A power equipment state judgment method and system based on multi-source data fusion

Abstract

The application provides a kind of power equipment state judgment method and system based on multi-source data fusion, it is related to power equipment state judgment technical field, comprising: respectively using with the temperature monitoring data, local discharge three-in-one monitoring data, environmental temperature and humidity smoke monitoring data corresponding exclusive sub-algorithm, independently carry out single dimension state determination, corresponding output temperature single state, local discharge single state and environmental single state, and generate each single state corresponding confidence and abnormal characteristic value;Each kind of single state result is converted into standardized evidence body, and the initial fusion state result of equipment is generated;Output corrected final equipment state level, risk assessment result and operation and maintenance suggestion;Output comprehensive research and judgment result containing the final equipment state level;The application improves the accuracy and practicality of final state level determination and operation and maintenance suggestion.

Description

Technical Field

[0001] This invention relates to the field of power equipment status assessment technology, and in particular to a power equipment status assessment method and system based on multi-source data fusion. Background Technology

[0002] In the field of condition monitoring and fault diagnosis of primary power equipment, existing technologies typically assess equipment operating status using threshold-over-limit alarms or independent judgments based on single monitoring methods. Temperature monitoring generally uses tiered alarms by setting fixed temperature thresholds, partial discharge monitoring usually uses single-type partial discharge detection data (UHF, ultrasonic, or ground wave) as the basis for judgment, and environmental monitoring mainly focuses on individual threshold judgments for temperature, humidity, or smoke. However, the actual operating status of primary power equipment is affected by the coupling of multiple physical fields. Data from a single monitoring dimension often cannot comprehensively characterize the true health level of the equipment, easily leading to misjudgments due to temporary sensor interference, missed judgments due to the insensitivity of a single monitoring method to certain fault types, and problems in making unified decisions when monitoring results conflict. In recent years, although some solutions have attempted to simply summarize multi-source data such as temperature and partial discharge for comprehensive judgment, they mostly use linear fusion methods such as weighted averaging or voting, failing to properly handle conflicts between different evidence sources and lacking in-depth utilization of the historical operating patterns of the equipment and their correlation with the external environment, resulting in insufficient accuracy and reliability of the final judgment results. Summary of the Invention

[0003] To address the aforementioned technical problems, the technical solution adopted by this invention is as follows: According to a first aspect of this application, a method for determining the status of power equipment based on multi-source data fusion is provided, comprising: Real-time acquisition of temperature monitoring data, partial discharge three-in-one monitoring data, and ambient temperature, humidity, and smoke monitoring data from primary power equipment; Each of the temperature monitoring data, partial discharge three-in-one monitoring data, and environmental temperature, humidity and smoke monitoring data is used to independently determine the single-dimensional state, output the temperature single-unit state, partial discharge single-unit state and environmental single-unit state, and generate the confidence level and abnormal feature value corresponding to each single-unit state. The results of various individual states are transformed into standardized evidence bodies, and multi-source evidence fusion calculations are performed based on the improved DS evidence theory algorithm to generate preliminary fusion state results of the equipment. Call the preset large model API, input the preliminary fusion status conclusion of the device, and combine the device's historical operating data, operation and maintenance rule base and fault case base to perform intelligent reasoning, and output the corrected final device status level, risk assessment result and operation and maintenance suggestions; The output includes a comprehensive assessment result of the final device status level.

[0004] According to another aspect of this application, a power equipment status assessment system based on multi-source data fusion is also provided, applied to the method described in the first aspect, the system comprising: The data acquisition module is used to collect real-time temperature monitoring data, partial discharge three-in-one monitoring data, and environmental temperature, humidity, and smoke monitoring data from primary power equipment. The primary unit analysis module includes a temperature monitoring submodule, a partial discharge three-in-one monitoring submodule, and an environmental temperature, humidity, and smoke monitoring submodule; The temperature monitoring submodule has a built-in hierarchical decision tree branching algorithm unit, which is used to make multi-level branching judgments based on the real-time temperature and short-term temperature rise rate of the equipment, and output the temperature unit status and the corresponding temperature confidence and temperature anomaly characteristic values. The partial discharge three-in-one monitoring submodule has a built-in weighted threshold fusion unit and a fault tree tracing unit. The weighted threshold fusion unit is used to perform multi-dimensional weighted fusion of UHF partial discharge data, ultrasonic partial discharge data and ground wave partial discharge data and output a comprehensive partial discharge risk score and the status of the partial discharge unit. The fault tree tracing unit is used to perform fault hierarchical tracing when the comprehensive partial discharge risk score exceeds the standard and output a partial discharge fault tracing label. The environmental temperature, humidity and smoke monitoring submodule has a built-in multi-parameter multi-level threshold determination unit, which is used to independently determine the threshold and multi-parameter linkage score of the real-time values ​​of environmental temperature, environmental humidity and smoke concentration, and output the status of individual environmental units and the corresponding environmental confidence and environmental anomaly characteristic values. The secondary fusion analysis module is connected to the primary single-unit analysis module. It has a built-in improved DS evidence theory fusion unit, which is used to convert the temperature single-unit state, partial discharge single-unit state and environmental single-unit state into standardized evidence bodies respectively. After assigning basic probability values ​​to the three sets of evidence sources, it sequentially performs two-evidence iterative synthesis calculation, outputs the probability assignment of each state, and selects the state corresponding to the maximum probability value as the preliminary fusion state result of the device. The large model interface module connects to the secondary fusion analysis module and is used to establish a communication connection with the external large model API. It packages and sends the preliminary fusion status results of the device, historical operating data of the device, operation and maintenance rule library, and fault case library to the large model API, and receives the intelligent inference results returned by the large model API. It performs confidence correction and risk assessment optimization on the preliminary fusion status results of the device, and outputs the final device status level, risk assessment results and operation and maintenance suggestions. The result output module is connected to the large model interface module and is used to output the comprehensive judgment result, which includes the final equipment status level, comprehensive judgment confidence, multi-source data judgment details and operation and maintenance handling strategy. The primary unit analysis module, the secondary fusion analysis module, the large model interface module, and the result output module are cascaded in sequence to form a closed-loop processing link of data acquisition, unit analysis, fusion decision-making, intelligent correction, and result output.

[0005] The present invention has at least the following beneficial effects: The present invention provides a power equipment status assessment method based on multi-source data fusion. First, it employs dedicated sub-algorithms corresponding to temperature, partial discharge, and environmental monitoring data for independent analysis. This allows for adaptability assessment based on the physical characteristics of various monitoring data, avoiding the problem of insufficient adaptability of a single general-purpose algorithm to different data types. Next, the three individual assessment results are transformed into standardized evidence bodies, and an improved DS evidence theory is introduced for multi-source fusion. In cases of evidence conflict, a conflict coefficient identification and weight adaptive adjustment mechanism reduces the impact of interfering evidence, enabling the fusion result to comprehensively reflect multi-dimensional information and effectively handle inconsistencies between evidence sources, thus improving the robustness of the fusion decision. Finally, it calls a large-scale model API to intelligently correct the data by combining historical equipment data, operation and maintenance rules, and fault cases. This fully utilizes equipment lifecycle data to verify the rationality of the initial fusion conclusions and optimize risk levels, overcoming the limitations of pure rule-based algorithms in trend deduction and causal relationship analysis, and improving the accuracy and practicality of the final status level assessment and operation and maintenance recommendations. Attached Figure Description

[0006] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0007] Figure 1 A flowchart of a power equipment status determination method based on multi-source data fusion provided in an embodiment of the present invention. Detailed Implementation

[0008] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0009] It should be noted that, based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Furthermore, this device and / or practice the method can be implemented using other structures and / or functionalities besides one or more of the aspects set forth herein.

[0010] The following will refer to Figure 1 The flowchart shown is a method for judging the status of power equipment based on multi-source data fusion, which introduces a method for judging the status of power equipment based on multi-source data fusion.

[0011] The power equipment status assessment method based on multi-source data fusion includes the following steps: The S100 collects real-time temperature monitoring data, partial discharge monitoring data, and ambient temperature, humidity, and smoke monitoring data from primary power equipment.

[0012] In this embodiment, the temperature monitoring targets cover key heat-generating components of primary power equipment, including switch moving and stationary contacts, busbar connections, cable joints, transformer windings, and cores. The temperature sensor employs a contact-type wireless temperature measurement scheme, transmitting the collected temperature data wirelessly to the acquisition device via a 470MHz frequency. The sensor can utilize an electromagnetic field self-powered supply for maintenance-free operation. After receiving the data, the acquisition device uploads the temperature data to the monitoring system via an RS485 bus or Ethernet. The acquired temperature parameters include real-time temperature values, the highest and lowest temperatures within the statistical period, and the acquisition period is typically on the order of seconds or minutes.

