Real-time data acquisition and diagnosis system of MCB distribution box based on edge computing

Abstract

The application is a kind of MCB distribution box real-time data acquisition and diagnosis system based on edge computing; its core is a dual-mode acquisition unit, which continuously monitors the transient fluctuation index of the line under low power consumption; only when the index exceeds the preset threshold, the system will instantaneously switch to high-precision mode to acquire compressed waveform data; after receiving the data, the edge diagnosis unit reconstructs the high-fidelity fault waveform and generates structured diagnostic results using the diagnostic model; the timing correlation positioning unit accurately locates the physical position of the fault based on the high-precision timestamps of multiple diagnostic results by calculating the arrival time difference; the cloud unit then aggregates all results, evaluates and optimizes the entire system, and issues new model parameters and trigger thresholds to the front end to form a closed-loop control; this design reduces the power consumption and network communication bandwidth pressure of the front-end device.

Description

Real-time data acquisition and diagnostic system for MCB distribution boxes based on edge computing Technical Field

[0001] This invention relates to the field of intelligent monitoring and diagnosis technology for power equipment, specifically to a real-time data acquisition and diagnosis system for MCB distribution boxes based on edge computing. Background Technology

[0002] In the current field of MCB distribution box safety operation and maintenance, line status monitoring requires processing high-frequency transient waveform data. In order to achieve fault diagnosis, existing solutions mostly adopt the mode of uploading all the raw waveform data collected at the front end to the cloud for centralized analysis.

[0003] This model has significant drawbacks: the continuous transmission of massive amounts of data puts enormous pressure on communication bandwidth and leads to excessive power consumption of the front-end acquisition equipment; the computational delay introduced by data transmission and cloud analysis causes a serious lag in fault response, making it difficult to achieve second-level location and handling; in addition, the fixed acquisition thresholds and diagnostic models cannot adapt to the dynamic changes in the line environment, resulting in a high false alarm or false alarm rate and insufficient reliability of the system during long-term operation.

[0004] Therefore, how to build a real-time diagnostic system that balances low power consumption, low latency, and high accuracy on the resource-constrained edge side, and enable it to have adaptive optimization capabilities, has become a key technical problem that urgently needs to be solved. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a real-time data acquisition and diagnostic system for MCB distribution boxes based on edge computing. Specifically, the technical solution of this invention includes:

[0006] The dual-modal acquisition unit is used to monitor the transient fluctuation index of the line waveform and selectively acquire compressed waveform data based on the comparison result of the transient fluctuation index and the preset trigger threshold.

[0007] The edge diagnostic unit is used to receive compressed waveform data to reconstruct high-fidelity fault waveforms and analyze the high-fidelity fault waveforms using a preset diagnostic model to generate structured diagnostic results.

[0008] The temporal correlation positioning unit is used to calculate the time difference of arrival and solve the physical location of the fault based on the high-precision event timestamps contained in multiple structured diagnostic results.

[0009] The cloud-based optimization control unit aggregates structured diagnostic results to evaluate system performance and generates optimized model parameters and optimized trigger thresholds, which are then distributed to the edge diagnostic unit and the dual-modal acquisition unit, respectively.

[0010] Preferably, the transient fluctuation index is generated as follows: the instantaneous value at the current sampling time and the instantaneous value at the previous sampling time are obtained, and the difference between the instantaneous value at the current sampling time and the instantaneous value at the previous sampling time is divided by a fixed time interval to obtain the transient fluctuation index.

[0011] Preferably, the process of selectively acquiring compressed waveform data includes: when the transient fluctuation index does not exceed the preset trigger threshold, the dual-mode acquisition unit uses an alert measurement matrix for continuous monitoring; when the transient fluctuation index exceeds the preset trigger threshold, the dual-mode acquisition unit switches to a focusing measurement matrix to acquire compressed waveform data.

[0012] Preferably, the process of reconstructing high-fidelity fault waveforms is as follows: the compressed waveform data and the focused measurement matrix are taken as input, and the high-fidelity fault waveform vector is decoded by solving a constrained L1 norm minimization problem.

