on the comprehensive data set to obtain a feature vector cluster;

[0146] The analysis and management module is used to construct and train the neural network YC-Net calculation model based on the genetic algorithm, collect and input the feature vectors into the trained YC-Net calculation model, and obtain the predicted line loss set of the distribution network in the current cycle time. The predicted line loss set consists of predicted line loss values, and each predicted line loss value corresponds to the sub-data in the feature vector;

[0147] Adjustment module: within the current cycle, the line loss rate difference is calculated between each predicted line loss value in the predicted line loss set and the corresponding actual line loss value. When the line loss rate difference exceeds the preset range, the coordinates of the power supply node are extracted, and the coordinates of the power supply node can be represented in a two-dimensional data model. The two-dimensional data model is displayed through a visual interface, and the power supply area corresponding to the power supply node is checked. When the line loss rate difference does not exceed the preset range, the line loss of the distribution network is normal.

[0148] The modules are connected to each other via wired and / or wireless means.

[0149] In addition, according to the implementation of the present application, the process described in the figure of a distribution network line loss intelligent measurement and management system can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, which stores machine-readable instructions, and the machine-readable instructions can be executed by a processor to execute instructions corresponding to the method steps provided in the present application. Of course, the architecture shown in the figure of a distribution network line loss intelligent measurement and management system is only exemplary. When implementing different devices, adaptive selection or adjustment can be performed according to actual needs.

[0150] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0151] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A distribution network line loss intelligent measurement and management system, characterized in that: The distribution network line loss intelligent measurement and management system includes: The data acquisition module is used to collect multi-dimensional heterogeneous data of the distribution network within the current cycle time; The multi-dimensional heterogeneous data includes power grid topology, voltage and current data information and environmental impact coefficient; The grid topology includes distribution lines, power supply nodes, and the power supply areas corresponding to the power supply nodes. The distribution lines, power supply areas, and power supply nodes are all obtained through the official website of the grid center or the information database of the power supply enterprise; Voltage and current data information is obtained through power and voltage sensors; The environmental impact coefficient is affected by temperature data, humidity data, wind speed data and distribution equipment function data. The temperature data, humidity data and wind speed data are obtained based on the regional weather information disclosed by the weather forecast. The distribution equipment function data is represented by the average depreciation rate of the power supply equipment in the power supply area. Environmental impact coefficient acquisition includes: ； in, and is the weight coefficient, , and are all impact factors, is the temperature data, Represents humidity data, Represents wind speed data, Indicates the functional data of the power distribution equipment. Indicates the area of ​​the power supply area, Indicates The natural index with base ; The data processing module is used to comprehensively analyze the multi-dimensional heterogeneous data, obtain a comprehensive data set, and perform cluster analysis on the comprehensive data set to obtain a feature vector cluster; The comprehensive analysis of multi-dimensional heterogeneous data includes: Constructing a comprehensive data set: constructing a two-dimensional data model according to the power grid topology; mapping the environmental impact coefficient to the corresponding power supply node in the two-dimensional space data model after normalization processing; adding time tags to the preprocessed voltage and current data information, and mapping the voltage and current data information after adding the time tags to the corresponding intersection points in the two-dimensional space data model, storing the environmental impact coefficient and the voltage and current data information as attribute fields of the two-dimensional space model, forming a comprehensive data set, which is composed of sub-data, each of which includes the voltage and current data information, time tags and environmental impact coefficients at the corresponding power supply node; The specific method of adding the time tag is as follows: obtaining the periodic time period collected by the power and voltage sensors, binary encoding the periodic time period, obtaining a string of binary characters, and associating the binary characters with the voltage and current data information within the periodic time period; Ways to build a two-dimensional data model include: According to the river network topology, the power grid topology structure is obtained, including distribution lines, power supply nodes and power supply areas corresponding to the power supply nodes. The distribution lines are taken as linear elements, the power supply nodes are converted into point elements, and the power supply areas are taken as surface elements. The power supply nodes and distribution lines are converted into a unified coordinate system to obtain the plane coordinates of each power supply node and distribution line. The voltage and current data information is preprocessed including: During the cycle time, the voltage and current data information at each power supply node is obtained, and the voltage and current data information is moved and averaged; Moving averages include: Step S21: Select the window size , set the original signal data points in the voltage and current data information to ,in, ; Step S22: Calculate the moving average and set the window size The values ​​of consecutive data points within are averaged to obtain the moving average value, which is calculated using the moving average processing formula; Moving average processing formula: ； Indicates at location The moving average From Location arrive Where The average of the data points, where Indicates that the original signal data point is at position The value at represents the moving scale, represents the smoothing coefficient, Represents the number of raw signal data points in the voltage and current data information; Step S23: applying moving average to replace the value of the original data point with the calculated moving average value; Step S24: repeating steps S22 to S23, applying moving average processing to all original signal data points in the voltage and current data information in sequence until all data points are smoothed; Step S25: repeating step S24, applying moving average processing to the voltage and current data information at each power supply node in turn, and obtaining smoothed voltage and current data information; The analysis and management module is used to construct and train the neural network YC-Net calculation model based on the genetic algorithm, collect and input the feature vectors into the trained YC-Net calculation model, and obtain the predicted line loss set of the distribution network in the current cycle time. The predicted line loss set consists of predicted line loss values, and each predicted line loss value corresponds to the sub-data in the feature vector; The YC-Net calculation model construction includes: Three parallel paths are used, and the three paths are set to 1×1, 3×3, and 5×5 convolution kernels respectively. The 1×1 convolution kernel is set before the 5×5 and 3×3 convolution kernels; The number of Inception modules is set to 3, the number of Inception convolution kernels is set to 512, and two fully connected layers are added above the softmax output layer. The fully connected layers use ReLU as the activation function, and the number of neurons in the fully connected layers is set to 2048 and 1024 respectively; The YC-Net calculation model defines the loss function as the Huber Loss loss function, which calculates the loss difference between the actual line loss value of each power supply node in the distribution network and the corresponding predicted line loss value; The adjustment module calculates the line loss rate difference between each predicted line loss value in the predicted line loss set and the corresponding actual line loss value within the current cycle. When the line loss rate difference exceeds the preset range, the coordinates of the power supply node are extracted, and the power supply area corresponding to the power supply node is checked. When the line loss rate difference does not exceed the preset range, the line loss of the distribution network is normal. The modules are connected to each other via wired and / or wireless means.

