A substation multi-device state end-cloud collaborative adaptive detection method, system, device and medium

Abstract

The present application relates to substation equipment state detection technical field, especially a kind of substation multi-device state end cloud collaborative adaptive detection method, system, equipment and medium.It is modular terminal to collect the operating state data of each to be detected equipment, first processing core is according to operating state fluctuation degree, electromagnetic interference intensity and end cloud credibility balance coefficient calculation resource scheduling weight, transmission to second processing core;Second processing core is according to resource scheduling weight adjustment characteristic fidelity adjustment factor, complete feature extraction and compression, obtain compressed feature parameter and local detection confidence score, transmission to cloud collaborative platform;Cloud collaborative platform executes global correlation analysis, combines with characteristic fidelity adjustment factor and local detection confidence score calculation end cloud credibility balance coefficient, issue to first processing core;First processing core is according to the resource scheduling weight reference parameter of updated end cloud credibility balance coefficient adjustment, complete closed-loop detection cycle.

Description

Technical Field

[0001] This invention relates to the field of substation equipment condition monitoring technology, and in particular to a method, system, equipment, and medium for multi-equipment condition monitoring in a substation via cloud-based collaborative adaptive monitoring. Background Technology

[0002] With the continuous increase in demand for intelligent operation and maintenance of substations, the two-level detection architecture combining modular edge computing terminals (hereinafter referred to as modular terminals) and cloud collaborative platforms has become the mainstream technical path for multi-equipment status detection in substations. In this architecture, modular terminals are deployed at the substation site to collect and process multi-dimensional operating data of main equipment such as transformers, circuit breakers, and instrument transformers in real time. The processing results are then uploaded to the cloud collaborative platform via communication links to complete global analysis, forming a collaborative detection system covering all equipment in the substation.

[0003] However, the existing solutions have the following three core technical problems.

[0004] First, the computing power allocation strategy is rigid. Existing modular terminals are limited by the on-site power supply capacity and hardware size, and the processor computing power has an inherent upper limit. Existing solutions generally adopt a fixed computing power allocation method: during the normal operation of the equipment, the fixed high computing power allocation results in the ineffective occupation of resources. When the equipment shows early signs of defects, the fixed allocation ratio cannot be increased in time, resulting in insufficient feature extraction accuracy and a high rate of missed detection of early defects.

[0005] Secondly, the feature compression ratio is fixed and cannot be adjusted. The existing solution uses a fixed feature compression ratio for the compressed data transmission, and fails to dynamically adjust it according to the real-time computing power status, the degree of anomaly precursors, and the intensity of electromagnetic interference (EMI) in the field: insufficient compression ratio in the normal stage leads to ineffective uplink bandwidth occupation, and excessive compression ratio in the abnormal stage leads to loss of defect feature information. Moreover, it cannot adapt to the dynamic and strong electromagnetic interference environment in the substation, resulting in a decrease in the accuracy of the detection data.

[0006] Third, the detection results from the terminal and the cloud lack adaptive fusion. Local detection on the terminal is limited to the perspective of a single device and cannot perceive the risk of associated defects across all devices on the entire site; global detection in the cloud is constrained by uplink bandwidth and the integrity of feature data is insufficient after compression and transmission; in the existing solution, the fusion weight of the detection results from both sides is fixed and cannot be dynamically adjusted according to the feature compression ratio and the confidence of both sides, so the overall detection accuracy is constrained by the lowest value of the capabilities of both ends.

[0007] The combination of these three problems makes it difficult for substation modular terminals to achieve high-precision equipment status monitoring around the clock under the constraints of limited computing power and bandwidth, and the risk of missing early defects persists. Summary of the Invention

[0008] In view of the problems existing in the prior art, the present invention is proposed.

[0009] Therefore, the problem to be solved by this invention is how to solve the core technical problems of high early defect detection rate caused by rigid computing power allocation, improper feature compression and fixed weight of edge-cloud fusion in existing solutions.

[0010] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, embodiments of the present invention provide a substation multi-device status end-cloud collaborative adaptive detection method, which includes: a modular terminal collecting operating status data of each device to be detected in the substation; a first processing core calculating resource scheduling weights based on the fluctuation degree of the operating status data, the intensity of on-site electromagnetic interference and the end-cloud reliability balance coefficient, and transmitting the data to a second processing core. The second processing core adjusts the feature fidelity adjustment factor according to the resource scheduling weight, completes feature extraction and compression, obtains compressed feature parameters and local detection confidence score, and transmits them to the cloud collaboration platform. The cloud-based collaborative platform performs a global correlation analysis on the compressed feature parameters to obtain a cloud detection confidence score. It then calculates the end-cloud confidence balance coefficient by combining the feature fidelity adjustment factor and the local detection confidence score, and sends it to the first processing core. The first processing core adjusts the baseline parameters of resource scheduling weights based on the updated edge-cloud trust balance coefficient, thus completing the closed-loop detection cycle.

[0011] As a preferred embodiment of the substation multi-equipment status terminal cloud collaborative adaptive detection method of the present invention, wherein: the first processing core calculates the resource scheduling weight, including obtaining the operating status fluctuation intensity based on the root mean square difference between the operating status data of each sampling point in the current sampling period and the period mean; The electromagnetic interference field strength collected on-site was normalized to obtain the quantized value of electromagnetic interference intensity. Based on the basic computing power allocation coefficient, the complementary values ​​of the intensity of operational status fluctuations and electromagnetic interference intensity and the end-cloud credibility balance coefficient of the previous period are weighted and superimposed with corresponding weight constants to obtain the resource scheduling weight.

[0012] As a preferred embodiment of the substation multi-equipment status terminal cloud collaborative adaptive detection method of the present invention, wherein: the second processing core adjusts the feature fidelity adjustment factor according to the resource scheduling weight, including, based on the preset feature compression ratio upper limit, multiplying the resource scheduling weight, the abnormal precursor probability of the current sampling period and the electromagnetic interference intensity quantization value by the corresponding weight constant and then summing them to obtain the comprehensive adjustment amount; The feature fidelity adjustment factor is obtained by subtracting the product of the comprehensive adjustment amount and the adjustable range of the compression ratio from the preset upper limit value of the feature compression ratio, within the range of the preset lower limit value to the upper limit value of the compression ratio.