[0013] Preferably, the structured diagnostic results include fault confidence, fault type, and high-precision event timestamps; wherein, the fault confidence generation process is as follows: input the high-fidelity fault waveform vector into the preset diagnostic model, and use the preset model parameters to perform forward propagation calculation on the high-fidelity fault waveform vector to obtain the fault confidence.

[0014] Preferably, the temporal correlation positioning unit is also used to: aggregate multiple fault events located to the same line segment, and calculate the precursor risk index of the line segment by combining the fault confidence level corresponding to each fault event and the preset risk weight, and upload the precursor risk index to the cloud optimization control unit.

[0015] Preferably, the process for evaluating system performance is as follows: obtain false alarm events and missed alarm events within the evaluation period, and combine them with preset false alarm cost weighting coefficients and missed alarm cost weighting coefficients to quantify the total diagnostic error cost of the system, which is used as the global loss function value for evaluation.

[0016] Preferably, the process of generating the optimized trigger threshold is as follows: when the total loss caused by missed events exceeds the preset risk tolerance, the trigger threshold is lowered; when the total loss caused by false alarm events exceeds the preset operation and maintenance efficiency target, the trigger threshold is raised.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] 1. This system constructs a dual-modal acquisition mechanism, which monitors with extremely low power consumption most of the time. Only when an abnormal transient fluctuation index is detected will it instantly switch to high-precision mode and acquire compressed waveform data. Compared with the traditional solution of uploading the entire original waveform, this greatly reduces the power consumption of the front-end device and the network communication bandwidth pressure.

[0019] 2. This system deploys the fault waveform reconstruction and diagnostic model on the edge side close to the data source, realizing localized real-time data processing; combined with multi-point high-precision timestamps for time series correlation analysis, it can quickly calculate the physical location of the fault, significantly shortening the response time from fault occurrence to accurate location, and improving fault handling efficiency from hours to seconds.

[0020] 3. This system establishes a closed-loop optimization control system that coordinates the cloud and the edge. The cloud platform aggregates diagnostic results, quantifies and evaluates the cost of false alarms and false alarms, and then generates optimized model parameters and acquisition trigger thresholds and sends them to the edge. This enables the system to have adaptive evolution capabilities and continuously optimize diagnostic accuracy, solving the problems of traditional solutions that cannot adapt to environmental changes and have low reliability due to the use of fixed models.

[0021] 4. This system aggregates and locates multiple fault events on the same line segment, and calculates the precursor risk index of the line by combining fault confidence and risk weight. This realizes the transformation from passive response to single fault events to proactive assessment of line health status. It can effectively identify potential high-risk sections and provide core decision-making basis for implementing predictive maintenance and preventing major accidents. Attached Figure Description

[0022] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0023] Figure 1 is a structural diagram of the system of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0025] Example 1:

[0026] Please refer to Figure 1, which shows the MCB distribution box real-time data acquisition and diagnostic system based on edge computing, including:

[0027] The dual-modal acquisition unit is used to monitor the transient fluctuation index of the line waveform and selectively acquire compressed waveform data based on the comparison result of the transient fluctuation index and the preset trigger threshold.

[0028] The edge diagnostic unit is used to receive compressed waveform data to reconstruct high-fidelity fault waveforms and analyze the high-fidelity fault waveforms using a preset diagnostic model to generate structured diagnostic results.

[0029] The temporal correlation positioning unit is used to calculate the time difference of arrival and solve the physical location of the fault based on the high-precision event timestamps contained in multiple structured diagnostic results.

[0030] The cloud-based optimization control unit aggregates structured diagnostic results to evaluate system performance and generates optimized model parameters and optimized trigger thresholds, which are then distributed to the edge diagnostic unit and the dual-modal acquisition unit, respectively.

[0031] This embodiment provides a real-time data acquisition and diagnostic system for MCB distribution boxes based on edge computing. It aims to achieve low-latency, high-precision, and low-power real-time monitoring, diagnosis, and location of line faults by deploying intelligent computing capabilities on the distribution box side close to the data source, and to complete the global optimization of the system through cloud collaboration. The system specifically includes a dual-modal acquisition unit, an edge diagnostic unit, a time-series correlation positioning unit, and a cloud optimization control unit.

