LSTM electric power Internet of Things ring main unit optimization method, system, device and medium

Abstract

The invention discloses an LSTM electric power Internet of Things ring main unit optimization method, system, device and medium, and belongs to the technical field of electric power Internet of Things management and data analysis, and the method comprises the steps: collecting communication state and network environment data in real time, and generating a multi-dimensional feature matrix through dynamic standard score standardization and derivative feature calculation; dynamically adjusting an LSTM forgetting gate weight based on the bandwidth fluctuation index, and outputting an equipment performance score; constructing a dynamic scoring algorithm and triggering a differentiated maintenance instruction in linkage with a multi-threshold scene strategy; a closed-loop feedback mechanism is adopted to screen high-deviation data, new scene training is enhanced through a dynamic loss function, and minute-level hot updating of the model is achieved; the method solves the problems of static threshold misjudgment, poor environment adaptability of the prediction model and single decision-making mechanism, remarkably improves the feature quality, the equipment evaluation precision and the resource allocation rationality, and strengthens the intelligent operation and maintenance capability of the system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power internet of things management and data analysis, in particular to an LSTM power internet of things ring main unit optimization method, system, device and medium. BACKGROUND

[0002] The operation and maintenance management of the power internet of things ring main unit is facing severe challenges in the dynamic network environment. The existing technology generally relies on static models or single indicators to develop maintenance strategies, such as triggering alarms based on fixed delay thresholds, or using offline models trained only with historical data for fault prediction. Such methods have significant limitations: they cannot respond to real-time network fluctuations, leading to maintenance lag for high-load devices and deterioration of communication quality; they lack the ability to analyze the coordinated impact of communication status and network environment, resulting in severe imbalance in resource allocation in multi-device concurrent scenarios. More critically, existing solutions do not establish a long-term dependence model between network fluctuations and device performance, making it impossible for the system to predict performance degradation trends when communication status is continuously abnormal, leading to frequent device failures. Although existing technologies attempt to introduce time series models to optimize data transmission, they do not work with maintenance strategies, and high-priority devices still fail to respond when network congestion occurs. The above-mentioned static mechanism defects, lack of time series correlation, and insufficient data fusion directly result in low operation and maintenance efficiency, increased costs, and decreased system stability.

[0003] The fundamental problem of traditional methods is the lack of closed-loop optimization capability and weak dynamic adaptability. Static standardized mechanisms have a significantly increased noise misjudgment rate during sudden traffic peaks, while offline training models cannot absorb real-time feedback data. For example, network topology mutations require long-term human intervention, during which the prediction function is completely disabled; at the same time, the fixed weight mechanism of existing time series models cannot perceive the nonlinear impact of network fluctuations on device performance, leading to increased prediction bias. The problem of relying on single threshold rules at the decision-making layer is particularly prominent, as it neither quantifies dynamic feature trend changes nor builds a multi-factor coordinated evaluation system, resulting in excessive or insufficient maintenance instructions. These defects are amplified in complex network environments: when bandwidth utilization is extremely high and topology changes dramatically, existing systems lack real-time optimization mechanisms, making it impossible to quickly adjust model parameters, ultimately leading to systematic operation and maintenance failures. Therefore, there is an urgent need for a dynamic maintenance solution that integrates real-time fluctuation perception, adaptive time series modeling, and closed-loop feedback. SUMMARY

[0004] In view of the above-mentioned problems, the present application is proposed.

[0005] Therefore, the technical problem solved by the present application is to construct an intelligent decision-making model by integrating real-time communication status (such as delay and packet loss rate) and network environment data (such as bandwidth fluctuation and topology change), aiming to solve the problems of ring main unit maintenance strategy lag and inefficient resource allocation in complex network environments, and promote the intelligent transformation of power distribution networks.