[0013] The integrated partial discharge (PD) monitoring system simultaneously acquires three types of PD signals: ultra-high frequency (UHF), ultrasonic (AE), and ground wave (TEV). The UHF sensor typically detects frequencies between 300MHz and 1500MHz and is magnetically mounted on the wall of the switchgear cable compartment. The ultrasonic sensor has a resonant frequency of 40kHz ± 1kHz, and the TEV sensor measures signals with a bandwidth of 3MHz to 100MHz. All three types of sensors are integrated into a single monitoring terminal. The PD analog signals acquired by the sensors are transmitted to the monitoring device via coaxial cable, where they undergo digital processing including amplification, filtering, detection, digital pulse extraction, digital filtering, and PD characteristic quantity calculation. Monitoring data includes the number of discharges, discharge magnitude, and PRPD / PRPS discharge spectra. The processed data is uploaded to the monitoring platform via RS485 bus (MODBUS-RTU protocol) or wired connection.

[0014] Environmental monitoring is achieved through temperature and humidity sensors and smoke sensors deployed in the power distribution room and switchgear. The temperature and humidity sensors collect ambient temperature and humidity data, while the smoke sensors detect fire hazards. After being aggregated by the data acquisition terminal, the environmental data is uploaded via RS485 digital signal communication using the Modbus protocol. Smoke concentration data is collected and uploaded synchronously with the temperature and humidity data.

[0015] S200 employs dedicated sub-algorithms corresponding to the temperature monitoring data, partial discharge three-in-one monitoring data, and environmental temperature, humidity, and smoke monitoring data to independently determine the single-dimensional state, outputting the temperature single-unit state, partial discharge single-unit state, and environmental single-unit state, and generating the confidence level and abnormal feature value corresponding to each single-unit state.

[0016] Furthermore, the dedicated sub-algorithm corresponding to the temperature monitoring data employs a hierarchical decision tree branching algorithm for analysis, specifically including: S210, using the real-time temperature of the equipment, the normal temperature threshold of the equipment, the warning temperature threshold of the equipment, the fault temperature threshold of the equipment, and the short-term temperature rise rate as the branching judgment features of the decision tree, the state filtering is performed layer by layer; wherein, the short-term temperature rise rate is calculated by dividing the difference between the highest temperature and the lowest temperature within the statistical period by the statistical duration.

[0017] This step first determines the four branching features of the decision tree: real-time equipment temperature T, normal temperature threshold T1, warning temperature threshold T2, fault temperature threshold T3, and short-term temperature rise rate V. T T1, T2, and T3 are fixed thresholds set in advance based on equipment type, rated parameters, and power operation and maintenance procedures.

[0018] Short-term temperature rise rate V T The calculation logic is as follows: taking the current time as the endpoint, a statistical window of 10 minutes is extracted backward, and the highest temperature T collected within that window is extracted. max With the lowest temperature T min The difference between the two values ​​is divided by 10 minutes to obtain the average temperature rise rate per minute. This rate reflects the drastic temperature changes of the equipment in a recent short period, rather than instantaneous fluctuations. The statistical window is updated using a sliding method and is recalculated for each data collection cycle.

[0019] S211, perform multi-level bifurcation determination based on the real-time temperature and the short-time temperature rise rate: The decision tree executes branching decisions layer by layer starting from the root node. Each node sets a decision condition. If the condition is met, the tree enters the corresponding sub-branch and eventually falls into one of the four state terminal nodes.

[0020] S212, when the real-time temperature does not exceed the normal temperature threshold and the short-term temperature rise rate does not exceed the first rate threshold, it is determined to be in a normal state.

[0021] To enter this branch, two conditions must be met simultaneously: the real-time temperature T does not exceed the normal threshold T1, and the short-term temperature rise rate V T The temperature change rate should not exceed 0.3℃ / min. The former indicates that the absolute temperature of the equipment is within a safe range, while the latter indicates that the temperature change is gradual and there is no obvious upward trend in the past 10 minutes. The combination of these two conditions indicates that the equipment is in a normal thermally stable state.

[0022] S213, when the real-time temperature is between the normal temperature threshold and the warning temperature threshold and the short-term temperature rise rate does not exceed the second rate threshold, it is determined to be in a state of alert.

[0023] When the temperature is between T1 and T2, meaning it has exceeded the normal range but has not yet reached the warning threshold, and the temperature rise rate does not exceed 0.5℃ / min, this branch is entered. This situation indicates that the equipment temperature is too high but the temperature rise is controllable, which may be affected by load fluctuations or ambient temperature. It requires attention but does not require emergency intervention at this time.

[0024] S214, when the real-time temperature is between the warning temperature threshold and the fault temperature threshold, or when the short-term temperature rise rate exceeds the second rate threshold, it is determined to be an abnormal state.

[0025] This branch is conditional or logical, and will enter if either of the following conditions is met: the temperature is between T2 and T3 (exceeding the warning threshold but not reaching the fault threshold), or the temperature rise rate exceeds 0.5℃ / min. The former corresponds to a persistently high temperature that has reached the warning range, while the latter corresponds to a rapid temperature rise, even if the current temperature is not necessarily high, indicating a potential for rapid deterioration. If either of these two conditions is met, the situation is considered abnormal and an investigation is required.

[0026] S215, when the real-time temperature exceeds the fault temperature threshold, it is determined to be a serious fault state; wherein, the first rate threshold is lower than the second rate threshold.

[0027] When the real-time temperature T directly exceeds the fault threshold T3, regardless of the rate of temperature rise, it is directly determined to be a serious fault state. This branch is placed at the bottom of the decision tree and has the highest priority. Once triggered, it means that the equipment has reached or exceeded the maximum operating temperature allowed by the design, and there is a risk of burnout or insulation breakdown.

[0028] In each branch, the difference between the first rate threshold (0.3℃ / min) and the second rate threshold (0.5℃ / min) reflects the graded assessment of the severity of temperature rise: 0.3℃ / min corresponds to the upper limit of normal fluctuation under stable operating conditions, while 0.5℃ / min corresponds to the critical line of temperature rise rate that requires vigilance.

[0029] This decision tree branching algorithm incorporates both the absolute temperature level and the short-term temperature rise rate into its judgment criteria, enabling it to distinguish between two different physical scenarios: slow temperature rise and rapid temperature rise. The former is often caused by increased load or environmental changes, while the latter often corresponds to emergencies such as a sudden increase in contact resistance or internal faults. By setting two different rate thresholds, it achieves a graded response to the temperature rise trend. The layer-by-layer branching structure of the decision tree makes each judgment path clear and traceable, allowing on-site maintenance personnel to directly locate the cause of the anomaly based on the hit branch. The algorithm requires no complex calculations, has a fast judgment speed, is suitable for real-time execution at the edge, and has good compatibility with existing power equipment temperature setting management standards.

[0030] Furthermore, the dedicated sub-algorithm corresponding to the partial discharge three-in-one monitoring data employs a multi-dimensional weighted threshold fusion algorithm for analysis, specifically including: S220 extracts real-time monitoring values ​​from three dimensions: UHF partial discharge data, ultrasonic partial discharge data, and ground wave partial discharge data, and assigns corresponding weight values ​​to each dimension, with the UHF dimension having the highest weight.

[0031] This step extracts real-time monitoring values ​​X1, X2, and X3 from the partial discharge monitoring terminal, corresponding to monitoring readings for UHF, ultrasound, and ground electromagnetic waves, respectively. The physical meanings and dimensions of these three types of data are different—UHF detects the amplitude of electromagnetic wave signals (unit: dBm), ultrasound detects the amplitude of mechanical vibration signals (unit: dB), and ground electromagnetic waves detect the amplitude of transient voltage signals to ground (unit: dBmV). Since the three types of data cannot be directly compared or added, normalization processing is required in subsequent steps.

[0032] The weighting of the three dimensions is as follows: UHF w1=0.4, ultrasonic w2=0.3, and ground wave w3=0.3. The reason for the highest weight for UHF is that the UHF method has the highest sensitivity for detecting internal insulation defects in equipment (such as metal particles, floating potentials, and internal air gap discharges), relatively small signal propagation attenuation, and strong resistance to low-frequency interference, effectively capturing early insulation degradation characteristics. Ultrasonic waves are sensitive to surface discharges and mechanical loosening but are easily affected by environmental noise; ground waves are effective for internal discharges in switchgear but have high requirements for the propagation path. These two methods serve as a supplement to UHF.