[0032] The dual-mode acquisition unit is designed to reliably capture critical fault transient events at the cost of extremely low system power consumption. The unit consists of sensor nodes deployed on each MCB line. It continuously monitors a key physical index, namely the transient fluctuation index, and instantaneously switches between a low-power alert mode and a high-precision focusing mode based on the comparison result of the index with a dynamically adjustable preset trigger threshold. It selectively acquires compressed waveform data containing rich fault characteristics only when necessary.

[0033] The edge diagnostic unit is designed to perform real-time fault analysis and judgment locally in the distribution box where the data is generated, avoiding the high latency and high bandwidth costs associated with uploading massive amounts of raw waveform data to the cloud. This unit uses a high-performance processor built into the edge computing gateway of the distribution box and a preset diagnostic model as its carrier. It is responsible for receiving compressed waveform data collected by the dual-modal acquisition unit, using a reconstruction algorithm based on compressed sensing theory to accurately restore the compressed data into a high-fidelity fault waveform vector, and then inputting this waveform into a preset artificial intelligence diagnostic model for analysis. Finally, it generates a structured diagnostic result that includes fault type, confidence level, and event timestamp.

[0034] The temporal correlation positioning unit (TCR) goes beyond fault diagnosis of a single node. By analyzing the timing differences in the responses of multiple nodes to the same physical event, it achieves precise spatial positioning of the fault source. Implemented in an edge computing gateway, this unit utilizes the microsecond-level high-precision time synchronization maintained between all sensing nodes. When a transient event occurs, its disturbance signal propagates along the line and is captured sequentially by multiple dual-mode acquisition units. The TCR collects the high-precision event timestamps contained in the structured diagnostic results reported by these nodes, accurately calculates the time difference of arrival of the disturbance signal between different nodes, and then solves a system of equations composed of multiple hyperboloids to inversely calculate the physical location of the fault in the transient event. In practical implementation, the TCR can also incorporate the topology information of the power distribution network and combine it with a signal propagation model to correct the calculated time difference of arrival, compensating for positioning errors caused by non-ideal factors such as line branches and impedance changes, thereby improving the accuracy and robustness of the positioning.

[0035] The cloud-based optimization control unit is designed to evaluate and optimize the performance of the entire system from a global perspective, ensuring that the system can operate adaptively in its optimal state over the long term. Deployed on a cloud server, this unit serves as the central control module of the system. It is responsible for periodically aggregating structured diagnostic results and location information from one or more distribution boxes, combining them with manually verified ground data to quantify and assess the cost of false alarms and missed alarms. Based on the evaluation results, this unit executes optimization algorithms to generate optimized model parameters and optimized trigger thresholds, which are then transmitted via secure wireless channels to the edge diagnostic unit of the edge computing gateway and the dual-modal acquisition unit of the line node, thereby completing the iterative upgrade of the edge AI model and the front-end perception strategy.

[0036] This embodiment constructs a complete technical system from front-end intelligent sensing and real-time edge diagnosis to cloud-based closed-loop optimization through the collaborative work of the four major units mentioned above. Compared with existing technologies, it reduces the power consumption and communication data volume of the system during normal operation by several orders of magnitude without sacrificing diagnostic accuracy. By performing real-time diagnosis and precise positioning at the edge, the fault response time is shortened from hours to seconds. Furthermore, through continuous learning and optimization on the cloud platform, the entire system has the ability to adapt and evolve, significantly improving the safety, reliability, and operation and maintenance efficiency of the MCB power distribution system.

[0037] Example 2:

[0038] The transient fluctuation index is generated as follows: the instantaneous value at the current sampling time and the instantaneous value at the previous sampling time are obtained, and the difference between the instantaneous value at the current sampling time and the instantaneous value at the previous sampling time is divided by a fixed time interval to obtain the transient fluctuation index.

[0039] Based on the system described in Example 1, this embodiment specifically defines the generation process of the transient fluctuation index in the dual-mode acquisition unit. The transient fluctuation index is a physical index used to quantify the drastic instantaneous changes in line voltage or current waveforms in real time. Its function is to serve as the core hardware basis for triggering the data acquisition mode to switch from low-power warning mode to high-precision focusing mode. Its source is the finite difference approximation calculation based on the first derivative of the signal.

