A power distribution network stability operation and maintenance method based on LSTM

Abstract

The application provides a power distribution network stability operation and maintenance method based on LSTM, comprising the following steps: collecting power distribution network related data, filtering abnormal values and filling missing values, and dividing continuous data into time series data; a long short-term memory neural network model is established to predict the probability of failure, and the long short-term memory neural network is optimized and corrected; the stability risk of the power distribution network is evaluated and reported to the background monitoring system in real time; when the risk degree exceeds the set threshold, the system sends a risk alarm to the operation and maintenance personnel; according to the prediction result of the neural network, the real-time monitoring data and the environmental change, the genetic algorithm is used for scheduling optimization management of the power distribution system. The application has the beneficial effects that: the power distribution network data is monitored in real time, the failure cause and position are identified, and the failure elimination efficiency is improved; the load is predicted according to the historical data, the operation efficiency of the power distribution network is optimized, the demand for manual operation is reduced, and the operation and maintenance efficiency and safety are improved.

Description

Technical Field

[0001] This invention belongs to the field of power grid operation and maintenance, and in particular relates to a distribution network stability operation and maintenance method based on LSTM. Background Technology

[0002] The stability operation and maintenance of distribution networks is an important issue in modern power systems, involving a series of tasks such as monitoring, diagnosing, predicting and controlling various equipment and loads in the distribution network.

[0003] In existing technologies, traditional methods for maintaining the stability of power distribution networks mainly rely on experience, rules, and manual adjustments. This approach has the following drawbacks:

[0004] Human error: Traditional methods rely on human experience and judgment, which carries the risk of human error. For example, during troubleshooting, personnel may overlook important information or make incorrect decisions;

[0005] Inefficiency: Traditional methods require a large amount of manual intervention, which is inefficient, especially in large-scale and complex power distribution networks;

[0006] Unable to meet the needs of large-scale complex distribution networks: With the development of power systems, the scale and complexity of distribution networks are constantly increasing, and traditional methods are difficult to meet the needs of large-scale complex distribution networks;

[0007] Lack of intelligence: Traditional methods lack the support of intelligent technologies and cannot automatically diagnose and predict faults. Summary of the Invention

[0008] In view of this, the present invention aims to propose an LSTM-based method for the operation and maintenance of distribution network stability, in order to solve at least one of the above-mentioned technical problems.

[0009] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0010] The first aspect of this invention provides a method for the operation and maintenance of distribution network stability based on LSTM, comprising the following steps:

[0011] S1. Collect historical electricity consumption data, power supply equipment operation data and distribution network topology, filter outliers and fill in missing values, and divide continuous data into time series data.

[0012] S2. Establish a long short-term memory neural network model, train it on historical data, predict the probability of failure, and optimize and correct the long short-term memory neural network by monitoring the working parameters in real time.

[0013] S3. Based on the output of the long short-term memory neural network model, combined with ambient temperature, weather forecast, and electricity load prediction, the system assesses the stability risk of the power distribution network and reports it to the background monitoring system in real time. When the risk level exceeds the set threshold, the system sends a risk alarm to the operation and maintenance personnel.

[0014] S4. Based on the prediction results of the neural network, real-time monitoring data, and environmental changes, a genetic algorithm is used to optimize the scheduling of the power distribution system.

[0015] Furthermore, the process of filtering outliers and missing values ​​is as follows:

[0016] S101. Filtering outliers: Calculate the mean, standard deviation, and outliers, and use the 3σ rule to identify outliers. Then, use linear interpolation to estimate and fill out the outliers.

[0017] S102. Filling in missing values: Based on the proportion of missing features and the importance of the features, delete samples or features containing missing values, or use the KNN method to predict and fill in the missing values.

[0018] Furthermore, the process of dividing continuous data into time series data is as follows:

[0019] S103. Based on historical electricity consumption, select different time intervals to process electricity consumption data;

[0020] S104. Determine the time step size based on the time span of historical information required by the model; use the same time step size each time to slide a window to extract a continuous time series data segment, and repeat this process to obtain more time series data; convert the continuous data within each window into a sequence.

[0021] S105. Standardize the data, perform translation and scaling operations between multiple time series, and perform data segmentation and packaging operations to obtain the power distribution network dataset.

