Industrial data acquisition system and control method for intelligent box variable environment monitoring

Abstract

This invention relates to the fields of smart power distribution networks and industrial Internet of Things (IoT) technology, specifically to an industrial data acquisition system and control method for environmental monitoring of smart transformer substations. It includes multi-source sensor data acquisition, edge adaptive collaborative filtering, and a hierarchical response control unit. The system dynamically adjusts the acquisition frequency based on the gradient of environmental feature changes and the sampling adjustment rate calculated from multi-modal sensor data. Its core is to dynamically extract features and perform on-site filtering using multi-parameter collaborative weights, intercepting redundancy and outputting key abrupt change data, thereby triggering local control with a hierarchical response. This invention abandons the fixed sampling mechanism and utilizes edge adaptive computing to solve the problem of bandwidth occupation by invalid normal data, effectively increasing sensing density at the initial stage of anomalies and suppressing redundancy in steady state, thus achieving optimized allocation of system resources and communication bandwidth.

Description

Technical Field

[0001] This invention relates to the fields of smart power distribution networks and industrial Internet of Things (IoT) technology, specifically to an industrial data acquisition system and control method for environmental monitoring of smart transformer substations. Background Technology

[0002] In intelligent transformer substation environmental monitoring environments, multi-source sensors continuously generate multimodal sensing data including temperature, humidity, partial discharge, and electrical status. To monitor and analyze this data, existing solutions generally adopt a centralized processing architecture with a fixed cycle. This involves acquiring all parameters at a static, fixed frequency through underlying acquisition nodes and transmitting them in full to the main control system for centralized anomaly calculation and control decision-making. Although this solution has a certain state awareness capability in conventional steady-state monitoring scenarios, its high dependence on uplink communication bandwidth and central computing power, coupled with the inability of fixed sampling mechanisms and equal resource allocation to match the variable rate of anomaly evolution in the field, results in the invalid accumulation of massive amounts of repetitive normal data and link congestion, making it easy for key abrupt change characteristics to be diluted. At the same time, relying solely on the main control system to issue control commands makes the closed-loop link lengthy, making it difficult to take rapid primary protection actions against short-term sudden thermal runaway or insulation degradation risks.

[0003] Therefore, how to achieve adaptive adjustment of multimodal data acquisition frequency, local filtering of redundant transmission burden, and improvement of edge-side control response agility have become urgent technical problems to be solved. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides an industrial data acquisition system and control method for intelligent transformer substation environmental monitoring. Specifically, the technical solution of this invention is as follows:

[0005] An industrial data acquisition system for intelligent transformer substation environmental monitoring includes: a multi-source sensor data acquisition unit with underlying acquisition nodes, an edge adaptive collaborative filtering unit, and a hierarchical response control unit;

[0006] The multi-source sensor data acquisition unit is used to acquire multimodal sensor data through the underlying acquisition nodes;

[0007] The edge adaptive collaborative filtering unit is used to obtain the rate of change of each parameter based on the ratio of the difference between each parameter in adjacent time steps to the time interval in the multimodal sensing data, and calculate the gradient of environmental feature change accordingly. Based on the ratio of the gradient of environmental feature change to the preset benchmark gradient, the edge adaptive sampling adjustment rate is calculated to dynamically adjust the sampling frequency.

[0008] It is also used to calculate the dynamic allocation value of multi-parameter collaborative weights based on the normalization result of the change rate of each parameter, and extract feature data accordingly.

[0009] The feature data is filtered in place to intercept redundant data and output key mutation data, and the ratio of the amount of intercepted redundant data to the total amount of data is used as the redundant data filtering ratio in place.

[0010] The graded response control unit is used to calculate the graded response time delay based on the ratio between the amplitude of the key mutation data and the preset amplitude benchmark, thereby triggering the local primary control logic and outputting the reported data to the main control system.

[0011] The triggering operation includes: when the amplitude of the key mutation data is greater than the preset amplitude abnormality threshold, the local primary control logic is triggered to generate a control command to start the intelligent transformer substation heat dissipation device or cut off a predetermined secondary electrical circuit.

[0012] When the amplitude is less than or equal to the preset amplitude abnormality threshold, a flag signal is generated to maintain the current control state.

[0013] Optionally, the multimodal sensing data includes at least one or more of the following: temperature parameters, humidity parameters, smoke parameters, partial discharge parameters, and electrical parameters.

[0014] Optionally, the operation of the edge adaptive collaborative filtering unit to calculate the dynamic allocation value of multi-parameter collaborative weights includes: based on the normalized result of the rate of change of different types of sensor data in the multimodal sensing data, and the communication priority mapping result obtained by inputting the rate of change into a preset communication priority mapping function, combined with a preset computing resource weight adjustment coefficient and a communication priority weight adjustment coefficient, to generate the dynamic allocation value of multi-parameter collaborative weights for different types of sensor data in real time; specifically, for the i-th type of sensor data, the dynamically allocation value of multi-parameter collaborative weights generated in real time... The calculation formula is:

[0015]

[0016] in, For the first Rate of change of sensor data This is a communication priority mapping function corresponding to the rate of change of parameters. This is the normalized function of the rate of change of the parameter. To calculate the resource weight adjustment coefficient, This refers to the communication priority weight adjustment coefficient; the calculation resource weight adjustment coefficient Communication priority weight adjustment coefficient These are empirical constants that are pre-calibrated and stored locally based on the sensor type corresponding to the underlying acquisition node;

[0017] The normalized function of the rate of change of the parameter The Min-Max normalization method is used to map it to the [0,1] interval; the communication priority mapping function Using the step mapping rule, when the rate of change When the rate of change exceeds a preset danger threshold, the mapping function takes a value of 2. When the value of the mapping function is less than or equal to the threshold of the rate of change of danger, the value of the mapping function is 1.

[0018] The dangerous change rate threshold is pre-calibrated based on the lower limit critical value of the change rate that triggers the alarm level in the same parameter in the historical fault samples of the target intelligent transformer or the physical tolerance limit of the equipment insulation material.

[0019] Optionally, the edge adaptive collaborative filtering unit is used to perform on-site filtering calculations on the feature data, including: applying a time series similarity association algorithm to identify invalid normal data and repetitive normal data in the feature data, and calculating the percentage of the sum of the intercepted invalid normal data and the repetitive normal data to the total amount of collected data, as the on-site filtering ratio of the redundant data.

[0020] Optionally, the operation of the edge adaptive collaborative filtering unit to calculate the edge adaptive sampling adjustment rate includes: calculating the response time and adjustment range of the sensor sampling frequency based on the ratio of the environmental feature change gradient to the preset maximum allowable gradient of the system, and generating the edge adaptive sampling adjustment rate.

[0021] Optionally, the system also includes a signal-to-noise ratio arbitration gating unit; the signal-to-noise ratio arbitration gating unit is used to calculate the variance of the key mutation data output by the edge adaptive collaborative filtering unit within a preset time window, as a systematic anomaly index;

[0022] The systematic anomaly index is compared with a preset variance anomaly threshold; when the systematic anomaly index is greater than the preset variance anomaly threshold, the execution of the hierarchical response control unit is triggered; when the systematic anomaly index is less than or equal to the preset variance anomaly threshold, the execution of the hierarchical response control unit is blocked.

[0023] Optionally, the edge adaptive collaborative filtering unit is further configured to segment the multimodal sensing data into multiple data sub-layers. The segmentation operation is performed based on a preset quantile of a specified parameter in the multimodal sensing data. The specified parameter is an assessment parameter characterizing the risk of the intelligent transformer substation's operating status.