[0040] This process is implemented through an ultra-low-power analog comparator circuit built into the sensing node; low-pass filtering or moving average filtering is applied to the raw instantaneous value sequence acquired by the sensing front end to suppress high-frequency noise interference; and the instantaneous value at the current sampling moment is obtained. and the instantaneous value at the previous sampling time These two instantaneous values ​​can be the voltage or current of the line, transmitted by the sensing front end at fixed microsecond intervals. The sample is obtained by sampling; the absolute value of the difference between the instantaneous value at the current sampling moment and the instantaneous value at the previous sampling moment is divided by a fixed time interval. The transient fluctuation index is obtained. Its calculation method is represented in the mathematical model as follows:

[0041] ;

[0042] in, Transient fluctuation index, whose dimensions are voltage / time or current / time, reflects the rate of change of the signal;

[0043] The instantaneous value at the current sampling moment, which is derived from the line voltage or current value collected in real time by the sensing front end;

[0044] The instantaneous value at the previous sampling moment is read from the storage unit when calculating the current exponent;

[0045] A fixed time interval, measured in microseconds, is generated by a preset sampling clock frequency of the node.

[0046] The underlying logic of this calculation method is that when the line is running stably, the instantaneous values ​​of adjacent sampling points differ very little, and the calculated values ​​are... The value fluctuates low around zero; however, once transient events with physical precursors such as electric arcs or surges occur, the waveform will be drastically distorted, leading to... A sudden surge, thus The value increases dramatically; by adopting this calculation method based on finite difference approximation, this embodiment can directly and in real time generate a physical index that accurately reflects the dynamic changes of the waveform at the hardware level with extremely low computational complexity. Compared with the traditional frequency domain analysis method that requires complex calculations, it greatly reduces the latency and power consumption of transient event detection, and provides a reliable and efficient physical criterion for subsequent event-driven acquisition.

[0047] Example 3:

[0048] The process of selectively acquiring compressed waveform data includes: when the transient fluctuation index does not exceed the preset trigger threshold, the dual-mode acquisition unit uses an alert measurement matrix for continuous monitoring; when the transient fluctuation index exceeds the preset trigger threshold, the dual-mode acquisition unit switches to a focusing measurement matrix to acquire compressed waveform data.

[0049] Based on the system described in Example 1, this embodiment specifically defines the process of selectively acquiring compressed waveform data in the dual-modal acquisition unit; the core of this process lies in establishing a dual-modal working mechanism, which aims to dynamically balance system power consumption and data acquisition accuracy.

[0050] When the transient fluctuation index does not exceed the preset trigger threshold, i.e. When this occurs, it indicates that the line is in normal operating condition, and the dual-mode acquisition unit uses a warning measurement matrix for continuous monitoring; the warning measurement matrix... It is a specially designed measurement matrix with structured sparse characteristics, whose function is to maintain basic monitoring of the steady-state energy changes of the power frequency signal on the line with extremely low power consumption at the microampere level; specifically, this warning measurement matrix... It can be constructed by cyclically shifting a pre-defined low-pass filter kernel vector whose length is much smaller than that of the original signal, ensuring that its hardware implementation requires very few multiply-accumulate operations;

[0051] When the transient fluctuation index exceeds the preset trigger threshold, i.e. When this occurs, it indicates that the system has detected a potential transient event; the preset trigger threshold... This is a critical value used to distinguish between normal fluctuations and abnormal transients. It is derived from statistical analysis of massive amounts of historical transient event waveform data, selecting the percentile that maximizes the distinction between normal and abnormal fluctuations. It can be dynamically adjusted remotely by the cloud-based optimization control unit. During initial calibration, this threshold is specifically calculated by collecting line waveform data under at least 24 hours of normal equipment operation, calculating the corresponding transient fluctuation index sequence, and selecting the 99.9 percentile of the statistical distribution of this sequence as the initial trigger threshold. The selection of the 99.9 percentile aims to achieve a balance between high sensitivity and low false alarm rate, ensuring that the vast majority of normal operating fluctuations will not trigger high-precision acquisition, while maximizing the capture of true precursors to transient events.

