Partial discharge monitoring strategy optimization method and system based on dynamic resource allocation

Abstract

The present application relates to the technical field of power system operation or management, and particularly to a partial discharge monitoring strategy optimization method and system based on dynamic resource allocation, which initializes monitoring parameters by constructing a three-level monitoring system including sensor nodes, aggregation nodes and a cloud processing center, periodically or triggeredly acquires system state information, calculates the risk level of each monitoring point in combination with a dynamic risk evaluation model, constructs a resource allocation optimization model with maximum efficiency based on the risk level and resource constraints, solves the optimal scheme by using an improved multi-objective particle swarm algorithm, issues it to each node to adjust the monitoring behavior, forms a closed-loop optimization, and dynamically updates the model parameters through an online learning mechanism. The present application realizes dynamic matching of risk and resources, improves monitoring accuracy and resource utilization, enhances the adaptability of the system to changes in equipment state and environment, and is suitable for partial discharge monitoring scenarios of various power equipment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system operation or management, in particular to a partial discharge monitoring strategy optimization method and system based on dynamic resource allocation. BACKGROUND

[0002] The current partial discharge monitoring technology faces many bottleneck problems that need to be solved in practical application, which is embodied in the following aspects:

[0003] In the operation and maintenance of the power system, partial discharge monitoring is a key means to evaluate the insulation state of the equipment. It can detect the discharge signal caused by internal insulation defects of the equipment, and can detect the insulation deterioration trend in advance, which is of great significance to avoid equipment failure and ensure stable operation of the system. With the increase of types of power equipment and the complexity of operating conditions, the existing partial discharge monitoring technology gradually shows many limitations in practical application.

[0004] At present, most monitoring strategies use a fixed parameter configuration mode, that is, fixed monitoring frequency, collection accuracy and other parameters are preset for the monitoring point, and remain unchanged throughout the process. This mode is difficult to adapt to the dynamic state of the equipment. When the equipment has abnormal discharge quantity, insulation state change and other conditions, the fixed parameters cannot be adjusted in time. If the initial parameter setting is too low, the discharge signal of the high-risk equipment may not be captured in time, forming a monitoring blind area. If the parameters are generally too high, it will lead to excessive consumption of resources in low-risk equipment areas, causing waste of computing resources, communication bandwidth and energy, and it is difficult to balance monitoring efficiency and resource cost.

[0005] At the resource allocation level, the existing scheme lacks collaborative optimization logic based on the actual risk of the equipment, and often uses "average allocation" or "experience allocation" method, without dynamically binding the real-time risk level of the equipment with the resource demand. This makes it possible for high-risk equipment to have insufficient monitoring accuracy due to insufficient resources, and low-risk equipment to occupy too many resources, resulting in low overall monitoring efficiency of the system.

[0006] At the same time, the existing system has weak adaptability to dynamic scenarios. During equipment operation, the insulation state will continue to evolve with factors such as operating time, environmental interference (such as electromagnetic interference, temperature and humidity changes), and the traditional monitoring strategy lacks an effective closed-loop adjustment mechanism, which cannot dynamically update the monitoring strategy according to the equipment state change, environmental interference intensity, etc. After long-term operation, parameter mismatch may occur, leading to risk misjudgment or omission.

[0007] In addition, some schemes do not fully consider the differences in equipment types, historical operating characteristics and industry specification requirements during the initial parameter configuration stage, but rely only on general parameters to start monitoring, which may cause unreasonable parameters at the initial stage of system startup, affecting the basic monitoring effect, and requiring frequent manual intervention for adjustment, increasing the operation and maintenance cost.

[0008] Therefore, a partial discharge monitoring strategy optimization method and system based on dynamic resource allocation are proposed to solve the above problems. SUMMARY

[0009] The purpose of the present application is to provide a partial discharge monitoring strategy optimization method and system based on dynamic resource allocation to solve the problems raised in the background art.

[0010] To achieve the above purpose, the present application provides the following technical solutions:

[0011] The partial discharge monitoring strategy optimization method based on dynamic resource allocation comprises the following steps:

[0012] S1, a partial discharge monitoring system is constructed, which comprises sensor nodes deployed at multiple monitoring points, a gathering node and a cloud processing center; the sensor nodes are used to collect partial discharge signals; the gathering node is used to receive and preprocess the data of the sensor nodes; the cloud processing center is used to execute an optimization algorithm of the monitoring strategy and issue control instructions;

[0013] S2, the monitoring strategy is initialized, and initial monitoring frequencies and data acquisition accuracies are configured for each monitoring point;

[0014] S3, the cloud processing center periodically or triggeredly acquires current system state information, which comprises real-time discharge amount data, historical discharge trend data, equipment working condition data, available resource states of the sensor nodes and the communication network of each monitoring point;

[0015] S4, based on the system state information, a real-time discharge risk level of each monitoring point is calculated through a dynamic risk evaluation model; the dynamic risk evaluation model fuses real-time discharge amount, discharge trend change rate and equipment health state parameters, and outputs a quantitative risk evaluation value;

[0016] S5, based on resource constraint conditions and the real-time discharge risk level of each monitoring point, a dynamic resource allocation optimization model is constructed, with the goal of maximizing the overall monitoring efficiency of the system; the resource constraint conditions comprise total computing resource constraint, total communication bandwidth constraint and total energy budget constraint; the monitoring efficiency is jointly defined by risk coverage rate, state awareness accuracy and resource consumption efficiency;

[0017] S6, a multi-objective optimization algorithm is used to solve the dynamic resource allocation optimization model, and an optimal resource allocation scheme for each monitoring point in the next period is obtained; the multi-objective optimization algorithm is an improved multi-objective particle swarm optimization algorithm based on chaotic mapping and adaptive crossover and mutation; the resource allocation scheme at least comprises computing resource shares, communication bandwidth shares allocated to each monitoring point and monitoring task execution parameters determined therefrom, the monitoring task execution parameters comprising monitoring frequencies and data acquisition accuracies;

[0018] S7, distribute the resource allocation scheme to the corresponding sink nodes and sensor nodes, and adjust the monitoring behaviors of the monitoring points;

[0019] S8, repeat steps S3 to S7 to realize dynamic closed-loop optimization of the monitoring strategy.

[0020] As a preferred scheme, the calculation process of the dynamic risk evaluation model in step S4 is as follows: the ratio of the real-time discharge amount to the discharge amount reference threshold is multiplied by the real-time discharge amount weight coefficient, the absolute value of the discharge amount change rate is multiplied by the discharge amount change rate weight coefficient, and the output value of the mapping function of the influence of the equipment health state on the risk is multiplied by the equipment health state weight coefficient, and the sum of the three is the real-time discharge risk evaluation value; wherein the sum of the real-time discharge amount weight coefficient, the discharge amount change rate weight coefficient and the equipment health state weight coefficient is one.

[0021] As a preferred scheme, the objective function of the dynamic resource allocation optimization model in step S5 is as follows: maximize the overall monitoring efficiency of the system, which is equal to the sum of the risk coverage efficiency and the perception accuracy efficiency of each monitoring point minus the total resource consumption cost of the system; wherein the risk coverage efficiency is the product of the risk evaluation value and the logarithm value of the monitoring frequency multiplied by the risk coverage weight factor, and the perception accuracy efficiency is the product of the data acquisition accuracy and the logarithm value multiplied by the perception accuracy weight factor.

[0022] The resource constraint conditions include: the total consumption of the computing resources of each monitoring point does not exceed the upper limit of the total computing resources of the system, the total consumption of the bandwidth resources of each monitoring point does not exceed the upper limit of the total communication bandwidth of the system, and the total consumption of the energy resources of each monitoring point does not exceed the upper limit of the total energy budget of the system; wherein the resource consumption of each monitoring point is the product of the single resource consumption coefficient of the monitoring point and the monitoring frequency and the data acquisition accuracy level.

[0023] As a preferred scheme, the improved multi-objective particle swarm optimization algorithm based on chaotic mapping and adaptive crossover mutation in step S6 has the following calculation process:

[0024] Chaotic initialization: generate a chaotic sequence by using the logistic mapping, and map the chaotic sequence to the value interval of the decision variable to generate an initial population;

[0025] Adaptive inertia weight: the inertia weight decreases nonlinearly with the increase of the iteration number, and the initial inertia weight is greater than the final inertia weight;

[0026] Particle velocity and position update: the new velocity is equal to the current inertia weight multiplied by the current velocity, plus the individual learning factor multiplied by the random number multiplied by the difference between the individual historical optimal position and the current position, plus the social learning factor multiplied by the random number multiplied by the difference between the global optimal position and the current position; the new position is equal to the current position plus the new velocity;

[0027] Adaptive crossover and mutation: the crossover probability and mutation probability are adaptively adjusted according to the particle fitness value, and the closer the fitness value is to the particle with the maximum fitness value, the smaller the crossover probability and mutation probability are;

[0028] External archive update and guiding particle selection: the non-dominated sorting and congestion calculation are used to maintain the external archive, and the guiding particle is selected from the non-dominated layer;

[0029] The output of the improved multi-objective particle swarm optimization algorithm based on chaotic mapping and adaptive crossover and mutation is a resource allocation scheme solution set satisfying the Pareto optimality.