[0006] To solve the above technical problems, the application provides the following technical solutions: an LSTM power Internet of Things ring network cabinet optimization method, which comprises the following steps: Real-time collection of ring network cabinet communication state data and network environment data, real-time calculation of sliding window data, determination of abnormal values through dynamic standard score standardization and historical data interpolation correction, calculation of dynamic delay change rate and bandwidth fluctuation index, and generation of a multi-dimensional feature matrix; Inputting the multi-dimensional feature matrix into a long short-term memory network, real-time adjustment of the forgetting gate weight based on the bandwidth fluctuation index, and outputting the device performance score and the key influence factor; According to the real-time network state, the dynamic delay change rate, the bandwidth fluctuation index and the packet loss rate are adjusted to construct a dynamic scoring algorithm, and a multi-threshold linkage strategy is used to define the delay and bandwidth overload scene, and a differentiated maintenance instruction is triggered; After collecting the device state feedback data after the execution of the maintenance instruction, the data with a prediction deviation exceeding a set threshold is screened, a dynamic loss function is used to strengthen the weight of new scene data, and model training is performed; when the accuracy of the verification set improves by more than a set threshold, the model version is updated and rollback is supported.

[0007] As a preferred scheme of the LSTM power Internet of Things ring network cabinet optimization method, wherein the real-time calculation of the sliding window data comprises setting a fixed time length of the sliding window, and real-time interception of the communication state data and the network environment data in the window; the mean and the standard deviation of the data in the sliding window are calculated, the standard score is calculated through the dynamic Z-score formula, and if the absolute value of the standard score exceeds a preset abnormal threshold, it is determined as an abnormal value; and the historical data is interpolated and corrected.

[0008] As a preferred scheme of the LSTM power Internet of Things ring network cabinet optimization method, wherein the multi-dimensional feature matrix comprises converting the original communication data into a high-precision multi-dimensional feature matrix through a dynamic sliding window Z-score standardization and a historical data interpolation correction mechanism; The original communication data comprises real-time delay, packet loss rate, bandwidth and topology state. The high-precision multi-dimensional feature matrix comprises delay, packet loss rate, bandwidth, topology state, dynamic delay change rate and bandwidth fluctuation index.

[0009] As a preferred scheme of the LSTM power Internet of Things ring network cabinet optimization method, wherein the output of the device performance score and the key influence factor comprises dynamic adjustment of the LSTM gate weight based on the bandwidth fluctuation index, construction of an adaptive time sequence fusion model, and output of the device performance score and the key influence factor.

[0010] As a preferred scheme of the LSTM power Internet of Things ring main unit optimization method, the dynamic score algorithm comprises calculating a comprehensive score through a dynamic weight coefficient by comprehensively considering a dynamic delay change rate, a bandwidth fluctuation index and a current packet loss rate.

[0011] As a preferred scheme of the LSTM power Internet of Things ring main unit optimization method, the multi-threshold linkage strategy comprises triggering a high-priority maintenance instruction and performing link switching when the delay exceeds a first threshold value and the bandwidth utilization rate exceeds a second threshold value, and triggering a communication path priority dynamic adjustment instruction when a topology change is detected and the bandwidth allocation is uneven.

[0012] As a preferred scheme of the LSTM power Internet of Things ring main unit optimization method, the dynamic loss function comprises fusing historical and new loss values, weighting and summing the historical data loss value and the new data loss value, and adjusting the weight according to the scene to select new data with a prediction deviation exceeding a set threshold value to participate in calculation.

[0013] The application provides an LSTM power Internet of Things ring main unit optimization system, which generates a multi-dimensional feature matrix fusing communication states and network environments in real time through a dynamic feature construction module, adjusts the LSTM forgetting gate weight based on the bandwidth fluctuation index using a time series adaptive prediction module, outputs a device performance score, generates a differentiated maintenance instruction by combining a dynamic weighted score and a scenario-based rule library through a multi-threshold decision module, and realizes model minute-level hot updating driven by incremental learning using a closed-loop optimization module, thereby solving the technical problems that a static model cannot adapt to network fluctuations, maintenance strategies lag and a closed-loop mechanism is missing.

[0014] To solve the above technical problems, the application provides the following technical scheme: an LSTM power Internet of Things ring main unit optimization system, comprising: A dynamic feature construction module acquires ring main unit communication state data and network environment data in real time, calculates data in a sliding window in real time, determines abnormal values through dynamic standard score standardization and performs historical data interpolation correction, calculates a dynamic delay change rate and a bandwidth fluctuation index, and generates a multi-dimensional feature matrix; A time series adaptive prediction module inputs the multi-dimensional feature matrix into a long short-term memory network, adjusts the forgetting gate weight based on the bandwidth fluctuation index in real time, and outputs a device performance score and a key influence factor; A multi-threshold decision module constructs a dynamic score algorithm according to a real-time network state, adjusts a dynamic delay change rate, a bandwidth fluctuation index and a packet loss rate, and defines a delay and bandwidth overload scene using a multi-threshold linkage strategy to trigger a differentiated maintenance instruction; A closed-loop optimization module collects device state feedback data after maintenance instruction execution, screens data with prediction deviation exceeding a set threshold, strengthens new scene data weight using a dynamic loss function, performs model training, and updates model version and supports rollback when verification set accuracy rate improvement exceeds a set threshold.