[0052] In this situation, the system immediately generates an interrupt signal, triggering the dual-mode acquisition unit to switch to a focused measurement matrix to acquire compressed waveform data; the focused measurement matrix It is a measurement matrix with higher randomness, designed to satisfy the isometry constraints of compressed sensing theory. Its function is to ensure high-fidelity capture of sparse, non-stationary transient fault signals; specifically, this focused measurement matrix... It is a distribution whose elements are independently and identically distributed from the standard normal distribution. A matrix is ​​randomly selected from the data and normalized before its first use. This matrix is ​​synchronously generated and stored in the edge computing gateway and the dual-modal acquisition unit to ensure consistency during reconstruction. After switching, the unit immediately performs a high-speed compressed sampling to obtain the compressed waveform data vector at the moment of the fault. ;

[0053] This embodiment introduces a warning measurement matrix and a focused measurement matrix, and dynamically switches them based on the transient fluctuation index to construct an efficient event-driven acquisition mechanism. This mechanism enables the system to monitor with near-dormant power consumption for most of the stable operating time, and only switches to high-precision acquisition mode instantaneously within a millisecond window when an abnormal event is detected. Compared with the traditional scheme of continuous high-speed sampling, this data acquisition method greatly reduces data transmission bandwidth and node power consumption, while ensuring reliable capture of key fault information.

[0054] Example 4:

[0055] The process of reconstructing high-fidelity fault waveforms is as follows: the compressed waveform data and the focused measurement matrix are taken as input, and the high-fidelity fault waveform vector is decoded by solving the constrained L1 norm minimization problem.

[0056] This embodiment, based on the system described in Embodiment 1, specifically defines the process of reconstructing high-fidelity fault waveforms in the edge diagnostic unit. The physical basis of this process is that signals generated by transient events such as line faults are usually highly non-stationary in the time domain, but in a certain transform domain, their energy is concentrated on a few coefficients, exhibiting good sparsity. For example, in the wavelet transform domain, the energy of transient signals can usually be characterized by a few key wavelet coefficients. This characteristic is a prerequisite for accurate reconstruction using compressed sensing theory. This process aims to utilize the stronger computing power of the edge computing gateway to accurately recover the high-resolution original fault waveform from the low-dimensional compressed waveform data reported by the dual-modal acquisition unit, providing high-quality data input for subsequent AI model diagnosis.

[0057] The mathematical essence of this process originates from compressed sensing theory; specifically, it involves compressing waveform data vectors collected and reported by sensing nodes. and the focusing measurement matrix used by the node during acquisition, and As input, the problem is decoded into a high-fidelity fault waveform vector by solving a constrained L1 norm minimization problem. The mathematical model for this optimization problem is expressed as follows:

[0058] ;

[0059] in, : The high-fidelity fault waveform vector to be solved;

[0060] Signal vector The L1 norm of the algorithm is used to promote sparsity of solutions in optimization.

[0061] : Compressed waveform data, which originates from the output of the dual-mode acquisition unit;

[0062] : Focus measurement matrix, which is shared by the sensing nodes and the edge computing gateway;

[0063] The edge diagnostic unit solves this problem by running convex optimization algorithms such as basis tracing. Considering real-time requirements, this embodiment prefers computationally efficient algorithms such as iterative hard thresholding or approximate message passing. At the same time, the edge computing gateway is configured with a task priority management mechanism to ensure that reconstruction tasks are processed first when concurrent transient events occur, thus guaranteeing low latency in fault diagnosis. This embodiment adopts a reconstruction algorithm based on the L1 norm minimization problem, enabling the system to accurately recover high-fidelity original fault waveforms at the edge when only a very small amount of compressed data is collected at the front end. This solves the huge pressure on power consumption, storage, and transmission bandwidth caused by the high sampling rate at the front end in traditional solutions, making high-precision fault diagnosis possible on resource-constrained edge devices.

[0064] Example 5:

[0065] The structured diagnostic results include fault confidence, fault type, and high-precision event timestamps. The fault confidence generation process is as follows: the high-fidelity fault waveform vector is input into the preset diagnostic model, and the high-fidelity fault waveform vector is forward-propagated using the preset model parameters to obtain the fault confidence.