[0030] As a preferred solution, the final executed resource allocation scheme is selected from the Pareto optimal solution set according to the preset decision rule in step S6, specifically: calculating the comprehensive evaluation index of each scheme, which is equal to the product of the normalized system overall monitoring effectiveness and the monitoring effectiveness decision weight coefficient plus the difference between one minus the product of the normalized system total resource consumption cost and the resource cost decision weight coefficient; selecting the resource allocation scheme with the maximum comprehensive evaluation index as the final execution scheme; wherein the sum of the monitoring effectiveness decision weight coefficient and the resource cost decision weight coefficient is one.

[0031] As a preferred solution, the adjustment of the monitoring behavior of each monitoring point in step S7 specifically includes: the aggregation node generates specific scheduling instructions according to the received monitoring task execution parameters and issues them to the corresponding sensor node; the sensor node adjusts its signal collection frequency, sampling rate and signal preprocessing algorithm complexity according to the scheduling instructions.

[0032] As a preferred solution, the trigger condition for trigger type acquisition of the current system state information in step S3 includes: the real-time discharge amount of any monitoring point exceeds the preset threshold, a device operating state mutation signal is received, or the sensor node resource availability rate is lower than the safety threshold.

[0033] As a preferred solution, the method further comprises:

[0034] S9, by an online learning mechanism, dynamically adjusting the risk evaluation weight coefficient in the dynamic risk evaluation model and the effectiveness balance weight factor in the objective function of the dynamic resource allocation optimization model according to historical monitoring data and resource allocation effect.

[0035] The steps of the local discharge monitoring strategy optimization system based on dynamic resource allocation and the execution method of the optimization system.

[0036] As can be seen from the above technical solutions provided by the present application, the local discharge monitoring strategy optimization method and system based on dynamic resource allocation provided by the present application have the following beneficial effects:

[0037] Improving the accuracy and risk sensitivity of partial discharge monitoring: By integrating multi-dimensional parameters such as real-time discharge quantity, discharge change rate, and equipment health status through a dynamic risk assessment model, the risk level of each monitoring point is quantitatively calculated. Compared with the traditional single threshold judgment method, it can more accurately capture the trend of equipment insulation degradation (such as sudden discharge, gradual aging, etc.). At the same time, by dynamically adjusting the model weight coefficients and mapping function parameters through an online learning mechanism, it can adapt to dynamic changes such as equipment aging and environmental interference, avoid risk misjudgment or omission, and significantly improve the timeliness and accuracy of fault warning.

[0038] Achieving efficient resource allocation and cost control: This invention constructs a dynamic resource allocation optimization model with the goal of "maximizing the overall monitoring efficiency of the system". It combines a multi-objective optimization algorithm to solve the optimal solution under constraints such as computing resources, communication bandwidth, and energy. By binding high-risk points with high resource requirements and appropriately reducing resource input for low-risk points, it avoids "one-size-fits-all" resource waste. In actual operation, it can reduce the total resource consumption cost of the system while ensuring the monitoring coverage quality of high-risk areas.

[0039] Enhance the system's dynamic adaptability and long-term stability: Through a closed-loop iteration of "state perception - risk calculation - resource optimization - behavior adjustment", the system can periodically or trigger-based responses to changes in equipment operating conditions and resource status (such as sudden increases in discharge or insufficient sensor power), and adjust the monitoring frequency and acquisition accuracy in real time to avoid the limitations of fixed strategies. On the other hand, the online learning mechanism continuously iterates model parameters (such as risk assessment weights and resource consumption coefficients) through historical data, enabling the system to adapt to scenarios such as the evolution of equipment characteristics (such as accelerated insulation aging) and changes in environmental interference (such as enhanced electromagnetic interference) in the long term, without the need for frequent manual intervention and ensuring the stability of long-term operation. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the steps in the partial discharge monitoring strategy optimization method based on dynamic resource allocation of the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0042] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific embodiments.

[0043] like Figure 1 As shown, this embodiment of the invention provides a method for optimizing partial discharge monitoring strategies based on dynamic resource allocation, including the following steps:

[0044] S1, a partial discharge monitoring system is constructed, the partial discharge monitoring system comprising sensor nodes deployed at a plurality of monitoring points, a convergence node and a cloud processing center; the sensor nodes are used to collect partial discharge signals; the convergence node is used to receive and pre-process data of the sensor nodes; the cloud processing center is used to execute an optimization algorithm of a monitoring strategy and issue control instructions;

[0045] S2, an initial monitoring strategy is initialized, and initial monitoring frequencies and data acquisition accuracies are configured for the monitoring points;

[0046] S3, the cloud processing center periodically or triggeredly acquires current system state information, the system state information comprising real-time discharge amount data, historical discharge trend data, device working condition data, available resource states of the sensor nodes and a communication network of the monitoring points;

[0047] S4, based on the system state information, a real-time discharge risk level of each monitoring point is calculated through a dynamic risk evaluation model; the dynamic risk evaluation model fuses real-time discharge amount, discharge trend change rate and device health state parameters, and outputs a quantitative risk evaluation value;

[0048] S5, based on resource constraint conditions and the real-time discharge risk level of each monitoring point, a dynamic resource allocation optimization model with the maximum monitoring efficiency of the whole system as an objective is constructed; the resource constraint conditions comprise total computing resource constraint, total communication bandwidth constraint and total energy budget constraint; the monitoring efficiency is jointly defined by risk coverage rate, state sensing accuracy and resource consumption efficiency;

[0049] S6, a multi-objective optimization algorithm is used to solve the dynamic resource allocation optimization model, and an optimal resource allocation scheme allocated to each monitoring point in the next period is obtained; the multi-objective optimization algorithm is an improved multi-objective particle swarm optimization algorithm based on chaotic mapping and adaptive crossover and mutation; the resource allocation scheme at least comprises a computing resource share, a communication bandwidth share allocated to each monitoring point and monitoring task execution parameters determined therefrom, the monitoring task execution parameters comprising monitoring frequencies and data acquisition accuracies;

[0050] S7, the resource allocation scheme is issued to the corresponding convergence node and sensor node, and monitoring behaviors of the monitoring points are adjusted;

[0051] S8, steps S3 to S7 are repeated, and dynamic closed-loop optimization of the monitoring strategy is realized;

[0052] S9, through an online learning mechanism, risk evaluation weight coefficients in the dynamic risk evaluation model and efficiency balance weight factors in an objective function of the dynamic resource allocation optimization model are dynamically adjusted according to historical monitoring data and resource allocation effects.

[0053] The core of step S1 in this embodiment is to build a basic system architecture for partial discharge monitoring, to provide hardware and operation carrier for subsequent monitoring strategy implementation and dynamic optimization, mainly through "node deployment - level configuration - system joint debugging" to realize the construction of complete monitoring link:

[0054] First of all, the selection and deployment of sensor nodes need to be combined with the discharge characteristics of monitoring equipment (such as transformers, switch cabinets, etc.), select appropriate sensors (such as ultra-high frequency, ultrasonic sensors, etc.), and determine the monitoring point position according to the weak insulation parts of the equipment and the field intensity concentration area, and complete the initial parameter (such as sampling rate, acquisition threshold) configuration and function verification of the node, to ensure the accuracy and coverage of signal acquisition;

[0055] Secondly, the construction and configuration of the aggregation node, taking the edge computing terminal as the core, build the hardware foundation containing processor, communication module, storage unit, at the same time implant data preprocessing function (such as noise filtering, feature compression, integrity check), and set the communication scheduling rules (such as time division multiple access mechanism) with the sensor, realize the efficient reception, processing and transfer of sensor data;

[0056] Thirdly, the architecture construction of cloud processing center, deploy server cluster and database system, divide the function division of application, database, algorithm server, at the same time deploy state perception, optimization algorithm, instruction generation and other core software modules, develop two-way communication interface with aggregation node, form the intelligent processing center of "data reception - algorithm running - instruction issuing";

[0057] Finally, through system joint debugging to verify the coordination of each link, test the communication link stability of sensor and aggregation node, aggregation node and cloud, check the integrity and timeliness of data flow, confirm that the running state (such as power consumption, temperature) of each node meets the design standard, finally build a three-level monitoring system of "sensor acquisition - aggregation node preprocessing - cloud intelligent optimization", provide hardware support and operation environment for subsequent steps.