[0015] The application provides a computer device, including a memory and a processor, and the memory stores a computer program.

[0016] The application provides a computer readable storage medium, which stores a computer program.

[0017] The application effectively eliminates the misjudgment problem of sudden network fluctuations caused by static threshold through a dynamic data processing mechanism, significantly improves feature quality, and uses network perception type time series modeling technology to enable the prediction model to adapt to complex environmental changes, greatly enhance the accuracy of device performance evaluation under topology reconstruction and other sudden scenarios, and realize dynamic optimization of maintenance strategies based on a multi-dimensional collaborative decision mechanism. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0019] Figure 1 A general flowchart of an LSTM power Internet of Things ring network cabinet optimization method provided by an embodiment of the application.

[0020] Figure 2 An actual application example diagram of an LSTM power Internet of Things ring network cabinet optimization method provided by an embodiment of the application. DETAILED DESCRIPTION

[0021] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the protection scope of the present application.

[0022] Embodiment 1, reference Figure 1 For an embodiment of the present application, the embodiment provides an LSTM power Internet of Things ring network cabinet optimization method, comprising: S10: Real-time acquisition of ring network cabinet communication state data and network environment data, real-time calculation of sliding window data, determination of abnormal values by dynamic standard score standardization and history data interpolation correction, calculation of dynamic delay change rate and bandwidth fluctuation index, generation of multi-dimensional feature matrix; through calculation of delay difference value in adjacent time window, combined with time interval, defined as dynamic delay change rate, wherein the dynamic delay change rate = , wherein, is the delay value at the current moment, is the delay value at the last moment, is the time interval; based on the bandwidth data in the sliding window, the standard deviation of the bandwidth in each time period is calculated, defined as the bandwidth fluctuation index, and the bandwidth fluctuation index = , wherein, is the standard deviation of the bandwidth, is the mean of the bandwidth.

[0023] It should be noted that through the dynamic sliding window Z-score (dynamic standard score standardization) standardization and history data interpolation correction mechanism, the original communication data is converted into a high-precision 6-dimensional feature matrix, i.e. a high-precision multi-dimensional feature matrix, solving the noise misjudgment problem caused by static threshold and standardization parameters in traditional methods, and reducing the feature extraction error rate.

[0024] Further, the dynamic Z-score standardization calculates the mean and the standard deviation of the data in the real-time calculation window (such as 5 minutes) as follows: , wherein, is the original data (such as delay, packet loss rate, bandwidth, etc. real-time acquisition value), is the mean of the data in the sliding window (5 minutes), is the standard deviation of the data in the sliding window, is the standardized value, used for abnormal value determination; If the delay data of the device in the 5-minute window is , the calculated , , then ; since , it is not determined as an abnormal value; when , it is determined as an abnormal value, real-time elimination and correction by historical data interpolation, wherein represents a standardized value in dynamic Z-score standardization calculation.

[0025] Further, the feature extraction mines multi-dimensional dynamic features, such as calculating dynamic delay change rate (reflecting delay fluctuation trend) and network bandwidth fluctuation index (quantifying bandwidth stability); Specifically, the delay change trend of adjacent time windows is calculated to obtain the dynamic delay change rate, denoted as: , wherein is the delay value of the current time window (such as 1 minute), is the delay value of the previous time window, is the time window interval (such as 1 minute), is the dynamic delay change rate; if , , then , reflecting that the delay shows an upward trend; Quantify the bandwidth stability, using standardization calculation, denoted as: , wherein is the bandwidth value of the time window, is the bandwidth mean value in the sliding window, is the number of data points in the window; the output result is a standardized feature matrix, containing 6-dimensional features (delay, packet loss rate, bandwidth, topology state, , ) to provide high-quality input for subsequent model construction; If the bandwidth in the window is , then , the calculated quantifies the bandwidth fluctuation intensity; if the delay data of a device in a certain period of time has a sudden abnormal increase, the cleaning process will eliminate the abnormal value and correct it through historical data interpolation, ensuring the accuracy of feature extraction.