[0022] Furthermore, in S201, the input and output layers of the long short-term memory neural network model are constructed to receive time series data and output prediction results;

[0023] S202. Construct the hidden layer of the Long Short-Term Memory Neural Network Model. The hidden layer uses a multi-layer neural network structure to predict future load and failure conditions.

[0024] S203. Use the memory cell structure of LSTM to capture the correlation between time series data.

[0025] Furthermore, the process of optimizing and correcting the Long Short-Term Memory Neural Network is as follows:

[0026] S204. Obtain AC phase difference, frequency offset, and harmonic content data within a certain time window from real-time monitoring data, and convert them into a data format that meets the input requirements of a long short-term memory neural network.

[0027] S205. Train the Long Short-Term Memory Neural Network using the monitoring dataset, and optimize the Long Short-Term Memory Neural Network using the network pruning method;

[0028] S206. Collect and record in real time the error between the predicted value and the actual value generated by the Long Short-Term Memory Neural Network, and correct the Long Short-Term Memory Neural Network based on the collected prediction error data.

[0029] Furthermore, the process of using genetic algorithms to optimize the scheduling of the power distribution system is as follows:

[0030] S401. Establish a data model of the power distribution system, design a fitness function to evaluate the performance of each individual based on the objective function, and determine the constraints related to the problem.

[0031] S402. Initialize the genome population and use mutation and crossover operations to generate the next generation genome;

[0032] S403. By calculating the fitness of each individual and ranking them according to their fitness, select the best genomes in the population, perform crossover and mutation operations, and obtain a new genome population.

[0033] S404. Repeat step S403 until the genomic population satisfies the convergence strategy and the optimal solution of the power distribution system data model is obtained.

[0034] S405. For different scenarios and objectives, the optimal solution obtained in step S404 is used as a task allocation scheme to optimize the scheduling of the power distribution system.

[0035] The second aspect of this invention provides an LSTM-based distribution network stability operation and maintenance system. This system designs and develops its front-end and back-end according to the requirements of distribution network stability operation and maintenance; it uses RESTful APIs to build API interfaces, providing data query, data upload, and fault alarm services; and it uses a relational database to store the collected data.

[0036] Using the High Charts visualization library, add real-time data monitoring charts, fault alarm prompts, historical data query and analysis modules to the front-end page, and call the corresponding back-end APIs to perform real-time monitoring, fault alarms, historical data query and analysis operations.

[0037] Furthermore, refined monitoring and control are carried out for local areas of the power distribution network, and the Internet of Things (IoT) technology is used to monitor the status of equipment in real time and remotely control it.

[0038] The specific process is as follows:

[0039] Select IoT devices that comply with power grid standards and install them at key locations in the power distribution network to collect relevant data;

[0040] Establish a stable and reliable communication network to transmit the collected data to the cloud or local server;

[0041] The system processes and stores the collected data and provides API interfaces for applications to call.

[0042] Design corresponding business logic and interface according to user needs, use API interface to obtain data, and perform real-time status monitoring and remote control of equipment.

[0043] Furthermore, the specific solution for adding real-time data monitoring charts, fault alarm prompts, historical data query and analysis modules to the front-end page using the High Charts visualization library is as follows:

[0044] Use the High Charts visualization library to display real-time load, voltage, and current data of the distribution network, and perform data analysis and comparison through charts;

[0045] Add a fault alarm module to the front-end page. When an abnormal situation occurs in the power distribution network, it will automatically send alarm information to the user to remind the user to deal with it in time.

[0046] Use the High Charts visualization library to display historical data in the form of line charts and bar charts, and perform data queries and analysis through the interface.

[0047] Furthermore, security measures will be implemented for the distribution network stability operation and maintenance system. The specific implementation plan is as follows:

[0048] The collected data is encrypted before transmission, and the SSL / TLS protocol is used to ensure the security of data transmission.

[0049] User authentication and access control are implemented to prevent unauthorized users from accessing the system;

[0050] Firewall and intrusion detection system technologies are used to prevent the system from being attacked by networks;

[0051] Regularly back up the database to quickly restore the system in case of system failure or data loss.