[0033] S221, calculate the degree of exceedance of the real-time monitoring value of each dimension of partial discharge index relative to its normal threshold and fault threshold, and normalize the degree of exceedance of each dimension to a unified numerical range.

[0034] Real-time monitoring values ​​for each dimension need to be converted into dimensionless hyperscaling I first. i This unifies the values ​​to the [0,1] interval. For the i-th type of partial discharge index, the overscaling calculation formula is: ; The physical meaning of each parameter in the formula is: X i X represents the real-time monitoring reading of the i-th type of partial discharge indicator; i0 This refers to the normal threshold for this type of indicator, that is, the upper limit that this indicator will not exceed under normal operating conditions; X imax This is the fault threshold for this type of indicator; exceeding this value indicates a clear insulation defect.

[0035] Overscaling I i The meaning of the value is: when the real-time monitored value does not exceed the normal threshold, I i ≤0, truncated to 0 during normalization; when the real-time monitored value exceeds the fault threshold, I i ≥1, truncated to 1 during normalization; when I is in between, i This reflects the relative position of the current value between the normal threshold and the fault threshold. The specific values ​​of the normal threshold and fault threshold for the three types of indicators are set according to the equipment type, voltage level, and industry standards—UHF is calibrated with reference to the background noise level and historical baseline, ultrasonic is calibrated with the environmental vibration background, and ground wave is calibrated with the typical background value of the switchgear.

[0036] S222: The normalized overscaling of each dimension is multiplied by its corresponding weight value and then summed to calculate the comprehensive risk score of partial discharge.

[0037] After obtaining the normalized overscaling values ​​I1, I2, and I3 for the three dimensions, the overscaling values ​​of each dimension are multiplied by their corresponding weights and then summed to obtain the comprehensive risk score for partial discharge: ; The rating is S. p The value range is [0,1]. When the real-time monitoring value of a dimension is exactly equal to its normal threshold, the overscale of that dimension is 0 and contributes nothing to the overall score; when all dimensions are within the normal range, S... p S approaches 0; when all dimensions reach or exceed their respective failure thresholds, p Approaching 1. The essence of the comprehensive score is the weighted sum of the risk levels of the three dimensions. The more a dimension exceeds the standard and the greater its weight, the more significant the boost to the final score.

[0038] S223. Based on the numerical range of the partial discharge comprehensive risk score, determine the status of the individual partial discharge unit. The lower the score, the better the status, and the higher the score, the worse the status.

[0039] The partial discharge comprehensive risk score S was calculated. p Then, the partial discharge cell status is determined based on the following four intervals: When S p When the value is ≤0.2, it is considered to be in a normal state. This range represents that the data for all three types of partial discharge are within the normal threshold range or have only very small fluctuations, without the characteristics of continuous discharge pulses.

[0040] When 0.2 p When the value is ≤0.4, it is considered a state of alert. This range corresponds to a situation where the partial discharge data of a single type slightly exceeds the limit but has not formed a continuous discharge mode. In actual field operations, anomalies in single-type data are often temporarily triggered by factors such as environmental electromagnetic interference and vibrations from nearby operations, rather than insulation defects in the equipment itself. Therefore, this range is set as a state of alert, prompting continued observation but not escalation for the time being.

[0041] When 0.4 p A value ≤0.7 is considered an abnormal state. This range typically indicates that partial discharge data in at least two dimensions exceed the limit simultaneously, or that the exceedance of a high-weight dimension (such as UHF) is significantly higher than normal. Multi-dimensional synchronous anomalies suggest that the signal source is more likely to be a genuine discharge inside the equipment rather than external interference, requiring the initiation of fault tracing analysis and timely retesting for confirmation.

[0042] When S p A value >0.7 indicates a severe fault condition. This range corresponds to three types of partial discharge synchronization anomalies or severe exceedances in a certain dimension. Specifically, it manifests as high discharge frequency, amplitude far exceeding the fault threshold, and discharge patterns exhibiting typical insulation defect characteristics, indicating a continuous risk of insulation degradation, requiring immediate maintenance measures.

[0043] This weighted threshold fusion algorithm calculates and normalizes the overscaling of three heterogeneous partial discharge data types—UHF, ultrasonic, and ground wave—eliminating the technical barriers of different dimensions and sensitivities among the three detection methods. This allows for comparison and synthesis of the three types of data within a unified framework. The UHF dimension is given the highest weight, aligning with practical engineering experience that the UHF method has the highest sensitivity and optimal signal-to-noise ratio in detecting internal insulation defects in equipment such as switchgear. Meanwhile, ultrasonic and ground wave data maintain a weight ratio comparable to UHF, avoiding over-reliance on a single method. The comprehensive score is calculated using a weighted summation method, where anomalies in one dimension directly increase the comprehensive score. When multiple dimensions show simultaneous anomalies, the score superposition effect is significant, distinguishing between internal discharge and external environmental interference, and effectively reducing the false alarm rate of partial discharge monitoring.

[0044] ​​Furthermore, the dedicated sub-algorithm corresponding to the partial discharge three-in-one monitoring data, after completing the weighted threshold fusion, also includes a fault tree hierarchical tracing step: Using the partial discharge comprehensive risk score exceeding the preset abnormal threshold as the top event, the fault tree is branched down layer by layer to the insulation aging bottom event, the partial discharge bottom event, the electromagnetic interference bottom event, and the equipment poor contact bottom event; Based on the membership degree of the fault characteristics corresponding to each event, the fault cause corresponding to the current partial discharge monitoring data is matched in reverse, and the partial discharge fault source label is output.

[0045] Furthermore, the dedicated sub-algorithm corresponding to the environmental temperature, humidity, and smoke monitoring data employs a multi-parameter, multi-level threshold judgment algorithm for analysis, specifically including: S230 collects real-time values ​​of ambient temperature, ambient humidity, and smoke concentration as three independent parameters.

[0046] This step extracts real-time readings of three environmental parameters from environmental monitoring terminals deployed inside the power distribution room and switchgear: real-time ambient temperature value X1 (unit: °C), real-time ambient humidity value X2 (unit: %RH), and real-time smoke concentration value X3. The sensors for these three parameters are integrated into the same environmental monitoring unit, and data is collected and uploaded synchronously. The temperature range is typically -40℃ to +125℃, the humidity range is 0 to 100%RH, and the smoke concentration is expressed in mg / m³. 3 Output in ppm or ppm.

[0047] S231, determine whether each independent parameter is within its own normal threshold range. For parameters that exceed the normal threshold range, calculate the relative deviation from the standard optimal value, and use it as the single parameter risk value of that parameter.

[0048] For each type of environmental parameter, a normal threshold range [X] is pre-set based on the power equipment operating environment standards and the operating condition requirements of the equipment manufacturer. jmin ,X jmax and the standard optimal value X jst d.

[0049] The normal threshold ranges for each parameter are set as follows: the normal range for ambient temperature is based on the equipment's operating ambient temperature requirements (usually -5℃ to +40℃ or -10℃ to +50℃, depending on the specific equipment type and installation location); the normal range for ambient humidity is based on insulation withstand capability and critical condensation humidity (usually 10% to 90%RH or 20% to 80%RH); and the normal range for smoke concentration is based on the lower limit of the warning value that does not trigger the fire alarm device (usually 50% of the alarm threshold of the smoke detector is used as the upper limit of normal).

[0050] Standard optimal value X jstThe setting of 'd' signifies the ideal value for this parameter that is most conducive to the safe and stable operation of the equipment. The standard optimal value for ambient temperature is generally 25℃, corresponding to the most suitable operating environment temperature for the equipment; the standard optimal value for ambient humidity is generally taken as the midpoint between 45% and 55%RH, such as 50%RH. This humidity range is neither too dry, leading to static electricity buildup, nor too humid, leading to the risk of condensation; the standard optimal value for smoke concentration is 0 mg / m³. 3 That is, there is no smoke present.

[0051] Single-parameter risk value E j The calculation logic is as follows: First, determine the current real-time monitoring value X. j Is it within the normal threshold range [X]? jmin ,X jmax ]Inside.

[0052] If X j If the value is within this range, the risk value of this parameter is 0, indicating that the current environmental parameters are within the allowable range for safe operation of the equipment and do not pose any threat.

[0053] If X j If the value exceeds this range (whether it is below the lower limit or above the upper limit), the absolute deviation of the current value from the standard optimal value is calculated, and then divided by the absolute value of the difference between the normal upper limit and the standard optimal value of the parameter to obtain the relative deviation as the risk value of the parameter.