[0066] Based on the system described in Example 1, this embodiment specifically defines the generation process of the structured diagnostic results and their key components generated by the edge diagnostic unit. The structured diagnostic results are a standardized data format, which transforms complex waveform analysis results into machine-readable, easily transmitted, and easily processed information units. In this embodiment, the structured diagnostic results explicitly include fault confidence, fault type, and high-precision event timestamps.

[0067] The process of generating fault confidence is as follows: The high-fidelity fault waveform vector reconstructed in the previous step is... After normalization, the data is then input into a preset diagnostic model; in this embodiment, the diagnostic model is a lightweight convolutional neural network. It is pre-trained and embedded in the edge computing gateway; the network consists of several stacked one-dimensional convolutional layers, pooling layers, and fully connected layers, and its output layer uses the Softmax activation function, capable of outputting a probability distribution vector for various preset fault types; as a concrete example, this... The model may include: two one-dimensional convolutional layers with kernel sizes of 5 and 3 respectively, and ReLU activation function; a max pooling layer with stride of 2 connected after each convolutional layer; and a fully connected layer with a final output size of M;

[0068] Using preset model parameters, forward propagation calculations are performed on the high-fidelity fault waveform vector to obtain the confidence vector for each fault type; preset model parameters It is the set of all weight parameters within the CNN model, obtained through supervised learning training on a cloud platform using massive amounts of labeled, real fault waveform data; the mathematical expression for this forward propagation calculation is:

[0069] ;

[0070] in, Fault confidence vector, which is a A probability distribution vector of dimension 1. This is the preset total number of fault types, for each element. Represents the input waveform Belongs to the The confidence level for each fault type; in the structured diagnostic results, the fault type is the type corresponding to the element with the highest probability value in the vector, and the fault confidence level is the highest probability value. For example, if the output vector is [0.1, 0.8, 0.1], corresponding to the three types [normal, arc fault, surge] respectively, then in the generated structured diagnostic results, the fault type is arc fault and the fault confidence is 0.8.

[0071] CNN model functions;

[0072] The input high-fidelity fault waveform vector originates from the output of the reconstruction process described in Example 4;

[0073] The model's internal weight parameter set;

[0074] By structuring the diagnostic results, and especially by introducing the quantitative indicator of fault confidence, this embodiment greatly enhances the value of diagnostic information. It provides a more refined diagnostic conclusion with probability assessment, enabling subsequent decisions to be based on more reliable and richer information, and significantly improving the usability and intelligence level of the overall system's diagnostic results.

[0075] Example 6:

[0076] The temporal correlation positioning unit is also used to: aggregate multiple fault events located to the same line segment, and calculate the precursor risk index of the line segment by combining the fault confidence level and preset risk weight corresponding to each fault event, and upload the precursor risk index to the cloud optimization control unit.

[0077] This embodiment expands the functionality of the time-series correlation localization unit based on the system described in Embodiment 1. After locating the physical location of a single transient event, the time-series correlation localization unit is also used to aggregate multiple fault events located on the same line segment, and combine the fault confidence level corresponding to each fault event. and preset risk weights The precursor risk index of the line segment was calculated. The early warning risk index is then uploaded to the cloud-based optimization and control unit.

[0078] Early warning risk index It is used to quantify a specific line segment This comprehensive indicator of the potential risk of serious failures in the future serves to provide direct data support for predictive maintenance; its calculation formula is as follows:

[0079] in, Line Section In the statistical period The cumulative early warning risk index within the region is a dimensionless value;

[0080] The statistical period is determined by the system operation and maintenance strategy.

[0081] :cycle The internal location was determined to be a line segment. The total number of transient events;

[0082] : No. The fault confidence of this event is derived from the output of the edge diagnostic unit;

[0083] : with the The risk weights associated with the sub-event failure type are pre-set by domain experts based on power safety regulations and historical failure data analysis; for example, they can be set at... The interval is defined as follows: instantaneous overvoltage has a weight of 1, while arc faults with fire hazards have a weight of 5.