[0013] As a preferred embodiment of the substation multi-equipment status terminal-cloud collaborative adaptive detection method described in this invention, the calculation of the terminal-cloud credibility balance coefficient includes obtaining the terminal local detection comprehensive credibility by multiplying the feature fidelity adjustment factor and the local detection confidence score. The overall credibility of cloud-based global detection is obtained by multiplying the complementary value of the feature fidelity adjustment factor with the cloud detection confidence score. The edge-cloud credibility balance coefficient is calculated using the overall credibility of local terminal detection as the numerator and the sum of the overall credibility of local terminal detection and the overall credibility of global cloud detection as the denominator.

[0014] As a preferred embodiment of the substation multi-equipment status terminal cloud collaborative adaptive detection method described in this invention, the method further includes, before the modular terminal collects the operating status data, reading the type code in the built-in storage chip of the access acquisition unit, and retrieving the corresponding type of data standardization processing parameters from the standardized processing parameter database according to the type code. The first processing core performs zero-mean standardization on the running status data according to the data standardization processing parameters, and transmits the processed data to the second processing core. When the acquisition unit is unplugged, the first processing core stops the data processing operation of the corresponding acquisition unit and releases computing resources.

[0015] As a preferred embodiment of the substation multi-equipment status terminal cloud collaborative adaptive detection method of the present invention, it further includes: dynamically adjusting the power supply voltage and clock frequency of the first processing core and the second processing core according to the numerical range of the resource scheduling weight; When the resource scheduling weight is lower than the first preset threshold, the operating voltage and clock frequency of the first processing core and the second processing core are reduced. When the resource scheduling weight is higher than the second preset threshold, the power supply voltage and clock frequency of the first processing core and the second processing core are adjusted to the rated operating value.

[0016] As a preferred embodiment of the substation multi-equipment status terminal cloud collaborative adaptive detection method described in this invention, the method further includes: the cloud collaborative platform receiving the operating status data and compressed feature parameters transmitted by the modular terminal, and using the optimization objective function of maximizing detection accuracy and minimizing computing power and bandwidth resource consumption, iteratively calculating the optimal values ​​of the resource scheduling weight and the weight constants of each correction term of the feature fidelity adjustment factor. The optimized weight constants are distributed to the first and second processing cores according to the preset optimization cycle; real-time optimization is triggered after the testing equipment completes maintenance.

[0017] Secondly, the present invention provides a substation multi-equipment status end-cloud collaborative adaptive detection system, which includes a data acquisition module, a modular terminal for acquiring the operating status data of each device to be detected in the substation, and a first processing core for calculating resource scheduling weights based on the fluctuation of the operating status data, the intensity of on-site electromagnetic interference and the end-cloud reliability balance coefficient, and transmitting the data to a second processing core. The transmission module, the second processing core, adjusts the feature fidelity adjustment factor according to the resource scheduling weight, completes feature extraction and compression, obtains compressed feature parameters and local detection confidence score, and transmits them to the cloud collaborative platform; The analysis module performs a global correlation analysis on the compressed feature parameters on the cloud collaboration platform to obtain the cloud detection confidence score. It then calculates the end-cloud confidence balance coefficient by combining the feature fidelity adjustment factor and the local detection confidence score, and sends it to the first processing core. The closed-loop detection module's first processing core adjusts the baseline parameters of resource scheduling weights based on the updated end-to-cloud trustworthiness balance coefficient to complete the closed-loop detection cycle.

[0018] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the computer program instructions, when executed by the processor, implement the steps of the substation multi-device status terminal cloud collaborative adaptive detection method as described in the first aspect of the present invention.

[0019] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, they implement the steps of the substation multi-device status terminal cloud collaborative adaptive detection method as described in the first aspect of the present invention.

[0020] The beneficial effects of this invention are as follows: This invention solves the core problem of high early defect detection rate in substation modular terminals under limited computing power and bandwidth constraints by constructing a closed loop with dynamic linkage of three variables: resource scheduling weight, feature fidelity adjustment factor, and end-to-cloud reliability balance coefficient. The first processing core incorporates three dimensions—operating status fluctuation, on-site electromagnetic interference intensity, and the end-to-cloud reliability balance coefficient fed back from the cloud in the previous cycle—into computing power allocation, transforming terminal computing power from a fixed proportion to on-demand response, avoiding the dilemma of ineffective resource occupation during normal periods and insufficient computing power during abnormal periods under a fixed allocation strategy. The second processing core adjusts the feature fidelity adjustment factor based on a three-dimensional linkage of computing power level, probability of anomaly precursors, and electromagnetic interference intensity. When the risk of anomalies increases, the feature compression ratio is proactively reduced, ensuring that defective features are fully preserved during transmission. When interference is strong, the compression ratio is appropriately increased to reduce bandwidth usage of redundant interference data, achieving a dynamic balance between bandwidth utilization and feature integrity. The edge-cloud credibility balance coefficient incorporates the feature compression ratio and bilateral confidence into the weight allocation. The fusion weight comes from the real-time capability assessment of the edge-cloud rather than a fixed ratio. The weight constant is periodically iteratively optimized to ensure that the scheduling parameters are adaptively updated in the long term according to the equipment's operating status, without the need for manual intervention. The resource scheduling weight directly drives the tiered adjustment of dual-core power supply and clock frequency. During the normal monitoring phase, the overall power consumption is reduced, and during the anomaly analysis phase, full-load computing power is restored, ensuring long-term low-power stable operation of the substation. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart of a multi-device status terminal cloud collaborative adaptive detection method for substations; Figure 2 A computer equipment diagram for a multi-device status terminal cloud collaborative adaptive detection method in substations; Figure 3 A closed-loop data flow diagram of multi-module collaborative computing power feature weights for a multi-equipment status terminal cloud collaborative adaptive detection method in substations; Figure 4 The time-series interaction diagram of the single-sampling-cycle algorithm for the multi-equipment status terminal cloud collaborative adaptive detection method in substations; Figure 5 This is a hardware module reliability verification and operation state transition diagram for a multi-device status cloud collaborative adaptive detection method in substations. Detailed Implementation

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0024] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0026] Example 1 Reference Figure 1 This is the first embodiment of the present invention, which provides a substation multi-device status terminal-cloud collaborative adaptive detection method, including, S100: The modular terminal collects the operating status data of each device to be tested in the substation. The first processing core calculates the resource scheduling weight based on the fluctuation of the operating status data, the intensity of electromagnetic interference on site, and the end-to-cloud reliability balance coefficient, and transmits it to the second processing core.