[0052] Compared with existing technologies, the LSTM-based distribution network stability operation and maintenance method described in this invention has the following advantages:

[0053] The LSTM-based distribution network stability operation and maintenance method described in this invention has the following functions:

[0054] Real-time monitoring: By using LSTM models to monitor distribution network data in real time, anomalies can be detected in a timely manner, and corresponding measures can be taken to prevent the fault from escalating.

[0055] Fault diagnosis: Based on the LSTM model, analysis of historical data can help identify the cause and location of faults, and improve the efficiency of troubleshooting.

[0056] Predictive capability: The LSTM model has strong time series forecasting capabilities and can perform short-term and long-term load forecasting, voltage and current fluctuation forecasting, etc., based on historical data.

[0057] Optimized scheduling: Based on the prediction results of the LSTM model, intelligent scheduling strategies can be used to optimize the operating efficiency of the distribution network and reduce energy consumption costs.

[0058] Simplified operation: LSTM-based automated systems can reduce the need for manual operations and improve operational efficiency and security. Attached Figure Description

[0059] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0060] Figure 1 This is a flowchart illustrating a distribution network stability operation and maintenance method based on LSTM, as described in an embodiment of the present invention. Detailed Implementation

[0061] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0062] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0063] Example 1:

[0064] A distribution network stability operation and maintenance method based on LSTM includes the following steps:

[0065] S1. Collect historical electricity consumption data, power supply equipment operation data and distribution network topology, filter outliers and fill in missing values, and divide continuous data into time series data.

[0066] In practical applications, due to various reasons for data collection and transmission, problems such as outliers and missing values often occur. These outliers and missing values may have a greater impact on model training and prediction, so it is necessary to filter and process them.

[0067] The process of filtering outliers and missing values is as follows:

[0068] S101. Filter outliers: Calculate the mean, standard deviation, outliers, and use the 3σ rule to identify outliers, and use the linear interpolation method to estimate and fill the outliers.

[0069] The 3σ rule (Three Sigma Rule), also known as the 68 - 95 - 99.7 rule, is a commonly used rule in statistics to describe the proportion of data contained within a certain interval in a normal distribution.

[0070] Specifically, the 3σ rule states that in a normal distribution with a mean of μ and a standard deviation of σ, approximately 68% of the data is within the range of ±1σ from the mean, approximately 95% of the data is within the range of ±2σ from the mean, and approximately 99.7% of the data is within the range of ±3σ from the mean.

[0071] Linear interpolation is a method of estimating intermediate positions by connecting a straight line between two adjacent data points given the data points. Its basic idea is to assume a linear relationship between two adjacent data points, and the value at the intermediate position can be estimated by simple proportional calculation on this straight line.

[0072] Specifically, for the given data points (x0, y0) and (x1, y1), where x0 < x1, and it is required to perform linear interpolation at a certain position x between x0 and x1, the formula can be used: y = y0 + (y1 - y0) / (x1 - x0) * (x - x0);

[0073] Among them, y represents the value estimated at x. The meaning of this formula is to divide the line segment between x0, y0 and x1, y1 into two segments according to the proportion, and then obtain the corresponding y value according to the proportion of x - x0.

[0074] S102. Fill missing values: According to the feature missing ratio and feature importance, delete samples or features containing missing values, or use the KNN method to predict and fill the missing values.

[0075] The process of dividing continuous data into time - series data is as follows:

[0076] S103. Select different time intervals to process electricity consumption data according to historical electricity consumption.

[0077] S104. Determine the time step size based on the time span of historical information required by the model (e.g., if the data of the past 5 days is selected as the input feature, then the time step size is 5); use the same time step size each time to slide a window to extract a continuous time series data segment, and repeat this process to obtain more time series data; convert the continuous data in each window into a sequence (e.g., convert the hourly electricity consumption of the past 5 days into a time series of length 120 (24*5)).

[0078] S105. Standardize the data, perform translation and scaling operations between multiple time series, and perform data segmentation and packaging operations to obtain the power distribution network dataset.

[0079] S2. Establish a long short-term memory neural network model, train it on historical data, predict the probability of failure, and optimize and correct the long short-term memory neural network by monitoring the working parameters in real time.