[0058] In this embodiment, the core of step S2 is to set the initial monitoring parameter benchmark for each monitoring point, by combining equipment characteristics and industry standard configuration to configure the basic monitoring strategy, to provide the initial operation basis for system start, and at the same time to lay the reference basis for subsequent dynamic resource allocation, mainly including five parts:

[0059] First, the determination of initial parameter benchmark, need to combine the equipment type difference analysis (such as the difference of insulation characteristics between high voltage and low voltage equipment), historical data statistical modeling (based on the discharge data of similar equipment in the past three years to determine the normal benchmark) and industry standard adjustment (according to the power equipment monitoring guide to check the parameter compliance), form the targeted benchmark value;

[0060] Second, the classification configuration of monitoring frequency is set according to the importance of the equipment (high frequency monitoring for primary equipment and low frequency monitoring for tertiary equipment), the upper and lower limits of frequency adjustment are preset (to avoid over-limit in subsequent dynamic adjustment), and a trigger type acquisition threshold is added (such as automatic continuous acquisition when the single discharge capacity exceeds the standard);

[0061] Third, the initial configuration of data acquisition accuracy is divided into three levels, corresponding to different sampling rates, sampling bits and signal resolution (high-risk equipment is adapted to high accuracy), and the resource consumption verification is associated (to ensure that the total resource consumption of the initial configuration does not exceed 60% of the rated capacity of the system), and the accuracy level is corrected for high interference environment;

[0062] Fourth, parameter writing and validation, the cloud generates standardized configuration instructions (including monitoring point ID, frequency, accuracy, etc.), which are transmitted to the aggregation node and sensor through an encrypted channel, the node stores the parameters and feeds back the confirmation, and the cloud triggers the first acquisition to verify the parameter effectiveness (such as sampling rate deviation, frequency execution error, etc.), and reconfigures the abnormal node;

[0063] Fifth, the initial strategy benchmark database is established, the cloud archives the initial parameters of each monitoring point (including configuration time, equipment type, resource consumption benchmark, etc.), evaluates the benchmark performance indicators (such as risk coverage rate, data integrity rate), and marks the strategy version, forming a traceable initial parameter archive, providing a comparison benchmark for subsequent optimization.

[0064] In this embodiment, the trigger conditions for triggering the acquisition of the current system state information in step S3 include that the real-time discharge capacity of any monitoring point exceeds the preset threshold, a device operating state mutation signal is received, or the resource availability of the sensor node is lower than the safety threshold;

[0065] Further, the function of step S3 is to perceive the system operating state and equipment discharge characteristics in real time, to comprehensively acquire multi-dimensional state information through the combination of periodic acquisition and trigger response, to provide accurate and timely data input for dynamic risk assessment and resource allocation optimization, and to ensure the scientificity and timeliness of subsequent decision-making; the following are the detailed steps:

[0066] Step S3-1: Periodic information acquisition mechanism is established:

[0067] Acquisition cycle classification setting: according to the type and importance of the equipment, the acquisition cycle level is divided:

[0068] Primary equipment (such as main transformer): the basic acquisition cycle is set to 5 minutes / time, and each acquisition contains 3 continuous power frequency cycles (60ms) of discharge signal;

[0069] Secondary equipment (such as GIS switch cabinet): the basic acquisition cycle is set to 15 minutes / time, and each acquisition contains 2 power frequency cycles (40ms);

[0070] Level 3 equipment (such as cable branch boxes): The basic data acquisition cycle is set to 30 minutes / time, and each acquisition includes 1 power frequency cycle (20ms).

[0071] The cycle length can be dynamically adjusted via cloud commands (adjustment step is 5 minutes, range is 1-60 minutes).

[0072] Periodic data collection content: The core information to be obtained in each periodic collection includes:

[0073] Discharge characteristic data: Real-time discharge quantity Discharge pulse frequency and discharge phase distribution (the proportion of pulse distribution within the 0-360° range);

[0074] Equipment operating data: Equipment operating voltage (kV), load current (A), winding / casing temperature (°C), ambient temperature and humidity (°C / %RH);

[0075] Node resource status: sensor remaining battery power, storage space utilization, communication module signal strength (dBm), and aggregation node CPU utilization;

[0076] Dynamic frequency adaptation for data acquisition: When the device is in a special operating state (such as load rate > 90% or ambient humidity > 90%), the data acquisition cycle is automatically shortened by 50% (such as from 5 minutes to 2.5 minutes for Level 1 devices), and returns to the original cycle within 10 minutes after the state is restored, ensuring information density under high-risk operating conditions.

[0077] Step S3-2: Setting Triggered Information Collection Conditions:

[0078] Level 1 Trigger Condition (Emergency Response): High-density data acquisition is immediately triggered when any of the following conditions are met:

[0079] Real-time discharge quantity ( This refers to the discharge threshold of equipment, such as transformers. );

[0080] Rate of change of discharge ( for Time monitoring point The rate of change of discharge quantity reflects the rate of change of discharge quantity over time (discharge quantity increases by more than 500pC within 10 seconds);

[0081] Sensor node battery level ≤15% or communication signal strength ≤-100dBm (resource critical state).

[0082] Triggered continuous acquisition mode: 1 acquisition every 2 seconds for 1 minute, then 1 acquisition every 10 seconds for 5 minutes until the state returns to normal;

[0083] Secondary trigger condition (attention response): When the following conditions are met, trigger enhanced acquisition:

[0084] Real-time discharge amount is in interval;

[0085] Device temperature exceeds rated value by 5°C (e.g. transformer top oil temperature exceeds 85°C);

[0086] Aggregation node cache usage rate ≥ 80% (data congestion warning);

[0087] After triggering, the acquisition period is shortened to 1 / 2 of the original period (e.g. from 15 minutes to 7.5 minutes for secondary devices), and after 3 cycles, evaluate whether to resume;

[0088] Trigger signal priority processing: When multiple trigger conditions are met at the same time, prioritize "primary trigger > secondary trigger > periodic acquisition" to ensure resource allocation in emergency situations (e.g. pause periodic acquisition and exclusive communication channel when primary trigger is triggered);

[0089] Step S3-3: Multi-dimensional state information classification processing:

[0090] Real-time discharge data processing: After the original discharge signal collected by the sensor is preprocessed by the aggregation node, the characteristic parameters are extracted:

[0091] Peak detection algorithm is used to identify the maximum discharge amount in each power frequency cycle with an error of within ±3%;

[0092] Calculate discharge phase distribution entropy (reflecting the randomness of discharge, the higher the entropy value, the closer to the fault state) through phase window statistics (every 10° as a window);

[0093] Slide average of the discharge amount collected for 3 times in a row to get the smooth value , filter out transient interference (such as false peaks caused by electromagnetic pulses);

[0094] Device operating condition data standardization: normalize various operating condition parameters to the [0, 1] interval for subsequent model calculations:

[0095] Temperature normalization:

[0096] ( is the normalized temperature value; is the measured temperature; The minimum / maximum temperature allowed for the device;

[0097] Load rate calculation: Load rate for the device; Real-time current, Rated current;

[0098] Ambient temperature correction: when humidity > 85%, multiply the discharge capacity data by a correction factor of 1.2 (because high humidity can exacerbate discharge);

[0099] Resource state data quantification: convert node resource state into a calculable quantitative index:

[0100] Resource availability: calculate the ratio of remaining resources to total resources (such as power availability Where, Power availability; Remaining power; Total power;

[0101] Communication quality score: based on the packet loss rate of the last 10 transmissions Calculate, (Where, Communication quality score; Packet loss rate of the last 10 transmissions) (full score is 100 points, packet loss rate > 2096 is scored as 0);

[0102] Calculate load index: (Where, Load index; CPU usage; Memory usage) (average of CPU and memory usage, reflecting computing pressure);

[0103] Step S3-4: data transmission and integrity check:

[0104] Layered transmission protocol adaptation: select communication protocol according to data type:

[0105] Real-time discharge data and trigger-type collection data: use MQTT protocol (QoS=2, ensure that messages are delivered exactly once), transmit through 5G network, transmission delay control ≤100ms;

[0106] Periodic operating condition data and resource state data: use LoRaWAN protocol (ClassA mode), balance low power consumption and coverage, transmission period and collection period are synchronized;

[0107] Large file historical data (such as continuous 1 hour discharge waveform): use HTTP protocol for fragmented upload, each fragment size ≤1MB, support breakpoint resume;​

[0108] Data encryption and authentication: All transmitted data uses end-to-end encryption.

[0109] Sensor → Aggregation Node: AES-128 symmetric encryption is used, and the key is dynamically generated and updated periodically by the aggregation node (updated every 24 hours).

