Fault prediction method for mining high-voltage power distribution device

Abstract

The invention relates to a fault prediction method for a mining high-voltage power distribution device. Temperature, current, voltage and gas leakage sensors are installed in the mining high-voltage power distribution device; integrating the multi-dimensional data by using a multi-sensor data fusion algorithm to form a real-time operation portrait of the power distribution device; extracting time sequence characteristics of temperature and voltage key variables by combining an operation rule of the high-voltage power distribution device; capturing abnormal fluctuation of the key characteristic parameters by using a sequence analysis model, and identifying a specific fault type in the high-voltage environment; constructing a feature mode database based on a time sequence, and matching new data with a historical fault mode; applying a feature selection algorithm to extract representative fault prediction features from the sensor data; establishing a fault prediction risk level model according to the hazard degrees of different fault modes; based on the fault risk level predicted in real time, an optimized self-adaptive maintenance scheme is generated, and the maintenance period and strategy are dynamically adjusted in combination with the actual operation state of the equipment.

Description

Technical Field

[0001] The invention relates to a fault prediction method for a high-voltage power distribution device for mining. Background Art

[0002] Although the fault prediction method for high-voltage power distribution devices for mines has achieved certain results in practical applications, it still has many shortcomings and disadvantages compared with the present invention, mainly in terms of fault identification accuracy, adaptability, response speed, and effectiveness of maintenance decisions. First of all, the traditional fault prediction method for high-voltage power distribution devices for mines is usually based on a single sensor or limited multi-sensor data for analysis. The singleness of the data source leads to the system's weak ability to capture fault states. For example, only using temperature or voltage sensors to monitor the operating status of the distribution device cannot fully reflect the multi-dimensional changes of the equipment in a high-voltage environment. The data dimension monitored by a single sensor is limited and is easily affected by external interference, such as changes in ambient temperature and humidity, resulting in unstable data quality and affecting the accuracy and reliability of fault prediction. In addition, traditional methods mostly use fixed weights to fuse and analyze sensor data, and do not dynamically adjust the weights according to the health status of the sensor, resulting in the system being unable to automatically reduce the weight of the sensor data when data drift or noise occurs in some sensors, causing data fusion deviations and affecting the accuracy of the operation portrait.

[0003] Secondly, in terms of fault mode identification, traditional methods usually use simple time series data analysis methods to detect abnormal fluctuations, and lack the ability to deeply identify specific fault modes. Existing methods often rely on fixed window time series analysis, and the length and weight of the sliding window cannot be dynamically adjusted according to the changes in fault characteristics, resulting in difficulty in accurately identifying the early characteristics of these faults when dealing with different fault modes such as insulation aging, temperature anomalies, and load overload. For example, for fault characteristics such as insulation aging, which are mainly manifested as high-frequency signal fluctuations, traditional time series analysis methods find it difficult to capture changes in high-frequency components, thereby delaying the timing of fault warning.

[0004] In addition, the existing fault prediction methods have poor adaptability in complex mine environments, mainly reflected in the in-depth understanding and processing of equipment operation data. Traditional methods usually use linear analysis and simple weighted average methods to fuse data, which makes it difficult to fully extract the deep correlation of high-dimensional data and ignores the nonlinear correlation between different sensors and the law of data fluctuation in complex environments. This method is often ineffective when applied in mines with changeable environments and high-dimensional data.

[0005] In terms of the effectiveness of maintenance decisions, traditional methods are often based on fixed maintenance strategies and lack the ability to dynamically adjust maintenance cycles and strategies based on real-time fault prediction results. Fixed maintenance cycles cannot flexibly respond to different operating conditions of equipment, resulting in minor faults that may lead to premature maintenance of equipment, while major faults are not handled in a timely manner, increasing maintenance costs and the risk of equipment downtime. Summary of the invention

[0006] The purpose of the present invention is to provide a fault prediction method for a high-voltage power distribution device for mining, thereby solving some of the drawbacks and shortcomings pointed out in the background technology.

[0007] The present invention solves the above-mentioned technical problems by adopting the following technical solutions: A fault prediction method for a high-voltage power distribution device for mining, comprising: S1, multi-dimensional sensor data fusion based on high-voltage power distribution characteristics:

[0008] S1.1. Install temperature, current, voltage and gas leakage sensors in the mine high voltage distribution device;

[0009] S1.2. Use multi-sensor data fusion algorithm to integrate multi-dimensional data to form a real-time operation portrait of the distribution device;

[0010] S2. Time series analysis of specific failure modes:

[0011] S2.1. Combine the operation rules of high-voltage power distribution equipment to extract the time series characteristics of key variables such as temperature and voltage;

[0012] S2.2. Use sequence analysis models to capture abnormal fluctuations in key characteristic parameters and identify fault types unique to high-voltage environments, including insulation aging, temperature anomalies, and load overload;

[0013] S2.3, build a time series-based feature pattern database to match new data with historical failure patterns and quickly identify known failure trends;

[0014] S3. Key feature extraction and risk level analysis in fault prediction:

[0015] S3.1, apply feature selection algorithm to extract representative fault prediction features from sensor data;

[0016] S3.2. According to the degree of harm of different fault modes, a fault prediction risk level model is established, which is divided into three levels: low, medium and high, to evaluate the potential threat of faults to high-voltage distribution equipment;

[0017] S4, Fault prediction and adaptive maintenance optimization closed loop:

[0018] S4.1. Generate optimized adaptive maintenance plans based on the real-time predicted fault risk level, and dynamically adjust maintenance cycles and strategies based on the actual operating status of the equipment.