[0080] Real-time monitoring of operating parameters involves acquiring and synchronizing real-time data to a computer or other data processing system via corresponding sensors. This invention employs the AC phase difference method for monitoring.

[0081] The AC phase difference can be determined by measuring the time difference between the reverse voltage peaks between two AC power supplies, using two millivoltmeters that can increase the mechanical voltage drop, and measuring the two signals in the oscillation circuit separately.

[0082] The present invention can also select the following methods for monitoring:

[0083] Frequency offset method:

[0084] Frequency offset can be determined by measuring the difference between the frequency generated by the power supply output waveform and the standard sampling rate (e.g., 50Hz), using equipment such as a spectrum analyzer and a lock-in amplifier.

[0085] Harmonic content method:

[0086] Harmonic content can be determined by measuring the proportion of harmonic components of different frequencies in the output waveform of an oscillating circuit, using equipment such as a power meter and a spectrum analyzer.

[0087] S201. Construct the input and output layers of the Long Short-Term Memory Neural Network model to receive time series data and output prediction results;

[0088] S202. Construct the hidden layer of the Long Short-Term Memory Neural Network Model. The hidden layer uses a multi-layer neural network structure to predict future load and failure conditions.

[0089] The hidden layers use a multi-layer network structure, with 128, 128 and 64 hidden units in each layer, respectively.

[0090] The return sequences of the first to N-1 layers are set to True, indicating that these N-1 layers are stateful and can transmit time step information;

[0091] Setting the return sequence of the Nth layer to False indicates that the last layer no longer needs time step information and only needs to return a fixed-length output.

[0092] After each layer, a BatchNormalization layer is added to adjust the mean and variance of the input data, and a Dropout layer is added to randomly remove some neurons to avoid overfitting;

[0093] A fully connected layer was added at the end of the model, and a linear activation function was used for the output.

[0094] By increasing the number of layers and hidden units in each layer of the Long Short-Term Memory Neural Network (LSTM) model, and combining this with regularization techniques, the expressive power of the LSTM network can be improved, thereby more accurately predicting future loads and fault conditions and achieving stable operation and maintenance of the distribution network.

[0095] S203. Capture the correlation between time series data by using the memory unit structure of the Long Short-Term Memory Neural Network Model.

[0096] The process of optimizing and correcting long short-term memory neural networks is as follows:

[0097] S204. Obtain AC phase difference, frequency offset, and harmonic content data within a certain time window from real-time monitoring data, and convert them into a data format that meets the input requirements of a long short-term memory neural network.

[0098] S205. Train the Long Short-Term Memory Neural Network using the monitoring dataset, and optimize the Long Short-Term Memory Neural Network using the network pruning method;

[0099] S206. Collect and record in real time the error between the predicted value and the actual value generated by the Long Short-Term Memory Neural Network, and correct the Long Short-Term Memory Neural Network based on the collected prediction error data.

[0100] S3. Based on the output of the long short-term memory neural network model, combined with ambient temperature, weather forecast, and electricity load prediction, the system assesses the stability risk of the power distribution network and reports it to the background monitoring system in real time. When the risk level exceeds the set threshold, the system sends a risk alarm to the operation and maintenance personnel.

[0101] The thresholds set and their basis are as follows:

[0102] 1) Ambient temperature threshold: Generally speaking, for information technology equipment such as computers and servers, the temperature within their operating range should be controlled within a reasonable range, otherwise it may lead to equipment failure or reduce their performance life.

[0103] For example, Intel recommends that servers operate at a temperature of around 30°C to 35°C. Other manufacturers also have their own temperature specifications. The specific temperature threshold should be determined according to the product manual.

[0104] 2) Electricity load threshold: By analyzing the electricity load of various sectors such as freight, commerce, and residential on the municipal power dashboard big data and predicting future trends, a relatively accurate electricity load curve is obtained, and the required reserve power for some special situations is set to determine the electricity load threshold.

[0105] 3) Power load thresholds: Based on the operating environment, the geographical location of the customer unit, and weather and seasonal factors (such as seasonal temperature changes, seasonal rainfall changes, and seasonal sunshine duration changes), load warning thresholds are comprehensively formulated for different seasons.