[0027] S200: The second processing core adjusts the feature fidelity adjustment factor based on the resource scheduling weight, completes feature extraction and compression, obtains compressed feature parameters and local detection confidence score, and transmits them to the cloud collaborative platform.

[0028] S300: The cloud-based collaborative platform performs global correlation analysis on the compressed feature parameters to obtain the cloud detection confidence score. It then calculates the end-cloud confidence balance coefficient by combining the feature fidelity adjustment factor and the local detection confidence score, and sends it to the first processing core.

[0029] S400: The first processing core adjusts the baseline parameters of resource scheduling weights based on the updated end-to-cloud trustworthiness balance coefficient to complete the closed-loop detection cycle.

[0030] In S100, the first processing core dynamically calculates resource scheduling weights based on the fluctuation level of operating status, the intensity of electromagnetic interference, and the end-cloud confidence balance coefficient of the previous cycle, enabling on-demand allocation of terminal computing power. In S200, the second processing core adjusts the feature fidelity adjustment factor based on the resource scheduling weight, the probability of anomalies, and the intensity of electromagnetic interference in a three-dimensional linkage, completing feature extraction and adaptive compression. In S300, the cloud-based collaborative platform performs global correlation analysis, calculates the end-cloud confidence balance coefficient by combining the feature fidelity adjustment factor and the end-cloud dual-side confidence, and sends it to the first processing core. In S400, the first processing core adjusts the resource scheduling weight benchmark parameters based on the updated end-cloud confidence balance coefficient, completing the closed-loop cycle of three-variable linkage, and realizing adaptive high-precision detection of early defects in multiple devices in the substation.

[0031] Example 2 Reference Figure 1 - Figure 5 This is the second embodiment of the present invention.

[0032] In this embodiment, in step S100, the modular terminal collects the operating status data of each device to be tested in the substation. The first processing core calculates the resource scheduling weight based on the fluctuation of the operating status data, the intensity of on-site electromagnetic interference, and the end-to-cloud reliability balance coefficient, and transmits it to the second processing core, including the following steps A1-A3: A1: Before collecting operational status data in the modular terminal, it also includes reading the type code in the built-in storage chip of the access acquisition unit and retrieving the corresponding type of data standardization processing parameters from the standardized processing parameter database based on the type code; The first processing core performs zero-mean standardization on the running status data based on the data standardization processing parameters, and then transmits the processed data to the second processing core. When the acquisition unit is unplugged, the first processing core stops the data processing operation of the corresponding acquisition unit and releases computing resources.

[0033] Specifically, the modular terminal adopts a hot-swappable acquisition module group at the hardware level. The acquisition module group consists of two or more acquisition units. Each acquisition unit is connected to the main control board of the modular terminal through a standardized board-to-board connector. The connector supports hot-swapping operation. The plugging and unplugging process does not cause electrical shock to the main control board and other connected acquisition units. Each acquisition unit is equipped with an independent isolation power supply and signal isolation circuit to prevent electrical faults of a single device under test from spreading to the terminal body and other acquisition units, thus ensuring the safety and stability of the overall operation of the acquisition module group.

[0034] Each acquisition unit is configured with dedicated sensing circuits and signal conditioning circuits according to the type of equipment to be tested. For power transformers, the acquisition unit integrates partial discharge sensing channels, winding temperature sensing channels, core grounding current sensing channels, and tank vibration sensing channels. For high-voltage circuit breakers, the acquisition unit integrates opening and closing coil current sensing channels, contact travel sensing channels, and mechanical vibration sensing channels. For instrument transformers, the acquisition unit integrates secondary voltage sampling channels, secondary current sampling channels, and dielectric loss sensing channels. The operating status data acquired by each acquisition unit is amplified, filtered, and converted from analog to digital by the unit's internal signal conditioning circuit before being transmitted to the first processing core via an isolated communication bus.

[0035] When the acquisition unit is connected, the acquisition module identification unit reads the type code in the internal storage chip of the acquisition unit through the isolated communication bus. The type code is a globally unique identification code that stores the functional type, compatible device type and hardware version information of the acquisition unit. The identification process takes no more than 200ms, achieving plug-and-play functionality.

[0036] The acquisition module identification unit retrieves the data standardization processing parameters corresponding to the acquisition unit type from the standardized processing parameter database based on the read type code. These parameters include the baseline average value, baseline standard deviation, data filtering parameters, and outlier removal threshold. These parameters are pre-trained from massive historical operating data of the corresponding type of device. The retrieved data standardization processing parameters are then transmitted to the data standardization processing unit of the first processing core. Upon receiving the data standardization processing parameters, the data standardization processing unit automatically updates the processing configuration of the corresponding acquisition channel, eliminating the need for manual settings and avoiding data processing deviations caused by manual configuration errors.

[0037] After receiving the operating status data of the acquisition unit within a complete sampling period, the data standardization processing unit calculates the arithmetic mean and standard deviation of all sampled data within that period: the arithmetic mean is the ratio of the sum of all sampled data within that period to the total number of sampled data, and the standard deviation is the square root of the sum of the squares of the differences between all sampled data and the arithmetic mean.