[0026] S20: Input the multi-dimensional feature matrix into the long short-term memory network, adjust the forgetting gate weight based on the bandwidth fluctuation index in real time, and output the device performance score and key influence factors.

[0027] It should be noted that the bandwidth fluctuation index is used to dynamically adjust the LSTM gate weight (input: 6-dimensional feature matrix output by S10), a multi-dimensional LSTM time series fusion model is constructed using the long short-term memory (LSTM) algorithm, including a feature fusion input layer, a dynamic weight LSTM layer, and an output layer; and an output device performance score and a key influence factor (output: sub-score and sensitivity label) are obtained, which solves the problem that the existing LSTM model cannot fuse multi-dimensional network environment data, and increases the prediction accuracy in the topology mutation scenario; wherein the key influence factor, i.e., the sensitivity label, is, for example, delay sudden increase sensitivity.

[0028] Further, the feature fusion input layer: the input dimension is a 6-dimensional feature matrix after preprocessing (from S10); the data format is sliding window data within a time window (such as 30 minutes); The dynamic weight LSTM layer: in the LSTM gate mechanism, the calculation of the forget gate , the input gate and the output gate is represented as: , wherein, is the hidden state of the previous moment, is the current input feature vector, , , is the weight matrix of the forget gate, the input gate and the output gate, , , is the bias term, is the activation function, , , represent the forget gate (Forget Gate), the input gate (Input Gate) and the output gate (Output Gate), respectively; The output layer: outputs the device performance score (in points) and the key influence factor (such as "delay sudden increase sensitivity").

[0029] The device performance score is a score between 0 and 100 calculated by the LSTM model based on multiple network features (such as delay, packet loss rate, bandwidth, etc.), representing the overall performance of the device; the calculation steps are as follows: (1) Input data: the LSTM model receives delay, packet loss rate, bandwidth, topology, etc. as input; (2) LSTM learning: the LSTM model automatically identifies the influence of each feature on device performance by learning the relationship between data; (3) Score calculation: The equipment performance score is calculated using the following weighted formula: , in, , , , It is the weight coefficient automatically learned by the LSTM model, indicating the impact of each feature on device performance.

[0030] Key impact factors (such as latency spike sensitivity) are used to indicate the sensitivity of a device to network changes. The calculation method is as follows: (1) Input data: Similar to device performance scoring, the input data of the LSTM model includes network characteristics such as latency, bandwidth, and packet loss rate, but with special attention paid to how these characteristics affect device performance when the network fluctuates; (2) LSTM learning: The LSTM model learns how devices respond to different network changes, especially sudden increases in latency and bandwidth fluctuations, by training on historical data. LSTM can identify which network changes will cause a sharp drop in device performance and map these changes to sensitivity labels. (3) Sensitivity label calculation: For example, the latency spike sensitivity is determined by calculating the response change of the device when the latency spike occurs; the calculation formula is as follows: , in, is the change in device response time, is the increase in latency. When latency increases significantly, the device's response time will usually change as well. This ratio indicates the device's sensitivity to latency changes. (4) Output sensitivity label: The LSTM model outputs sensitivity labels based on the response time changes of the device, which are usually divided into multiple levels. For example, high sensitivity is defined as the device reacts strongly to a sudden increase in delay and its performance drops significantly. Medium sensitivity is defined as the device reacts to delay changes to a certain extent, but does not affect normal operation. Low sensitivity is defined as the device can adapt well to delay fluctuations and its performance does not change much.

[0031] Furthermore, according to the bandwidth fluctuation index ( )Adjustment Weight, expressed as: , in, is the adjustment coefficient (default ), controls the impact of bandwidth fluctuations on memory, is the current bandwidth fluctuation index, is the original forget gate weight, is the adjusted forgetting gate weight, indicating the degree of historical information retention of the LSTM forgetting gate after bandwidth fluctuation index adjustment; when , , the characteristics of the high fluctuation period are strengthened.