[0106] Based on the climate characteristics of Suzhou City, the following electricity load thresholds are set:

[0107] Summer (May 9 - September 19): Summer temperatures rise, and air conditioning and other cooling equipment are used extensively. The power load threshold is set at 350,000 kilowatts to ensure sufficient power supply.

[0108] Winter (November 30 - February 26): Winter temperatures drop and Suzhou does not provide centralized heating, leading to the extensive use of heating equipment. However, winter temperatures in Suzhou do not fall below -5 degrees Celsius, so the electricity load threshold can be appropriately lowered. The electricity load threshold is set at 250,000 kilowatts to meet people's heating needs.

[0109] Based on the precipitation characteristics of Suzhou City, the following electricity load thresholds are set:

[0110] Rainy season (February 4 - December 14): This season lasts for 10 months, with heavy rainfall from February to December, and at least 13 mm of rainfall in the relevant 31 days. Municipal drainage power consumption increases, and the power load threshold will be increased by 10,000 to 30,000 kilowatts based on the historical rainfall time distribution.

[0111] Low rainfall period (December 14 - February 4): This period lasts 1.7 months, from December to February. During this time, there is no need to increase the electricity load threshold.

[0112] It should be noted that July is the month with the most rainfall, with an average rainfall of 169 mm, mostly in the form of showers. To prevent sudden large-scale rainfall, the electricity load threshold for the entire month of July will be increased by 30,000 kilowatts.

[0113] Based on the sunshine characteristics of Suzhou City, the following electricity load thresholds are set:

[0114] Summer (June-September): During this season, the hours of sunshine are longer, and the electricity consumption for urban lighting decreases. Based on historical sunshine hours and intensity, the electricity load threshold will be reduced by 10,000 to 30,000 kilowatts.

[0115] Winter (October to May of the following year): During this season, the daylight hours are shorter, and the electricity consumption for urban lighting increases. Based on historical daylight hours and intensity, the electricity load threshold will be increased by 20,000 to 50,000 kilowatts.

[0116] S4. Based on the prediction results of the neural network, real-time monitoring data, and environmental changes, a genetic algorithm is used to optimize the scheduling of the power distribution system.

[0117] The process of using genetic algorithms to optimize the scheduling and management of power distribution systems is as follows:

[0118] S401. Establish a data model of the power distribution system, design a fitness function to evaluate the performance of each individual based on the objective function, and determine the constraints related to the problem.

[0119] S402. Initialize the genome population and use mutation and crossover operations to generate the next generation genome;

[0120] S403. By calculating the fitness of each individual and ranking them according to their fitness, select the best genomes in the population, perform crossover and mutation operations, and obtain a new genome population.

[0121] S404. Repeat step S403 until the genomic population satisfies the convergence strategy and the optimal solution of the power distribution system data model is obtained.

[0122] S405. For different scenarios and objectives, the optimal solution obtained in step S404 is used as a task allocation scheme to optimize the scheduling of the power distribution system.

[0123] Genetic Algorithm (GA) is an optimization algorithm based on evolutionary principles. It searches for the optimal solution by simulating the reproduction and heredity processes of biological populations. In a genetic algorithm, an initial genome population needs to be initialized, and operations such as mutation and crossover are used to generate the next generation genome. Its computational method is as follows:

[0124] 1) Initialize the genome population:

[0125] Initialize an M*L matrix P, with elements p (ij) Represents the j-th gene position of the i-th genome, and its value range is usually (0,1) or a certain real number interval;

[0126] 2) Mutation operation:

[0127] Mutation operations typically involve randomly mutating a gene locus with a certain probability to increase diversity. Assuming the mutation probability is pm, then for the j-th gene locus in the i-th genome, the following formula can be used:

[0128] P' (ij) =1-p (ij) ~B(pm);

[0129] Where B(pm) represents the Bernoulli distribution with probability pm producing 1 and probability (1-pm) producing 0, that is, when the probability of mutation is less than 1, p (ij) The probability of inverting is pm; otherwise, it remains unchanged, i.e., the original value is taken.