[0038] After subtracting the arithmetic mean from the original data of each sampling point, the difference is divided by the standard deviation to obtain the operating status data after zero-mean standardization. The processed data has a mean of 0 and a standard deviation of 1, eliminating the differences in dimensions and amplitude ranges between data from different types of acquisition units. Regardless of whether the original data is current, voltage, temperature, or vibration acceleration, it is converted into standardized data of a unified dimension, ensuring that the operating status data of different dimensions contribute equally in the subsequent feature extraction process, thereby improving the accuracy of feature extraction.

[0039] After standardization, the data standardization unit stores the arithmetic mean and standard deviation of the current sampling period into the non-volatile memory of the first processing core for use in subsequent periods for data trend analysis and outlier identification. When the acquisition unit is detected to be disconnected, the acquisition module identification unit sends a channel shutdown command to the data standardization unit. The first processing core then stops the data processing operation of the corresponding channel, clears the temporary running data of that channel, and releases the corresponding computing resources to ensure efficient utilization of terminal system resources.

[0040] A2: The first core processing calculation resource scheduling weight includes obtaining the intensity of the running status fluctuation based on the root mean square of the difference between the running status data of each sampling point in the current sampling period and the period average. The electromagnetic interference field strength collected on-site was normalized to obtain the quantized value of electromagnetic interference intensity. Based on the basic computing power allocation coefficient, the complementary values ​​of the intensity of operational status fluctuations and electromagnetic interference intensity and the end-cloud credibility balance coefficient of the previous period are weighted and superimposed with corresponding weight constants to obtain the resource scheduling weight.

[0041] Reference Figure 3 Specifically, the first processing core is implemented using a low-power microcontroller, responsible for basic data processing, resource scheduling weight calculation, and system state management. It establishes a data interaction channel with the second processing core via a high-speed parallel bus, with a data transmission latency of no more than 1 millisecond. In this embodiment, the resource scheduling weight corresponds to the total computing power allocation coefficient. This represents the proportion of the total computing power allocated to the dual-core heterogeneous processing architecture during the current sampling period to the total computing power resources of the terminal. The higher the value, the more computing power resources are allocated.

[0042] The value is determined by three correction dimensions: the intensity of operational status fluctuations, the quantified value of electromagnetic interference intensity, and the cloud-edge reliability balance coefficient of the previous cycle. It is calculated using the dynamic allocation formula of the terminal's dual-core computing power, as follows: In the formula, Let t be the total computing power allocation coefficient of the modular terminal at time t, i.e., the resource scheduling weight. It is dimensionless and ranges from 0 to 1. The basic computing power allocation coefficient ranges from 0.2 to 0.4. It is the minimum computing power ratio required to ensure the basic operation of the terminal, data collection and data transmission, and to ensure that the basic operating status can be maintained under any working conditions. , , The three are weighting constants, all of which are dimensionless, and their sum does not exceed 0.8, ensuring that the final value of the total computing power allocation coefficient does not exceed 1; N is the number of sampling points per cycle, which is determined by the sampling frequency and the sampling period. Let be the running status value collected at the i-th sampling point at time t. It is the arithmetic mean of the operating status collected within the sampling period corresponding to time t, and the two have completely consistent dimensions; The background electromagnetic interference intensity at time t is normalized and mapped to the 0-1 interval, which is the quantized value of electromagnetic interference intensity. It is obtained by linear mapping of the field interference field strength collected by the electromagnetic interference detection module built into the terminal. The higher the value, the stronger the field electromagnetic interference intensity. Δt is the end-to-cloud reliability balance coefficient of the previous sampling period; Δt is the system sampling period, ranging from 100ms to 1s.

[0043] First correction term of the formula ( The second correction term of the formula () adopts the root mean square form of the coefficient of variation, which quantifies the intensity of the fluctuation in the operating status within the current sampling period by the root mean square of the square of the relative change rate of each sampling point data with respect to the period mean. The higher the fluctuation intensity, the greater the deviation of the equipment operation from the normal benchmark, and the more computing power is needed for high-precision feature extraction and anomaly analysis; The third correction term of the formula introduces a quantitative value for electromagnetic interference intensity. The stronger the electromagnetic interference at the scene, the lower the signal-to-noise ratio of the collected data, requiring more computing power for data filtering, denoising, and feature enhancement. The item introduces a complementary amount from the previous cycle's edge-cloud confidence balance coefficient. When the local detection confidence in the previous cycle is insufficient ( When the value is low, the complementary amount ( The weight constant is relatively large, so we proactively increase the allocation of local computing power and enhance local feature extraction capabilities to achieve a reverse closed-loop drive of cloud detection results on terminal computing power allocation. , , The initial value is determined by the cloud-based collaborative platform through iterative optimization and then sent out. After receiving it, the terminal directly writes it into the non-volatile memory of the first processing core. It can also be dynamically updated with each optimization cycle, without the need for manual on-site configuration. After the calculation is completed, the first processing core synchronously transmits the value to the second processing core and the power management hardware, and... Stored in non-volatile memory for reference in the next sampling cycle.

[0044] A3: Calculate the end-to-cloud credibility balance coefficient, including the product of the feature fidelity adjustment factor and the local detection confidence score, to obtain the overall credibility of the terminal local detection; The overall credibility of cloud-based global detection is obtained by multiplying the complementary value of the feature fidelity adjustment factor with the cloud detection confidence score. The edge-cloud credibility balance coefficient is calculated using the overall credibility of local terminal detection as the numerator and the sum of the overall credibility of local terminal detection and the overall credibility of global cloud detection as the denominator.

[0045] Specifically, the edge-cloud trust balance coefficient This represents the weight of the local detection results from the terminal in the current sampling period in the final detection conclusion of the end-to-cloud collaboration. The value ranges from 0 to 1, with higher values ​​indicating a greater weight for local detection. The calculation depends on the feature fidelity adjustment factor. (Output from step B1 of step S200), Terminal local detection confidence score (Output from the second core feature extraction operation) and cloud-based global detection confidence score (Output from global correlation analysis operation of cloud collaboration platform), calculated using cloud-terminal collaborative detection weight closed-loop allocation formula, the specific calculation formula is as follows: In the formula, is the edge-cloud credibility balance coefficient at time t, with dimensionless dimensions and a value range of 0 to 1; k7 and k8 are weight constants, both with dimensionless dimensions and a value range of 0.5 to 1.5, which are used to adjust the contribution ratio of local detection and cloud detection to the overall credibility. The feature fidelity adjustment factor at time t is the feature compression ratio, which is dimensionless. Let t be the confidence score of the local detection at the terminal, which is dimensionless and ranges from 0 to 1; Let t be the global detection confidence score in the cloud at time t. It is dimensionless and ranges from 0 to 1.