[0024] An industrial data acquisition and control method for intelligent transformer substation environmental monitoring includes the following steps:

[0025] S1. Acquire multimodal sensing data through the bottom-level acquisition nodes of the multi-source sensing data acquisition unit;

[0026] S2. Calculate the gradient of environmental feature changes based on the ratio of the difference between each parameter in the multimodal sensing data within adjacent time steps to the time interval.

[0027] S3. Calculate the edge adaptive sampling adjustment rate based on the ratio of the environmental feature change gradient to the preset benchmark gradient, and dynamically adjust the sampling frequency based on the edge adaptive sampling adjustment rate.

[0028] S4. Based on the normalization result of the rate of change of the multimodal sensing data, calculate the dynamic allocation value of the multi-parameter collaborative weight, and extract feature data from the multimodal sensing data in a weighted manner based on the dynamic allocation value of the multi-parameter collaborative weight.

[0029] S5. The edge-adaptive collaborative filtering unit performs edge-side local filtering calculation on the feature data, obtains the ratio of the amount of intercepted redundant data to the total amount of data as the redundant data local filtering ratio, and based on the redundant data local filtering ratio, intercepts redundant data and outputs key mutation data.

[0030] S6. Based on the ratio between the magnitude of the critical mutation data and the preset amplitude benchmark, calculate the graded response time delay, and based on the graded response time delay, trigger the local primary control logic and output the reported data to the main control system; when the magnitude of the critical mutation data is greater than the preset amplitude abnormality threshold, trigger the local primary control logic to generate a control command for starting the intelligent transformer substation heat dissipation device or cutting off a predetermined secondary electrical circuit; when the magnitude of the critical mutation data is less than or equal to the preset amplitude abnormality threshold, generate a flag signal to maintain the current control state.

[0031] The present invention has the following beneficial effects:

[0032] 1. This invention calculates the gradient of environmental feature changes and the edge adaptive sampling adjustment rate through an edge adaptive collaborative filtering unit, dynamically adjusting the acquisition frequency of the multi-source sensor data acquisition unit; it extracts feature data by combining the dynamic allocation values ​​of multi-parameter collaborative weights and performs on-site filtering calculations, intercepting redundant data and outputting key mutation data; this mechanism effectively solves the problem of invalid normal data occupying bandwidth due to existing fixed sampling, can improve the sensing density in the initial stage of anomalies and suppress redundant communication in steady state, and achieves optimized allocation of system resources and communication bandwidth;

[0033] 2. This invention calculates the graded response time delay based on key mutation data through a graded response control unit. When the data amplitude exceeds a preset amplitude anomaly threshold, it directly triggers the local primary control logic to generate control commands for regulating the heat dissipation device or electrical circuit. This mechanism breaks the high dependence of existing solutions on the backbone link and centralized decision-making, moves the control closed loop forward to the edge side, avoids the dilution of key mutation data by massive background information, realizes low-latency rapid identification and local protection of transient high-risk anomalies, shortens the response cycle of control commands, and improves the accuracy of high-risk anomaly interception. Attached Figure Description

[0034] Figure 1 This is a structural diagram of the system of the present invention;

[0035] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0036] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0037] Example 1:

[0038] Please see Figure 1 An industrial data acquisition system for intelligent transformer substation environmental monitoring includes:

[0039] It features a multi-source sensor data acquisition unit with underlying acquisition nodes, an edge adaptive collaborative filtering unit, and a hierarchical response control unit;

[0040] The multi-source sensor data acquisition unit is used to acquire multimodal sensor data through the underlying acquisition nodes;

[0041] The edge adaptive collaborative filtering unit is used to obtain the rate of change of each parameter based on the ratio of the difference between each parameter in adjacent time steps to the time interval, and calculate the gradient of environmental feature change accordingly. Based on the ratio of the gradient of environmental feature change to the preset benchmark gradient, the edge adaptive sampling adjustment rate is calculated to dynamically adjust the sampling frequency.

[0042] It is also used to calculate the dynamic allocation value of multi-parameter collaborative weights based on the normalization results of the change rate of each parameter, and to extract feature data in a weighted manner accordingly;

[0043] Feature data is filtered in place to intercept redundant data and output key mutation data. The ratio of the amount of intercepted redundant data to the total amount of data is used as the in-place filtering ratio of redundant data.

[0044] The graded response control unit is used to calculate the graded response time delay based on the ratio between the amplitude of key mutation data and the preset amplitude benchmark, thereby triggering the local primary control logic and outputting the reported data to the main control system.

[0045] The triggering operation includes: when the amplitude of the key mutation data exceeds the preset amplitude abnormality threshold, the local primary control logic is triggered to generate a control command to start the intelligent transformer substation heat dissipation device or cut off the predetermined secondary electrical circuit;

[0046] When the amplitude is less than or equal to the preset amplitude abnormality threshold, a flag signal is generated to maintain the current control state;

[0047] This embodiment provides a mechanism for an industrial data acquisition system for environmental monitoring of intelligent prefabricated substations. Specifically, the system is deployed in an outdoor intelligent prefabricated substation in an industrial park. The interior of the prefabricated substation is configured with a temperature rise sensitive area, a busbar connection area, a cable terminal area, and a ventilation and heat exchange area. Each area is equipped with a bottom-level acquisition node.

[0048] The bottom-level acquisition nodes are connected to the edge controller via a wired bus or short-range industrial wireless link. The edge controller integrates data caching, local analysis and control output interfaces. The main control system is set up on the station-level monitoring platform or the power distribution automation master station.

[0049] The multi-source sensor data acquisition unit is responsible for continuously acquiring environmental and electrical status information from different physical locations inside the transformer substation. Multimodal does not only refer to different data formats, but also reflects different physical mechanisms: temperature reflects the balance between conductor heating and heat dissipation, humidity reflects the risk of condensation on the insulation surface, smoke reflects overheating carbonization or arc products, partial discharge reflects the development of insulation defects, and electrical parameters reflect load, voltage quality and circuit stress.

[0050] After receiving this data, the edge adaptive collaborative filtering unit does not send it all up at a fixed pace, but first calculates the rate of change within adjacent time steps.

[0051] If the rate of change of a certain type of parameter exceeds the first preset threshold within a preset time window, the first preset threshold is obtained by extracting the 90th to 95th percentile range of the rate of change of the same type of parameter during the most recent year's historical fault-free steady-state operation of the smart transformer, then it indicates that the field condition is deviating from the steady state, such as increased contact resistance causing local hot spots, or the start of active partial discharge after being damp.

[0052] At this point, the system maps this change to a higher gradient of environmental feature changes; specifically, the gradient of environmental feature changes. The calculation formula is:

[0053]

[0054] in, The total number of parameter types in the acquired multimodal sensing data. For the first The rate of change of sensor data within adjacent time steps Configure static weight values ​​for the corresponding sensor type, and satisfy the following conditions: By using the vector L2 norm calculation, the rate of change of multi-source heterogeneity is fused into a single environmental feature change gradient index.

[0055] The logic for setting the preset benchmark gradient is as follows: extract the 95th percentile of the gradient of the changes in various environmental and electrical parameters of the target intelligent transformer during the recent month of fault-free steady-state operation as the preset benchmark gradient corresponding to that parameter.