[0054] Taking ambient temperature as an example: if the standard optimal value is 25℃, and the normal range is -5℃ to +40℃, and the current actual temperature is 50℃, the actual value exceeds the normal range. Its absolute deviation from the standard optimal value is |50-25|=25℃, and the difference between the normal upper limit and the standard optimal value is |40-25|=15℃. Dividing these two values ​​gives a deviation of 25 / 15≈1.67. A deviation exceeding 1 indicates that the current value has significantly deviated from the normal range. Similarly, if the current temperature is below the lower limit, such as -10℃, the absolute deviation is |-10-25|=35℃, and the relative deviation is 35 / 15≈2.33. For smoke concentration, since the standard optimal value is 0, the calculation of its risk value essentially degenerates into the ratio of the concentration value to the upper limit threshold.

[0055] S232 assigns weight values ​​to ambient temperature, ambient humidity, and smoke concentration, respectively; among them, smoke concentration has the highest weight.

[0056] S233: Multiply the individual parameter risk values ​​of each parameter by their corresponding weight values ​​and sum them up to obtain the comprehensive environmental risk score.

[0057] Weights were assigned to the three environmental parameters: ambient temperature weight α1=0.3, ambient humidity weight α2=0.3, and smoke concentration weight α3=0.4. The sum of the three weights was 1.

[0058] The reason why smoke concentration is given the highest weight is that the presence of smoke in power distribution rooms and switchgear directly indicates an emergency situation of equipment short circuit and fire or overheating and combustion of insulation materials, which is far more dangerous than simply exceeding the limits of ambient temperature and humidity. Poor temperature and humidity mainly affect the insulation performance of equipment and accelerate aging, which is a progressive hazard; smoke, on the other hand, is a direct manifestation of fire hazard and requires the highest priority response.

[0059] After obtaining the individual parameter risk values ​​E1, E2, and E3 for each parameter, multiply each parameter risk value by its corresponding weight and sum them up to obtain the comprehensive environmental risk score S. e Its value range is [0, +∞). Since the risk value of each individual parameter may exceed 1 (when it is severely exceeded), the comprehensive score may also exceed 1.

[0060] Environmental Comprehensive Risk Score e The calculation formula is: S e =0.3×E1+0.3×E2+0.4×E3.

[0061] S234. Based on the comprehensive environmental risk score, the system adopts a judgment logic of single-parameter anomaly warning and multi-parameter linkage upgrade to output the status of individual environmental units.

[0062] After obtaining the comprehensive environmental risk score S e Based on this, and combined with the intermediate judgment results of whether each single parameter exceeds the standard independently, the final environmental unit status is output using the judgment logic of "single parameter anomaly warning and multi-parameter linkage upgrade".

[0063] The core rule of this judgment logic is: when S e When the value is 0 and all parameters are within the normal threshold range, it is considered to be in a normal state, indicating that the current environmental conditions are good and no operation is required.

[0064] When only one parameter exceeds the normal threshold range and S e When at a low level (e.g., 0) e If the value is ≤0.4, it is considered a state of alert. This situation corresponds to a slight anomaly in a single environmental factor, such as an ambient temperature that is slightly higher than normal but not reaching a dangerous level, or humidity that is slightly above the limit but not approaching the condensation threshold. Being classified as a state of alert means that the trend of this parameter's change needs to be monitored, but no coordinated action will be triggered at this time.

[0065] When any two parameters simultaneously exceed the normal threshold range, or when a single parameter severely exceeds the limit, causing S... e When significantly elevated (e.g., 0.4) e ​​A value ≤0.8 is considered an abnormal state. Typical scenarios include: high temperature and high humidity exceeding the standard simultaneously, or smoke concentration reaching more than 50% of the fire alarm threshold but not yet triggering multi-parameter linkage. In this case, maintenance personnel need to be arranged to check the operating status of the environmental control equipment on-site.

[0066] When S e When the smoke concentration exceeds 0.8 and multiple parameters are abnormally linked—that is, when severe temperature and humidity conditions combined with excessive smoke concentration, or when the smoke concentration alone reaches the fire alarm threshold and the ambient temperature remains high—it is considered a serious fault state. This situation carries a real risk of internal short circuits causing fires or insulation breakdowns leading to fires, requiring the triggering of fire alarm linkage and immediate emergency response. If the smoke concentration alone is severely exceeded (e.g., E3 > 1.5 and continues to rise), even if the temperature and humidity parameters are normal, it is directly classified as a serious fault state, reflecting the priority logic of smoke parameter as a "veto factor."

[0067] This multi-parameter, multi-level threshold judgment algorithm independently judges and quantifies the deviation of three environmental parameters: temperature, humidity, and smoke. This allows the degree of abnormality of each parameter to be expressed as a continuous numerical value rather than a simple "exceeding / not exceeding the limit" method, avoiding the jittering and misjudgment problem that occurs near the critical value in the fixed threshold method. Smoke concentration is given the highest weight and has a "one-vote veto" effect when it is severely exceeded, ensuring that fire hazards receive the highest level of response priority, which is in line with the actual safety needs of power distribution sites where fire risks are far higher than ordinary temperature and humidity fluctuations. The judgment logic that triggers an alert state only for single parameter anomalies and triggers state upgrades for multi-parameter linkage effectively distinguishes between general fluctuations in the equipment operating environment and dangerous conditions that truly require human intervention, reducing the invalid alarm rate of environmental monitoring.

[0068] S300 transforms various individual state results into standardized evidence bodies, performs multi-source evidence fusion calculations based on the improved DS evidence theory algorithm, and generates preliminary fusion state results for the equipment.

[0069] This step is the core of the entire secondary fusion analysis. Its function is to perform cross-dimensional correlation analysis on the three independent individual state results output from the primary analysis, resolve potential conflicts or inconsistencies between the individual conclusions, and output a comprehensive preliminary fusion state result. This step adopts an improved DS evidence theory algorithm, which overcomes the defect of traditional DS algorithms in distorting fusion results when dealing with highly conflicting evidence by introducing conflict coefficient identification and adaptive weight correction mechanisms.

[0070] Furthermore, the multi-source evidence fusion calculation based on the improved DS evidence theory algorithm specifically includes: S310 sets up an identification framework that includes four states: normal state, attention state, abnormal state, and critical fault state.

[0071] First, a device status identification framework Θ = {Normal State, Attention State, Abnormal State, Critical Failure State} is established. This framework encompasses four mutually exclusive and complete assumptions about device operating states. The completeness requirement of the identification framework is that regardless of which state the fusion result ultimately falls into, that state must belong to this set; there cannot be a fusion conclusion falling outside the framework. The four states have a clear hierarchical progression—from normal to critical failure, with the risk level increasing progressively. This provides an orderly basis for subsequent probability allocation and maximum probability decision-making.

[0072] S311, the temperature-dependent monomer state, the partial discharge-dependent monomer state, and the environmental monomer state are respectively regarded as three independent evidence sources, and basic probability values ​​are assigned to each state level according to the confidence level corresponding to each monomer state.

[0073] The temperature-dependent cell status, partial discharge-dependent cell status, and environmental cell status output from the first-level assessment step are defined as three independent evidence sources: E1 (temperature), E2 (partial discharge), and E3 (environment). The data content of each evidence source includes: the status level determined by the evidence source, the corresponding confidence level, and the anomalous characteristic value.

[0074] Basic Probability Assignment (BPA) transforms the opinions of each evidence source regarding the four states into probability allocations. Specifically, for a given evidence source, the confidence level corresponding to its output state level is denoted as C (ranging from 0 to 1). The basic probability that the evidence source considers the current device to be in that state is C. The remaining probabilities 1-C are assigned to the other three states in the identification framework in an average distribution. Taking temperature evidence as an example: if the temperature unit is determined to be in an abnormal state with a confidence level of 0.8, the probability of assigning the abnormal state to the temperature evidence source is 0.8, and the probabilities of assigning the normal state, the attention state, and the severe fault state are each (1-0.8) / 3≈0.0667. If the confidence level is 1.0, the remaining probability is 0, and the evidence source definitively points to a certain state.

[0075] After assigning basic probability values ​​to each source of evidence, three sets of probability allocation functions m1 (temperature), m2 (partial discharge), and m3 (environment) are obtained. Each function is a probability distribution defined on the identification frame Θ, which satisfies that the sum of the probability assignments of the source of evidence for all four states is equal to 1.

[0076] S312, first synthesize the temperature evidence source and the partial discharge evidence source into two pieces of evidence, calculate the conflict coefficient between the two pieces of evidence, and then perform weighted fusion on the basic probability assignment of the two pieces of evidence based on the conflict coefficient to obtain the intermediate fusion result.