[0019] Furthermore, the method for forming a real-time operation portrait of a power distribution device includes:

[0020] The dynamic weight adaptive fusion algorithm based on the health status of the sensor is introduced. i and data quality factor Q i To dynamically adjust the weight of each sensor, the specific calculation method is as follows:

[0021]

[0022] Among them, H i represents the health metric of sensor i, which is calculated by integrating the rate of change of temperature and voltage over time t′; T i (t′) and V i (t′) represent the temperature and voltage data of sensor i at time t′ respectively; function f(T i (t′)) represents the quality function of temperature data, which is used to evaluate abnormal temperature fluctuations. The function g(V i (t′)) represents the quality function of voltage data, which is used to capture the high-order changes of voltage; coefficients α and β are adjustment parameters, which adjust the weights of temperature and voltage data according to the sensor type and the credibility of the data.

[0023] Furthermore, the method for forming a real-time operation portrait of a power distribution device includes:

[0024] The spatial and temporal correlation of sensor data is analyzed through spatiotemporal clustering to identify anomalies in fault detection. In spatiotemporal correlation analysis, the spatial correlation weight W is used. 空间 , is calculated comprehensively through geographical location distance and time correlation. The specific formula is as follows:

[0025]

[0026] Among them, W 空间 represents the spatial association weight in the fusion result; D 地理 (x) is the geographical location distance of the sensor, h(D 地理 (x)) is a function of distance, which is used to measure the impact of geographical distance on data fusion; T 关联 (t′) is the correlation in the time dimension, k(T 关联 (t′)) represents the time correlation function, which is used to capture the rate of change of data in time; the coefficients σ and γ are adjustment parameters, corresponding to the adjustment of spatial and temporal correlation weights, respectively.

[0027] Furthermore, the method for forming a real-time operation portrait of a power distribution device includes:

[0028] A deep neural network model is used for feature extraction and fusion. A deep neural network is used to perform nonlinear feature mapping on multidimensional data. The feature vector F of each sensor is generated through the feature mapping function φ(x). i , and then fuse at the feature level; finally, the fused feature vector F 融合 The calculation is as follows:

[0029]

[0030] Among them, F 融合 is the fused high-dimensional running state feature vector; W i represents the dynamic weight of each sensor; φ(F i (x)) is the feature vector generated by the neural network, K(x) is the kernel function, which is used to amplify the weight of key features in the fusion process; the symbol * represents the convolution operation, which is used to capture the local correlation in the feature vector.

[0031] Furthermore, the time series analysis method of the specific failure mode includes:

[0032] By calculating the change rate of temperature and voltage sensor data in real time, the length and weight of the sliding window are dynamically adjusted to adapt to the changes in different fault characteristics, including temperature anomalies and load overload. The weight calculation formula of the adaptive weighted sliding window is as follows:

[0033]

[0034] Where: W i (t) represents the window weight at time t, which is used to quantify the contribution of the current data of sensor i to fault identification; T(t′) and V(t′) are the time series data of temperature and voltage, respectively, reflecting the changes of these parameters in the high-voltage environment; functions P(T) and Q(V) are the response functions of temperature and voltage data, respectively used to quantify the sensitivity of temperature and voltage data to abnormal fluctuations; α and β are adjustment parameters used to balance the different effects of temperature and voltage on fault prediction; Δt represents the length of the sliding window, and the integration operation within the window calculates the rate of change of temperature and voltage over a period of time.

[0035] Furthermore, the time series analysis method of the specific failure mode includes:

[0036] Wavelet transform is introduced to perform multi-level decomposition of sensor signals; the signal is decomposed into high-frequency and low-frequency components to identify the specific high-frequency signal fluctuations of insulation aging; the early characteristics of insulation aging are captured by analyzing the decomposed high-frequency components. The specific high-frequency signal characteristic calculation formula is as follows:

[0037]

[0038] in:

[0039] S 高频 (t) represents the high-frequency signal characteristics at time t, which is used to identify early signs of insulation aging; G k (x) is the wavelet basis function. The wavelet basis function of the kth layer is used to extract the high-frequency components of the signal and emphasize the high-frequency fluctuation characteristics in the data; V k (t) is the voltage signal sequence of the kth sensor at time t, reflecting the insulation aging characteristics; h(V k (t)) is the transformation function of the voltage signal, which represents the nonlinear fluctuation of the data during the insulation aging process; γ k is the weight of each component, representing the importance of each layer of wavelet components.

[0040] Furthermore, the time series analysis method of the specific failure mode includes:

[0041] Through the multi-dimensional trend prediction model, the trend changes of each feature are jointly analyzed to generate the fused trend feature value for anomaly detection; the formula for multi-feature fusion trend detection is as follows:

[0042]

[0043] Where: T 融合 Represents the trend feature value after fusion, which is used to capture the complex fault trend of load overload or temperature anomaly; F i (t′) is the time series trend data of the i-th feature, reflecting the change pattern of the fault type; λ i is the weight coefficient of each feature, indicating the importance of different features in the fusion process; t0 to t represents the detection time range, which integrates the trend changes of each feature to form an overall anomaly detection standard.