[0046] molecular Quantifying the overall reliability of local detection: The lower the compression level (the lower the feature compression level, the more complete the transmitted feature data), the higher the basic reliability of local detection. The higher the value, the stronger the reliability of the local detection results; the product of the two is the comprehensive reliability quantification of the local detection.

[0047] Additional terms in denominator Quantifying the overall reliability of cloud-based detection: The higher the value, the lower the integrity of local feature data. The larger the size, the higher the demand for cloud-based global analysis. The higher the value, the stronger the reliability of the cloud analysis results; the product of the two is the comprehensive reliability quantification of the cloud detection. Always maintained between 0 and 1, the corresponding cloud detection weight coefficient is 1- The sum of the two is always 1, ensuring the integrity of the weight allocation. The calculation process of this formula is completed by the collaborative weight allocation module of the cloud-based collaborative platform, and the calculation result is... The value is transmitted to the first processing core via the downlink communication link, and the first processing core uses this value as part of the resource scheduling weight calculation formula for the next sampling period. This drives the next iteration of the three-variable closed loop.

[0048] In this embodiment, in step S200, the second processing core adjusts the feature fidelity adjustment factor according to the resource scheduling weight, completes feature extraction and compression, obtains compressed feature parameters and local detection confidence scores, and transmits them to the cloud collaborative platform, including the following step B1: B1: The second processing core adjusts the feature fidelity adjustment factor based on the resource scheduling weight, including multiplying the resource scheduling weight, the probability of abnormal precursors in the current sampling period and the quantized value of electromagnetic interference intensity by the corresponding weight constants and summing them to obtain the comprehensive adjustment amount, based on the preset feature compression ratio upper limit. The feature fidelity adjustment factor is obtained by subtracting the product of the overall adjustment amount and the adjustable range of the compression ratio from the preset upper limit of the feature compression ratio, within the range of the preset lower limit to the upper limit of the compression ratio.

[0049] Specifically, the second processing core is implemented using an embedded neural network processor, and establishes a data interaction channel with the first processing core through a high-speed parallel bus, receiving standardized operating status data and total computing power allocation coefficients transmitted by the first processing core. Next, the feature extraction stage begins. The second processing core calls a lightweight one-dimensional convolutional neural network deployed in the neural network acceleration engine to complete the feature extraction operation. This network structure consists of an input layer, three convolutional layers, two pooling layers, two fully connected layers, and an output layer.

[0050] The number of input layer nodes is consistent with the length of the running status data after single-sampling period standardization. Three convolutional layers use 3×1 one-dimensional convolutional kernels with a stride of 1, responsible for extracting local temporal features from the running status data and capturing fine-grained patterns of device status data changes over time. Two pooling layers use 2×1 max-pooling kernels to downsample and reduce the dimensionality of the features extracted by the convolutional layers, reducing the computational cost of the fully connected layers while retaining the most significant feature responses. Two fully connected layers integrate and perform non-linear mapping on the dimensionality-reduced features. The output layer has three independent branches: the first branch outputs the original feature data, which is a high-dimensional feature vector after dimensionality reduction by the network; the dimension can be determined according to... The size is dynamically adjusted; the second branch outputs the probability of abnormal precursors in the current sampling period. The value ranges from 0 to 1, representing a quantitative value indicating the deviation between the current operating status data and the equipment's normal operating baseline state. A higher value indicates a greater likelihood of abnormal defects in the equipment. The third branch outputs a local detection confidence score at the terminal. This is a comprehensive quantitative value representing the completeness, signal-to-noise ratio, and feature discrimination of the current feature data. The value ranges from 0 to 1, and a higher value indicates a stronger reliability of the local detection results.

[0051] The aforementioned one-dimensional convolutional neural network was pre-trained using a supervised method. The training dataset consisted of operational data from the entire lifecycle of various types of substation equipment, covering three categories of labeled samples: normal operating state, early defect state, and fault state. Each type of state data corresponds to... Label values ​​(normal state label values ​​are close to 0, early defect state label values ​​are in the range of 0.5 to 0.8, and fault state label values ​​are close to 1) and Tag value.

[0052] The training process employs mini-batch stochastic gradient descent for iterative parameter updates. The loss function is a weighted sum of three parts: the mean squared error loss from feature reconstruction in the first branch, the binary cross-entropy classification loss in the second branch, and the mean squared error loss from confidence regression in the third branch. The initial learning rate is set to 0.001, and a cosine annealing strategy is used to gradually decrease the learning rate with each training epoch. The batch size is set to 64. The training stops when the validation set loss stops decreasing after 20 consecutive epochs. The optimal network weight parameters are then output and stored in the non-volatile memory of the second processing core as the inference model deployed in the field.

[0053] Complete feature extraction and obtain the raw feature data (output of the first branch). (Second branch output) and After the third branch output, the second processing core enters the feature compression stage, and the feature compression unit receives the data transmitted by the first processing core. With the output of its own feature extraction operation Combined with the electromagnetic interference detection module built into the terminal, the data collected in the current period... The feature fidelity adjustment factor is determined by optimizing the feature compression ratio formula of the detection data. The specific calculation formula is as follows: In the formula, The feature fidelity adjustment factor at time t, i.e., the feature compression ratio, is dimensionless and its value range is [value range missing]. arrive This is the upper limit of the characteristic compression ratio, which is dimensionless and ranges from 0.8 to 0.95, representing the highest allowable compression ratio. This is the lower limit of the characteristic compression ratio, which is dimensionless and ranges from 0.1 to 0.3. It represents the lowest allowable compression ratio. , , These are weighting constants, all dimensionless, and their sum is 1. This ensures that the overall adjustment coefficient remains between 0 and 1, so that the final... The calculation result does not exceed arrive interval; This is the total computing power allocation coefficient calculated by the first processing core in the current cycle; The probability of anomalies output by the second core feature extraction operation in the current cycle; This is the quantized value of the electromagnetic interference intensity for the current period.