[0077] The fusion sequence adopts an iterative strategy of "core evidence first, auxiliary evidence later," first synthesizing the temperature evidence source E1 and the partial discharge evidence source E2. The reason for choosing to fuse temperature and partial discharge first is that temperature reflects the heating state of the equipment, and partial discharge reflects the insulation state of the equipment. Both are direct representations of the equipment's physical state and have a high correlation. Fusing the physical evidence internally first is conducive to forming a stable and reliable intermediate conclusion.

[0078] The core formula for combining two pieces of evidence is the standard DS combination rule. Given the basic probability assignments m1 and m2 for two sources of evidence, their combination result in a certain state A is: ; Here, B and C respectively traverse all state combinations in the identification framework Θ that intersect with A and whose intersection is exactly A. K in the denominator is the conflict coefficient, defined as: ; The physical meaning of the conflict coefficient K is the sum of the probabilities that the two sets of evidence assign all probabilities to incompatible states. The larger the value of K, the more serious the disagreement between the two sources of evidence—for example, the temperature evidence source assigns all probabilities to the "normal state," while the partial discharge evidence source assigns all probabilities to the "severe fault state." In this case, B∩C is an empty set, and K approaches 1. K=0 represents that the two sets of evidence are completely consistent, and K=1 represents that the two sets of evidence are completely conflicting (at this time, the denominator of the synthesis formula is zero, and they cannot be merged, requiring the triggering of a correction mechanism).

[0079] S313, the intermediate fusion result and the environmental evidence source are combined again to form two pieces of evidence, and after iterative fusion, the final probability values ​​of each state are output.

[0080] After the initial fusion of temperature and partial discharge, the intermediate fusion result m is obtained. 12 (Each of the four states has a probability assignment). Then, the intermediate evidence and the environmental evidence source m3 are fused together in a second iteration using the same two-evidence synthesis formula to obtain the final probability assignments for each state m(Θ)={m(normal),m(attention),m(abnormal),m(serious failure)}.

[0081] The design consideration for placing environmental evidence sources in the final fusion is that environmental parameters are external interference factors, have a low weight in determining equipment status, and are prone to introducing uncertainty and interference noise. By first completing the internal fusion of the two ontological evidence sources—temperature and partial discharge—to form a stable intermediate conclusion, and then introducing environmental evidence for fine-tuning, it is possible to effectively prevent excessive disturbance of the core conclusions by environmental noise.

[0082] S314, Select the state corresponding to the maximum probability as the initial fusion state result of the device.

[0083] After obtaining the final probability assignments for each state, the fusion conclusion is selected using the maximum probability decision rule: ; That is, the state with the highest probability among the four states is selected as the initial fusion state result of the device. This decision rule is based on the maximum trust principle of the DS evidence theory—among all possible hypotheses, the hypothesis with the highest degree of evidence support is selected as the final judgment result.

[0084] Furthermore, during the synthesis of the two pieces of evidence, when the conflict coefficient between the two sets of evidence exceeds a preset conflict threshold, an adaptive weight correction is triggered. S320 reduces the adaptive weight value corresponding to conflicting evidence, thereby weakening the impact of that evidence on the fusion result.

[0085] S321, at the same time, increases the weighting of the average probability assignment of multiple pieces of evidence to smooth out the impact of conflicting evidence.

[0086] S322, the value of the adaptive weight is inversely proportional to the conflict coefficient, that is, the more intense the conflict, the smaller the value of the adaptive weight.

[0087] When the conflict coefficient K is detected to be greater than the preset conflict threshold K0 (e.g., K0=0.85) during the synthesis of two pieces of evidence, it indicates that the judgment opinions of the two sets of evidence are highly inconsistent. At this time, directly using the standard DS synthesis formula may lead to a counterintuitive phenomenon in the fusion result—the high probability assignment of one source of evidence is abnormally amplified or distorted due to the excessive conflict coefficient.

[0088] This triggers the adaptive weight adjustment mechanism. The core formula for the adjustment is: ; The physical meanings of the parameters in the formula are as follows: m'(A) is the basic probability assignment to state A after correction; m(A) is the basic probability assignment to state A originally assigned to this source of evidence; m avg (A) is the arithmetic mean of the probabilities assigned to state A by all evidence sources involved in the fusion; ω is the adaptive weight, with a value range of (0,1).

[0089] The adaptive weight ω is inversely proportional to the conflict coefficient K. Specifically, the mapping rule is as follows: when K=K0, ω=0.8, meaning the original evidence still dominates, and the average weight is only used for auxiliary smoothing; as K increases to near 1, ω gradually decreases to near 0.2, at which point the credibility of the original evidence is significantly weakened, and the fusion result relies more on the average opinion of each evidence source. The design logic of this inverse relationship is: the more intense the conflict, the greater the divergence among the evidence sources, and the credibility of any single evidence source should be suppressed, thus adopting the majority average opinion as a more robust fusion benchmark.

[0090] Taking a real-world scenario as an example: the temperature evidence source determines the device to be in "normal condition" (probability 0.9), while the partial discharge evidence source determines it to be in "severe fault condition" (probability 0.85). The conflict coefficient K between the two is close to 0.765 (0.9 × 0.85 = 0.765). If the preset threshold K0 = 0.7, a correction is triggered. Let the current conflict coefficient K = 0.765, corresponding to ω = 0.6. Then, for the "severe fault condition," the corrected probability is m'(severe fault) = 0.6 × 0.85 + (1 - 0.6) × (0.85 + 0.05) / 2 = 0.51 + 0.18 = 0.69—the original probability of severe fault from the partial discharge evidence source is corrected from 0.85 to 0.69, while the original probability of severe fault from the temperature evidence source was only 0.05, which is raised to 0.18 after averaging. The revised results synthesized the opinions of the two sets of evidence: although the probability of severe failure was lowered, it was still significantly higher than the probability of normal state (the revised probability of normal state is about 0.6×0.05+0.4×0.475=0.22). The final decision was still dominated by the partial discharge evidence, but the extreme nature of the conclusion was smoothed out, avoiding the risk of a single piece of evidence completely suppressing the conclusion under highly conflicting circumstances.

[0091] This improved DS evidence fusion algorithm constructs a complete identification framework encompassing four device states. It quantifies the judgments of the three evidence sources—temperature, partial discharge, and environment—into comparable probability distributions through basic probability assignment, achieving a unified expression and computable fusion of cross-dimensional heterogeneous evidence. The iterative order of fusing temperature and partial discharge evidence first, followed by introducing environmental evidence, ensures that evidence with stronger characterization capabilities for the device's physical state is internally consistent before incorporating environmental interference factors as auxiliary correction terms. This effectively reduces the contamination of the core fusion conclusion by external environmental noise. By introducing a real-time calculation mechanism for the conflict coefficient K and a dynamic adjustment mechanism for the adaptive weight ω, the algorithm reduces the original probability weight of conflicting evidence and increases the proportion of average assignment across multiple pieces of evidence when two sets of evidence are highly conflicting. This avoids the anti-intuitive fusion defects of traditional DS algorithms in conflict scenarios, ensuring that the fusion results maintain rationality and stability under different levels of evidence consistency.

[0092] S400: Call the preset large model API, input the preliminary fusion status conclusion of the device, and combine the device's historical operating data, operation and maintenance rule base and fault case base to perform intelligent reasoning, and output the corrected final device status level, risk assessment result and operation and maintenance suggestions.

[0093] This step is the final decision-making layer in the entire analysis process. Its role is to introduce the semantic understanding and causal reasoning capabilities of the large model based on the preliminary conclusions output by the secondary DS fusion, to verify the rationality of the fusion conclusions, optimize the risk level, and generate operation and maintenance decisions. Before S400 is executed, evidence source preprocessing (S410-S411) needs to be completed to ensure that the data input to the DS fusion engine has a unified format and reasonable weight allocation.

[0094] Furthermore, prior to the multi-source evidence fusion calculation, an evidence source preprocessing step is also included: S410 assigns fusion weights to each individual state evidence source according to a preset weight allocation rule; among them, the partial discharge state has the highest fusion weight, followed by the temperature state, and the environmental state has the lowest.

[0095] Based on the ability of each monitoring dimension to characterize the equipment status and its fault sensitivity, a fusion weight is preset for the three types of evidence sources: partial discharge individual status has the highest weight (0.5), temperature individual status is the second highest (0.3), and environmental individual status has the lowest weight (0.2). The sum of the three weights is 1.