[0130] 3) Crossover operation:

[0131] Crossover operations typically involve exchanging a random gene locus in two genomes with a certain probability, resulting in a new offspring genome. Assuming the crossover probability is pc, for the i-th and j-th genomes, a single-point crossover operation at crossover position k can be performed using the following formula:

[0132] p” (ik) =p (ik) ~B(pc)+(1-p (ik) )~B(1-pc) ；

[0133] p” (jk) =p (jk) ~B(pc)+(1-p (jk) )~B(1-pc);

[0134] In this case, B(pc) is the same as the above mutation operation, that is, when the crossover probability is less than 1, the i-th and j-th genomes are exchanged at position k with a certain probability; otherwise, no crossover operation is performed.

[0135] Example 2:

[0136] A system based on LSTM for distribution network stability operation and maintenance is proposed. According to the needs of distribution network stability operation and maintenance, the system front-end and back-end are designed and developed. RESTful API is used to build API interfaces to provide data query, data upload and fault alarm services, and relational database is used to store various collected data.

[0137] Using the HighCharts visualization library, add real-time data monitoring charts, fault alarm prompts, historical data query and analysis modules to the front-end page, and call the corresponding back-end APIs to perform real-time monitoring, fault alarms, historical data query and analysis operations.

[0138] For localized areas of the power distribution network, refined monitoring and control are implemented, and IoT technology is used to monitor and remotely control equipment status in real time.

[0139] The specific process is as follows:

[0140] Select IoT devices that comply with power grid standards and install them at key locations in the power distribution network to collect relevant data;

[0141] Establish a stable and reliable communication network to transmit the collected data to the cloud or local server;

[0142] The system processes and stores the collected data and provides API interfaces for applications to call.

[0143] Design corresponding business logic and interface according to user needs, use API interface to obtain data, and perform real-time status monitoring and remote control of equipment.

[0144] The specific solution for adding real-time data monitoring charts, fault alarm prompts, historical data query and analysis modules to the front-end page using the HighCharts visualization library is as follows:

[0145] Use the HighCharts visualization library to display real-time load, voltage, and current data of the distribution network, and perform data analysis and comparison through charts;

[0146] Add a fault alarm module to the front-end page. When an abnormal situation occurs in the power distribution network, it will automatically send alarm information to the user to remind the user to deal with it in time.

[0147] Use the HighCharts visualization library to display historical data in the form of line charts and bar charts, and perform data query and analysis through the interface.

[0148] The following is a specific implementation plan for ensuring the security of the power distribution network stability operation and maintenance system:

[0149] The collected data is encrypted before transmission, and the SSL / TLS protocol is used to ensure the security of data transmission.

[0150] User authentication and access control are implemented to prevent unauthorized users from accessing the system;

[0151] Firewall and intrusion detection system technologies are used to prevent the system from being attacked by networks;

[0152] Regularly back up the database to quickly restore the system in case of system failure or data loss.

[0153] Example 3:

[0154] To address the problem of widespread and prolonged power outages in distribution networks, it is necessary to construct a Long Short-Term Memory (LSTM) neural network structure for prediction of long-term data series. The following aspects need to be considered when establishing an LSM neural network model:

[0155] For long-term series: a relatively long window length should be set to eliminate random, irrelevant fluctuations, and an overlapping sliding window method should be used to increase data utilization.

[0156] Processing of input features: The input features need to take into account multiple factors, including load conditions, temperature, weather, equipment status, etc., and then normalize them.

[0157] Consider multiple time scales: combine the Long Short-Term Memory Neural Network model with subnetworks with lower time resolution to form a multi-level hybrid architecture, leveraging the long-term dependency of the Long Short-Term Memory Neural Network model and the advantages of other methods in local details;

[0158] Adjusting the time step: reducing (smoothing) noise signals and increasing the number of features through pre-trained neural networks.

[0159] Design a loss function: Choose an appropriate error metric and add a regularization term to the loss function to overcome the overfitting problem.

[0160] The following methods can be used to eliminate random, unrelated changes:

[0161] Remove seasonal variation: This is done by breaking down the data into three parts: trend, period, and residual, and removing seasonal variation from them.