[0054] Formula To adjust the baseline, the weighted sum of the three correction terms is compared with the adjustable range of the compression ratio ( The product of these factors together constitutes the downward adjustment of the benchmark value: The higher the value, the more abundant the total computing power allocated to the dual cores, the lower the corresponding feature fidelity adjustment factor, and the more complete the original feature details are preserved, ensuring high-precision analysis under abnormal conditions. The higher the value, the greater the risk of equipment defects. The corresponding feature fidelity adjustment factor should be lower to prevent the loss of key defect features due to compression operations. The higher the elevation, the stronger the electromagnetic interference at the site, and the lower the proportion of effective feature information in the collected data. The form of this correction term weakens its contribution, increases the corresponding feature compression ratio, and reduces the uplink transmission of redundant interference data.

[0055] After the calculation is completed, the feature compression unit uses a sparse coding-based feature compression algorithm to compress the original feature data output from the first branch: first, a sparse transformation is performed on the original feature data to remove redundant information, and then... The sparse transformation result is quantized and encoded to generate compressed feature parameters; the compressed feature data can be fully recovered by the corresponding decoding algorithm, and the equipment defect features are not lost due to the compression operation.

[0056] Compression feature parameters and The data is synchronously transmitted to the cloud-based collaborative platform via a communication link. The calculation process of this formula is completed in real time by the feature compression unit of the second processing core. The feature compression operation is directly used in the current cycle to ensure that the compression ratio is optimally matched with the current operating conditions, computing power and electromagnetic environment.

[0057] In this embodiment, in step S300, the cloud-based collaborative platform performs a global correlation analysis on the compressed feature parameters to obtain a cloud-based detection confidence score. It then calculates the end-cloud confidence balance coefficient by combining the feature fidelity adjustment factor and the local detection confidence score, and sends it to the first processing core. This includes the following C1 step: C1: The cloud-based collaborative platform receives the operational status data and compressed feature parameters transmitted by the modular terminal. Taking the maximization of detection accuracy and the minimization of computing power and bandwidth resource consumption as the optimization objective function, iteratively calculates the optimal values ​​of the resource scheduling weight and the weight constants of each correction term of the feature fidelity adjustment factor. The optimized weight constants are distributed to the first and second processing cores according to the preset optimization cycle; real-time optimization is triggered after the testing equipment completes maintenance.

[0058] Specifically, the global feature analysis unit of the cloud-based collaborative platform is deployed in a distributed computing cluster, capable of synchronously processing compressed feature parameters uploaded by all modular terminals within the substation. The global feature analysis unit first decodes the received compressed feature parameters to recover the core feature data of the operating status of each device under test. Then, it inputs the feature data into a pre-trained device status assessment model to obtain the independent status assessment result for each device. Simultaneously, it analyzes the status correlation between multiple devices within the same electrical bay, at the same voltage level, and with interconnected electrical connections, based on the substation's electrical connection topology. This identifies the impact of changes in the status of a single device on related devices and captures early signs of cascading defects across devices. The global feature analysis unit outputs a cloud-based global detection confidence score. Taking into account the completeness of feature data, the consistency of multi-device correlation analysis, and the accuracy of historical detection results, the value ranges from 0 to 1, with higher values ​​indicating stronger reliability of the cloud-based global analysis results. Simultaneously, device defect early warning signals are output, divided into three levels: warning, alarm, and emergency, corresponding to three detection conclusions: early defect precursors, clear defect characteristics, and serious fault risk, respectively. This notifies maintenance personnel to take corresponding maintenance measures. The output is then fed into the collaborative weight allocation module, and combined with... and The end-to-cloud credibility balance coefficient calculation formula is obtained according to step A3. The data is then transmitted to the first processing core via a downlink communication link.

[0059] Based on this, the weight constant iterative optimization unit performs an adaptive update operation on the weight constants. The optimization unit can simultaneously aggregate multi-cycle operating status data and compressed feature parameters uploaded by multiple modular terminals within the substation, establishing a station-wide equipment operation database. The optimization unit cleans and labels the aggregated data, including the actual operating status category of the equipment, the accuracy score of the detection results, the computing power resource usage, and the uplink bandwidth usage. A joint objective function is constructed, maximizing detection accuracy and minimizing computing power and bandwidth resource usage. An adaptive gradient descent optimization algorithm is used, with the partial derivatives of the objective function with respect to each weight constant as the optimization direction, to optimize the resource scheduling weight calculation formula. , , In the formula for determining the feature fidelity adjustment factor , , Perform iterative updates. Stop iterating when the change in the objective function between two consecutive iterations is less than the preset convergence threshold, and output the optimal combination of weight constants for the current round.

[0060] The weight constant iterative optimization unit supports two working modes: global optimization mode and single-terminal customized optimization mode. The global optimization mode performs joint optimization based on the operating data of all modular terminals in the entire station and outputs a general weight constant applicable to terminals of the same type throughout the station, which is suitable for scenarios with similar field environments and equipment types. The single-terminal customized optimization mode performs independent optimization based on the long-term historical data of a single terminal corresponding to the device under test and outputs a unique weight constant adapted to the operating characteristics of the device, which is suitable for main devices with high importance and special operating characteristics.

[0061] After optimization, the cloud-based collaborative platform distributes the updated weight constants to the first and second processing cores of the corresponding modular terminals via the downlink communication link. After receiving the data, the terminals automatically write the weight constants and store the parameters before and after the update into non-volatile memory, supporting parameter traceability and effect verification.

[0062] The regular optimization cycle for the weight constant is set to 30 days. An immediate optimization is automatically triggered after the testing equipment completes maintenance to ensure that the weight constant is always optimally matched with the current operating status of the equipment and the on-site environment, thus guaranteeing the accuracy of resource scheduling weight calculation and feature compression ratio determination and the detection effect during long-term operation.