[0110] Aggregation Node → Cloud: TLS 1.3 protocol is used for encrypted transmission, combined with the device's unique certificate for two-way authentication to prevent data tampering and spoofing attacks;

[0111] Integrity verification and retransmission mechanism:

[0112] Each data packet is appended with a 32-bit CRC checksum. After receiving the data, the cloud verifies the checksum. If the checksum does not match, the packet is marked as "damaged" and a retransmission is requested.

[0113] Set a retransmission threshold: If the same data packet fails to be retransmitted after 3 attempts, start the backup communication channel (such as switching from LoRa to 4G) and record the channel quality degradation alarm.

[0114] For data with high time-series requirements (such as discharge phase data), if the transmission delay exceeds 500ms, it is marked as "timeout data" and is only used for historical analysis and not involved in real-time risk calculation;

[0115] Step S3-5: Status information fusion, storage, and visualization:

[0116] Multi-source data fusion storage: The cloud database adopts a hybrid storage architecture of "time-series + relational".

[0117] Time Series Database (InfluxDB): Stores parameters such as real-time discharge amount, temperature, and current that change over time. It is indexed by "monitoring point ID + timestamp" and supports millisecond-level queries.

[0118] Relational database (MySQL): Stores basic device information (model, commissioning time), resource configuration parameters (initial frequency, precision level), and trigger event records (trigger time, cause, processing result), and links time-series data through foreign keys;

[0119] Data lifecycle management: Real-time data is retained for 7 days (high precision), historical data is retained for 1 year by "hourly aggregation", and expired data is automatically archived to cold storage (can be queried but does not participate in real-time calculation);

[0120] Status information visualization: A multi-dimensional monitoring interface built on a cloud platform.

[0121] Real-time status dashboard: Displays the location of each monitoring point in the form of a topology map, and marks the status with color (green: normal; yellow: watch; red: emergency). Hovering the mouse displays key parameters such as real-time discharge amount, temperature, and resource availability.

[0122] Trend curve analysis: Plot the discharge volume change curve and temperature curve over the past 24 hours, and automatically mark the trigger event points (e.g., Level 1 triggering occurs at 10:23).

[0123] Resource Status Dashboard: Statistics on the distribution of power consumption, communication quality, and computing load for each node, and a heat map showing resource-scarce areas (red indicates resource utilization > 80%).

[0124] Real-time push notification of abnormal information: When an abnormal data is detected (such as three consecutive failed data collections or discharge exceeding the threshold), the system automatically generates an alarm message and pushes it to maintenance personnel via SMS and APP. The message includes: monitoring point ID, abnormality type, occurrence time, and current status parameters, ensuring timely maintenance response (target response time ≤ 15 minutes).

[0125] In this embodiment, the calculation process of the dynamic risk assessment model in step S4 is as follows: multiply the ratio of the real-time discharge quantity to the discharge quantity reference threshold by the real-time discharge quantity weight coefficient, add the absolute value of the discharge quantity change rate multiplied by the discharge change rate weight coefficient, and add the output value of the mapping function of the equipment health status on the risk multiplied by the equipment health status weight coefficient. The sum of the three is the real-time discharge risk assessment value; wherein, the sum of the real-time discharge quantity weight coefficient, the discharge change rate weight coefficient, and the equipment health status weight coefficient is one.

[0126] Furthermore, step S4 transforms system state information into quantified risk indicators. By integrating multi-dimensional state parameters through a dynamic risk assessment model, it accurately calculates the real-time discharge risk level of each monitoring point, providing targeted decision-making basis for subsequent resource allocation optimization. The detailed steps are as follows:

[0127] Step S4-1: Preprocessing of model input parameters:

[0128] Real-time discharge quantity smoothing: For the real-time discharge quantity acquired by S3 The moving average method was used to eliminate transient interference, and three consecutive data collections were used. Calculate the mean The formula is (in, for Time monitoring point Smoothed discharge amount; , They are respectively , Time monitoring point the real-time discharge amount) ; if there is data missing, the last valid collection value is supplemented;

[0129] Discharge amount change rate calculation: based on the smoothed discharge amount, the first-order difference method is used to calculate the discharge amount change rate , the formula is (wherein, T is the collection period; is the smoothed discharge amount of the monitoring point at the moment), and the absolute value is taken as the model input, reflecting the change trend of the discharge amount;

[0130] Device health status parameter normalization: the device operating condition data (such as temperature normalized value, load rate, etc.) processed by S3 is fused into the device health status parameter , and the weighted summation method is used to calculate: (wherein, , , are the normalized temperature, load rate, and operating condition parameters, respectively; , , are the fusion weights, and the sum is 1), ensuring the value range is [0, 1];

[0131] Step S4-2: Dynamic risk evaluation model parameter configuration

[0132] Initial setting of weight coefficient: according to the type and importance of the device corresponding to the monitoring point, the risk evaluation weight coefficients α, β, and γ are set:

[0133] For devices sensitive to insulation aging (such as cable joints), set α = 0.4, β = 0.3, and γ = 0.3, focusing on the real-time discharge amount and the device health status;

[0134] For devices sensitive to sudden discharge (such as switch cabinets), set α = 0.3, β = 0.4, and γ = 0.3, focusing on the discharge amount change rate;

[0135] The weight coefficient can be dynamically adjusted later through the online learning mechanism of S9;

[0136] Mapping function Selection: according to the influence characteristics of the device health status on the discharge risk, select the appropriate mapping function: if the impact of health status deterioration on risk presents linear growth, use the linear function ( is the proportional coefficient, taking the value 1.2-1.5); if it presents nonlinear saturated growth (such as risk growth slowing down when the health status is extremely poor), use the Sigmoid function (m, n are shape parameters, m takes 5-8, n takes 0.5), ensuring​ The value range matches other risk components;

[0137] Discharge amount reference threshold Determination: Based on the historical failure data of the equipment, the minimum discharge amount when the equipment has obvious insulation deterioration is determined as If there is no historical data, the standard value of similar equipment is referred to (such as the minimum discharge amount of the first-class equipment 1.2 times the typical value, and 1.5 times for the second-class equipment);

[0138] Step S4-3: Real-time discharge risk evaluation value calculation:

[0139] Model formula substitution calculation: Substitute the preprocessed parameters into the dynamic risk evaluation model, and the calculation formula is: Wherein, is the real-time discharge risk evaluation value of the monitoring point at time α, β, γ are risk evaluation weight coefficients, and the sum is 1; is the smoothed discharge amount of the monitoring point at time is the discharge amount reference threshold; is the discharge amount change rate of the monitoring point at time is the mapping function of the equipment health state on the risk; is the equipment health state parameter of the monitoring point at time

[0140] Calculation result range verification: If the calculation result exceeds the range [0, 1], it is truncated (0 is taken when ≤0, and 1 is taken when ≥1); If the result is abnormal due to parameter abnormality (such as is 0 but a high risk value appears), it is marked as “invalid value” and parameter re-inspection is triggered;

[0141] Step S4-4: Risk level division and output:

[0142] Grade threshold setting: According to the distribution characteristics of , combined with historical risk-failure correlation data, three risk levels are divided:

[0143] Low risk: , corresponding to the normal operation state of the equipment, the discharge risk is controllable;

[0144] Medium risk: , corresponding to the existence of potential insulation problems of the equipment, which needs to be strengthened;

[0145] High risk:​​​​​​​​ , the corresponding device may have a significant discharge failure, and resources need to be allocated in priority; the threshold value can be dynamically calibrated according to the actual failure condition through the online learning mechanism of S9;

[0146] Risk level association output: generate a binary result of "risk evaluation value + level" for each monitoring point (such as corresponding to "medium risk"), and associate the monitoring point ID, calculation timestamp, and store it to the cloud database, and push it to the input interface of the dynamic resource allocation optimization model as the core input parameter of S5;

[0147] Step S4-5: result anomaly verification and correction:

[0148] Cross-time consistency verification: compare the calculated in the last three times, if the difference between the adjacent two results is greater than or equal to 0.4 and there is no obvious working condition change (such as device load, environment does not mutate), it is determined as "fluctuation anomaly", and the discharge amount change rate and are recalculated, and if necessary, the last period weight coefficient is used for recalculation;

[0149] Same device multiple monitoring point cooperative verification: for multiple monitoring points of the same device (such as 3 monitoring points of a transformer), if the risk level of a monitoring point is 2 levels or more higher than that of the other two points (such as single high risk, other two low risk), combined with the sensor resource state (such as signal strength, power) of the point, whether it is caused by sensor abnormality is investigated, and if it is confirmed to be abnormal, the average risk level of other monitoring points of the same device is used instead.