[0056] The system internally sets a preset benchmark gradient to characterize fluctuations in the normal state. By calculating the ratio of the gradient of environmental feature changes to the preset benchmark gradient, an edge adaptive sampling adjustment rate that can quantify the severity of deviation is obtained. Specifically, the edge adaptive sampling adjustment rate... The calculation formula is:

[0057]

[0058] in, The gradient represents the current environmental characteristics change. As a preset baseline gradient, The system's preset scaling factor; the system's preset scaling factor The range of values ​​is limited to The specific value is dynamically calibrated by forward mapping between the upper limit of the hardware processing capability of the underlying acquisition node and the current communication bandwidth margin of the backbone link, so as to ensure that the adjusted sampling frequency does not exceed the physical upper limit of the system.

[0059] Based on this edge adaptive sampling rate, the sampling frequency is dynamically adjusted accordingly, for example, by increasing the sampling frequency, so that the bottom node can capture continuous details in the early stage of anomaly formation; if the parameters are stable for a long time and the ratio of the environmental feature change gradient to the preset benchmark gradient is low, the sampling frequency is reduced to reduce the invalid data burden on the edge end and the main control end.

[0060] The edge-adaptive collaborative filtering unit does not assign weights equally to all parameters, but rather generates dynamic multi-parameter collaborative weight allocation values ​​based on the relative activity of each parameter's rate of change. An example analysis process is provided below:

[0061] Suppose that within a certain time period, there exists a temperature data set T, a humidity data set H, a partial discharge data set P, and a current data set I. Among these, T and P show a continuous increase, the rate of change of H is lower than a preset normal fluctuation threshold, and I only experiences normal load fluctuations. In this case, the system will allocate more computing resources and communication priority to T and P, and the extracted feature data will be mainly based on T and P, while H and I will only retain background reference information. The feature data formed in this way is closer to the anomaly mechanism itself and can reflect the real evolution chain of heating—insulation degradation—discharge enhancement.

[0062] After obtaining the feature data, the system performs on-site filtering at the edge. Redundant data mainly includes two types: one is invalid normal data that maintains a steady state for multiple consecutive cycles and does not show any risk indication; the other is repetitive normal data that is highly repeated in a short period of time and does not change the fault judgment result. The edge intercepts these two types of data and only retains key mutation data.

[0063] Key mutation data typically manifest as a sudden increase in temperature slope, a sharp increase in discharge pulse activity, the initial appearance of abnormal smoke signals, or a temporal coupling response between multiple parameters; the graded response control unit determines the response time delay based on the amplitude level of the key mutation data; the graded response time delay is negatively correlated with the amplitude of the key mutation data;

[0064] The specific calculation logic is as follows: the graded response time delay is equal to the preset maximum delay constant divided by the ratio of the critical mutation data amplitude to the preset amplitude benchmark; when the amplitude of the critical mutation data is within the preset deviation range, the system allocates a preset basic observation time window to avoid excessive action on short-term disturbances;

[0065] Among them, the preset amplitude benchmark is obtained by calibrating the average value of the sudden change amplitude of similar sensor data during the historical steady-state operation of the target smart transformer; the preset amplitude anomaly threshold is a safety limit value pre-set based on the physical tolerance limit of the monitored smart transformer equipment or the limit critical value of historical fault samples.

[0066] If the critical mutation data exceeds the preset amplitude abnormality threshold, the local primary control logic immediately generates control commands, such as starting the fan, opening the forced cooling channel, limiting the load of non-critical circuits, or switching the state of electrical circuits; if the threshold is not exceeded, a flag signal to maintain the current control state is output, and the event is processed as a record.

[0067] As a boundary processing mechanism, various abnormal situations may occur during field operation. If individual bottom-level acquisition nodes temporarily lose connection, the edge controller can call the historical stable value of the adjacent area as a short-term placeholder and attach the missing measurement mark to the data frame to prevent the communication failure from being mistakenly identified as an environmental anomaly. If multiple parameters change abruptly at the same time but the time synchronization is obviously distorted, it is first determined to be link jitter or sampling clock offset. Time alignment is performed first and then collaborative judgment is entered.

[0068] If a parameter shows a single spike and lacks the support of other parameters, the spike is only temporarily stored as a candidate anomaly and is not directly triggered to control, so as to avoid malfunctions caused by sensor transient pulse interference. If the edge buffer is close to the limit, key mutation data and control-related logs are retained first, and only sampled summaries of background normal data are retained.

[0069] During the peak summer load period at the same industrial park's transformer substation, the temperature in the busbar connection area first showed a slow increase, and the partial discharge sensor showed increased activity over several monitoring cycles. At the same time, the humidity in the ventilation area was also high. The edge end determined that this was not simply due to increased load, because a normal increase in load usually manifests as a synchronous change in current and temperature. However, this was accompanied by signs of discharge and a damp background. Therefore, the sampling frequency in the connection area and cable terminal area was increased, while the upload ratio in the steady-state temperature and humidity area was reduced.

[0070] When the critical mutation data continues to amplify and exceeds the abnormal threshold, the local heat dissipation device is immediately activated and an event packet is reported to the main control system; if the temperature rise slows down and the discharge activity decreases after subsequent heat dissipation, the system returns to a lower sampling frequency and maintains the current loop state.

[0071] The purpose of this step is to move the acquisition-analysis-filtering-response closed loop forward to the edge, so that the system can improve the sensing density in the initial stage of anomalies and suppress redundant communication in the steady-state operation stage, thereby achieving low-latency identification and deterministic control of environmental risks in smart transformer substations.

[0072] Multimodal sensing data includes at least one or more of the following parameters: temperature, humidity, smoke, partial discharge, and electrical parameters.

[0073] This embodiment provides a configuration mechanism for multimodal sensing data; specifically, in the aforementioned scenario of intelligent transformer substations in industrial parks, the bottom-level acquisition nodes are configured with at least one or more of the following parameters: temperature, humidity, smoke, partial discharge, and electrical parameters, to cover the physical signs of common failure modes inside transformer substations.

[0074] Temperature parameters are usually obtained by contact thermistors or infrared temperature measurement components, mainly reflecting problems such as conductor heating, increased contact resistance at the junctions, and blockage of heat dissipation paths; humidity parameters are used to reflect the tendency of condensation inside the enclosure and the risk of moisture on the insulation surface. Especially in environments where the day-night temperature difference exceeds the set temperature difference threshold and the ventilation volume is lower than the set ventilation standard, this parameter has a leading significance for insulation degradation.

[0075] Smoke parameters are used to identify carbonization of insulation materials, overheating and volatilization of the coating layer, and micro-arc byproducts. These often appear later than the temperature rise, but once they appear, it usually indicates that the fault has entered a more dangerous stage. Partial discharge parameters are an important basis for the early identification of insulation defects and can be collected by ultrasound, ultra-high frequency or pulsed current.

[0076] Electrical parameters include, but are not limited to, current, voltage, power, power factor, and circuit switch status, used to distinguish between normal thermal changes caused by load variations and abnormal thermal changes caused by device defects;

[0077] If only a single parameter is used, such as monitoring only temperature, normal temperature rise may be mistaken for abnormality during high-load seasons; if only partial discharge is monitored, pure thermal faults caused by ventilation failure may be missed.

[0078] Therefore, multimodal combinations can form cross-validation in terms of physical mechanisms. Taking an exemplary analysis process as an example: assuming the temperature data set is T, the humidity data set is H, the smoke data set is S, the partial discharge data set is P, and the electrical parameter data set is E; when T increases and E increases synchronously, P remains stable, and S does not change, the system is more likely to identify it as an interpretable temperature rise caused by the load; when T increases and P increases and H is high, it is more likely to correspond to the development of defects after the insulation is damp.