[0044] Beneficial effects of the present invention: 1. Improve the accuracy of fault identification: This method can monitor the operating status of the power distribution device in real time and form an accurate operating profile through multi-dimensional sensor data fusion and dynamic weight adaptive algorithm. With the help of dynamic adjustment of health metrics and data quality coefficients, the system can identify the reliability of different sensor data and automatically adjust the weights when drift or noise occurs to ensure the robustness of the fused data. This dynamic adaptability significantly improves the accuracy of fault identification.

[0045] 2. Rapidly capture specific fault modes: This method combines spatiotemporal clustering analysis with time series analysis of specific fault modes to accurately identify fault characteristics such as insulation aging, temperature anomalies, and load overload in power distribution devices. By extracting high-frequency components in the signal through wavelet transform, the system can capture specific high-frequency fluctuation signals at the early stage of insulation aging, thereby providing timely warnings before faults occur and avoiding safety hazards caused by faults.

[0046] 3. Enhance the ability to adapt to complex environments: In the high-pressure environment of mines, the complexity of equipment operation and the high dimensionality of data often make it difficult for traditional methods to accurately identify faults. This method uses nonlinear feature mapping of deep neural networks to achieve feature extraction and fusion of multidimensional data, which can effectively process the deep correlation between different sensor data, capture complex fault trends, and improve the coverage of fault prediction and the ability to cope with complex environments.

[0047] 4. Dynamically adjust maintenance strategies to achieve adaptive closed-loop optimization: This method generates optimized adaptive maintenance plans in real time based on the risk level of fault prediction, and dynamically adjusts maintenance cycles and strategies to ensure that maintenance decisions for equipment in different states are reasonable and effective. The optimization of the adaptive maintenance closed loop makes the preventive maintenance of equipment more accurate, avoids unnecessary maintenance interventions, and at the same time increases the service life of equipment and reduces the maintenance cost of mines. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The present invention is a flow chart of a fault prediction method for a high-voltage power distribution device for mining.

[0049] Figure 2 A flow chart of a method for forming a real-time operation profile of a power distribution device according to the present invention.

[0050] Figure 3 This is a flow chart of the time series analysis method for a specific fault mode of the present invention. DETAILED DESCRIPTION

[0051] The specific implementation modes of the present invention will be described in detail below in conjunction with the accompanying drawings.

[0052] Combined with Figure 1Flowchart, a fault prediction method for a high-voltage power distribution device for mining of the present invention, the first step is to construct a real-time operation portrait based on the high-voltage power distribution characteristics through multi-dimensional sensor data fusion technology, so as to monitor the operation status of the device and predict faults in advance. In this method, S1.1 installs a variety of sensors such as temperature, current, voltage and gas leakage at key positions of the high-voltage power distribution device for mining, with the purpose of capturing various state changes that occur in the high-voltage power distribution device during operation. These sensors can collect multi-dimensional data of the power distribution device in real time, including temperature rise, current fluctuation, voltage anomaly and potential signs of gas leakage, etc. These data reflect different aspects of the operation of the device and are also the key basis for fault prediction. S1.2 uses a multi-sensor data fusion algorithm to integrate these multi-dimensional data. Traditional methods are usually based on monitoring of a single data source, which is easy to miss some important information, while multi-sensor data fusion can comprehensively analyze various sensor data in time and space dimensions to achieve a comprehensive understanding of the operation status. The data fusion algorithm first pre-processes the data of each sensor, including filtering out noise, eliminating data drift, etc., and then integrates the data of different sensors through weighted fusion or deep learning models to generate a comprehensive real-time operation portrait. This operation portrait not only reflects the comprehensive status of the current device, but also has the ability to dynamically adjust and update in real time, and can effectively monitor the health status of the power distribution device in the high-voltage environment of the mine. Once a sensor data fluctuates abnormally, the system will identify potential risks and issue warnings through the fused data portrait, providing support for subsequent fault prediction and maintenance, thereby extending the life of the equipment and ensuring the safety and continuity of mine operations.

[0053] The second step is to systematically monitor and identify abnormal conditions of the equipment through time series analysis technology of specific fault modes. S2.1 Combined with the operating rules of high-voltage power distribution devices, the time series characteristics of key variables such as temperature and voltage are extracted. The changes in temperature and voltage are often periodic or regular in high-voltage equipment. In normal operation, the change pattern of these variables is relatively stable. By analyzing historical data, the typical fluctuation range and trend of temperature and voltage under normal operation are determined to form time series characteristics, which provide a reference for subsequent abnormal detection. S2.2 Use the sequence analysis model to monitor the fluctuations of these key characteristic parameters in real time, focusing on capturing abnormal changes in temperature and voltage. Typical faults in high-voltage environments include insulation aging, temperature anomalies, and load overload, which often lead to abnormal fluctuations in temperature or voltage curves. The sequence analysis model compares the current time series characteristics of the equipment with the standard characteristic pattern. When a parameter exceeds the normal fluctuation range, the system can identify this anomaly and classify it as a specific fault type. For example, a gradual increase in temperature indicates the aging of the insulation material, a sharp change in voltage is related to load overload, and a higher frequency of current changes is a sign of abnormal temperature. S2.3 builds a feature pattern database based on time series, matches the new data monitored with historical fault patterns, and quickly identifies known fault trends. By building this feature pattern database, a variety of known fault time series feature patterns can be stored, and the real-time data can be compared through a matching algorithm to quickly determine whether the current equipment operation status meets the known fault characteristics.