[0150] In this embodiment, the objective function of the dynamic resource allocation optimization model in step S5 is to maximize the overall monitoring efficiency of the system, which is equal to the sum of the risk coverage efficiency and the perception accuracy efficiency of each monitoring point minus the total resource consumption cost of the system; wherein the risk coverage efficiency is the product of the risk evaluation value and the logarithm value of the monitoring frequency multiplied by the risk coverage weight factor, and the perception accuracy efficiency is the product of the data acquisition accuracy and the logarithm value multiplied by the perception accuracy weight factor;

[0151] The resource constraint conditions include: the total consumption of the calculation resources of each monitoring point does not exceed the upper limit of the total calculation resources of the system, the total consumption of the bandwidth resources of each monitoring point does not exceed the upper limit of the total communication bandwidth of the system, and the total consumption of the energy resources of each monitoring point does not exceed the upper limit of the total energy budget of the system; wherein the resource consumption of each monitoring point is the product of the single resource consumption coefficient of the monitoring point and the monitoring frequency and the data acquisition accuracy level;

[0152] Further, the role of step S5 is to build an optimization framework of "risk-resource-performance" linkage, based on the real-time discharge risk level output by S4 and the resource state information obtained by S3, to clarify the resource constraint boundary and system performance target, and to convert the monitoring strategy optimization problem into a quantifiable mathematical model for algorithm solving in S6, providing clear target orientation and constraint basis. The following are the detailed steps:

[0153] Step S5-1: Quantitative definition of system overall monitoring performance:

[0154] Risk coverage component definition: Quantify the requirement of "high-risk point priority monitoring" into a risk coverage item, the formula is (Wherein, is the total number of monitoring points; is the risk coverage weight factor; is the real-time discharge risk evaluation value of the monitoring point ; is the monitoring frequency improvement coefficient of performance; is the monitoring frequency of the monitoring point ); Through the logarithmic function, it reflects that "the higher the risk, the greater the marginal performance of frequency improvement" (such as high-risk point , the performance gain from 2 to 4 is 4 times that of low-risk point );

[0155] State awareness accuracy component definition: Quantify the contribution of data acquisition accuracy to state awareness as an awareness accuracy item, the formula is (Wherein, is the awareness accuracy weight factor; is the improvement coefficient of acquisition accuracy on performance; is the data acquisition accuracy level of the monitoring point ); Similarly, a logarithmic function is used to avoid resource waste caused by excessive accuracy improvement (such as the performance gain from accuracy level 3 to 4, which is lower than the gain from level 1 to 2);

[0156] Resource consumption efficiency component definition: Quantify the loss of overall performance due to resource cost as a cost item, represented by the total resource consumption cost of the system, the formula is (Wherein, is the resource cost weight factor); Through the integration of computing resources, communication bandwidth, and energy consumption: (Wherein, , , are the total computing resources, bandwidth, and energy consumption, respectively; , , ​Resource type weight, sum to 1)

[0157] Overall performance function integration: integrate the above three parts into the overall monitoring performance objective function of the system: , ensure that the objective function reflects the dual demands of "improving risk coverage and perception accuracy" and "reducing resource consumption";

[0158] Step S5-2: definition and quantification of resource constraints:

[0159] Total computing resource constraint setting: based on the upper limit of the computing power of the cloud and the aggregation node, determine the computing resource constraint formula: (Where, is the monitoring point The computing resource consumption coefficient of single collection processing is related to the sensor type; , Monitoring frequency, collection accuracy level, etc. is the upper limit of the total computing resources of the system);

[0160] Determine by real-time monitoring of server CPU / memory usage (e.g. take 80% of the total computing power as the upper limit, reserve 20% for sudden computing);

[0161] Total communication bandwidth constraint setting: according to the actual bearing capacity of the communication network, determine the bandwidth constraint formula: (Where, is the monitoring point The bandwidth consumption coefficient of single data transmission is related to the data compression rate; is the upper limit of the total communication bandwidth of the system); Set reference to nearly 1 hour network average throughput (e.g. take 90% of the average throughput to avoid network congestion);

[0162] Total energy budget constraint setting: combined with the power supply capability of sensors and aggregation nodes (such as battery capacity, mains stability), determine the energy constraint formula: (Where, is the monitoring point The energy consumption coefficient of single collection transmission is related to the transmission power; is the upper limit of the total energy budget of the system);

[0163] For battery-powered nodes, calculate "daily energy consumption x remaining endurance days" (e.g. take 7 times the daily energy consumption when the remaining endurance is 7 days);

[0164] Dynamic adaptation of constraint parameters: set dynamic adjustment mechanism for each constraint condition: when the usage rate of a certain type of resource is continuously below 50% for 3 periods (such as the usage rate of computing resources is only 30%), the upper limit of the corresponding constraint is increased by 10% (such as from 1000 units to 1100 units); if the usage rate is continuously above 90% for 3 periods, it is reduced by 10%, ensuring that the constraint matches the actual resource state;

[0165] Step S5-3: determination of performance balance weight and consumption coefficient:

[0166] Performance balance weight factor 、 、 Setting: set according to system operation target difference:

[0167] Fault warning priority scenario (such as equipment maintenance period): set = 0.4, = 0.3, = 0.3, focusing on risk coverage and sensing accuracy;

[0168] Resource energy saving priority scenario (such as low battery node): set = 0.2, = 0.2, = 0.6, focusing on controlling resource consumption;

[0169] The weight factor can be dynamically adjusted according to the historical performance improvement rate through the online learning mechanism of S9 (such as appropriately increasing the weight under the condition that the performance improvement is significant under the weight);

[0170] Resource consumption coefficient 、 、 Calibration: determine the basic value through experiments and historical data statistics:

[0171] For each monitoring point, continuously collect 10 times of resource consumption data (such as computing resource consumption, bandwidth occupation, energy consumption) under different 、 , adopt linear regression fitting to obtain the consumption coefficient (such as , solve 、 by least square method;

[0172] For monitoring points with large environmental interference (such as high electromagnetic interference area), multiply the coefficient by a correction factor of 1.1-1.3 to compensate for additional resource loss;

[0173] Monitoring efficiency coefficient 、 Setting: Reflecting the efficiency of monitoring frequency on risk coverage, high-risk points are set in areas where (frequency promotion is more effective), and low-risk points are set in areas where ; Reflecting the efficiency of collection accuracy on perception, monitoring points with ambiguous insulation status of equipment are set (accuracy promotion is more necessary), and those with clear status are set ;

[0174] Step S5-4: Integration and verification of dynamic resource allocation optimization model:

[0175] Model variables and boundaries are clear: the decision variables of the model are the , (resource limits need to be met , , , , , adjustment intervals preset for S2); the constraint conditions are the three resource constraints of step S5-2; the objective function is the overall performance function of step S5-1 , forming a complete "variable-constraint-target" model framework;

[0176] Model feasibility verification: select historical typical scenarios (such as 3 high-risk points + 5 medium-risk points + 10 low-risk points), and substitute the actual resource state into the model to check whether there is a feasible solution: if (i.e. still exceeding the resource upper limit under the minimum configuration), trigger resource emergency scheduling (such as temporarily closing unnecessary collection of 2 low-risk points), and mark it as "resource overload" state; if there is a feasible solution, record the constraint tightness of the current model (such as the usage rate of resource constraints is 75%);

[0177] Model and algorithm adaptability check: confirm that the objective function and constraint conditions of the model meet the solution requirements of the improved multi-objective particle swarm optimization algorithm used in S6: whether the objective function is derivable (the logarithmic function is derivable, meeting the algorithm gradient calculation requirement), whether the constraint is linear (the resource constraint is a linear inequality, which is convenient for algorithm processing); if there is a nonlinear constraint (such as the communication delay constraint introduced later), linearization processing (such as approximation with piecewise linear function) needs to be performed in advance to ensure that the algorithm can be efficiently solved;

[0178] Step S5-5: Real-time updating mechanism of model parameters:

[0179] Parameter update trigger condition: set the trigger scenario for model parameter update:

[0180] Periodic update: update the resource constraint upper limit , , ) and the consumption coefficient ( , , ); event-triggered update: when a monitoring point sensor is replaced (such as replacing a UHF sensor), the communication network is upgraded (such as switching from 4G to 5G), or the device working condition changes (such as the load rate rising from 60% to 90%), the corresponding parameters are immediately updated (such as recalibrating or adjusting ;

[0181] Parameter update process: after receiving the parameter update trigger signal, the cloud processing center automatically retrieves the latest resource state data (such as server monitoring data, sensor energy consumption data), recalculates and replaces the corresponding parameters in the model, and at the same time, retains the parameter update record (including update time, reason, old value, new value) to facilitate model tracing and rollback (such as when the updated model has no feasible solution, it can be rolled back to the previous version of the parameter).