[0162] Smoothing: The original data is smoothed by averaging, weighting, or exponential smoothing to reduce the impact of noise;

[0163] State-space model: By introducing the concept of hidden states, the influence of noise is reduced and the predictive ability of the model is improved;

[0164] Siamese Neural Network: This method combines two neural networks to process the same input, eliminating the influence of irrelevant variations on the prediction results.

[0165] The method for normalizing information such as load conditions, temperature, weather, and equipment status is as follows:

[0166] Normalization is generally divided into two types: linear normalization and standardization.

[0167] Linear normalization: Scaling the eigenvalues ​​to the interval [0,1] or [-1,1] using the min-max normalization method, the formula is: x'=(x-min(x)) / (max(x)-min(x));

[0168] Standardization: This refers to converting feature values ​​into data with a mean of 0 and a variance of 1, using the z-score normalization method. The formula is: x' = (x - mean(x)) / std(x).

[0169] The specific method for considering multiple time scales is as follows:

[0170] Define different sliding window sizes, and set the window size to multiple time scales such as 5min, 15min, 30min and 60min when predicting load conditions;

[0171] The input data is divided into multiple time series using the sliding window method, and each time series is normalized.

[0172] For each time series, a corresponding LSTM model is built, and it is trained and evaluated using training and validation sets at different time scales;

[0173] When predicting the output, the prediction results of multiple models are weighted and averaged to obtain the final prediction result.

[0174] The specific method for adding a regularization term to the loss function is as follows:

[0175] Adding a regularization term to the loss function can effectively prevent the LSTM model from overfitting. Generally, there are two main forms: L1 regularization and L2 regularization.

[0176] L1 regularization, also known as Lasso regularization, is based on the idea of ​​weighted summing of the absolute values ​​of each element in the model's parameter vector, which serves as the regularization term. The specific method is as follows:

[0177] Loss function = original loss function + λ * ∑|θi|;

[0178] Where θi is the i-th element in the parameter vector of the model, and λ is the regularization coefficient;

[0179] L2 regularization, also known as Ridge regularization, is similar to L1 regularization, but it simply adds the squares of each element in the parameter vector in a weighted manner as the regularization term. The specific method is as follows:

[0180] Loss function = original loss function + λ * ∑(θi)2;

[0181] Where θi is the i-th element in the model's parameter vector, and λ is the regularization coefficient.

[0182] Those skilled in the art will recognize that the units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0183] In the several embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the division of units described above is merely a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. The aforementioned units may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention according to actual needs.

[0184] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

[0185] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A distribution network stability operation and maintenance method based on LSTM, characterized in that: Includes the following steps: S1. Collect historical electricity consumption data, power supply equipment operation data and distribution network topology, filter outliers and fill in missing values, and divide continuous data into time series data. S2. Establish a long short-term memory neural network model, train it on historical data, predict the probability of failure, and optimize and correct the long short-term memory neural network by monitoring the working parameters in real time. S3. Based on the output of the long short-term memory neural network model, combined with ambient temperature, weather forecast, and electricity load prediction, the system assesses the stability risk of the power distribution network and reports it to the background monitoring system in real time. When the risk level exceeds the set threshold, the system sends a risk alarm to the operation and maintenance personnel. S4. Based on the prediction results of the neural network, monitor data and environmental changes in real time, and use genetic algorithms to optimize the scheduling of the power distribution system. The process of optimizing and correcting long short-term memory neural networks is as follows: S204. Obtain AC phase difference, frequency offset, and harmonic content data within a certain time window from real-time monitoring data, and convert them into a data format that meets the input requirements of a long short-term memory neural network. S205. Train the Long Short-Term Memory Neural Network using the monitoring dataset, and optimize the Long Short-Term Memory Neural Network using the network pruning method; S206. Collect and record in real time the error between the predicted value and the actual value generated by the Long Short-Term Memory Neural Network, and correct the Long Short-Term Memory Neural Network based on the collected prediction error data; The process of using genetic algorithms to optimize the scheduling and management of power distribution systems is as follows: S401. Establish a data model of the power distribution system, design a fitness function to evaluate the performance of each individual based on the objective function, and determine the constraints related to the problem. S402. Initialize the genome population and use mutation and crossover operations to generate the next generation genome; S403. By calculating the fitness of each individual and ranking them according to their fitness, select the best genomes in the population, perform crossover and mutation operations, and obtain a new genome population. S404. Repeat step S403 until the genomic population satisfies the convergence strategy and the optimal solution of the power distribution system data model is obtained. S405. For different scenarios and objectives, the optimal solution obtained in step S404 is used as a task allocation scheme to optimize the scheduling of the power distribution system.