[0054] The third step is to systematically identify potential faults and assess their risk levels through key feature extraction and risk level analysis, providing a basis for maintenance decisions. S3.1 Apply feature selection algorithms to extract the most representative fault prediction features from the data collected by sensors. Mine high-voltage power distribution devices are usually equipped with a variety of sensors to continuously monitor temperature, voltage, current, gas leakage and other data. The amount of raw data generated by these sensors is large and multi-dimensional. If they are not properly screened, the complexity of the analysis will increase and the prediction efficiency will be affected. Through the feature selection algorithm, the system can effectively identify key features directly related to the occurrence of faults, such as temperature change rate, voltage fluctuation frequency, current amplitude abnormality, etc., so that the prediction model focuses on these key data and improves the accuracy and response speed of fault prediction. S3.2 After extracting the key features, the system will establish a fault prediction risk level model based on the degree of harm of different fault modes. The model divides fault risks into three levels: low, medium and high, and comprehensively evaluates the potential threats of faults to high-voltage power distribution devices. Low risk means that the equipment is in a slightly abnormal state, which will not have a direct impact on the current operation, but requires continuous monitoring; medium risk means that the equipment has shown an obvious failure trend and preventive maintenance needs to be arranged to prevent further deterioration; high risk means that the operating status of the equipment has seriously deviated from the normal range, posing a direct threat to mine safety and production continuity, and requires immediate maintenance. Through this risk level model, the system can issue early warnings in a timely manner according to the characteristics and impacts of different faults, and guide maintenance personnel to take targeted measures to ensure the safe and reliable operation of mine high-voltage power distribution equipment.

[0055] The fourth step is to dynamically adjust and optimize the equipment maintenance strategy by building a closed loop of fault prediction and adaptive maintenance optimization to ensure the efficient and safe operation of the equipment in the mine environment. S4.1 Generate an optimized adaptive maintenance plan based on the real-time predicted fault risk level obtained from the previous analysis. This plan combines the current operating status of the equipment, fault risk assessment and historical operating data, and continuously adjusts the maintenance strategy in an adaptive manner so that the maintenance work can accurately respond to the different risk levels of the equipment. For example, when the fault risk level of the equipment is assessed as low, the system will recommend extending the maintenance cycle and using conventional inspection methods; when the fault risk reaches medium, the system will generate a preventive maintenance plan, recommend shortening the maintenance cycle, and increase the focus on specific fault characteristics. Inspection; if the equipment is at a high risk level, the system will automatically adjust the maintenance plan, immediately arrange a comprehensive overhaul, and issue an emergency maintenance order to prevent the fault from further deteriorating and ensure the safety of mine production. In addition, the dynamic adjustment of the maintenance cycle and strategy will also be optimized based on the real-time operating data feedback of the equipment, that is, after each maintenance, the system will collect and analyze new operating data, combine the latest status information of the equipment with the risk prediction model, and update the maintenance plan to keep it consistent with the current status of the equipment.

[0056] Embodiment 1:

[0057] Combined with Figure 2 In this embodiment, a high-voltage power distribution device in a mine environment is equipped with a variety of sensors to monitor temperature, voltage, current and other data in real time to help capture early signs of faults. It is assumed that 10 temperature sensors and 10 voltage sensors are installed on a power distribution device in a mine, and the sampling frequency of sensor data is once per second to ensure the real-time nature of monitoring data. In order to achieve adaptive fusion of sensor data, the health metric H is used. i and data quality factor Q i To dynamically adjust the weight of each sensor to ensure that when the sensor drifts or errors occur, the system can promptly reduce the impact of its data on the overall operating profile.

[0058] The health metric H of sensor i i The calculation formula is set as:

[0059]

[0060] In this formula, the health metric H i It is obtained by integrating the rate of change of temperature and voltage, which helps to evaluate the health status of the sensor. The value range of α and β is set to 0.1≤α≤0.5 and 0.1≤β≤0.3. The specific value can be fine-tuned according to the type of sensor, data reliability, and stability of different sensors.

[0061] In a certain mining high-voltage power distribution device, the temperature and voltage data collected by sensor i in the time interval t = 0 to t = 10 seconds are as follows: Temperature T i (t′) grows steadily during the initial period, denoted as T i (t′)=20+0.5·t′(unit: degrees Celsius); and the voltage V i (t′) shows slight fluctuations in the same interval, represented by V i (t') = 220 + 0.2 sin (0.1 t') (unit: volt). On this basis, the quality function f (T i (t′)) = T i (t′)·0.05 is used to reflect the linear change of temperature and calculate its contribution to the health metric when α=0.3. The quality function of voltage data g(V i (t′))=V i (t′) 2 0.001, representing the change in the squared voltage term to capture higher-order fluctuations in the data, and setting β = 0.2 before calculating its contribution to the health metric.