[0063] In this embodiment, step S400, the first processing core adjusts the baseline parameters of resource scheduling weights based on the updated end-to-cloud trustworthiness balance coefficient to complete the closed-loop detection cycle, including the following step D1: D1: Also includes dynamically adjusting the power supply voltage and clock frequency of the first and second processing cores based on the numerical range of the resource scheduling weight. When the resource scheduling weight is lower than the first preset threshold, reduce the operating voltage and clock frequency of the first processing core and the second processing core; When the resource scheduling weight is higher than the second preset threshold, the power supply voltage and clock frequency of the first processing core and the second processing core will be adjusted to the rated operating value.

[0064] Reference Figure 5 Specifically, the modular terminal has a built-in multi-channel programmable power management chip. The chip integrates multiple buck converters and linear regulators to provide independent power supply channels for the first processing core, the second processing core, and the acquisition module group. The output voltage of each power supply channel can be dynamically adjusted via a serial bus, with an adjustment range of 0.8V to 3.3V and an adjustment step of 12.5mV, which can accurately match the core power supply configuration under different computing power requirements.

[0065] The power management hardware establishes a direct connection with the system control bus of the first processing core, and receives the total computing power allocation coefficient output by the first processing core in real time. (i.e., resource scheduling weight), based on Within the specified numerical range, the power supply voltage and clock frequency are dynamically adjusted according to preset grading rules.

[0066] when When the power consumption is below the first preset threshold of 0.3, it indicates that the current state is a normal monitoring state with low computing power requirements. The power management hardware will reduce the clock frequency of the first processing core to the basic operating frequency, adjust the power supply voltage to the minimum stable operating voltage, and at the same time shut down the neural network acceleration engine of the second processing core, retaining only the basic data transmission channel and reducing the clock frequency of the second processing core to the standby frequency. In this state, the overall power consumption of the terminal can be reduced to less than 20% of the full load operating state, meeting the long-term low power consumption operation requirements of the substation.

[0067] when When the value is in the range of 0.3 to 0.7, it indicates that the device is in a monitoring state with medium computing power requirements. The device's operating status fluctuates to a certain extent. The power management hardware increases the clock frequency of the first processing core to 50% of the rated frequency, adjusts the corresponding power supply voltage, and at the same time activates the neural network acceleration engine of the second processing core, setting its clock frequency to 40% of the rated frequency. This maintains a low power consumption level while meeting the current requirements for feature extraction and data processing.

[0068] when When the value exceeds the second preset threshold of 0.7, it indicates that the device is in an abnormal analysis state with high computing power requirements. The device shows obvious signs of abnormal operation. The power management hardware will increase the clock frequency of the first and second processing cores to the rated full load frequency, adjust the power supply voltage to the rated operating voltage, and activate all hardware acceleration engines to ensure sufficient computing power to complete high-precision feature extraction and abnormal analysis, and ensure accurate identification of device defects.

[0069] The power management hardware is also equipped with overvoltage protection, overcurrent protection and overheat protection functions. When an abnormal state such as overvoltage, overcurrent or chip junction temperature is detected in the power supply channel, the protection mechanism is immediately triggered to cut off the power supply to the corresponding channel. At the same time, an abnormal alarm signal is sent to the first processing core to improve the safety and reliability of operation.

[0070] At the end of each sampling cycle, the power management hardware uploads the current power configuration parameters and terminal power consumption data to the data storage unit of the cloud collaboration platform for subsequent power consumption analysis and long-term optimization. After completing the dynamic adjustment of dual-core power supply and clock frequency, the first processing core synchronously adjusts the end-cloud reliability balance coefficient received in this cycle. Update the resource scheduling weight calculation formula for the next cycle. The parameters complete a full closed-loop detection cycle involving the three variables of resource scheduling weight, feature fidelity adjustment factor, and edge-cloud credibility balance coefficient, and then proceed to the data acquisition and processing flow of the next sampling cycle.

[0071] Example 3 Reference Figure 2 This is the third embodiment of the present invention.

[0072] The above is a schematic scheme of a substation multi-device status terminal cloud collaborative adaptive detection method. It should be noted that the technical solution of this substation multi-device status terminal cloud collaborative adaptive detection system belongs to the same concept as the technical solution of the aforementioned substation multi-device status terminal cloud collaborative adaptive detection method. Details not described in detail in this embodiment of the substation multi-device status terminal cloud collaborative adaptive detection system can be found in the description of the aforementioned substation multi-device status terminal cloud collaborative adaptive detection method.

[0073] This embodiment also provides a substation multi-equipment status terminal-cloud collaborative adaptive detection system, including: The acquisition module and modular terminal collect the operating status data of each device to be tested in the substation. The first processing core calculates the resource scheduling weight based on the fluctuation of the operating status data, the intensity of on-site electromagnetic interference, and the end-to-cloud reliability balance coefficient, and transmits it to the second processing core. The transmission module, the second processing core, adjusts the feature fidelity adjustment factor according to the resource scheduling weight, completes feature extraction and compression, obtains compressed feature parameters and local detection confidence score, and transmits them to the cloud collaborative platform; The analysis module performs global correlation analysis on the compressed feature parameters on the cloud collaboration platform to obtain the cloud detection confidence score. It then calculates the end-cloud confidence balance coefficient by combining the feature fidelity adjustment factor and the local detection confidence score, and sends it to the first processing core. The closed-loop detection module's first processing core adjusts the baseline parameters of resource scheduling weights based on the updated end-to-cloud trustworthiness balance coefficient to complete the closed-loop detection cycle.

[0074] This embodiment also provides an electronic device suitable for multi-device status terminal cloud collaborative adaptive detection in substations, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the multi-device status terminal cloud collaborative adaptive detection method for substations as proposed in the above embodiment.

[0075] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the substation multi-device status terminal cloud collaborative adaptive detection method proposed in the above embodiment.