[0084] This improvement enhances the invention's capability from a passive response to single, isolated fault events to a continuous, proactive, and forward-looking assessment of line health. By calculating a precursor risk index, the system can identify high-risk line sections that frequently experience low-energy micro-discharges but have not yet developed into serious faults. This provides core data for predictive maintenance, enabling intervention before faults occur and thus avoiding potential power outages and safety incidents.

[0085] Example 7:

[0086] The process of evaluating system performance is as follows: obtain false alarm events and missed alarm events within the evaluation period, and combine them with preset false alarm cost weighting coefficients and missed alarm cost weighting coefficients to quantify the total diagnostic error cost of the system, which is used as the global loss function value for evaluation.

[0087] This embodiment, based on the system described in Embodiment 1, specifically defines the process of evaluating system performance in the cloud-based optimization control unit; this process is implemented by constructing a global loss function; false alarm events and missed alarm events within the evaluation period are obtained, combined with a preset false alarm cost weighting coefficient. and the underreporting cost weighting coefficient The total diagnostic error cost of the system is quantified. The global loss function value is used for evaluation; the false positive cost weighting coefficient and the false negative cost weighting coefficient are two strategic parameters, which are configured by the system administrator based on the security level and operational cost requirements of the application scenario; the mathematical model of this global loss function is:

[0088] ;

[0089] Of which, the total loss caused by false alarm events for Total losses caused by unreported events (i.e., the total loss due to underreporting in Example 8) is .

[0090] in, The global loss function value is a dimensionless scalar.

[0091] The total number of events during the assessment period;

[0092] and : Indicator function, if the first If the event is confirmed as a false alarm or a missed alarm, its value is 1; otherwise, it is 0.

[0093] : False alarm cost weighting coefficient;

[0094] : Weighting coefficient for underreported costs;

[0095] By defining the total diagnostic error cost, which directly correlates false positives and false negatives with actual operating costs, as a global loss function, this embodiment provides a unified and business-value-based quantitative indicator for system performance evaluation. This enables subsequent optimization processes to directly target minimizing total operating costs, thereby ensuring that the direction of system iteration remains highly consistent with the user's core business needs.

[0096] Example 8:

[0097] The process of generating the optimized trigger threshold is as follows: when the total loss caused by missed events exceeds the preset risk tolerance, the trigger threshold is lowered; when the total loss caused by false alarm events exceeds the preset operation and maintenance efficiency target, the trigger threshold is raised.

[0098] Based on the system described in Example 1, this embodiment specifically defines the process of generating the optimized trigger threshold in the cloud-based optimization control unit. This process is a key execution link for realizing the closed loop of the system's adaptive capability, and it is executed through a heuristic adjustment logic based on preset costs and goals.

[0099] When the total loss caused by missed events exceeds the preset risk tolerance, the trigger threshold is lowered. The preset risk tolerance is a management parameter set by the system administrator based on the operation and maintenance strategy of the specific application scenario. When the total loss from missed events exceeds this tolerance, the cloud optimization control unit will generate an optimized trigger threshold that is lower than the current value. And it is issued to improve the sensitivity of the dual-modal acquisition unit; this downsizing process follows the following update rules: ,in, This is the current threshold. The total loss is caused by unreported events. It is the risk tolerance threshold. It is a preset hyperparameter used to control the down-rate, for example, 0.1;

[0100] When the total loss caused by false alarms exceeds the preset operational efficiency target, the trigger threshold is increased. The preset operational efficiency target is a management parameter related to operational costs, also preset by the system administrator. When the total loss from false alarms is too high, the cloud-based optimization control unit will generate an optimized trigger threshold that is higher than the current value. And it is issued to reduce the false alarm rate; the adjustment process follows the following update rules: ,in, The total loss is caused by false alarm events. It is the target threshold for operational efficiency. It is a preset hyperparameter used to control the up-rate, for example, 0.05;

[0101] This embodiment achieves intelligent, closed-loop control of the front-end acquisition strategy by establishing a heuristic adjustment logic directly related to risk and cost. It transforms the trigger threshold from a static parameter into an adaptive parameter that dynamically optimizes based on actual operational performance. This adaptive adjustment capability ensures that the system can automatically find an optimal balance between excessively high and low sensitivity during long-term operation, thus minimizing the overall diagnostic error cost. It tends to be minimized.