[0062] Substitute the above temperature and voltage data into the formula for integration to calculate the health metric value H of sensor i during this period. i Under the conditions of α = 0.3 and β = 0.2, the contribution of the temperature change rate is obtained by integral calculation:

[0063]

[0064] The contribution of higher-order changes in voltage fluctuations is:

[0065]

[0066] The final health metric H i ≈0.079, this value represents the health level of the sensor during this period. i and data quality factor Q i By combining the sensors, the system can adjust the weights of each sensor in real time according to the health status of each sensor, forming a real-time operation profile of the integrated high-voltage distribution device.

[0067] In applications, as time goes by and the sensors age, some sensors gradually experience deviations or signal attenuation, which directly affects H i The value of the sensor is gradually reduced. At this time, the system will automatically reduce the weight W of the sensor. i , making the overall fusion result insensitive to sensors with poor data quality.

[0068] In order to further optimize the fault prediction method of the mine high-voltage power distribution device, spatiotemporal clustering analysis is introduced to more accurately capture the spatial and temporal correlation of sensor data, so as to more effectively identify abnormal situations in fault detection. It is assumed that the 10 temperature sensors and 10 voltage sensors arranged in the current mine environment are located at different key positions of the power distribution device, and these positions have a certain distance and circuit association in the spatial layout of the mine equipment. The system uses spatiotemporal correlation analysis to calculate the spatial correlation weight W. 空间 When clustering, the geographical distance and time correlation of sensors are comprehensively considered to form an effective clustering of the spatial characteristics of each sensor data.

[0069] The formula for spatiotemporal correlation analysis is as follows:

[0070]

[0071] Among them, W 空间 It is the weight result of the sensor based on spatiotemporal correlation during the fusion process, reflecting the influence of geographic location and time correlation on data fusion. 地理(x) represents the geographical distance of the sensors in space. For example, assume that two sensors are located between the main distribution cabinet and the auxiliary distribution cabinet of a mine, with a distance of 10 meters;

[0072]

[0073] Where λ = 5 is the attenuation coefficient, which is used to measure the importance of geographical distance to the data - that is, the greater the distance, the smaller the weight. By integrating the spatial dimension, the spatial contribution of this distance to the sensor fusion effect is obtained.

[0074] Let’s look at the time-related part. 关联 (t′) represents the data association in the time dimension, taking the sensor data sequence in the current time window. Assume that the data sampled by the sensor in the past 5 minutes show obvious correlation fluctuations, and the time correlation function k(T 关联 (t′)) = cos(0.1·t′), which is used to reflect the correlation period of the sensor reading. The second-order derivative operation is performed on the time change to capture the acceleration fluctuation of the data. The coefficients σ = 0.2 and γ = 0.3 control the influence weights of geographical distance and time correlation respectively, and the weights are adjusted according to the reliability of the sensor data in the mine environment. By substituting into the calculation, the integral result of the spatial part is approximately The result of the time part is Therefore, it can be concluded that W 空间 ≈0.634+0.15=0.784, which indicates the comprehensive weight influence of the sensor under the current geographical location and time association conditions.

[0075] Through the above calculations, the system can more intuitively determine which sensors have stronger correlations in space and time, and accordingly adjust the weights appropriately in the real-time operation portrait. This spatiotemporal correlation analysis can more effectively identify data anomalies caused by geographical distance and operation time series, thereby having a more real-time response capability in detecting faults such as current overload and temperature surge. It is assumed that after a long period of operation, the sensors near the main distribution cabinet show increased temperature fluctuations and faster frequency, while the sensor data at the auxiliary distribution cabinet is relatively stable. Through the higher spatiotemporal weight W 空间 The system can identify the initial signs of load concentration and insulation aging in the main distribution cabinet, thereby providing early warning and reminding maintenance personnel to conduct equipment inspection and maintenance.

[0076] In order to further improve the fault prediction accuracy of high-voltage power distribution equipment in mines, a deep neural network model is introduced for feature extraction and fusion. The real-time operation portrait is generated with the help of nonlinear feature mapping, and the deep correlation between complex data is captured through feature fusion. The temperature and voltage sensor data in the mine are set to be processed by a deep neural network to identify abnormal equipment status. The multi-dimensional data collected by the current sensor is set to be input into the neural network after preprocessing, and the nonlinear feature mapping function φ(x) of the deep neural network is used to extract key features. The deep neural network performs feature extraction on the input data of each sensor and generates a feature vector F i , that is, φ(F i (x)), this feature vector captures the complex internal relationship between temperature, voltage and other data. When processing temperature sensor data, the temperature fluctuation characteristics are mapped to F1(x) = φ(T), reflecting the potential failure risk during the temperature rise process, while the voltage sensor data is mapped to F2(x) = φ(V), which is used to reveal the relationship between voltage fluctuation and load change. The dynamic weight W of each sensor i It will be adjusted according to the previous spatiotemporal clustering analysis results, setting W1 = 0.6 and W2 = 0.4, reflecting the priority of the current temperature signal in the fusion process.