[0079] As a boundary processing mechanism, in actual deployment, adaptive configuration is allowed based on the site's computing and storage resources, space constraints, or risk level. If a box becomes a low-risk, low-capacity scenario, temperature and electrical parameters can be deployed as a basic combination. If it is located in a high-humidity coastal or chemical industrial park environment, humidity and smoke parameters should be added first. If it is a critical power supply node, a partial discharge channel should be deployed at the same time.

[0080] For sites that do not yet have a certain type of sensor, the system can still operate, but the edge will reduce the confidence level of the judgment of the relevant fault mode and indicate the monitoring blind zone in the report to avoid the main control system from mistakenly believing that full sensing has been covered; for example, in the aforementioned box substation, temperature and partial discharge nodes are configured in the cable terminal area, temperature and current nodes are configured in the busbar area, and humidity and smoke nodes are configured on the top of the box.

[0081] After entering the rainy season, the humidity at the top remained high for a long time. The activity of partial discharge in the cable terminal area changed before the smoke parameters, while the current in the busbar area did not increase abnormally. Based on this, the system judged that the risk was mainly in the damp insulation area rather than the load side, and subsequently increased the sampling and response level for this area.

[0082] The purpose of this step is to form a joint perception of heat, humidity, discharge and electrical stress by using a multimodal parameter layout that matches the fault mechanism, thereby enabling the distinguishable monitoring of different types of transformer substation anomalies.

[0083] The operation of the edge adaptive collaborative filtering unit to calculate the dynamic allocation value of multi-parameter collaborative weights includes: normalizing the rate of change of different types of sensor data in multimodal sensing data, and obtaining the communication priority mapping result after inputting the rate of change into a preset communication priority mapping function, and combining the preset computing resource weight adjustment coefficient and communication priority weight adjustment coefficient to generate the dynamic allocation value of multi-parameter collaborative weights for different types of sensor data in real time; specifically, for the i-th type of sensor data, the real-time generated dynamic allocation value of multi-parameter collaborative weights... The calculation formula is:

[0084]

[0085] in, For the first Rate of change of sensor data This is a communication priority mapping function corresponding to the rate of change of parameters. This is the normalized function of the rate of change of the parameter. To calculate the resource weight adjustment coefficient, Calculate the communication priority weight adjustment coefficient; calculate the resource weight adjustment coefficient. Communication priority weight adjustment coefficient These are empirical constants that are pre-calibrated and stored locally based on the sensor type corresponding to the underlying acquisition node;

[0086] Normalization function of the rate of change of parameters The Min-Max normalization method is used to map it to the [0,1] interval; the communication priority mapping function... Using the step mapping rule, when the rate of change When the rate of change exceeds the preset danger threshold, the mapping function takes a value of 2. When the value of the mapping function is less than or equal to the threshold of the rate of change of danger, the value of the mapping function is 1.

[0087] The dangerous change rate threshold is pre-calibrated based on the lower limit critical value of the change rate that triggers the alarm level in the same parameter in the historical fault samples of the target intelligent transformer or the physical tolerance limit of the equipment insulation material;

[0088] This embodiment provides a multi-parameter collaborative weight dynamic allocation mechanism; specifically, based on the aforementioned system, if all sensor data are processed on an average basis, during abnormal concentrated outbreaks, important information and background information may compete for resources in the same processing channel, resulting in the dilution of truly critical discharge, smoke, or hot spot evolution information.

[0089] Therefore, this embodiment introduces a joint allocation method of computing resource weight coefficient and communication priority weight coefficient to generate dynamic allocation values ​​of multi-parameter collaborative weights. The computing resource weight coefficient reflects the processing resource quota allocated by the edge controller for a certain type of parameter in a certain period of time, such as cache ratio, processing cycle and feature extraction frequency. The communication priority weight coefficient reflects the order in which this type of parameter is preferentially sent under limited bandwidth.

[0090] The multi-parameter collaborative weight dynamic allocation value formed by multiplying the two is not an abstract statistical quantity, but an engineering quantitative expression of the importance of the parameter in fault identification. Taking a simplified analysis model as an example: in a certain period of time, the rate of change of temperature data group T is higher than the preset benchmark value, the rate of change of partial discharge data group P is higher than the rate of change of temperature data group T, the humidity data group H is stable, and the electrical data group E is within the normal fluctuation range. Then, at the edge, a resource allocation order can be formed with P being the highest, T the second highest, E the next highest, and H the lowest.

[0091] Thus, under the same processing resource conditions, P and T will be extracted and uploaded more frequently, while H will mainly be used as a background parameter to retain the summary; this allocation is consistent with the evolution mechanism of transformer substation anomalies; for insulation aging type faults, partial discharge is often a sensitive sign that appears earlier than smoke; for contact loose faults, the relationship between temperature and current is more valuable for explanation; for surface discharge caused by moisture condensation, humidity parameters may be given higher weight in the early stage.

[0092] Therefore, the dynamic allocation value of the multi-parameter collaborative weight is not a fixed configuration, but rather it dynamically shifts according to the relative activity of the rate of change. If the rate of change of a certain type of sensor data is higher than the set abnormal upper limit, but the sensor has long-term noise exceeding the preset signal-to-noise ratio threshold or calibration failure, then its dynamic allocation value of the multi-parameter collaborative weight will not be directly amplified to the highest value, but will first be constrained by the health status threshold to avoid abnormal sensors occupying resources for a long time.

[0093] If the rate of change of all parameters is close to steady state, the system will revert to the basic allocation template to ensure that all channels maintain the minimum necessary monitoring; if the communication link is temporarily congested, the original fragments of the data corresponding to the dynamic allocation value of high multi-parameter collaborative weight will be retained first, and only the statistical summary or status label of the data corresponding to the dynamic allocation value of low multi-parameter collaborative weight will be uploaded.

[0094] During the evening peak operation of the aforementioned transformer substation, partial discharge activity increased in the cable terminal area, and the temperature rise in the busbar area began to rise, while the humidity on the top of the box remained high but did not continue to increase. At this time, the edge end allocated more processing cycles to partial discharge and temperature data, using them as the main channel for fault tracing; humidity data was retained as background conditions to interpret the insulation environment; and only the load profile was retained in the ordinary current waveform.

[0095] In this way, even if the main control link is congested during peak hours, it can ensure that the feature data packets corresponding to the highest communication priority are transmitted first. The purpose of this mechanism is to concentrate limited edge computing power and bandwidth on the parameters that best characterize the current risk evolution, thereby achieving a simultaneous improvement in anomaly identification efficiency and communication efficiency.

[0096] The edge adaptive collaborative filtering unit is used to perform on-site filtering calculations on feature data, including: applying a time series similarity association algorithm to identify invalid normal data and repetitive normal data in the feature data, and calculating the percentage of the sum of the intercepted invalid normal data and repetitive normal data to the total amount of collected data, which is used as the on-site filtering ratio of redundant data.

[0097] Specifically, the operation of identification using time series similarity association algorithms includes: calculating the similarity of changes between time series segments of different parameters in multimodal sensing data based on Pearson correlation coefficient or dynamic time warping algorithm; when the similarity of changes of multiple parameters within a preset time window is higher than a set association threshold, and the current amplitude of each parameter is within a preset safe steady-state range, the range of the set association threshold is set as follows: to ;

[0098] If the preset safe steady-state range is within the envelope range of 10% above or below the factory rated operating parameter tolerance of this type of equipment, then the characteristic data is determined to be invalid normal data or repetitive normal data.