[0096] The reason partial discharge (PD) has the highest weight is that it is a direct indicator of equipment insulation degradation. Synchronous anomalies in three types of signals—ultra-high frequency, ultrasonic, and ground wave—often point to actual insulation defects within the equipment, making it a core indicator for assessing equipment health. Temperature, as a reflection of equipment operating conditions, is sensitive to load changes and faults such as poor contact, but is easily affected by ambient temperature and load fluctuations, hence its second-highest weight. Environmental monitoring reflects the external operating conditions of the equipment and has the weakest direct ability to determine the equipment's physical condition, but it can serve as an auxiliary basis for interference identification and fault tracing, hence its lowest weight.

[0097] S411 encapsulates the state level, confidence level, and abnormal characteristic values ​​in each individual state result into a standardized evidence body according to a unified data format, eliminating the dimensional differences between various monitoring data types.

[0098] Each individual state result is packaged into a standardized evidence body according to a unified data format. The data structure of each evidence body contains at least three fields: a state level field, which records one of the four states determined by the evidence source; a confidence level field, which records the degree of certainty of the evidence source regarding its determination result, taken from the confidence level value output by the first-level analysis step; and an anomaly feature value field, which records the key anomaly indicators extracted by the evidence source during the determination process—for example, temperature evidence sources record real-time temperature values ​​and temperature rise rates, partial discharge evidence sources record partial discharge amplitudes and discharge frequencies in various dimensions, and environmental evidence sources record specific values ​​of temperature and humidity and smoke concentration values.

[0099] The core function of standardized packaging is to eliminate the differences in dimensions and formats among the three types of monitoring data. Temperature data is in °C, partial discharge data is in dBm, dB, and dBmV, and environmental data is in °C, %RH, and mg / m³. 3 The three types of data have completely different physical meanings and numerical ranges, making them unsuitable for direct input into the same fusion model. By encapsulating them with a unified data structure, the judgment conclusions of each evidence source are abstracted into a triplet form of "state + confidence + feature," enabling subsequent DS fusion calculations to be performed at a unified abstract level without needing to concern themselves with the physical units and dimensions of the underlying data.

[0100] After completing the secondary DS fusion and outputting the preliminary fusion status results of the devices, the system calls the preset large model API to enter the intelligent inference stage. During the call, the system packages and sends the following data to the large model API: the preliminary fusion status conclusion of the devices (including the probability assignment of each of the four states and the final selected state), the historical operating status sequence of the devices (covering historical data of temperature, partial discharge, and environment over the past few weeks or months), historical fault records (the types of faults that have occurred in the past and the handling records), historical maintenance data (the time, content, and results of each maintenance), and the preset operation and maintenance rule base and fault case base.

[0101] Furthermore, the intelligent reasoning process of the large model API includes the following dimensions: Historical data correlation reasoning: Extract the equipment's historical operating status sequence, historical fault records, and historical maintenance data to determine whether the current abnormal state is a persistent defect or periodic interference.

[0102] The large model extracts the historical operating status sequence of the equipment and performs time-series comparison analysis between the current abnormal state and historical data. The specific judgment logic is as follows: if the current abnormal characteristics are similar to the evolution trajectory before a certain historical fault occurred—for example, the temperature shows a continuous upward trend and the partial discharge amplitude increases week by week—it is judged as a persistent defect and needs to be dealt with first; if the current abnormality matches the periodic fluctuations in historical data—for example, the temperature briefly exceeds the standard during the high temperature period every summer but then recovers on its own—it is judged as periodic interference, and the warning level can be reduced.

[0103] Environmental correlation simulation: The current ambient temperature, humidity and smoke concentration data are combined with the preliminary fusion status results of the equipment to distinguish between real internal faults and false triggering by external environmental interference.

[0104] The large model jointly analyzes the current ambient temperature, humidity, and smoke concentration data with the preliminary fusion conclusions. The reasoning logic is as follows: if the partial discharge anomaly increases synchronously with the ambient humidity and the UHF signal characteristics match the surface discharge mode, it may be a surface discharge caused by a humid environment, belonging to an environment-induced anomaly; if the partial discharge anomaly still exists when the environmental data is normal and the discharge spectrum shows internal discharge characteristics, it is determined to be a true internal fault of the equipment. The goal of this dimension of reasoning is to distinguish between "equipment malfunction" and "false triggering caused by the external environment," avoiding misjudgments due to environmental interference.

[0105] Fault source tracing reasoning: Input the partial discharge anomaly features and temperature anomaly features into the large model API, match them with the power equipment fault knowledge base, and output the fault location, fault type classification, and fault development trend prediction.

[0106] The large-scale model inputs partial discharge anomaly features (amplitude of various dimensions, discharge frequency, PRPD spectrum features) and temperature anomaly features (real-time temperature, temperature rise rate) and matches them with a power equipment fault knowledge base. The fault knowledge base pre-stores the "feature-cause-location" mapping relationships for various typical faults—for example, "strong UHF signal + weak ultrasonic signal + normal temperature" corresponds to "internal floating potential discharge," and "strong UHF signal + strong ultrasonic signal + high temperature" corresponds to "internal air gap discharge in insulation accompanied by heating." The large-scale model retrieves the fault entry from the knowledge base that best matches the current features through semantic matching and similarity calculation, outputting fault location (e.g., "cable joint phase A"), fault type classification (e.g., "insulation aging type partial discharge"), and fault development trend prediction (e.g., "the degradation rate is expected to accelerate within the next two weeks").

[0107] Risk level confidence correction: Based on preset industry operation and maintenance standards and similar equipment failure cases, the probability values ​​of each state in the preliminary fusion state results of the equipment are corrected to optimize the final state level.

[0108] The large model, incorporating pre-defined industry operation and maintenance standards and historical failure cases of similar equipment, corrects the probability assignments for each state in the DS fusion output. The correction logic is as follows: if the large model determines, through historical data analysis and fault tracing, that the current anomaly has a high risk of development, the probability assignment for the corresponding severe state is increased; otherwise, it is appropriately decreased. The corrected probability assignments are used to determine the final state level.

[0109] Maintenance suggestion generation: Based on the final equipment status level and fault tracing results, match the preset handling strategy library, and output maintenance suggestion text including maintenance priority, inspection cycle adjustment suggestions and early warning handling measures.

[0110] The large model matches a pre-defined handling strategy library based on the final determined equipment status level and fault tracing results. The handling strategy library contains pre-set handling plan templates for each combination of "status level + fault type". After selecting a matching item from the template, the large model instantiates and populates it with the specific parameters of the current equipment (equipment number, installation location, fault characteristic value), generating maintenance suggestion text containing the following: maintenance priority (urgent / high / medium / low), inspection cycle adjustment suggestions (e.g., "adjust the inspection cycle from once a month to once a week"), and early warning handling measures (e.g., "suggest arranging partial discharge retesting within 72 hours"). The generated suggestion text is output in natural language, directly readable and executable by maintenance personnel.

[0111] This large-scale model's API-assisted reasoning step introduces an independent intelligent reasoning stage after the initial conclusions of the DS fusion output. Through historical data correlation reasoning, it examines the current anomaly within the operational trajectory of the equipment's entire lifecycle, enabling the identification of the essential difference between persistent defects and periodic interference, thus compensating for the lack of time dimension consideration in single fusion judgments. Environmental correlation inference incorporates external operating condition data into the analysis scope, effectively distinguishing between genuine internal equipment faults and environmentally induced false triggers, reducing invalid alarms caused by environmental interference. Fault source tracing reasoning matches the power equipment fault knowledge base, transforming the fusion conclusion from "the equipment is in..." to... The "abnormal state" is further refined to "what fault caused the equipment to be in an abnormal state," giving the assessment results actionable engineering value. The risk level confidence correction stage combines industry standards and similar cases to review and optimize the fusion probability assignment, ensuring that the final state level determination is based on both data-driven fusion results and knowledge-driven expert experience. The operation and maintenance suggestion generation stage automatically converts the assessment conclusions into natural language text containing maintenance priorities, inspection cycle adjustments, and handling measures, directly serving on-site operation and maintenance decisions, and realizing a closed-loop process from "data collection" to "state determination" to "handling suggestions."

[0112] S500, outputs a comprehensive assessment result including the final device status level.

[0113] The comprehensive assessment results should include at least the following five pieces of information: The first item is the final status level of the equipment. This status level is selected from four levels: normal state, alert state, abnormal state, and critical fault state, and serves as the final conclusion of this assessment. The final status level is determined based on the probability assignments for each state after the large model is corrected, and the state corresponding to the highest probability is taken as the final output. This status level is the most direct basis for maintenance personnel to judge the current health status of the equipment.