[0077] After generating the feature vectors of each sensor, the system will fuse these features at the feature level and calculate the final fused feature vector F 融合 To construct a high-dimensional operating status portrait of the device. This fusion process relies on the following formula:

[0078]

[0079] In this formula:

[0080] F 融合 Represents the fused high-dimensional operating status feature vector, which performs weighted convolution on the feature vectors of each sensor to generate the overall device status.

[0081] W i is the dynamic weight of each sensor, reflecting the importance of each sensor. Here W1 = 0.6 means that the weight of the temperature sensor is high, because the previous spatiotemporal analysis shows that the temperature fluctuation is more significant, indicating the risk of insulation aging caused by load concentration.

[0082] φ(F i (x)) is a feature vector generated by a neural network. For the temperature feature vector F1(x) = φ(T) and the voltage feature vector F2(x) = φ(V), the abnormal patterns of temperature and voltage are reflected respectively during the fusion process.

[0083] Kernel Function The weights used to amplify key features are used to extract local correlations in sensor features through convolution operations*, helping to more accurately identify the characteristics of sudden abnormalities, especially when the load increases sharply and the temperature rises sharply.

[0084] By substituting and calculating the data, the temperature feature F1(x) captured by the temperature sensor is set to reach a peak value of 0.8 at x=0, and the voltage feature F2(x) is set to take a value of 0.5 at x=0. Under the effect of , the weight amplification effect of temperature feature is more obvious. In the specific calculation, the fusion feature value

[0085]

[0086] The fusion results show that under the current weight and data distribution, temperature anomalies have a greater impact on the overall profile. Based on this, the system determines that the operating status of the distribution device tends to be temperature anomalies, and there are signs of insulation aging or overload, triggering an early warning.

[0087] The feature extraction and fusion through deep neural network make the real-time operation portrait of the equipment more sensitive and adaptable. When the system detects that the fusion feature value continues to increase to above 0.75, combined with the previous spatiotemporal analysis, the temperature fluctuation in the main distribution cabinet area has increased dramatically. The system will further identify that the insulation of the equipment has begun to age, and then issue an emergency warning to the maintenance personnel.

[0088] Embodiment 2:

[0089] Combined with Figure 3 In this embodiment, in order to further improve the fault prediction accuracy of the high-voltage power distribution device for mines, a time series analysis method for specific fault modes is introduced. By calculating the change rate of the temperature and voltage sensor data in real time, the length and weight of the sliding window are dynamically adjusted to adapt to the changes in different fault characteristics, such as temperature anomalies and load overloads. Assume that the temperature and voltage data of a high-voltage power distribution device in a mine are unstable during operation. The system needs to monitor the fluctuations of these data and identify potential faults through sliding window analysis. For this purpose, an adaptive weighted sliding window is set, and the window weight calculation formula is as follows:

[0090]

[0091] In this formula:

[0092] W i (t) represents the window weight at time t, which is used to quantify the importance of the current data of sensor i in fault identification. As the window weight changes, the system can dynamically respond to abnormal data at different time points.

[0093] T(t′) and V(t′) are the time series data of temperature and voltage, respectively, reflecting the temperature and voltage changes of the equipment under high voltage environment. It is assumed that the currently monitored temperature gradually rises within a certain period of time, which is expressed as T(t′)=25+0.3·t′ (unit: Celsius), while the voltage data fluctuates slightly, which is V(t′)=220+0.15·sin(0.2·t′) (unit: volt).

[0094] Functions P(T) and Q(V) are the response functions of temperature and voltage, respectively, and are used to evaluate the sensitivity of these data to fault prediction. Set P(T) = 0.05·T and Q(V) = 0.002·V 2 , where changes in temperature data are more likely to cause fault alarms, so the weight coefficient α of temperature data is set to 0.4≤α≤0.6, while voltage data is slightly less sensitive to faults, and the corresponding coefficient β ranges from 0.1≤β≤0.3.

[0095] Δt represents the length of the sliding window, which is currently set to 10 seconds. The integral operation within the window is used to calculate the rate of change of temperature and voltage during this period of time, allowing the system to adjust the prediction weight based on the real-time fluctuation of the data.

[0096] By substituting the data for calculation, set the current time t = 10 seconds, set α = 0.5 and β = 0.2, and enter the data changes of temperature and voltage: the rate of change of temperature is 0.5·0.05·0.3≈0.0075, and the second-order rate of change of voltage is Integrate to obtain the window weight within the time window

[0097] Therefore, the sliding window weight W i (t) = 0.0759 reflects the comprehensive response level of the current sensor within the 10-second window, indicating that temperature fluctuations contribute significantly to the risk of failure. Through this adaptive weighted sliding window analysis, when the temperature change rate increases significantly and the voltage fluctuates frequently, the system will increase the weight of the sliding window in real time to identify the characteristics of temperature anomalies and load overloads. In subsequent monitoring, the system detects that the temperature continues to rise and the voltage fluctuations intensify, and the window weight W i (t) It accumulates continuously and eventually triggers a risk warning, reminding maintenance personnel to conduct equipment inspection and troubleshooting.