[0099] When the similarity of the change is less than or equal to the set correlation threshold, or when the current amplitude of any parameter exceeds the preset safe steady-state range, the feature data is determined to be non-redundant data containing newly added information and is retained.

[0100] This embodiment provides an on-site filtering mechanism for redundant data. Specifically, in the aforementioned system, even if weight allocation has been completed, if the edge end still sends up a large amount of normal data unchanged, the main control system will still accumulate a massive amount of low-value records during long-term operation. Therefore, this embodiment further introduces the identification and interception of invalid normal data and duplicate normal data.

[0101] Invalid normal data usually refers to data that does not change the state conclusion under the current fault judgment framework, such as continuous temperature and humidity readings in a stable environment for a long time; repeating normal data refers to data in which multiple sampling results are approximately consistent within a short period of time and the trend of change has not changed.

[0102] The purpose of time series similarity association algorithms is not to perform conventional user recommendation calculations, but to use the linkage relationship between multiple parameters to determine whether a batch of data still has new information. Taking an exemplary analysis process as an example: Suppose that the consecutive feature segments are F1, F2, F3, and F4, where F1 and F2 both show stable temperature, stable humidity, no smoke, and no discharge, F3 still maintains the same relationship, and F4 begins to show local discharge enhancement. Then most of the content in F1 to F3 can be marked as normal background, and only the sampled summary is retained, while F4 is retained as key mutation data.

[0103] The ratio of the amount of intercepted data to the total amount of collected data can be used to form the local filtering ratio of redundant data, which is used to characterize the filtering strength at the edge. Setting this filtering ratio has engineering significance. If the local filtering ratio of redundant data is lower than the set lower threshold, it indicates that the edge is not intercepting normal data enough and the communication resource consumption is still high. If the filtering ratio is higher than the set upper threshold, it may over-compress background information, which is not conducive to the main control end understanding the abnormal context.

[0104] Therefore, the system can set the filtering target range according to the site risk level. High-risk nodes should retain more context, while ordinary nodes can more actively compress repetitive normal data. As a boundary processing mechanism, if some normal data changes little but is in the transition phase before and after an abnormal event, it should not be completely deleted, but should be retained in the form of window summary for subsequent tracing of the cause and effect of the abnormality.

[0105] If the collaborative relationship assessment fails, for example, due to abnormal time alignment of multimodal data or loss of connection of some sensors, the system will pause deep filtering and only perform basic deduplication to avoid accidentally deleting key evidence. If a certain anomaly does not return to a steady state for a long time after it occurs, the corresponding data will not be regarded as normal duplicate data, but will continue to be sent as continuous risk data.

[0106] During the daily operation of the aforementioned transformer substation, the temperature, current, and humidity in the busbar area maintain an interpretable and stable relationship during most daytime and nighttime periods. The edge end retains only hourly profiles and extreme value summaries of these periodic repetitive data.

[0107] As evening approached, partial discharge in the cable terminal area began to intensify, and the temperature rise also accelerated. The previously compressed transmission strategy immediately switched to retaining the original key segments, thus allowing the main control system to see the complete process of the anomaly initially emerging from a stable background.

[0108] The purpose of this mechanism is to eliminate data burdens that do not contribute to state judgment at the edge as much as possible, while retaining mutation information that can truly change the fault conclusion, thereby achieving a balance between optimized allocation of communication resources and fault traceability.

[0109] The operation of the edge adaptive collaborative filtering unit to calculate the edge adaptive sampling adjustment rate includes: calculating the response time and adjustment range of the sensor sampling frequency based on the ratio of the gradient of environmental feature change to the preset maximum allowable gradient of the system, and generating the edge adaptive sampling adjustment rate;

[0110] The preset maximum allowable gradient of the system is a fixed extreme value that is pre-calibrated and stored in the local database of the edge controller, based on the thermal stability limit parameters of the monitored electrical equipment inside the smart transformer, the insulation material tolerance test data, or the limit evolution rate of historical fault samples.

[0111] This embodiment provides an edge adaptive sampling adjustment mechanism. Specifically, in the aforementioned system, if only the rate of change of a certain parameter is obtained, but the increase in the sampling frequency and the response speed are not clearly defined, two problems are likely to occur: first, the response delay is greater than the anomaly evolution time, missing early details of the anomaly; second, the sampling frequency is too high, exceeding the computing power or storage limit of the edge controller. Therefore, this embodiment determines the response time and adjustment range of the sampling frequency by using the relationship between the gradient of environmental feature changes and the maximum allowable gradient of the system.

[0112] The gradient of environmental characteristic changes reflects the urgency of the current state change, while the maximum allowable gradient of the system can be understood as the upper bound of the highest acceptable risk evolution rate on site. When the difference between the two is greater than the preset difference threshold, it indicates that the risk change is still within a controllable range, and the sampling increase should be restrained. When the two are close, it indicates that the failure evolution rate has approached the upper limit of the system tolerance, and the sampling should be increased rapidly and the adjustment range expanded.

[0113] Taking a simplified analysis model as an example: if the temperature segment T1 to T2 changes slightly during a certain period, while the partial discharge segment P1 to P2 changes significantly, then the sampling increase for channel P is greater than that for channel T; if T also starts to rise rapidly afterward, then T and P enter the high-frequency sampling region together; this process does not require complex redundant calculations, and its core is to match the sampling frequency with the rate of change of the fault mechanism.

[0114] This design conforms to the physical laws of the transformer substation site; thermal faults caused by poor contact usually go through a process of slow temperature rise - local aggravation - increased risk of thermal runaway; insulation defects may manifest as enhanced discharge activity first, followed by temperature rise and smoke appearance; if sampling is always fixed at low frequency, the system can only obtain low-resolution data change trends; if high frequency is always maintained, the edge end will process a large amount of meaningless steady-state data for a long time; the dynamic adjustment mechanism makes the sampling frequency match the evolution rate of the current environmental characteristic change gradient.

[0115] If the gradient change exceeds the upper limit allowed by the system, it indicates that the site may have entered a stage of rapid deterioration. At this time, we no longer rely solely on gradual frequency modulation, but directly enter the emergency sampling mode to capture site details at the highest allowed frequency and shorten the local response delay.

[0116] If the gradient of change is too low or close to zero, the sampling will drop back to the basic monitoring frequency, but the minimum keep-alive sampling will still be maintained to prevent the device from losing its sensing capability due to complete reduction in sampling. If a certain type of sensor has a physical sampling limit, the edge end will maintain high priority uploading after reaching the limit and will no longer continue to increase the frequency.

[0117] In the aforementioned transformer substation, only a slow increase in humidity at the top was detected in the initial stage. At this time, the system adjusted the sampling frequency according to the first preset frequency modulation step size. The partial discharge activity in the cable terminal area increased significantly in a short period of time, and the system quickly increased the sampling frequency of the discharge channel in this area. After that, the temperature in the connection area also began to rise rapidly, and the temperature channel entered rapid sampling simultaneously.

[0118] Thus, the edge end can first capture the precursors of discharge and then record the thermal expansion process to form a continuous chain of evidence. The purpose of this mechanism is to make the sampling rhythm consistent with the speed of on-site risk evolution, so as to achieve fine-grained capture of key stages and avoid ineffective occupation of acquisition and processing resources during long-term steady state.

[0119] The system also includes a signal-to-noise ratio arbitration gating unit; the signal-to-noise ratio arbitration gating unit is used to calculate the variance of the key mutation data output by the edge adaptive collaborative filtering unit within a preset time window, as a systematic anomaly index;

[0120] The systemic anomaly index is compared with a preset variance threshold; when the systemic anomaly index is greater than the preset variance threshold, the execution of the hierarchical response control unit is triggered; when the systemic anomaly index is less than or equal to the preset variance threshold, the execution of the hierarchical response control unit is blocked.