[0182] In this embodiment, the improved multi-objective particle swarm optimization algorithm based on chaotic mapping and adaptive crossover and mutation in step S6 includes the following calculation processes:

[0183] Chaotic initialization: a chaotic sequence is generated using the logistic map, and the chaotic sequence is mapped to the value interval of the decision variable to generate an initial population;

[0184] Adaptive inertia weight: the inertia weight decreases nonlinearly with the increase of the iteration number, and the initial inertia weight is greater than the final inertia weight;

[0185] Particle velocity and position update: the new velocity is equal to the current inertia weight multiplied by the current velocity plus the individual learning factor multiplied by the random number multiplied by the difference between the individual historical optimal position and the current position plus the social learning factor multiplied by the random number multiplied by the difference between the global optimal position and the current position; the new position is equal to the current position plus the new velocity;

[0186] Adaptive crossover and mutation: the crossover probability and mutation probability are adaptively adjusted according to the particle fitness value, and the closer the fitness value is to the maximum fitness value, the smaller the crossover probability and mutation probability of the particle;

[0187] External archive update and guide particle selection: the non-dominated sorting and congestion degree calculation are used to maintain the external archive, and the guide particles are selected from the non-dominated layer;

[0188] The output of the improved multi-objective particle swarm optimization algorithm based on chaotic mapping and adaptive crossover and mutation is a set of resource allocation scheme solutions that meet the Pareto optimality;

[0189] In step S6, the final executed resource allocation scheme is selected from the set of Pareto optimal solutions according to a preset decision rule. Specifically, a comprehensive evaluation index of each scheme is calculated, which is equal to the normalized system overall monitoring effectiveness multiplied by the monitoring effectiveness decision weight coefficient plus the difference between one minus the normalized system total resource consumption cost multiplied by the resource cost decision weight coefficient; the resource allocation scheme with the maximum comprehensive evaluation index is selected as the final execution scheme; wherein the sum of the monitoring effectiveness decision weight coefficient and the resource cost decision weight coefficient is one;

[0190] Further, the role of step S6 is to solve the dynamic resource allocation optimization model through the improved multi-objective optimization algorithm, generate the Pareto optimal solution set considering the system monitoring effectiveness and resource consumption under the premise of meeting the resource constraint condition, and select the final execution scheme according to the decision rule, thereby providing a quantitative basis for monitoring strategy adjustment; the following is the detailed steps:

[0191] Step S6-1: parameter initialization of improved multi-objective particle swarm optimization algorithm:

[0192] Particle swarm size and dimension setting: according to the number of monitoring points The particle dimension is determined to be (every monitoring point corresponds to 2 decision variables: monitoring frequency and collection accuracy level ); the particle swarm size is set to (ensure population diversity, such as when the size is 60), and each particle represents a resource allocation scheme ;

[0193] Decision variable value range definition: set the boundary for each decision variable: ( , are the upper and lower limits of the monitoring frequency preset in S2), ( , are the upper and lower limits of the accuracy level preset in S2), to avoid particles exceeding the actual feasible range;

[0194] Initial configuration of algorithm core parameters: set the maximum number of iterations (dynamically adjusted according to the number of monitoring points); initial inertia weight , final inertia weight ; learning factor ; initial crossover probability , initial mutation probability ; external archive capacity is set to 100 (maximum number of non-dominated solutions stored);

[0195] Step S6-2: initial population generation based on chaotic mapping:

[0196] Logistic chaotic sequence generation: A chaotic sequence is generated using the Logistic mapping, with the following formula: (in, For the first The chaotic value of the next iteration; These are the parameters for controlling chaos. For the first The chaotic value of the next iteration, the initial value Randomly select values ​​within (0,1) that are not equal to 0.25, 0.5, or 0.75 to generate a chaotic sequence of the same length as the particle dimension, ensuring both sequence traversability and randomness.

[0197] Mapping chaotic sequences to decision variables: converting chaotic values Mapping to the range of values ​​for the decision variable, the formula is: , (in, , The mapped monitoring frequency and accuracy level; , , , (For the boundary of decision variables), the mapping result is rounded (to an integer precision level) to form the initial particles;

[0198] Initial population feasibility verification: For each generated particle, the resource constraints (computational resources, bandwidth, energy constraints) of S5 are substituted for verification. If the constraints are not met, the decision variables outside the high-risk points are randomly adjusted (reducing the low-risk points). or Make it feasible, ensuring the proportion of feasible solutions in the initial population. ;

[0199] Step S6-3: Adaptive update of particle velocity and position:

[0200] Adaptive inertia weight calculation: The inertia weight is calculated based on the current iteration number k, using the following formula: (in, This represents the current inertia weight; , These are the initial and final inertia weights; This represents the current iteration number; (to maximize the number of iterations), so that the weights decrease non-linearly with each iteration (in the early stage, large weights enhance global search, and in the later stage, small weights enhance local optimization).

[0201] Particle velocity update: according to formula Update speed (of which, For particles No. Vidi The speed of each iteration; For the first The speed of each iteration; , For learning factors; , A random number between (0,1); For particles No. The optimal position of an individual in dimension; For particles No. Vidi The position of the next iteration; For the global guiding particle (position of dimension), and impose boundary constraints on velocity. , (20% of the decision variable interval);

[0202] Particle position update: according to formula Update location (where, For particles No. Vidi The position of the next iteration), and the truncation process for positions that exceed the boundary (e.g. Time to take Rounding is performed on integer variables such as precision levels;

[0203] Step S6-4: Adaptive crossover mutation and external archive update:

[0204] Adaptive crossover operation: The updated particle is crossed with an elite particle randomly selected from the external archive, with the crossover probability set according to... Calculate (where, The crossover probability; The initial crossover probability; , These are the maximum and minimum fitness values ​​of the current population; The fitness value of the particle to be crossed; (A very small constant), particles with low fitness (poor performance) are crossovered with a high probability to promote population evolution; the crossover method is arithmetic crossover: ( A random number in (0,1) For elite particles (Dimensional position);

[0205] Adaptive mutation operation: Mutate the crossover particles, with the mutation probability according to... Calculate (where, The mutation probability; The initial mutation probability; For the fitness value of the current particle, the particle with low fitness adopts high mutation probability to increase diversity; the mutation method adopts Gaussian mutation: For the Gaussian random number with mean 0 and variance , Linearly decreases from 0.2 to 0.05 with iteration);

[0206] External archive maintenance: merge the particles after crossover and mutation with the current external archive, filter the non-dominated solutions (i.e. there is no other solution that is better in all objectives) through non-dominated sorting, calculate the crowding distance of the non-dominated solutions (the average distance of the solution and its neighboring solutions in the same layer, the greater the distance, the smaller the crowding distance), and keep the solutions with large crowding distance until the archive capacity is filled, to ensure that the solutions in the archive are uniformly distributed and cover the entire Pareto front;

[0207] Step S6-5: guided particle selection and iteration termination judgment:

[0208] Guided particle selection based on risk weighting: select guided particles from the external archive When, introduce a risk level weighting factor: for each non-dominated solution, calculate its risk weighted value ( is the risk evaluation value of the monitoring point ), prefer to select larger solutions (pay more attention to high-risk point resource allocation), if the same, select the solution with large crowding distance, balance the risk sensitivity and the diversity of the solution;

[0209] Iteration termination condition judgment: terminate the iteration if any of the following conditions is met:

[0210] The maximum number of iterations is reached ;

[0211] The system performance of the optimal solution in the external archive changes by 20 consecutive iterations (convergent and stable);

[0212] The calculation time exceeds the preset threshold (such as 5 seconds, to ensure real-time performance);

[0213] Pareto optimal solution set output: the non-dominated solutions in the final external archive are taken as the Pareto optimal solution set, each solution contains the , and the corresponding system performance , total resource consumption , to provide alternative solutions for subsequent decision-making;

[0214] Step S6-6: decision selection of the final resource allocation scheme: ​

[0215] Comprehensive evaluation index calculation: for each scheme in the Pareto optimal solution set, calculate the comprehensive evaluation index (wherein, is the comprehensive evaluation index; , is the decision weight coefficient and ; is the normalized system performance; is the normalized total resource consumption); , According to the running mode adjustment: fault early warning mode , Energy saving mode ;

[0216] Optimal scheme selection and verification: select the largest scheme as the final execution scheme, and perform secondary verification under the resource constraint condition (to ensure actual feasibility). If there is a constraint violation, select a suboptimal scheme until a feasible scheme is found; the output scheme includes the calculation resource share , communication bandwidth share ) and corresponding , of each monitoring point, which serves as the basis for S7 adjustment.