2. The distribution network stability operation and maintenance method based on LSTM according to claim 1, characterized in that: The process of filtering outliers and missing values ​​is as follows: S101. Filtering outliers: Calculate the mean, standard deviation, and outliers, and use the 3σ rule to identify outliers. Then, use linear interpolation to estimate and fill out the outliers. S102. Filling in missing values: Based on the proportion of missing features and the importance of the features, delete samples or features containing missing values, or use the KNN method to predict and fill in the missing values.

3. The distribution network stability operation and maintenance method based on LSTM according to claim 1, characterized in that: The process of dividing continuous data into time series data is as follows: S103. Based on historical electricity consumption, select different time intervals to process electricity consumption data; S104. Determine the time step size based on the time span of historical information required by the model; use the same time step size each time to slide a window to extract a continuous time series data segment, and repeat this process to obtain more time series data; convert the continuous data within each window into a sequence. S105. Standardize the data, perform translation and scaling operations between multiple time series, and perform data segmentation and packaging operations to obtain the power distribution network dataset.

4. The distribution network stability operation and maintenance method based on LSTM according to claim 1, characterized in that: The process of establishing a Long Short-Term Memory (LSTM) neural network model is as follows: S201. Construct the input and output layers of the Long Short-Term Memory Neural Network model to receive time series data and output prediction results; S202. Construct the hidden layer of the Long Short-Term Memory Neural Network Model. The hidden layer uses a multi-layer neural network structure to predict future load and failure conditions. S203. Use the memory cell structure of LSTM to capture the correlation between time series data.

5. The distribution network stability operation and maintenance method based on LSTM according to claim 4, characterized in that: Based on the requirements of distribution network stability operation and maintenance, design and develop the front-end and back-end of the distribution network stability operation and maintenance system; We use RESTful APIs to build API interfaces, providing data query, data upload, and fault alarm services, and use relational databases to store the collected data. Using the High Charts visualization library, add real-time data monitoring charts, fault alarm prompts, historical data query and analysis modules to the front-end page, and call the corresponding back-end APIs to perform real-time monitoring, fault alarms, historical data query and analysis operations.

6. The distribution network stability operation and maintenance method based on LSTM according to claim 5, characterized in that: For localized areas of the power distribution network, refined monitoring and control are implemented, and IoT technology is used to monitor and remotely control equipment status in real time. The specific process is as follows: Select IoT devices that comply with power grid standards and install them at key locations in the power distribution network to collect relevant data; Establish a stable and reliable communication network to transmit the collected data to the cloud or local server; The system processes and stores the collected data and provides API interfaces for applications to call. Design corresponding business logic and interface according to user needs, use API interface to obtain data, and perform real-time status monitoring and remote control of equipment.

7. The distribution network stability operation and maintenance method based on LSTM according to claim 6, characterized in that: The specific solution for adding real-time data monitoring charts, fault alarm prompts, historical data query and analysis modules to the front-end page using the High Charts visualization library is as follows: Use the High Charts visualization library to display real-time load, voltage, and current data of the distribution network, and perform data analysis and comparison through charts; Add a fault alarm module to the front-end page. When an abnormal situation occurs in the power distribution network, it will automatically send alarm information to the user to remind the user to deal with it in time. Use the High Charts visualization library to display historical data in the form of line charts and bar charts, and perform data queries and analysis through the interface.

8. The distribution network stability operation and maintenance method based on LSTM according to claim 6, characterized in that: The following is a specific implementation plan for ensuring the security of the power distribution network stability operation and maintenance system: The collected data is encrypted before transmission, and the SSL / TLS protocol is used to ensure the security of data transmission. User authentication and access control are implemented to prevent unauthorized users from accessing the system; Firewall and intrusion detection system technologies are used to prevent the system from being attacked by networks; Regularly back up the database to quickly restore the system in case of system failure or data loss.

Images