[0121] This embodiment provides a signal-to-noise ratio arbitration gating mechanism. Specifically, in the aforementioned system, local actions can be triggered based on a single key mutation data. Although the response is fast, false actions may still occur under conditions of strong electromagnetic interference, sensor transient pulse interference, or instantaneous communication jitter. Therefore, this embodiment adds a signal-to-noise ratio arbitration gating unit to perform a gating judgment on whether the key mutation data has systematic abnormal significance.

[0122] The dispersion of key mutation data within a preset time window can be used to distinguish between persistent anomalies and sporadic spikes. If multiple related segments fluctuate around the same risk direction within a short window, it indicates that the anomaly is coherent, and its variance characteristics and distribution pattern are more likely to correspond to real fault activities. If only a single isolated spike appears and the remaining segments quickly return to the stable area, it is more likely to be noise interference.

[0123] Taking a simplified analysis model as an example: Suppose that four key segments K1, K2, K3, and K4 are obtained within the window, where K1, K2, and K3 all point to the simultaneous enhancement of partial discharge and temperature rise, and K4 maintains the same trend. Then the gating unit considers there to be a systematic anomaly and allows subsequent graded response to be executed. If only K2 is a single-point spike, while K1, K3, and K4 all return to stability, then the gating unit blocks execution and only retains the event record.

[0124] The background for introducing this gating is that the transformer substation is often in a high electromagnetic and heavy load switching environment. Partial discharge sensors, smoke sensors and even wireless links may be subject to short-term disturbances. If the edge terminal takes immediate action on all sudden changes, it may cause frequent start-stop of the wind turbine and accidental disconnection of non-critical circuits, which will affect the stability of power supply. By sending critical sudden changes to the gate control unit first, suspected abnormalities can be distinguished from executable abnormalities.

[0125] If the amount of data in the window is insufficient, for example, if some data acquisition nodes have just returned to the online state, the gating unit can adopt a conservative strategy: for general risks, the execution can be blocked first and supplementary data acquisition can continue, while for high-risk risks, manual review can be allowed and reported first; if there is a significant simultaneous change in multiple sensors, even if the window is short, the fast release channel can be used to avoid delaying the protection opportunity due to waiting for the complete window.

[0126] If the gate control unit detects a significant increase in the input noise background, it will automatically raise the access threshold and simultaneously report a noise environment deterioration flag. In the aforementioned transformer substation, thunderstorms caused a complex external electromagnetic environment, and the partial discharge sensor showed an isolated spike. However, temperature, humidity, smoke, and electrical parameters did not react in unison. Based on this, the gate control unit blocked the graded response and only recorded the log.

[0127] Several hours later, multiple consecutive discharge pulse enhancement segments reappeared in the cable terminal area, and the temperature in the connection area rose synchronously. The anomaly within the window was continuous. At this time, the gating unit released the access, and the local control logic was executed. The purpose of this mechanism is to maintain a fast response capability while suppressing false triggering caused by noise or glitches, thereby improving the stability of edge control.

[0128] The edge adaptive collaborative filtering unit is also used to segment multimodal sensing data into multiple data sub-layers. The segmentation operation is based on the preset quantile of a specified parameter in the multimodal sensing data. The specified parameter is an assessment parameter that characterizes the risk of the smart transformer substation's operating status.

[0129] For example, the preset quantile can be selected from the upper quartile and median within the statistical period. The edge adaptive collaborative filtering unit divides the data into three data sub-layers accordingly: values ​​above the upper quartile are classified into the high-risk state sub-layer, values ​​between the median and the upper quartile are classified into the attention state sub-layer, and values ​​below or equal to the median are classified into the normal background sub-layer.

[0130] This embodiment provides a data sub-layer segmentation mechanism based on quantiles. Specifically, in the aforementioned system, if all multimodal data are mixed and processed at the same level, a small amount of high-risk data is easily covered or ignored by a large amount of ordinary data. Especially during long-term operation, it is difficult for the edge end to take differentiated processing for different risk densities. Therefore, this embodiment introduces a method of data layering based on preset quantiles according to specified parameters.

[0131] The specified parameters can be temperature, partial discharge activity, humidity, smoke concentration, or a normalized comprehensive risk indicator; the significance of segmenting by quantiles is that it no longer relies on fixed absolute values ​​to divide the data, but forms a relative hierarchy based on the site's own operational distribution.

[0132] Taking a simplified analysis model as an example: the activity of partial discharge in a certain stage is divided into sub-layers L1, L2, and L3 according to low, medium, and high stages. L1 carries the normal background, L2 carries the state of concern, and L3 carries the high-risk state. Different processing strategies can be configured for different sub-layers at the edge. For example, L1 mainly stores the summary, L2 retains the trend characteristics, and L3 retains the original key fragments and sends them up first.

[0133] The reason for using quantiles is that the same absolute value may not represent the same risk meaning under different transformer substations, different seasons, and different load structures. For example, the temperature background of high-load sites in summer is generally higher than that of light-load sites in winter. If a uniform fixed quantile threshold is used, it is easy for high-load sites to remain in the pseudo-anomaly level for a long time.

[0134] Quantile-based sub-layering can better adapt to the operational distribution of individual sites, highlighting the parts that are truly off-ground in a relative sense. If the sample size of the specified parameters is insufficient, quantile stratification may be unstable. In this case, the edge end will first fall back to the preset static level template, and then switch to dynamic quantile stratification after accumulating enough operational data.

[0135] If the overall background shifts during a certain period, such as due to seasonal changes or changes in ventilation conditions after equipment maintenance, the system can recalculate the layer boundaries to avoid long-term distortion of the old boundaries. If there are obvious conflicts between multiple parameters, such as the temperature being at a low level while partial discharge is at a high level, the sub-layer containing the high-risk parameter will be the dominant processing level.

[0136] During the rainy season at the aforementioned transformer substation, the edge end is stratified using partial discharge activity as the specified parameter; most daily data falls in the low layer and is only summarized; a few humid periods enter the middle layer, and the system strengthens trend tracking; when the data from the cable terminal area enters the high layer one evening, the system immediately elevates the relevant temperature, humidity and electrical segments to the high-risk processing channel, thereby preventing discharge precursors from being masked by massive background data.

[0137] The purpose of this mechanism is to organize information hierarchically according to the data distribution of each site, thereby enabling differentiated processing and resource focus on data at different risk levels.

[0138] This embodiment provides an execution mechanism for local primary control logic. Specifically, in the aforementioned system, simply identifying and reporting anomalies is insufficient to address the rapid risks at the transformer substation site, especially since thermal runaway or insulation breakdown precursors may amplify rapidly in a short period of time. Therefore, this embodiment clarifies that the local primary control logic includes at least starting the heat dissipation device or cutting off secondary electrical circuits, and simultaneously reporting key mutation data to the main control system.

[0139] Activating the heat dissipation device is suitable for risk scenarios where temperature rise accumulation is the primary factor, such as contact overheating, air duct blockage, and excessive ambient heat load. Its physical logic is to prioritize restoring thermal balance and reduce the rate of continued degradation of conductors and insulation materials. Disconnecting secondary electrical circuits is suitable for scenarios where obvious electrical abnormalities have appeared, but it is still desirable to retain core power supply functions as much as possible.