[0217] In this embodiment, the adjustment of the monitoring behavior of each monitoring point in step S7 specifically includes: the aggregation node generates specific scheduling instructions according to the received monitoring task execution parameters and sends them to the corresponding sensor nodes; the sensor nodes adjust their signal acquisition frequency, sampling rate, and signal preprocessing algorithm complexity according to the scheduling instructions;

[0218] Further, the role of step S7 is to convert the optimal resource allocation scheme generated by S6 into actual monitoring behavior adjustment instructions, and through the collaborative execution of the cloud, aggregation nodes, and sensor nodes, to realize the dynamic adaptation of the monitoring strategy of each monitoring point and ensure that the resource allocation optimization result takes effect; the following are the detailed steps:

[0219] Step S7-1: Resource allocation scheme instruction packaging and delivery:

[0220] Scheme instruction standardized packaging: the cloud processing center converts the final resource allocation scheme (including the , , calculation resource share, bandwidth share, etc. of each monitoring point) into standardized instructions, which are stored in a JSON format structure, and the fields include: instruction ID (unique identifier), target monitoring point ID list, and (monitoring frequency), (collecting precision level), execution start time (accurate to seconds), instruction validity period (default 30 minutes); add a digital signature to the instruction (generated based on the device's unique key) to prevent tampering;

[0221] Hierarchical path planning: hierarchical distribution according to "cloud → aggregation node → sensor node":

[0222] From the cloud to the aggregation node: adopt encrypted HTTP protocol (TLS1.3) transmission, prefer 5G communication channel (time delay ≤100ms), if 5G signal is weak, switch to 4G; send by range division when distributing (e.g. one aggregation node corresponds to 10 monitoring points, then pack separately) to avoid instruction congestion;

[0223] From the aggregation node to the sensor node: adopt LoRaWAN protocol (Class C mode, support instant downlink), allocate independent time slots (time slot length 20ms) for each sensor node, and distribute according to the risk level of monitoring points (high-risk point instructions are sent first);

[0224] Tracking the progress of distribution: the cloud records the state of instruction distribution in real time ("to be sent", "sending", "delivered", "executed"), and the aggregation node and sensor node need to return "acknowledgment of receipt" (including instruction ID and local timestamp) within 1 second after receiving the instruction; if no acknowledgment is received within 3 seconds, the cloud automatically retransmits (up to 3 times), and if retransmission still fails, it is marked as "abnormal distribution" and triggers manual troubleshooting;

[0225] Step S7-2: instruction analysis and scheduling of the aggregation node:

[0226] Instruction integrity check: after receiving the instruction, the aggregation node first verifies the digital signature (compares the locally stored device public key) and the instruction format (checks whether the required fields are complete), and if the verification fails, it refuses to execute and returns "verification error"; after verification, the parameters of each monitoring point in the instruction are parsed, matched with the locally stored monitoring point ID list (invalid IDs are excluded), and a "monitoring point-parameter" mapping table is generated;

[0227] Local resource adaptation fine-tuning: the aggregation node fine-tunes the instruction parameters based on its real-time resource state (such as current CPU occupancy, remaining bandwidth):

[0228] If the computing resource share of a certain monitoring point exceeds the current remaining computing capacity of the aggregation node (e.g. the instruction requires a 20% share, but only 15% is left), the monitoring point's (e.g. from 5 times / minute to 4 times / minute), while remains unchanged (priority is given to accuracy);

[0229] If communication bandwidth is insufficient (e.g., the total bandwidth requirement of the instruction exceeds the currently available bandwidth by 10%), then for low-risk points ( )of Temporarily reduce by 20%, and record the fine-tuning amount (which will be fed back to the cloud for model optimization later);

[0230] Scheduling instruction generation and distribution: For each sensor node, the aggregation node generates detailed scheduling instructions, including: local sampling rate (based on...). Confirmed, such as (corresponding to 50MHz), acquisition period (1 / ), data compression algorithms (such as (Use lossless compression when high error rate, and lossy compression when low error rate); preprocessing task type (e.g., whether to enable phase distribution entropy calculation); distribute commands through dedicated communication ports of sensor nodes (e.g., UART interface) to ensure low bit error rate in command transmission. ;

[0231] Step S7-3: Adjusting the monitoring behavior of sensor nodes:

[0232] Parameter configuration update: After receiving the scheduling command, the sensor node writes the new parameters to its local configuration file (overwriting the original initial parameters) and synchronously updates the hardware registers.

[0233] Adjust the sampling frequency: Reset the sampling trigger interval via the timer (e.g., (Set a 12-second interval for each trigger per minute) to ensure trigger error. ms;

[0234] Adjust the acquisition precision: Press Switching the sampling circuit gain (e.g.) The gain is set to 20dB. Set the sampling depth to 10dB and update the ADC sampling bit depth (16-bit / 14-bit / 12-bit);

[0235] Adjust the preprocessing algorithm: If To improve accuracy, enable higher-order filtering algorithms (such as wavelet denoising); if Reduce or switch to basic filtering (such as mean filtering) to reduce energy consumption;

[0236] Post-adjustment self-check: After the sensor node completes the parameter update, it performs one trial acquisition: acquires a signal for one power frequency cycle, checks whether the actual sampling rate and accuracy are consistent with the command (if the sampling rate deviation is ≤5%, it is qualified), and measures the current energy consumption (compared with before adjustment, the deviation should be ≤10%). If the self-check is qualified, it records "Adjustment completed"; if it is unqualified, it reverts to the parameters before adjustment and returns "Adjustment failed" (the aggregation node reissues the command).

[0237] Historical parameter archiving: The sensor node will adjust the parameters before and after the adjustment. , The parameters (such as sampling rate, etc.) and adjustment timestamps are stored in non-volatile memory (such as Flash), and the most recent 10 adjustment records are retained to facilitate subsequent fault tracing (such as checking whether the parameters are applied correctly in case of abnormal discharge).

[0238] Step S7-4: Real-time verification of the adjustment effect:

[0239] Initial Data Acquisition Verification: After adjustment, the sensor nodes perform initial data acquisition according to the new parameters, and the aggregation node verifies the received data.

[0240] Integrity check: Checks the number of data frames (and) Matching, such as (If the frame rate is 1 minute, then 5 frames should be received within 1 minute).

[0241] Accuracy verification: Extract discharge quantity characteristics (such as peak value) from the data. ), and a comparison of features with those under the same working conditions before adjustment (high precision) The lower feature details should be richer, such as a more complete phase distribution);

[0242] Resource consumption verification: Calculate the actual bandwidth usage at this monitoring point after adjustment (e.g., data size per frame). The energy consumption (measured by a current sensor) was confirmed to be within the specified limits.

[0243] Status feedback chain construction: The aggregation node summarizes the verification results (including "qualified / unqualified" labels, actual parameter values, and resource consumption data) and uploads them to the cloud via 5G / 4G channels; the cloud compares the feedback results with the original allocation scheme and calculates the "parameter execution deviation rate" (e.g., ...). If the deviation rate is ≤10%, it is marked as “adjustment effective”; otherwise, it is included in the sample library of S9 online learning (for optimizing subsequent allocation models).

[0244] Step S7-5: Emergency handling of abnormal scenarios:

[0245] Sensor node unresponsive: If the aggregation node fails to receive data from a sensor node three consecutive times (even though the acknowledgment command has been delivered), it is determined that the node is "unresponsive" and the backup strategy is immediately activated.

[0246] If the monitoring point is a high-risk point ( Enable parameter enhancement schemes for other monitoring points on the same device (if any) (e.g., enhance adjacent monitoring points). (Increase by 50%), filling monitoring blind spots;

[0247] If it is a low-risk point, suspend the node collection task, allocate its resource share to other nodes, and push a "node failure" alarm to the operation and maintenance terminal;

[0248] Parameter adjustment conflict processing: if multiple sensor nodes appear resource competition after adjustment (such as simultaneously occupying the communication channel), the aggregation node reassigns the communication time slot through dynamic time division multiplexing (high-risk point time slot priority > low-risk point), and reduces the (10% temporarily reduced) until the conflict is resolved; the conflict processing process needs to be completed within 2 seconds to avoid affecting the monitoring continuity;

[0249] Emergency scene forced adjustment: if the S3 primary trigger condition is triggered during the adjustment process (such as a sudden increase in the discharge amount of a certain monitoring point), the aggregation node can temporarily interrupt the current adjustment process and execute the emergency collection instruction (such as increasing the point to the maximum allowed value) first, and then restore the original adjustment plan after the emergency state is resolved, to ensure the monitoring reliability in abnormal situations.