2. According to claim 1, the intelligent measurement and management system for distribution network line loss is characterized in that: The smoothing coefficient Ways to obtain include: Get the window size The value of consecutive data points within the extraction window size The data point with the largest inlier value The data point with the smallest sum , then set the window size Except for data points and data points The data points outside of ,and , and is the weight coefficient, then 。 3. The intelligent measurement and management system for distribution network line loss according to claim 2 is characterized in that: The cluster analysis of the comprehensive dataset includes: Using the improved k-means algorithm, the voltage and current data information at each power supply node is used as the feature data to perform cluster analysis on the sub-data of the comprehensive data set; Assign data points to the nearest cluster center: randomly initialize the cluster center, calculate the Euclidean distance between the voltage and current data information in each sub-data and the cluster center, and assign them to the cluster center with the nearest distance. All sub-data filled in each cluster center are regarded as a cluster; Updating cluster centers: Introducing cost function , set the cost threshold ,like , represents the cost function Greater than or equal to the cost threshold , then you need to update the cluster center, calculate the mean of all sub-data in each cluster, use this mean as the new cluster center, and distribute the sub-data. If , represents the cost function Less than cost threshold , cost function The lower the value, the closer it is to the cluster center, so there is no need to recalculate the cluster center. The cost function The calculation formula is: ； in, represents the number of cluster centers, represents the index of the cluster center, represents the number of sub-data in the comprehensive dataset, Indicates the index of the sub-data. Indicates Sub-data and The Euclidean distance between cluster centers; Construct feature vector collection: take all sub-data of each cluster as a feature vector, and combine all feature vectors into a feature vector collection.

4. The intelligent measurement and management system for distribution network line loss according to claim 3 is characterized in that: The hidden layer node settings of the YC-Net calculation model include: Constrain the hidden layer nodes and set the hidden layer node value constraints as: ； in, represents the number of hidden layer nodes, represents the number of output nodes, represents the number of input nodes, Represents the adjustment constant.

5. The intelligent measurement and management system for distribution network line loss according to claim 4 is characterized in that: After the hidden layer node values ​​of the YC-Net calculation model are determined, the index of the line loss prediction accuracy of the model is evaluated according to the fitness function, and the YC-Net calculation model with the best line loss prediction ability when the fitness function is larger is selected. The fitness function is expressed as: ； in, Represents the fitness function, which is used to measure the quality of the model. Indicates the index of the iteration number, that is, the current iteration number, is the number of iterations, is the index of the hidden layer node, which is used to traverse each hidden layer node. For the The first iteration The theoretical line loss value of each node is the expected loss obtained based on the actual distribution network using traditional power calculation methods. For the The first iteration The calculated line loss value of each node is the line loss value predicted by inputting the feature vector into the trained neural network model.

6. The intelligent measurement and management system for distribution network line losses according to claim 5 is characterized in that: The YC-Net calculation model training process includes: S11, collecting characteristic vectors of the distribution network within the historical period time t as a sample set, and the sample set has sufficient samples; S11, input each feature vector of each feature vector set in the sample set into the YC-Net measurement model in turn for forward propagation; S12. Obtain the predicted line loss value output by the YC-Net measurement model, calculate the loss value using the Huber Loss function, and update the weight and bias of the model through the back propagation algorithm; S13. Repeat S12 and S13 until the loss function value of the YC-Net calculation model no longer changes, stop training, and obtain the trained YC-Net calculation model.

Images