[0032] S30: According to the real-time network state, adjust the dynamic delay change rate, bandwidth fluctuation index and packet loss rate to construct a dynamic scoring algorithm, and use a multi-threshold linkage strategy to define the delay surge and bandwidth overload scenarios, and trigger differentiated maintenance instructions.

[0033] It should be noted that based on the dynamic scoring algorithm (input: performance score, bandwidth utilization and topology state), the differentiated maintenance instructions (output: JSON format instructions) are generated, the resource waste problem caused by single index decision is solved through the multi-threshold linkage strategy, and the resource allocation efficiency is improved.

[0034] Further, the dynamic scoring algorithm and the multi-threshold linkage strategy are designed; Specifically, the dynamic scoring algorithm is represented as: , wherein, is the dynamic scoring result, is the dynamic delay change rate, is the bandwidth fluctuation index, is the current packet loss rate, , , is the dynamic weight coefficient (initial value is , , ); when the topology changes (such as adding a device node), increase by 20% (i.e. ), to strengthen the delay sensitivity; when the bandwidth utilization is > 85%, increase by 30% (i.e. ), to strengthen the bandwidth fluctuation impact; The multi-threshold linkage strategy includes: scenario 1: delay surge and bandwidth utilization , triggering link switching + high-load device priority maintenance; scenario 2: topology change causes uneven bandwidth allocation, triggering dynamic adjustment of communication path priority (such as prioritizing core node bandwidth).

[0035] Further, the instructions are generated, and the JSON format instructions are output, for example: { "device ID": "R001", "maintenance action": "switch to backup link", "priority": 1, "Trigger condition": "Delay = 62 ms, bandwidth utilization = 93%" }.

[0036] S40: Collect device state feedback data after executing the maintenance instruction, screen data with a prediction deviation exceeding a set threshold, use a dynamic loss function to strengthen the weight of new scene data, perform model training, update the model version and support rollback when the verification set accuracy continuously improves by more than a set threshold.

[0037] It should be noted that the incremental learning is driven by real-time feedback data (input: device state after executing the instruction), the model is updated in minutes (output: new version of the model and parameters), the lag problem of traditional offline training is solved, and the prediction error in a sudden scene is reduced.

[0038] Further, a closed-loop feedback mechanism driven by incremental learning is designed, including incremental training and model version management. The incremental training includes preferentially selecting high-score error data (such as prediction deviation ), and using a loss function to incrementally train the random forest, which is expressed as: , wherein, is the loss value of the historical data, is the loss value of the new data, such as the dynamic weight coefficient after topology reconstruction, and the dynamic weight coefficient in a conventional scene; if new data is added after topology reconstruction, the model preferentially learns the characteristics of the new data, and the adaptation speed is improved . The model version management includes retaining historical model versions (such as , ), and supporting fast rollback; the version update trigger condition is that the verification set accuracy is improved by more than for three consecutive iterations; if the new version decreases in the verification set accuracy, it is automatically rolled back to , to ensure system stability.

[0039] Embodiment 2, referring to Figure 2 , provides an LSTM power Internet of Things ring network cabinet optimization method, in order to verify the beneficial effects of the present application, scientific demonstration is carried out through experiments.

[0040] ​A large power Internet of Things system includes multiple ring network cabinets for distributing power to different device nodes. Each device node in the system transmits information such as network transmission delay, packet loss rate, bandwidth, and topology changes. The ring network cabinet needs to monitor these data in real time and adjust the maintenance strategy according to the changes in the network environment to ensure the stable operation of the device.

[0041] As shown in Figure 2 , the dynamic maintenance optimization process of the application in the burst network fluctuation scenario is shown, including: Network fluctuation occurs and data is collected: the power Internet of Things system where the ring network cabinet is located has network fluctuation, which is characterized by sudden increase of delay from to , sharp increase of bandwidth from to , and change of network topology due to addition of 5 device nodes. The traditional static threshold method will have difficulty in making accurate maintenance decisions, because the traditional method usually operates according to fixed rules (for example, issuing a warning when the delay exceeds ), which cannot adapt to sudden network fluctuations or topology changes, resulting in excessive or insufficient maintenance.