[0114] The second item is the overall assessment confidence level. The overall assessment confidence level is calculated by assigning a probability value corresponding to the final selected state, output as a percentage or decimal. This value reflects the credibility of the assessment conclusion—a higher confidence level indicates more consistent support from various evidence sources for the conclusion, making the conclusion more reliable; a lower confidence level indicates some disagreement among evidence sources, resulting in greater uncertainty in the conclusion, requiring maintenance personnel to verify the conclusion in conjunction with the on-site situation during actual handling.

[0115] The third item is the multi-source data analysis details. This details list the intermediate results of the entire analysis process in tabular or structured data form, including at least: the state level of each individual component (the state determined by temperature, partial discharge, and environment), the confidence level of each individual component (the degree of confidence each individual algorithm has in its own judgment), the fusion weight of each individual component (temperature 0.3, partial discharge 0.5, environment 0.2), and the evidence probability assignments for each of the four states after DS fusion (normal state probability, attention state probability, abnormal state probability, and severe fault state probability). The purpose of the multi-source data analysis details is to make the formation process of the final conclusion completely traceable—operations personnel can check the data and judgment basis of each step item by item to confirm the rationality of the conclusion.

[0116] The fourth item is large-scale model fault tracing analysis and risk assessment. This item is presented in the form of natural language text output by the large-scale model, covering: fault location (e.g., "phase A cable joint"), fault type classification (e.g., "insulation aging type partial discharge"), fault development trend prediction (e.g., "the rate of degradation is expected to accelerate within the next two weeks"), and comprehensive risk level (e.g., "high risk"). This part transforms the quantified status level into specific fault descriptions and risk assessments, making it easier for maintenance personnel to understand "what went wrong with the equipment" and "how serious the problem is".

[0117] The fifth item is the operation and maintenance handling strategy and early warning level. This item is also presented in the form of natural language text output by the large model, and the content covers: maintenance priority (urgent / high / medium / low), inspection cycle adjustment suggestions (such as "adjust the inspection cycle from once a month to once a week"), and early warning handling measures (such as "it is recommended to arrange partial discharge retest within 72 hours"). This part directly translates the equipment status assessment conclusions into executable operation and maintenance work instructions.

[0118] The comprehensive assessment output process aggregates all calculation conclusions from three levels—first-level individual device assessment, second-level DS fusion, and large-scale model intelligent correction—into a single output structure. The final device status level provides a clear health status conclusion, and the comprehensive assessment confidence level reflects the credibility of the conclusion. The multi-source data assessment details fully record the entire process data from the input of each individual device status to the DS fusion probability output and then to the correction, enabling maintenance personnel to verify any intermediate conclusion by comparing it with the original data, ensuring that the final decision is traceable and verifiable. The fault source analysis and maintenance suggestions generated by the large model further transform the status level into specific fault descriptions, risk assessments, and handling measures, allowing maintenance personnel to directly execute corresponding operations without having to interpret the data themselves, thus shortening the response time from alarm to handling.

[0119] Furthermore, although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0120] Embodiments of the present invention also provide a power equipment status assessment system based on multi-source data fusion, applied to the method described in the above embodiments, the system comprising: The data acquisition module is used to collect real-time temperature monitoring data, partial discharge three-in-one monitoring data, and environmental temperature, humidity, and smoke monitoring data from primary power equipment.

[0121] The first-level single-unit analysis module includes a temperature monitoring submodule, a partial discharge three-in-one monitoring submodule, and an environmental temperature, humidity, and smoke monitoring submodule. Among them, the temperature monitoring submodule has a built-in hierarchical decision tree branching algorithm unit, which is used to perform multi-level branching judgment based on the real-time temperature and short-term temperature rise rate of the equipment, and output the temperature single-unit status and the corresponding temperature confidence and temperature anomaly characteristic value.

[0122] The partial discharge three-in-one monitoring submodule has a built-in weighted threshold fusion unit and a fault tree tracing unit. The weighted threshold fusion unit is used to perform multi-dimensional weighted fusion of UHF partial discharge data, ultrasonic partial discharge data and ground wave partial discharge data and output a comprehensive partial discharge risk score and the status of the partial discharge unit. The fault tree tracing unit is used to perform fault hierarchical tracing when the comprehensive partial discharge risk score exceeds the standard and output a partial discharge fault tracing label.

[0123] The environmental temperature, humidity and smoke monitoring submodule has a built-in multi-parameter multi-level threshold determination unit, which is used to independently determine the threshold and perform multi-parameter linkage scoring on the real-time values ​​of environmental temperature, environmental humidity and smoke concentration, and output the status of individual environmental units and the corresponding environmental confidence and environmental anomaly characteristic values.

[0124] The secondary fusion analysis module is connected to the primary single-unit analysis module. It has a built-in improved DS evidence theory fusion unit, which is used to convert the temperature single-unit state, partial discharge single-unit state and environmental single-unit state into standardized evidence bodies respectively. After assigning basic probability values ​​to the three sets of evidence sources, it sequentially performs two-evidence iterative synthesis calculation, outputs the probability values ​​of each state, and selects the state corresponding to the maximum probability as the preliminary fusion state result of the device.

[0125] The large model interface module connects to the secondary fusion analysis module and is used to establish a communication connection with the external large model API. It packages and sends the preliminary fusion status results of the device, historical operating data of the device, operation and maintenance rule library, and fault case library to the large model API, and receives the intelligent inference results returned by the large model API. It performs confidence correction and risk assessment optimization on the preliminary fusion status results of the device, and outputs the final device status level, risk assessment results and operation and maintenance suggestions.

[0126] The result output module, connected to the large model interface module, is used to output a comprehensive judgment result including the final device status level, comprehensive judgment confidence level, multi-source data judgment details, and operation and maintenance handling strategy. The first-level single-unit judgment module, the second-level fusion judgment module, the large model interface module, and the result output module are cascaded in sequence to form a closed-loop processing link of data acquisition, single-unit judgment, fusion decision-making, intelligent correction, and result output.

[0127] While specific embodiments of the invention have been described in detail by way of examples, those skilled in the art should understand that the examples are for illustrative purposes only and are not intended to limit the scope of the invention. Those skilled in the art should also understand that various modifications can be made to the embodiments without departing from the scope and spirit of the invention.

Claims

1. A method for determining the status of power equipment based on multi-source data fusion, characterized in that, include: Real-time acquisition of temperature monitoring data, partial discharge three-in-one monitoring data, and ambient temperature, humidity, and smoke monitoring data from primary power equipment; Each of the temperature monitoring data, partial discharge three-in-one monitoring data, and environmental temperature, humidity and smoke monitoring data is used to independently determine the single-dimensional state, output the temperature single-unit state, partial discharge single-unit state and environmental single-unit state, and generate the confidence level and abnormal feature value corresponding to each single-unit state. The results of various individual states are transformed into standardized evidence bodies. Multi-source evidence fusion calculations are performed based on the improved DS evidence theory algorithm to generate preliminary fusion state results of the equipment. Call the preset large model API, input the preliminary fusion status conclusion of the device, and combine the device's historical operating data, operation and maintenance rule base and fault case base to perform intelligent reasoning, and output the corrected final device status level, risk assessment result and operation and maintenance suggestions; The output includes a comprehensive assessment result of the final device status level.

2. The method for determining the status of power equipment based on multi-source data fusion according to claim 1, characterized in that, The dedicated sub-algorithm corresponding to the temperature monitoring data uses a hierarchical decision tree branching algorithm for analysis, specifically including: The decision tree uses real-time equipment temperature, normal equipment temperature threshold, equipment warning temperature threshold, equipment fault temperature threshold, and short-term temperature rise rate as branching characteristics to perform state filtering layer by layer; wherein, the short-term temperature rise rate is calculated by dividing the difference between the highest and lowest temperatures within the statistical period by the statistical duration. Multi-level bifurcation determination is performed based on the real-time temperature and the short-time temperature rise rate: When the real-time temperature does not exceed the normal temperature threshold and the short-term temperature rise rate does not exceed the first rate threshold, it is determined to be in a normal state. When the real-time temperature is between the normal temperature threshold and the warning temperature threshold and the short-term temperature rise rate does not exceed the second rate threshold, it is determined to be in a state of alert. When the real-time temperature is between the warning temperature threshold and the fault temperature threshold, or when the short-term temperature rise rate exceeds the second rate threshold, it is determined to be an abnormal state. When the real-time temperature exceeds the fault temperature threshold, it is determined to be a serious fault state; wherein, the first rate threshold is lower than the second rate threshold.