[0098] To further analyze and detect the failure modes of high-voltage power distribution devices used in mines, wavelet transform is introduced to perform multi-level decomposition of sensor signals, especially to decompose sensor signals into high-frequency and low-frequency components to capture specific fault characteristics. In the detection of insulation aging, insulation loss will gradually increase the high-frequency components in the signal. The multi-level decomposition of wavelet transform can effectively extract these high-frequency fluctuations and provide support for early fault identification.

[0099] Set the voltage sensor V during device operation. k The signal of (t) shows abnormal fluctuation at time t = 10 seconds. To this end, the high-frequency signal feature S is introduced 高频 (t) Calculation formula to extract the high frequency characteristics of the equipment to identify early signs of insulation aging:

[0100]

[0101] Among them, S 高频 (t) is the characteristic value of the high-frequency signal at time t, which reflects the fluctuation of the high-frequency component and is used to identify potential insulation aging risks.

[0102] In this formula:

[0103] G k (x) is the wavelet basis function of the kth layer, which is used to extract high-frequency components. Used to emphasize the high-frequency fluctuation characteristics in the signal.

[0104] V k (t) represents the voltage signal of sensor k at time t, and the voltage signal is assumed to be V k (t) = 220 + 0.15sin (0.2t), and its small fluctuation reflects the change of equipment voltage.

[0105] h(V k (t)) is the transformation function of the voltage signal. For example, if h(V k (t)) = log(1 + V k (t)), is used to nonlinearly amplify small fluctuations in the voltage signal, making the system more sensitive to subtle signs of insulation aging.

[0106] γ k For the weight of each component, set the value range to 0.1≤γ k ≤0.5, used to adjust the importance of each layer’s wavelet components. Set the current layer weight to γ k =0.3.

[0107] Bring in the data for calculation, first calculate V k(t) = 220 + 0.15 sin (0.2 · 10) ≈ 220.3. Calculate the second-order derivative and we get At this time, the wavelet basis function The value of is 1 (the center position when x=0). After calculating the integral, the current high-frequency signal characteristics are obtained.

[0108] As the system continues to operate, if the high-frequency component continues to increase, it indicates that early signs of insulation aging have appeared. Suppose that in the next test, the system detects S 高频 If (t) continues to increase, even reaching above 0.01, the system will mark the sensor data as high risk and send a warning signal.

[0109] In the fault prediction of high-voltage power distribution equipment for mines, the system introduces a multi-dimensional trend prediction model to further strengthen the real-time monitoring of the equipment's operating status. The model generates fused trend feature values ​​by jointly analyzing the trend changes of key parameters such as temperature and voltage to detect complex fault trends such as load overload and temperature anomaly. In the application, the temperature sensors and voltage sensors of mine equipment are set to continuously collect operating data, and the system analyzes these data in real time through the trend prediction model to form an overall abnormality detection standard.

[0110] The temperature data of the current period shows a slow upward trend, while the voltage data fluctuates slightly. Based on these data, the calculation formula of the multidimensional trend prediction model is expressed as:

[0111]

[0112] in:

[0113] T 融合 It is the fused trend feature value, which is used to capture the fault trend of current load overload or temperature abnormality.

[0114] F i (t′) represents the time series trend data of each sensor, reflecting the change pattern of each fault feature. For example, the temperature feature F1(t′)=25+0.5·t′ (unit: Celsius) represents the gradual increase of temperature, and the voltage feature F2(t′)=220+0.2·sin(0.1·t′) (unit: volt) shows the periodic fluctuation of voltage.

[0115] λ i It is the weight coefficient of each feature, which is used to adjust the importance of different features in the fusion process. The weight coefficient of temperature is set to λ1=0.6 and the weight coefficient of voltage is set to λ2=0.4 to reflect that the influence of temperature in the current monitoring is more significant.

[0116] t0 to t represents the monitoring time range, which is set here to 0 to 10 seconds.

[0117] Substituting specific data into the calculation, the change of temperature trend characteristics is F1(t′)=25+0.5·t′. Under the weight of λ1=0.6, the integral of the temperature part is The voltage trend characteristic F2(t′)=220+0.2·sin(0.1·t′) is weighted by λ2=0.4, and the integral of the voltage part is approximately

[0118] Therefore, the trend characteristic value T after fusion 融合 ≈165+880=1045. The results show that the comprehensive trend characteristics of the current temperature and voltage are quite obvious, especially the combination of temperature rise and voltage fluctuation indicates the gradual increase of equipment load.

[0119] As the system continues to monitor, if the fusion eigenvalue T 融合 If a preset threshold is exceeded (this threshold is set to 1000), the system will determine that there is a trend of overload or abnormal temperature, and immediately issue an early warning to the mine equipment management system, suggesting that the load and insulation status of the distribution device be checked. Through this multi-feature trend fusion detection method, the system can more comprehensively capture the trend changes of abnormal equipment signals, ensure early identification of faults, provide data support for preventive maintenance of equipment, effectively extend the service life of the distribution device, and ensure the safety and continuous operation of the mine.

[0120] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A fault prediction method for a high voltage power distribution device for a mine, characterized in that: S1. Multi-dimensional sensor data fusion based on high-voltage power distribution characteristics: S1.