[0140] The secondary circuits here can be predefined according to the power supply level of the site, such as auxiliary lighting, non-critical backup circuits, or load circuits that can be temporarily transferred; by disconnecting the secondary circuits first, the electrical stress in the fault area can be quickly reduced, giving the main control system time for subsequent higher-level decisions.

[0141] Uploading key mutation data while controlling the output ensures that the main control system does not only receive a result of an action already taken, but also obtains the chain of evidence that triggered the action, such as temperature rise trend, partial discharge enhancement segment, and smoke initial appearance sequence. In this way, the main control system can subsequently determine whether it is necessary to issue further maintenance work orders, adjust the load distribution between stations, or perform remote isolation.

[0142] If the cooling device start command has been issued but the feedback shows that the execution failed, such as the fan is stuck or the power supply is abnormal, the system will immediately escalate the reporting level and shift the control focus to load voltage drop or secondary circuit disconnection; if the abnormality continues to expand after the secondary electrical circuit is disconnected, the local system will only maintain the safety actions that have been executed and will not continue to expand the disconnection scope without authorization, but will wait for the main control system or protection device to take over, so as to prevent the overstepping of control from affecting the continuity of power supply.

[0143] If the key mutation data is insufficient to distinguish between thermal and electrical faults, then the heat dissipation action that has less impact on power supply and is reversible should be selected first. In the aforementioned box-type substation, the system first detected that the temperature in the busbar connection area was continuously rising and the partial discharge was not significant. Therefore, the local cooling fan was started first and the temperature rise key segment was sent up.

[0144] If partial discharge intensifies and smoke appears again in the cable terminal area, it indicates that the risk is no longer limited to insufficient heat dissipation. The local authorities further cut off the predefined secondary branches and transmit all key mutation data and executed actions to the main control system for higher-level handling.

[0145] The purpose of this mechanism is to allow the edge of the field to take controllable, limited, and risk-matched primary protective actions before the main control system completes centralized decision-making, thereby achieving risk expansion suppression and upper-level response coordination.

[0146] Example 2:

[0147] Please see Figure 2 An industrial data acquisition and control method for environmental monitoring of intelligent prefabricated substations includes the following steps:

[0148] S1. Acquire multimodal sensing data through the bottom-level acquisition nodes of the multi-source sensing data acquisition unit;

[0149] S2. Calculate the gradient of environmental feature changes based on the ratio of the difference between each parameter in adjacent time steps to the time interval in the multimodal sensing data.

[0150] S3. Calculate the edge adaptive sampling adjustment rate based on the ratio of the gradient of environmental feature changes to the preset benchmark gradient, and dynamically adjust the sampling frequency based on the edge adaptive sampling adjustment rate.

[0151] S4. Based on the normalization result of the rate of change of the multimodal sensing data, calculate the dynamic allocation value of the multi-parameter collaborative weight, and extract feature data from the multimodal sensing data in a weighted manner based on the dynamic allocation value of the multi-parameter collaborative weight.

[0152] S5. The edge adaptive collaborative filtering unit performs edge-side local filtering calculation on the feature data, obtains the ratio of the amount of intercepted redundant data to the total amount of data as the redundant data local filtering ratio, and based on the redundant data local filtering ratio, intercepts redundant data and outputs key mutation data.

[0153] S6. Based on the ratio between the magnitude of the key mutation data and the preset magnitude benchmark, calculate the graded response time delay, and based on the graded response time delay, trigger the local primary control logic and output the reported data to the main control system.

[0154] When the amplitude of the critical mutation data exceeds the preset amplitude abnormality threshold, the local primary control logic is triggered to generate control commands to start the intelligent transformer substation heat dissipation device or cut off the predetermined secondary electrical circuit; when the amplitude of the critical mutation data is less than or equal to the preset amplitude abnormality threshold, a flag signal is generated to maintain the current control state.

[0155] This embodiment provides an industrial data acquisition and control method for intelligent transformer substation environmental monitoring; specifically, the method can run on the edge controller of the aforementioned transformer substation, forming a complete step chain from bottom-level acquisition to local control and then to main control reporting;

[0156] In S1, the bottom-level acquisition nodes synchronously or quasi-synchronously acquire multimodal sensor data from multiple areas inside the transformer substation. In order to ensure that data from different physical locations and different sensor types can be analyzed in a unified manner, the edge terminal preferably completes timestamp alignment, node identification and basic integrity verification first.

[0157] Taking a simplified analysis model as an example: the temperature segment Tseg from the busbar area, the humidity segment Hseg from the top, the partial discharge segment Pseg from the cable terminal area, and the electrical segment Eseg from the loop side are encapsulated into a joint data packet within the same monitoring period;

[0158] In S2, the system forms an environmental characteristic change gradient based on the rate of change of each parameter within adjacent time steps. This gradient reflects the rate at which the field state shifts from a steady state to an abnormal state, rather than the absolute value at a single moment. If the gradient value is large, it usually indicates that a certain fault mechanism is actively evolving. If the change is slow, it may still be in the observable stage.

[0159] In S3, the system generates an edge adaptive sampling adjustment rate and dynamically adjusts the sampling frequency accordingly; sampling boosting is prioritized for areas that change rapidly and have high risk correlation, while sampling deflating is applied to areas that are stable for a long time, thus forming a sampling strategy at the edge that involves high-frequency sampling during abnormal periods and low-frequency sampling during normal periods.

[0160] In S4, the system assigns weights to different parameters based on the normalized rate of change and extracts feature data from multimodal data in a weighted manner. The feature data here is not simply a single parameter, but rather retains combined evidence that can explain the evolution of the fault, such as the enhancement of partial discharge accompanied by a rise in temperature or the initial appearance of smoke in a high humidity background.

[0161] In S5, the system performs on-site filtering of feature data; data with normal background, repetitive fragments, and data that do not contribute incrementally to the conclusion are intercepted, and only key mutation data are retained as the core input for subsequent control and reporting; the on-site filtering ratio of redundant data can be used as an operational indicator of the edge filtering effect for subsequent strategy tuning.

[0162] In S6, the system calculates the graded response time delay based on the level of the critical mutation data amplitude relative to the preset amplitude benchmark. If the critical mutation has exceeded the abnormal threshold, the local primary control logic generates a control command within a short delay. The action type can be to start the heat dissipation device or adjust the electrical circuit. If the abnormal threshold has not been exceeded, the system generates a flag signal to maintain the current control state, while continuing to observe and report according to the strategy.

[0163] If data is missing, node is disconnected, or link is congested in any step, the method will not terminate; for missing data in S1, a placeholder from the nearest period can be used and a data missing marker can be added; for incomplete inputs in S2 to S4, basic trend judgment can be performed in a downgraded manner.

[0164] For cases where filtering judgment is unstable in S5, the filtering intensity can be temporarily reduced; for control uncertainties in S6, conservative actions or only reporting without action can be prioritized; if the anomaly continues to amplify and is not relieved after local action, the method will automatically upgrade the event level, retain more original key fragments and prioritize sending them to the main control system.

[0165] During a high humidity and high load operation of the aforementioned transformer substation, S1 first detected that the top humidity was consistently high, the partial discharge in the terminal area began to increase, and the temperature in the connection area rose slightly; S2 identified that the discharge rate was faster than other parameters; S3 then increased the sampling frequency of the terminal area.

[0166] S4 assigns higher synergistic weight to partial discharge and temperature; S5 filters out a large amount of stable background data, retaining only abnormal evolution fragments; S6 activates the heat dissipation device and sends reported data to the main control system after partial discharge and temperature rise both cross the preset risk level.