[0076] The storage medium proposed in this embodiment and the method for implementing multi-device status terminal cloud collaborative adaptive detection in substations proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0077] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0078] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A multi-equipment status terminal cloud collaborative adaptive detection method for substations, characterized in that: This includes a modular terminal collecting operational status data from various devices to be tested within the substation; a first processing core calculating resource scheduling weights based on the fluctuation level of the operational status data, the intensity of on-site electromagnetic interference, and the end-to-cloud reliability balance coefficient; and transmitting the data to a second processing core. The second processing core adjusts the feature fidelity adjustment factor according to the resource scheduling weight, completes feature extraction and compression, obtains compressed feature parameters and local detection confidence score, and transmits them to the cloud collaboration platform. The cloud-based collaborative platform performs a global correlation analysis on the compressed feature parameters to obtain a cloud detection confidence score. It then calculates the end-cloud confidence balance coefficient by combining the feature fidelity adjustment factor and the local detection confidence score, and sends it to the first processing core. The first processing core adjusts the baseline parameters of resource scheduling weights based on the updated edge-cloud trust balance coefficient, thus completing the closed-loop detection cycle.

2. The substation multi-equipment status terminal cloud collaborative adaptive detection method as described in claim 1, characterized in that: The first processing core calculates resource scheduling weights, including obtaining the intensity of operational status fluctuations based on the root mean square difference between the operational status data of each sampling point in the current sampling period and the period mean. The electromagnetic interference field strength collected on-site was normalized to obtain the quantized value of electromagnetic interference intensity. Based on the basic computing power allocation coefficient, the complementary values ​​of the intensity of operational status fluctuations and electromagnetic interference intensity and the end-cloud credibility balance coefficient of the previous period are weighted and superimposed with corresponding weight constants to obtain the resource scheduling weight.

3. The substation multi-equipment status terminal-cloud collaborative adaptive detection method as described in claim 2, characterized in that: The second processing core adjusts the feature fidelity adjustment factor based on the resource scheduling weight, including multiplying the resource scheduling weight, the probability of abnormal precursors in the current sampling period, and the quantized value of electromagnetic interference intensity by the corresponding weight constants and summing them to obtain the comprehensive adjustment amount, based on the preset feature compression ratio upper limit. The feature fidelity adjustment factor is obtained by subtracting the product of the comprehensive adjustment amount and the adjustable range of the compression ratio from the preset upper limit value of the feature compression ratio, within the range of the preset lower limit value to the upper limit value of the compression ratio.

4. The substation multi-equipment status terminal cloud collaborative adaptive detection method as described in claim 3, characterized in that: The computational cloud credibility balance coefficient includes the product of the feature fidelity adjustment factor and the local detection confidence score to obtain the terminal local detection comprehensive credibility. The overall credibility of cloud-based global detection is obtained by multiplying the complementary value of the feature fidelity adjustment factor with the cloud detection confidence score. The edge-cloud credibility balance coefficient is calculated using the overall credibility of local terminal detection as the numerator and the sum of the overall credibility of local terminal detection and the overall credibility of global cloud detection as the denominator.

5. The substation multi-equipment status terminal cloud collaborative adaptive detection method as described in claim 4, characterized in that: Before the modular terminal collects the operating status data, the method further includes reading the type code in the built-in storage chip of the access collection unit and retrieving the corresponding type of data standardization processing parameters from the standardized processing parameter database according to the type code. The first processing core performs zero-mean standardization on the running status data according to the data standardization processing parameters, and transmits the processed data to the second processing core. When the acquisition unit is unplugged, the first processing core stops the data processing operation of the corresponding acquisition unit and releases computing resources.

6. The substation multi-equipment status terminal-cloud collaborative adaptive detection method as described in claim 5, characterized in that: It also includes dynamically adjusting the power supply voltage and clock frequency of the first processing core and the second processing core based on the numerical range of the resource scheduling weight; When the resource scheduling weight is lower than the first preset threshold, the operating voltage and clock frequency of the first processing core and the second processing core are reduced. When the resource scheduling weight is higher than the second preset threshold, the power supply voltage and clock frequency of the first processing core and the second processing core are adjusted to the rated operating value.

7. The substation multi-equipment status terminal-cloud collaborative adaptive detection method as described in claim 6, characterized in that: It also includes receiving the operation status data and compressed feature parameters transmitted by the modular terminal through the cloud collaboration platform, and using the optimization objective function of maximizing detection accuracy and minimizing computing power and bandwidth resource consumption, iteratively calculating the optimal values ​​of the resource scheduling weight and the weight constants of each correction term of the feature fidelity adjustment factor. The optimized weight constants are distributed to the first and second processing cores according to the preset optimization cycle. Real-time optimization is triggered once the testing equipment has completed maintenance.

8. A substation multi-equipment status terminal cloud collaborative adaptive detection system, based on the substation multi-equipment status terminal cloud collaborative adaptive detection method according to any one of claims 1 to 7, characterized in that: It also includes a data acquisition module, where modular terminals collect operating status data of each device to be tested in the substation. The first processing core calculates resource scheduling weights based on the fluctuation of the operating status data, the intensity of on-site electromagnetic interference, and the end-to-cloud reliability balance coefficient, and transmits the data to the second processing core. The transmission module, the second processing core, adjusts the feature fidelity adjustment factor according to the resource scheduling weight, completes feature extraction and compression, obtains compressed feature parameters and local detection confidence score, and transmits them to the cloud collaborative platform; The analysis module performs a global correlation analysis on the compressed feature parameters on the cloud collaboration platform to obtain the cloud detection confidence score. It then calculates the end-cloud confidence balance coefficient by combining the feature fidelity adjustment factor and the local detection confidence score, and sends it to the first processing core. The closed-loop detection module's first processing core adjusts the baseline parameters of resource scheduling weights based on the updated end-to-cloud trustworthiness balance coefficient to complete the closed-loop detection cycle.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the substation multi-equipment status terminal cloud collaborative adaptive detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the substation multi-equipment status terminal cloud collaborative adaptive detection method according to any one of claims 1 to 7.

Images