1. Install temperature, current, voltage and gas leakage sensors in the mine high voltage distribution device; S1.

2. Use multi-sensor data fusion algorithm to integrate multi-dimensional data to form a real-time operation portrait of the distribution device; S2. Time series analysis of specific failure modes: S2.

1. Combine the operation rules of high-voltage power distribution equipment to extract the time series characteristics of key variables such as temperature and voltage; S2.

2. Use sequence analysis models to capture abnormal fluctuations in key characteristic parameters and identify fault types unique to high-voltage environments, including insulation aging, temperature anomalies, and load overload; S2.3, build a time series-based feature pattern database to match new data with historical failure patterns and quickly identify known failure trends; S3. Key feature extraction and risk level analysis in fault prediction: S3.1, apply feature selection algorithm to extract representative fault prediction features from sensor data; S3.

2. According to the degree of harm of different fault modes, a fault prediction risk level model is established, which is divided into three levels: low, medium and high, to evaluate the potential threat of faults to high-voltage distribution equipment; S4, Fault prediction and adaptive maintenance optimization closed loop: S4.

1. Generate optimized adaptive maintenance plans based on the real-time predicted fault risk level, and dynamically adjust maintenance cycles and strategies based on the actual operating status of the equipment.

2. A method for predicting faults of a high voltage power distribution device for a mine according to claim 1, characterized in that The method for forming a real-time operation portrait of a power distribution device comprises: The dynamic weight adaptive fusion algorithm based on the health status of the sensor is introduced. i and data quality factor Q i To dynamically adjust the weight of each sensor, the specific calculation method is as follows: Among them, H i represents the health metric of sensor i, which is calculated by integrating the rate of change of temperature and voltage over time t′; T i (t′) and V i (t′) represent the temperature and voltage data of sensor i at time t′ respectively; function f(T i (t′)) represents the quality function of temperature data, which is used to evaluate abnormal temperature fluctuations. The function g(V i (t′)) represents the quality function of voltage data, which is used to capture the high-order changes of voltage; coefficients α and β are adjustment parameters, which adjust the weights of temperature and voltage data according to the sensor type and the credibility of the data.

3. A method for predicting a fault of a high voltage power distribution device for a mine according to claim 2, characterized in that The method for forming a real-time operation portrait of a power distribution device comprises: The spatial and temporal correlation of sensor data is analyzed through spatiotemporal clustering to identify anomalies in fault detection. In spatiotemporal correlation analysis, the spatial correlation weight W is used. 空间 , is calculated comprehensively through geographical location distance and time correlation. The specific formula is as follows: Among them, W 空间 represents the spatial association weight in the fusion result; D 地理 (x) is the geographical location distance of the sensor, h(D 地理 (x)) is a function of distance, which is used to measure the impact of geographical distance on data fusion; T 关联 (t′) is the correlation in the time dimension, k(T 关联 (t′)) represents the time correlation function, which is used to capture the rate of change of data in time; the coefficients σ and γ are adjustment parameters, corresponding to the adjustment of spatial and temporal correlation weights, respectively.

4. A fault prediction method for a high voltage power distribution device for mining according to claim 3, characterized in that The method for forming a real-time operation portrait of a power distribution device comprises: A deep neural network model is used for feature extraction and fusion. A deep neural network is used to perform nonlinear feature mapping on multidimensional data. The feature vector F of each sensor is generated through the feature mapping function φ(x). i , and then fuse at the feature level; finally, the fused feature vector F 融合 The calculation is as follows: Among them, F 融合 is the fused high-dimensional running state feature vector; W i represents the dynamic weight of each sensor; φ(F i (x)) is the feature vector generated by the neural network, K(x) is the kernel function, which is used to amplify the weight of key features in the fusion process; the symbol * represents the convolution operation, which is used to capture the local correlation in the feature vector.

5. A fault prediction method for a high voltage power distribution device for mining according to claim 1, characterized in that The time series analysis method for the specific fault mode includes: calculating the change rate of temperature and voltage sensor data in real time, and dynamically adjusting the length and weight of the sliding window to adapt to the changes of different fault characteristics, including temperature anomalies and load overload.

6. A method for predicting faults of a high voltage power distribution device for mining according to claim 5, characterized in that The time series analysis method for the specific failure mode includes: Wavelet transform is introduced to perform multi-level decomposition of sensor signals; the signals are decomposed into high-frequency and low-frequency components to identify specific high-frequency signal fluctuations of insulation aging; and the early characteristics of insulation aging are captured by analyzing the decomposed high-frequency components.

7. A method for predicting faults of a high voltage power distribution device for mining according to claim 6, characterized in that The time series analysis method for the specific failure mode includes: Through the multi-dimensional trend prediction model, the trend changes of each feature are jointly analyzed to generate the fused trend feature value for anomaly detection; the formula for multi-feature fusion trend detection is as follows: Where: T 融合 Represents the trend feature value after fusion, which is used to capture the complex fault trend of load overload or temperature anomaly; F i (t′) is the time series trend data of the i-th feature, reflecting the change pattern of the fault type; λ i is the weight coefficient of each feature, indicating the importance of different features in the fusion process; t0 to t represents the detection time range, which integrates the trend changes of each feature to form an overall anomaly detection standard.

Images