3. The method for determining the status of power equipment based on multi-source data fusion according to claim 1, characterized in that, The dedicated sub-algorithm corresponding to the partial discharge three-in-one monitoring data uses a multi-dimensional weighted threshold fusion algorithm for analysis, specifically including: Real-time monitoring values ​​from three dimensions—UHF partial discharge data, ultrasonic partial discharge data, and ground wave partial discharge data—are extracted, and corresponding weight values ​​are assigned to each dimension, with UHF having the highest weight. The excess of the real-time monitoring value of each dimension of partial discharge index relative to its normal threshold and fault threshold is calculated, and the excess of each dimension is normalized to a uniform numerical range. The normalized overscaling values ​​of each dimension are multiplied by their corresponding weight values ​​and then summed to obtain the comprehensive risk score of partial discharge. Based on the numerical range of the partial discharge comprehensive risk score, the status of a single partial discharge cell is determined. The lower the score, the better the status, and the higher the score, the worse the status.

4. The method for determining the status of power equipment based on multi-source data fusion according to claim 3, characterized in that, After completing the weighted threshold fusion, the dedicated sub-algorithm corresponding to the partial discharge three-in-one monitoring data also includes a fault tree hierarchical tracing step: Using the partial discharge comprehensive risk score exceeding the preset abnormal threshold as the top event, the fault tree is branched down layer by layer to the insulation aging bottom event, the partial discharge bottom event, the electromagnetic interference bottom event, and the equipment poor contact bottom event; Based on the membership degree of the fault characteristics corresponding to each event, the fault cause corresponding to the current partial discharge monitoring data is matched in reverse, and the partial discharge fault source label is output.

5. The method for determining the status of power equipment based on multi-source data fusion according to claim 1, characterized in that, The dedicated sub-algorithm corresponding to the environmental temperature, humidity, and smoke monitoring data employs a multi-parameter, multi-level threshold judgment algorithm for analysis, specifically including: Real-time values ​​of ambient temperature, ambient humidity, and smoke concentration were collected as three independent parameters. Determine whether each independent parameter is within its own normal threshold range. For parameters that exceed the normal threshold range, calculate the relative deviation from the standard optimal value, which is used as the single parameter risk value for that parameter. Weight values ​​were assigned to ambient temperature, ambient humidity, and smoke concentration, with smoke concentration having the highest weight. The comprehensive environmental risk score is calculated by multiplying the individual risk values ​​of each parameter by their corresponding weight values ​​and then summing them up. Based on the comprehensive environmental risk score, a judgment logic of single-parameter anomaly early warning and multi-parameter linkage upgrade is adopted to output the status of individual environmental units.

6. The method for determining the status of power equipment based on multi-source data fusion according to claim 1, characterized in that, The multi-source evidence fusion calculation based on the improved DS evidence theory algorithm specifically includes: Establish an identification framework that includes four states: normal state, attention state, abnormal state, and critical fault state; The temperature-dependent monomer state, the partial discharge-dependent monomer state, and the environmental monomer state are respectively regarded as three independent evidence sources, and basic probability values ​​are assigned to each state level according to the confidence level corresponding to each monomer state. First, the temperature evidence source and the partial discharge evidence source are combined into two pieces of evidence. The conflict coefficient between the two pieces of evidence is calculated. Based on the conflict coefficient, the basic probability values ​​of the two pieces of evidence are weighted and fused to obtain an intermediate fusion result. The intermediate fusion result is combined with the environmental evidence source again to synthesize the two pieces of evidence. After iterative fusion, the final probability values ​​of each state are output. The state corresponding to the maximum probability is selected as the initial fusion state result of the device.

7. The method for determining the status of power equipment based on multi-source data fusion according to claim 6, characterized in that, During the synthesis of the two pieces of evidence, when the conflict coefficient between the two sets of evidence exceeds a preset conflict threshold, an adaptive weight correction is triggered. Reduce the adaptive weight value corresponding to conflicting evidence to weaken the impact of the evidence on the fusion result; At the same time, the weighting of the average probability assignment of multiple pieces of evidence is increased to smooth out the impact of conflicting evidence; The value of the adaptive weight is inversely proportional to the conflict coefficient; that is, the more intense the conflict, the smaller the value of the adaptive weight.

8. The method for determining the status of power equipment based on multi-source data fusion according to claim 1, characterized in that, The intelligent reasoning process of the large model API includes the following dimensions: Historical data correlation reasoning: Extract the equipment's historical operating status sequence, historical fault records, and historical maintenance data to determine whether the current abnormal state is a persistent defect or periodic interference; Environmental correlation simulation: The current ambient temperature, humidity and smoke concentration data are combined with the preliminary fusion status results of the equipment to distinguish between real internal faults and false triggering by external environmental interference. Fault source tracing reasoning: Input the partial discharge anomaly features and temperature anomaly features into the large model API, match them with the power equipment fault knowledge base, and output the fault location, fault type classification and fault development trend prediction. Risk level confidence correction: Based on preset industry operation and maintenance standards and similar equipment failure cases, the probability values ​​of each state in the preliminary fusion state results of the equipment are corrected to optimize the final state level; Maintenance suggestion generation: Based on the final equipment status level and fault tracing results, match the preset handling strategy library, and output maintenance suggestion text including maintenance priority, inspection cycle adjustment suggestions and early warning handling measures.

9. The method for determining the status of power equipment based on multi-source data fusion according to claim 1, characterized in that, Prior to the multi-source evidence fusion calculation, an evidence source preprocessing step is also included: According to the preset weight allocation rules, each individual state evidence source is assigned a fusion weight; among them, the partial discharge individual state has the highest fusion weight, followed by the temperature individual state, and the environmental individual state has the lowest fusion weight. The status level, confidence level, and abnormal feature values ​​in each individual status result are packaged into a standardized evidence body according to a unified data format to eliminate the dimensional differences between various monitoring data types.

10. A power equipment status assessment system based on multi-source data fusion, characterized in that, The system, applied to the method of any one of claims 1 to 9, comprises: The data acquisition module is used to collect real-time temperature monitoring data, partial discharge three-in-one monitoring data, and environmental temperature, humidity, and smoke monitoring data from primary power equipment. The primary unit analysis module includes a temperature monitoring submodule, a partial discharge three-in-one monitoring submodule, and an environmental temperature, humidity, and smoke monitoring submodule; The temperature monitoring submodule has a built-in hierarchical decision tree branching algorithm unit, which is used to make multi-level branching judgments based on the real-time temperature and short-term temperature rise rate of the equipment, and output the temperature unit status and the corresponding temperature confidence and temperature anomaly characteristic values. The partial discharge three-in-one monitoring submodule has a built-in weighted threshold fusion unit and a fault tree tracing unit. The weighted threshold fusion unit is used to perform multi-dimensional weighted fusion of UHF partial discharge data, ultrasonic partial discharge data and ground wave partial discharge data and output a comprehensive partial discharge risk score and the status of the partial discharge unit. The fault tree tracing unit is used to perform fault hierarchical tracing when the comprehensive partial discharge risk score exceeds the standard and output a partial discharge fault tracing label. The environmental temperature, humidity and smoke monitoring submodule has a built-in multi-parameter multi-level threshold determination unit, which is used to independently determine the threshold and multi-parameter linkage score of the real-time values ​​of environmental temperature, environmental humidity and smoke concentration, and output the status of individual environmental units and the corresponding environmental confidence and environmental anomaly characteristic values. The secondary fusion analysis module is connected to the primary single-unit analysis module. It has a built-in improved DS evidence theory fusion unit, which is used to convert the temperature single-unit state, partial discharge single-unit state and environmental single-unit state into standardized evidence bodies respectively. After assigning basic probability values ​​to the three sets of evidence sources, it sequentially performs two-evidence iterative synthesis calculation, outputs the probability assignment of each state, and selects the state corresponding to the maximum probability value as the preliminary fusion state result of the device. The large model interface module connects to the secondary fusion analysis module and is used to establish a communication connection with the external large model API. It packages and sends the preliminary fusion status results of the device, historical operating data of the device, operation and maintenance rule library, and fault case library to the large model API, and receives the intelligent inference results returned by the large model API. It performs confidence correction and risk assessment optimization on the preliminary fusion status results of the device, and outputs the final device status level, risk assessment results and operation and maintenance suggestions. The result output module is connected to the large model interface module and is used to output the comprehensive judgment result, which includes the final equipment status level, comprehensive judgment confidence, multi-source data judgment details and operation and maintenance handling strategy. The primary unit analysis module, the secondary fusion analysis module, the large model interface module, and the result output module are cascaded in sequence to form a closed-loop processing link of data acquisition, unit analysis, fusion decision-making, intelligent correction, and result output.

Images