[0042] Data preprocessing and feature extraction: the maintenance system of the ring network cabinet adopts an intelligent decision-making method based on LSTM and real-time network fluctuation. When a network fluctuation event occurs, the system first performs data preprocessing, calculates the mean and standard deviation of delay and bandwidth in real time through dynamic Z-score standardization technology, and judges whether the delay fluctuation exceeds the normal fluctuation range based on the sliding window method. For example, the system will correct the sudden increase of delay through historical data interpolation to avoid misjudgment of the abnormal situation.

[0043] LSTM time series modeling and dynamic weight adjustment: the system fuses multi-dimensional data through LSTM model to learn the time series relationship between delay, bandwidth, packet loss rate, and topology change. The LSTM network can capture the long-term dependence of delay and bandwidth fluctuation on device performance and make performance prediction accordingly. As shown in Figure 2 , when the bandwidth fluctuation index exceeds the threshold, the system dynamically adjusts the forgetting gate weight (such as increasing the weight by 3 times) to improve the prediction accuracy of the model in complex network environment.

[0044] Dynamic scoring and maintenance instruction generation: when making device maintenance decisions, the system comprehensively evaluates the impact of multiple factors through a dynamic scoring model. The system not only considers sudden increases in delay, but also calculates the bandwidth fluctuation, packet loss rate, and the impact of topology changes in real time, and generates a comprehensive performance score. For example, when the delay suddenly increases and the bandwidth utilization rate is close to saturation, the system will prioritize high-load devices and generate a "switch link + high-priority maintenance" JSON instruction.

[0045] Perform maintenance and closed-loop feedback optimization: after executing the maintenance instruction, the system collects device state feedback data in real time. If the prediction deviation exceeds , the incremental learning mechanism is triggered to update the model parameters. For example, if the addition of a new device node causes a change in the topology, the system quickly adapts to the network environment through incremental learning, improving the accuracy of subsequent decisions.

[0046] Embodiment 3 is an embodiment of the present application, which provides an LSTM power Internet of Things ring network cabinet optimization system, comprising: A dynamic feature construction module collects ring network cabinet communication state data and network environment data in real time, calculates data in a sliding window in real time, determines abnormal values through dynamic standard score standardization and performs historical data interpolation correction, calculates dynamic delay change rate and bandwidth fluctuation index, and generates a multi-dimensional feature matrix; A time series adaptive prediction module inputs the multi-dimensional feature matrix into a long short-term memory network, adjusts the forgetting gate weight based on the bandwidth fluctuation index in real time, and outputs the device performance score and key influence factors; A multi-threshold decision module constructs a dynamic scoring algorithm based on the dynamic delay change rate, bandwidth fluctuation index, and packet loss rate according to the real-time network state, and defines delay and bandwidth overload scenarios using a multi-threshold linkage strategy to trigger differentiated maintenance instructions; A closed-loop optimization module collects device state feedback data after executing the maintenance instruction, filters data with a prediction deviation exceeding a set threshold, uses a dynamic loss function to strengthen the weight of new scenario data, and performs model training. When the validation set accuracy continuously improves by more than a set threshold, the model version is updated and rollback is supported.

[0047] The embodiment also provides an electronic device suitable for an LSTM power Internet of Things ring network cabinet optimization method, comprising a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize an LSTM power Internet of Things ring network cabinet optimization method as described in the above embodiment.

[0048] The embodiment also provides a storage medium having a computer program stored thereon, which is executed by a processor to realize an LSTM power Internet of Things ring network cabinet optimization method as described in the above embodiment.

[0049] The storage medium proposed in the embodiment belongs to the same inventive concept as the LSTM power IOT ring main unit optimization method proposed in the above embodiment. The technical details not described in detail in the embodiment can be seen in the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0050] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by software and necessary general hardware, and of course can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH, a hard disk or an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the method of each embodiment of the present application.