[0167] If the discharge continues after heat dissipation, the secondary circuit can be restricted according to the established strategy. The purpose of this method is to organize multimodal sensing, edge analysis, redundancy compression and hierarchical control into a continuous industrial processing link, so as to realize the forward identification of environmental anomalies in intelligent transformer substations, low-latency handling and high-value data reporting.

[0168] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. An industrial data acquisition system for environmental monitoring of intelligent prefabricated substations, characterized in that, include: It features a multi-source sensor data acquisition unit with underlying acquisition nodes, an edge adaptive collaborative filtering unit, and a hierarchical response control unit; The multi-source sensor data acquisition unit is used to acquire multimodal sensor data through the underlying acquisition nodes; The edge adaptive collaborative filtering unit is used to obtain the rate of change of each parameter based on the ratio of the difference between each parameter in adjacent time steps to the time interval in the multimodal sensing data, and calculate the gradient of environmental feature change accordingly. Based on the ratio of the gradient of environmental feature change to the preset benchmark gradient, the edge adaptive sampling adjustment rate is calculated to dynamically adjust the sampling frequency. It is also used to calculate the dynamic allocation value of multi-parameter collaborative weights based on the normalization result of the change rate of each parameter, and extract feature data accordingly. The feature data is filtered in place to intercept redundant data and output key mutation data, and the ratio of the amount of intercepted redundant data to the total amount of data is used as the redundant data filtering ratio in place. The graded response control unit is used to calculate the graded response time delay based on the ratio between the amplitude of the key mutation data and the preset amplitude benchmark, thereby triggering the local primary control logic and outputting the reported data to the main control system. The triggering operation includes: when the amplitude of the key mutation data is greater than the preset amplitude abnormality threshold, the local primary control logic is triggered to generate a control command to start the intelligent transformer substation heat dissipation device or cut off a predetermined secondary electrical circuit. When the amplitude is less than or equal to the preset amplitude abnormality threshold, a flag signal is generated to maintain the current control state.

2. The industrial data acquisition system for intelligent transformer substation environmental monitoring according to claim 1, characterized in that, The multimodal sensing data includes at least one or more of the following parameters: temperature, humidity, smoke, partial discharge, and electrical parameters.

3. The industrial data acquisition system for intelligent prefabricated substation environmental monitoring according to claim 1, characterized in that, The operation of the edge adaptive collaborative filtering unit to calculate the dynamic allocation value of multi-parameter collaborative weights includes: Based on the normalized results of the change rates of different types of sensor data in the multimodal sensing data, and the communication priority mapping results obtained by inputting the change rates into a preset communication priority mapping function, combined with the preset computational resource weight adjustment coefficient and communication priority weight adjustment coefficient, a multi-parameter collaborative weight dynamic allocation value for the different types of sensor data is generated in real time; specifically, for the i-th type of sensor data, the real-time generated multi-parameter collaborative weight dynamic allocation value... The calculation formula is: ； in, For the first Rate of change of sensor data This is a communication priority mapping function corresponding to the rate of change of parameters. This is the normalized function of the rate of change of the parameter. To calculate the resource weight adjustment coefficient, The communication priority weight adjustment coefficient; the calculation resource weight adjustment coefficient Communication priority weight adjustment coefficient These are empirical constants that are pre-calibrated and stored locally based on the sensor type corresponding to the underlying acquisition node; The normalized function of the rate of change of the parameter The Min-Max normalization method is used to map it to the [0,1] interval; the communication priority mapping function Using the step mapping rule, when the rate of change When the rate of change exceeds a preset danger threshold, the mapping function takes a value of 2. When the value of the mapping function is less than or equal to the threshold of the rate of change of danger, the value of the mapping function is 1. The dangerous change rate threshold is pre-calibrated based on the lower limit critical value of the change rate that triggers the alarm level in the same parameter in the historical fault samples of the target intelligent transformer or the physical tolerance limit of the equipment insulation material.

4. The industrial data acquisition system for intelligent transformer substation environmental monitoring according to claim 1, characterized in that, The edge adaptive collaborative filtering unit is used to perform on-site filtering calculations on feature data, including: applying a time series similarity association algorithm to identify invalid normal data and repetitive normal data in the feature data, calculating the percentage of the sum of the intercepted invalid normal data and the repetitive normal data to the total amount of collected data, and using this as the on-site filtering ratio of redundant data.

5. The industrial data acquisition system for intelligent transformer substation environmental monitoring according to claim 1, characterized in that, The operation of the edge adaptive collaborative filtering unit to calculate the edge adaptive sampling adjustment rate includes: calculating the response time and adjustment range of the sensor sampling frequency based on the ratio of the environmental feature change gradient to the preset maximum allowable gradient of the system, and generating the edge adaptive sampling adjustment rate.

6. The industrial data acquisition system for intelligent transformer substation environmental monitoring according to claim 1, characterized in that, The system also includes a signal-to-noise ratio arbitration gating unit; The signal-to-noise ratio arbitration gating unit is used to calculate the variance of the key mutation data output by the edge adaptive collaborative filtering unit within a preset time window, as a systematic anomaly index. The systematic anomaly index is compared with a preset variance anomaly threshold; when the systematic anomaly index is greater than the preset variance anomaly threshold, the execution of the hierarchical response control unit is triggered; when the systematic anomaly index is less than or equal to the preset variance anomaly threshold, the execution of the hierarchical response control unit is blocked.

7. The industrial data acquisition system for intelligent transformer substation environmental monitoring according to claim 1, characterized in that, The edge adaptive collaborative filtering unit is also used to segment the multimodal sensing data into multiple data sub-layers. The segmentation operation is based on a preset quantile of a specified parameter in the multimodal sensing data. The specified parameter is an assessment parameter characterizing the risk of the intelligent transformer substation's operating status.

8. An industrial data acquisition and control method for environmental monitoring of intelligent prefabricated substations, characterized in that, Includes the following steps: S1. Acquire multimodal sensing data through the bottom-level acquisition nodes of the multi-source sensing data acquisition unit; S2. Calculate the gradient of environmental feature changes based on the ratio of the difference between each parameter in the multimodal sensing data within adjacent time steps to the time interval. S3. Calculate the edge adaptive sampling adjustment rate based on the ratio of the environmental feature change gradient to the preset benchmark gradient, and dynamically adjust the sampling frequency based on the edge adaptive sampling adjustment rate. S4. Based on the normalization result of the rate of change of the multimodal sensing data, calculate the dynamic allocation value of the multi-parameter collaborative weight, and extract feature data from the multimodal sensing data in a weighted manner based on the dynamic allocation value of the multi-parameter collaborative weight. S5. The edge-adaptive collaborative filtering unit performs edge-side local filtering calculation on the feature data, obtains the ratio of the amount of intercepted redundant data to the total amount of data as the redundant data local filtering ratio, and based on the redundant data local filtering ratio, intercepts redundant data and outputs key mutation data. S6. Based on the ratio between the magnitude of the key mutation data and the preset magnitude benchmark, calculate the graded response time delay, and based on the graded response time delay, trigger the local primary control logic and output the reported data to the main control system. When the amplitude of the critical mutation data is greater than the preset amplitude abnormality threshold, the local primary control logic is triggered to generate a control command for starting the intelligent transformer substation heat dissipation device or cutting off a predetermined secondary electrical circuit; when the amplitude of the critical mutation data is less than or equal to the preset amplitude abnormality threshold, a flag signal is generated to maintain the current control state.

Images