[0102] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended 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.

Claims

1. A real-time data acquisition and diagnostic system for MCB distribution boxes based on edge computing, characterized in that, include: The dual-modal acquisition unit is used to monitor the transient fluctuation index of the line waveform and selectively acquire compressed waveform data based on the comparison result of the transient fluctuation index and the preset trigger threshold. The edge diagnostic unit is used to receive compressed waveform data to reconstruct high-fidelity fault waveforms and analyze the high-fidelity fault waveforms using a preset diagnostic model to generate structured diagnostic results. The temporal correlation positioning unit is used to calculate the time difference of arrival and solve the physical location of the fault based on the high-precision event timestamps contained in multiple structured diagnostic results. The cloud-based optimization control unit aggregates structured diagnostic results to evaluate system performance and generates optimized model parameters and trigger thresholds, which are then distributed to the edge diagnostic unit and the dual-modal acquisition unit, respectively. The system performance evaluation process is as follows: False alarms and false negatives within the evaluation period are acquired, and combined with preset false alarm cost weighting coefficients and false negative cost weighting coefficients, the total diagnostic error cost of the system is quantified and used as the global loss function value for evaluation. The mathematical model of this global loss function is as follows: Of which, the total loss caused by false alarm events for Total losses caused by unreported events for ;in, The global loss function value is a dimensionless scalar. The total number of events during the assessment period; and : Indicator function, if the first If the event is confirmed as a false alarm or a missed alarm, its value is 1; otherwise, it is 0. : False alarm cost weighting coefficient; The cost weighting coefficient for missed events; the process of generating the optimized trigger threshold is as follows: when the total loss caused by missed events exceeds the preset risk tolerance, the trigger threshold is lowered; when the total loss caused by false alarm events exceeds the preset operation and maintenance efficiency target, the trigger threshold is raised; this lowering process follows the following update rules: ,in, This is the current threshold. The total loss is caused by unreported events. It is the risk tolerance threshold. It is a preset hyperparameter used to control the down-rate; ,in, The total loss is caused by false alarm events. It is the target threshold for operational efficiency. It is a preset hyperparameter used to control the up-rate.

2. The MCB distribution box real-time data acquisition and diagnosis system based on edge computing according to claim 1, characterized in that, The transient fluctuation index is generated as follows: the instantaneous value at the current sampling time and the instantaneous value at the previous sampling time are obtained, and the difference between the instantaneous value at the current sampling time and the instantaneous value at the previous sampling time is divided by a fixed time interval to obtain the transient fluctuation index.

3. The MCB distribution box real-time data acquisition and diagnosis system based on edge computing according to claim 1, characterized in that, The process of selectively acquiring compressed waveform data includes: when the transient fluctuation index does not exceed the preset trigger threshold, the dual-mode acquisition unit uses an alert measurement matrix for continuous monitoring; when the transient fluctuation index exceeds the preset trigger threshold, the dual-mode acquisition unit switches to a focusing measurement matrix to acquire compressed waveform data.

4. The MCB distribution box real-time data acquisition and diagnosis system based on edge computing according to claim 1, characterized in that, The process of reconstructing high-fidelity fault waveforms is as follows: the compressed waveform data and the focused measurement matrix are taken as input, and the high-fidelity fault waveform is decoded by solving the constrained L1 norm minimization problem.

5. The MCB distribution box real-time data acquisition and diagnosis system based on edge computing according to claim 1, characterized in that, The structured diagnostic results include fault confidence, fault type, and high-precision event timestamps. The fault confidence generation process is as follows: the high-fidelity fault waveform is input into the preset diagnostic model, and the high-fidelity fault waveform is forward-propagated using the preset model parameters to obtain the fault confidence.

6. The MCB distribution box real-time data acquisition and diagnosis system based on edge computing according to claim 1, characterized in that, The temporal correlation positioning unit is also used to: aggregate multiple fault events located to the same line segment, and calculate the precursor risk index of the line segment by combining the fault confidence level and preset risk weight corresponding to each fault event, and upload the precursor risk index to the cloud optimization control unit.

Images