[0051] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. An LSTM power Internet of Things ring main unit optimization method, characterized by: include, Real-time collection of ring main unit communication status data and network environment data, real-time calculation of data within the sliding window, determination of outliers through dynamic standard score normalization and historical data interpolation and correction, calculation of dynamic delay change rate and bandwidth fluctuation index, and generation of a multi-dimensional feature matrix; The multi-dimensional feature matrix is ​​input into the long short-term memory network, and the forget gate weight is adjusted in real time based on the bandwidth fluctuation index to output the device performance score and key influencing factors; A dynamic scoring algorithm is built based on real-time network status by adjusting the dynamic delay change rate, bandwidth fluctuation index, and packet loss rate. A multi-threshold linkage strategy is used to define delay and bandwidth overload scenarios, triggering differentiated maintenance instructions. Collect equipment status feedback data after maintenance instructions are executed, filter out data with prediction deviations exceeding the set threshold, use dynamic loss functions to strengthen the weight of new scenario data, and perform model training. When the accuracy of the validation set exceeds the set threshold, update the model version and support rollback.

2. The LSTM power Internet of Things ring main unit optimization method according to claim 1, characterized in that: The real-time calculation of data in the sliding window includes setting a sliding window of fixed length, intercepting communication status data and network environment data in the window in real time; calculating the mean and standard deviation of the data in the sliding window, calculating the standard score through the dynamic Z-score formula, and determining it as an abnormal value if the absolute value of the standard score exceeds a preset abnormal threshold; and interpolating and correcting through historical data.

3. The LSTM power Internet of Things ring main unit optimization method according to claim 2, characterized in that: The multidimensional feature matrix includes converting the original communication data into a high-precision multidimensional feature matrix through a dynamic sliding window Z-score standardization and historical data interpolation correction mechanism; The raw communication data includes real-time delay, packet loss rate, bandwidth, and topology status; The high-precision multi-dimensional feature matrix includes delay, packet loss rate, bandwidth, topology status, dynamic delay change rate, and bandwidth fluctuation index.

4. The LSTM power Internet of Things ring main unit optimization method according to claim 3, characterized in that: The output device performance score and key influencing factors include dynamically adjusting LSTM gating weights based on the bandwidth fluctuation index, building an adaptive timing fusion model, and outputting device performance scores and key influencing factors.

5. The LSTM power Internet of Things ring main unit optimization method according to claim 4, characterized in that: The dynamic scoring algorithm is constructed by combining the dynamic delay change rate, the bandwidth fluctuation index and the current packet loss rate to calculate the comprehensive score using a dynamic weight coefficient.

6. The LSTM power Internet of Things ring main unit optimization method according to claim 5, characterized in that: The multi-threshold linkage strategy includes triggering a high-priority maintenance instruction and executing a link switch when the delay exceeds a first threshold and the bandwidth utilization exceeds a second threshold; and triggering a communication path priority dynamic adjustment instruction when a topology change is detected and the bandwidth is unevenly distributed.

7. The LSTM power Internet of Things ring main unit optimization method according to claim 6, characterized in that: The dynamic loss function includes fusing historical and new loss values, weighting and summing the historical data loss value and the new data loss value; adjusting the weight according to the scenario, and selecting new data whose prediction deviation exceeds the set threshold to participate in the calculation.

8. An LSTM power Internet of Things ring main unit optimization system, applying an LSTM power Internet of Things ring main unit optimization method according to any one of claims 1 to 7, characterized in that: include: The dynamic feature construction module collects ring main unit communication status data and network environment data in real time, calculates data within the sliding window in real time, determines outliers through dynamic standard score standardization and performs historical data interpolation and correction, calculates the dynamic delay change rate and bandwidth fluctuation index, and generates a multi-dimensional feature matrix; The time series adaptive prediction module inputs the multidimensional feature matrix into the long short-term memory network, adjusts the forget gate weight in real time based on the bandwidth fluctuation index, and outputs the device performance score and key influencing factors; The multi-threshold decision module builds a dynamic scoring algorithm based on real-time network status by adjusting the dynamic delay change rate, bandwidth fluctuation index, and packet loss rate. It also uses multi-threshold linkage strategies to define delay and bandwidth overload scenarios, triggering differentiated maintenance instructions. The closed-loop optimization module collects equipment status feedback data after maintenance instructions are executed, filters data whose prediction deviation exceeds the set threshold, uses a dynamic loss function to strengthen the weight of new scenario data, and performs model training. When the accuracy of the verification set exceeds the set threshold, the model version is updated and rollback is supported.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of an LSTM power Internet of Things ring network cabinet optimization method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of an LSTM power Internet of Things ring network cabinet optimization method described in any one of claims 1 to 7 are implemented.

Images