[0250] In this embodiment, step S9 realizes continuous optimization of model parameters through the construction of a dynamic learning mechanism, so that the system can adapt to device state evolution and environmental changes. The core includes four parts:

[0251] First, dynamic sample library construction: from historical data, select complete and representative samples (covering different risk levels, device states, and environmental conditions), and expand risk trend slope, resource efficiency, and other derived features. Old data is also eliminated according to timeliness (preferably within 3 months, and 10% of historical data is eliminated each month), providing high-quality input for learning;

[0252] Second, model parameter iteration: for the risk evaluation model, use stochastic gradient descent to optimize weight coefficients α, β, and γ, and adjust the mapping function parameters through maximum likelihood estimation; for the resource allocation model, use reinforcement learning (Q-Learning) to optimize the performance balance weight, and use the sliding window least squares method to calibrate the resource consumption coefficient, while adaptively adjusting the particle swarm algorithm control parameters according to the algorithm convergence speed;

[0253] Third, effectiveness verification: first test with an offline validation set (requiring risk prediction accuracy improvement ≥3% or resource efficiency improvement ≥2%), then select typical monitoring points for 24-hour online pilot, verify risk coverage, resource efficiency, and fault warning effect, and promote the whole system after meeting the standards;

[0254] Fourth, full-system synchronization: using RSA+AES encryption transmission according to priority (real-time synchronization of core parameters, daily / weekly synchronization of general parameters), pushing to each node through a chain synchronization mechanism, while establishing version management and rollback mechanism, and restoring to stable version if the efficiency decreases after updating, to ensure long-term reliable operation of the system.

[0255] A partial discharge monitoring strategy optimization system based on dynamic resource allocation, the steps of the optimization system execution method.

[0256] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications, changes, omissions, substitutions, and equivalents can be made by one of ordinary skill in the art without departing from the spirit and scope of the application, which is defined by the following claims and their equivalents.

Claims

1. A partial discharge monitoring strategy optimization method based on dynamic resource allocation, characterized in that: Includes the following steps: S1. Construct a partial discharge monitoring system, which includes sensor nodes, aggregation nodes, and a cloud processing center deployed at multiple monitoring points; sensor nodes are used to collect partial discharge signals; aggregation nodes are used to receive and preprocess the data from the sensor nodes; and the cloud processing center is used to execute the optimization algorithm of the monitoring strategy and issue control commands. S2. Initialize the monitoring strategy and configure the initial monitoring frequency and data acquisition accuracy for each monitoring point; S3. The cloud processing center periodically or triggeredly acquires the current system status information, which includes real-time discharge data of each monitoring point, historical discharge trend data, equipment operating data, and available resource status of sensor nodes and communication networks. S4. Based on system status information, calculate the real-time discharge risk level of each monitoring point through a dynamic risk assessment model; The dynamic risk assessment model integrates real-time discharge quantity, discharge trend change rate and equipment health status parameters to output a quantitative risk assessment value. S5. Based on resource constraints and the real-time discharge risk level of each monitoring point, a dynamic resource allocation optimization model is constructed with the goal of maximizing the overall monitoring efficiency of the system. Resource constraints include total computing resource constraints, total communication bandwidth constraints, and total energy budget constraints. Monitoring efficiency is jointly defined by risk coverage, state perception accuracy, and resource consumption efficiency. S6. A multi-objective optimization algorithm is used to solve the dynamic resource allocation optimization model to obtain the optimal resource allocation scheme for each monitoring point in the next cycle. The multi-objective optimization algorithm is an improved multi-objective particle swarm optimization algorithm based on chaotic mapping and adaptive crossover mutation. The resource allocation scheme includes at least the share of computing resources, the share of communication bandwidth, and the monitoring task execution parameters determined accordingly for each monitoring point. The monitoring task execution parameters include the monitoring frequency and data acquisition accuracy. S7. Distribute the resource allocation plan to the corresponding aggregation nodes and sensor nodes, and adjust the monitoring behavior of each monitoring point; S8. Repeat steps S3 to S7 to achieve dynamic closed-loop optimization of the monitoring strategy.

2. The partial discharge monitoring strategy optimization method based on dynamic resource allocation according to claim 1, characterized in that: The calculation process of the dynamic risk assessment model in step S4 is as follows: multiply the ratio of real-time discharge quantity to discharge quantity reference threshold by the real-time discharge quantity weight coefficient, add the absolute value of discharge quantity change rate multiplied by the discharge change rate weight coefficient, and add the output value of the mapping function of equipment health status on risk multiplied by the equipment health status weight coefficient. The sum of the three is the real-time discharge risk assessment value; wherein, the sum of the real-time discharge quantity weight coefficient, the discharge change rate weight coefficient, and the equipment health status weight coefficient is one.

3. The partial discharge monitoring strategy optimization method based on dynamic resource allocation according to claim 2, characterized in that: The objective function of the dynamic resource allocation optimization model in step S5 is to maximize the overall monitoring efficiency of the system. This efficiency is equal to the sum of the risk coverage efficiency and the perception accuracy efficiency of each monitoring point minus the total resource consumption cost of the system. Among them, the risk coverage efficiency is the product of the risk assessment value and the logarithm of the monitoring frequency, multiplied by the risk coverage weight factor, and the perception accuracy efficiency is the logarithm of the data acquisition accuracy multiplied by the perception accuracy weight factor. The resource constraints include: the total computational resource consumption of each monitoring point does not exceed the upper limit of the total computational resources of the system; the total bandwidth resource consumption of each monitoring point does not exceed the upper limit of the total communication bandwidth of the system; and the total energy resource consumption of each monitoring point does not exceed the upper limit of the total energy budget of the system. Among them, the resource consumption of each monitoring point is the product of the single resource consumption coefficient of the monitoring point and the monitoring frequency and data acquisition accuracy level.

4. The partial discharge monitoring strategy optimization method based on dynamic resource allocation according to claim 3, characterized in that: The improved multi-objective particle swarm optimization algorithm based on chaotic mapping and adaptive crossover mutation in step S6 includes the following calculation process: Chaotic initialization: A chaotic sequence is generated using logistic mapping, and the chaotic sequence is mapped to the value range of the decision variables to generate an initial population; Adaptive inertia weight: The inertia weight decreases nonlinearly with the number of iterations, and the initial inertia weight is greater than the final inertia weight; Particle velocity and position update: New velocity is based on the formula Calculate, where, As the current inertia weight, For particles No. Vidi The speed of each iteration; For the first The speed of each iteration; For individual learning factors, As a social learning factor; , A random number in the interval (0,1); For particles No. The optimal position of an individual in dimension; For particles No. Vidi The position of the next iteration; For the global guiding particle The position of the dimension; the new position according to the formula Calculate, where, For particles No. Vidi The position of the next iteration; Adaptive crossover and mutation: The crossover and mutation probabilities are adaptively adjusted based on the particle's fitness value. The closer the fitness value of a particle is to the maximum fitness value, the lower its crossover and mutation probabilities. External archive update and guide particle selection: The external archive is maintained using non-dominated sorting and crowding calculation, and guide particles are selected from the non-dominated layer; The output of the improved multi-objective particle swarm optimization algorithm based on chaotic mapping and adaptive crossover mutation is the solution set of resource allocation schemes that satisfy Pareto optimality.

5. The partial discharge monitoring strategy optimization method based on dynamic resource allocation according to claim 4, characterized in that: In step S6, the resource allocation scheme to be finally executed is selected from the Pareto optimal solution set according to the preset decision rules. Specifically, the comprehensive evaluation index of each scheme is calculated, which is based on the formula... Calculate, where, As a comprehensive evaluation indicator; For monitoring effectiveness decision-making weighting coefficients; For resource cost decision weighting coefficients; The normalized system performance; The total resource consumption is normalized; the resource allocation scheme with the highest comprehensive evaluation index is selected as the final implementation scheme; among which, the monitoring effectiveness decision weight coefficient is... With resource cost decision weighting coefficient The sum of them is one.

6. The partial discharge monitoring strategy optimization method based on dynamic resource allocation according to claim 1, characterized in that: The adjustment of the monitoring behavior of each monitoring point in step S7 specifically includes: the aggregation node generating specific scheduling instructions based on the received monitoring task execution parameters and sending them to the corresponding sensor nodes; and the sensor nodes adjusting their own signal acquisition frequency, sampling rate, and signal preprocessing algorithm complexity based on the scheduling instructions.

7. The partial discharge monitoring strategy optimization method based on dynamic resource allocation according to claim 1, characterized in that: The triggering conditions for obtaining the current system status information in step S3 include: the real-time discharge amount of any monitoring point exceeds a preset threshold, a sudden change signal in the operating status of the equipment is received, or the availability of sensor node resources is lower than a safety threshold.

8. The partial discharge monitoring strategy optimization method based on dynamic resource allocation according to claim 3, characterized in that: The method further includes: S9. Through an online learning mechanism, based on historical monitoring data and resource allocation effectiveness, dynamically adjust the risk assessment weight coefficients in the dynamic risk assessment model and the efficiency balance weight factors in the objective function of the dynamic resource allocation optimization model.

9. A partial discharge monitoring strategy optimization system based on dynamic resource allocation, characterized in that: The optimization system performs the steps of the method according to any one of claims 1-8.

Images