Power station equipment state degradation early warning method

Abstract

The invention provides a power station equipment state degradation early warning method. The method comprises the following steps that a charge sensor and a high-precision dielectric sensor are integrated in a lubricating oil circulating system of power station equipment; setting an acquisition time interval, dividing the target acquisition time period into acquisition time points according to the preset acquisition time interval, preliminarily acquiring charge data of the oil product of the power station equipment at each acquisition time point, and identifying an oil product degradation preliminary evaluation coefficient of the lubricating oil corresponding to the power station equipment based on the charge data; identifying whether the lubricating oil circulation system corresponding to the power station equipment has oil degradation; and meanwhile, based on the arranged high-precision dielectric sensors, dielectric data of the oil product of the power station equipment at each acquisition time point are synchronously extracted, the oil product state of the lubricating oil circulating system corresponding to the power station equipment is deeply evaluated, and corresponding early warning signals are sent. Through multi-parameter fusion analysis, the monitoring precision is improved, the initial sign of oil product degradation can be found earlier, and equipment faults caused by oil product performance reduction are avoided.

Description

Technical Field

[0001] This invention belongs to the field of power plant equipment status early warning technology, and specifically relates to a method for early warning of power plant equipment status deterioration. Background Technology

[0002] Oil quality monitoring in power plant equipment is crucial for ensuring safe operation and extending service life. Lubricating and insulating oils, widely used in power plant equipment, not only provide lubrication, cooling, and insulation but also directly impact operational efficiency and safety. As equipment operates over time, oil quality can deteriorate due to environmental factors such as high temperature, humidity, and impurities, leading to decreased lubrication and insulation performance, and potentially causing equipment failures or safety accidents. Regular oil monitoring allows for timely detection of deterioration trends, enabling proactive replacement or treatment to prevent major equipment malfunctions caused by oil issues. Furthermore, oil quality monitoring provides vital reference data on equipment operating status. Changes in oil parameters indirectly reflect operating conditions and potential hazards, facilitating condition-based maintenance and refined operation, reducing maintenance costs, and improving equipment reliability and economy. Therefore, oil quality monitoring is a key component in ensuring the long-term stable operation of power plant equipment.

[0003] However, current technologies for monitoring oil quality in power plant equipment still have some technical shortcomings and deficiencies, limiting the comprehensiveness and accuracy of monitoring. First, traditional oil quality monitoring typically relies on laboratory analysis, requiring manual sampling and testing, which is time-consuming and cannot achieve real-time online monitoring. This lag may cause equipment damage due to deterioration problems before test results are available. Second, existing oil quality monitoring technologies mostly focus on single indicators, such as viscosity, acid value, or breakdown voltage. However, these indicators often only reflect one aspect of oil deterioration, lacking a comprehensive assessment of the oil's overall performance and easily overlooking potential problems. Summary of the Invention

[0004] The main objective of this invention is to provide a method for early warning of equipment condition deterioration in power plants, addressing the aforementioned problems.

[0005] Therefore, the above-mentioned objective of the present invention is achieved through the following technical solution:

[0006] A method for early warning of equipment condition deterioration in power plants includes the following steps:

[0007] S1. Integrate a charge sensor and a high-precision dielectric sensor into the lubricating oil circulation system of power station equipment;

[0008] S2. Set the collection time interval, divide the target collection time period into each collection time point according to the preset collection time interval, and then initially collect the charge data of the power station equipment oil at each collection time point, and based on it, identify the preliminary evaluation coefficient of the oil deterioration of the corresponding lubricating oil of the power station equipment.

[0009] S3. Based on the preliminary assessment coefficient of the oil deterioration of the lubricating oil corresponding to the power station equipment, identify whether there is oil deterioration in the lubricating oil circulation system corresponding to the power station equipment. If so, proceed to step S4; otherwise, exit the method loop.

[0010] S4. Based on the deployed high-precision dielectric sensors, the dielectric data of the power station equipment oil at each collection time point are extracted synchronously to deeply evaluate the oil status of the corresponding lubricating oil circulation system of the power station equipment and send corresponding early warning signals.

[0011] While adopting the above technical solutions, the present invention may also adopt or combine the following technical solutions:

[0012] As a preferred technical solution of the present invention: in step S2, the charge data of the oil in the power plant equipment at each collection time point includes charge density, charge distribution uniformity, charge change rate and charge fluctuation amplitude.

[0013] As a preferred technical solution of the present invention: Step S2, identifying the preliminary assessment coefficient of oil deterioration of the lubricating oil corresponding to the power station equipment specifically includes the following steps:

[0014] S21. The z-score standardization method is used to preprocess the charge density, charge distribution uniformity, charge change rate and charge fluctuation amplitude of the oil in the power station equipment at each collection time point, and the SOM model is trained at the same time until the SOM model converges.

[0015] S22. Map the charge density of the preprocessed power plant equipment oil at each sampling time point to the trained SOM model to obtain the weight vector of the neurons in the SOM model that maps the charge density of the preprocessed power plant equipment oil at each sampling time point. This weight vector is then denoted as α. h h is the number of each collection time point, and the value of h ranges from 1 to H, where H is the total number of collection time points;

[0016] S23. Export the charge data of the power plant equipment at each normal time point under normal operating conditions from the equipment database. This gives the charge density of the power plant equipment at each normal time point under normal operating conditions. Similarly, input this data into the SOM model to obtain the weight vector of the charge density mapping neuron of the power plant equipment at each normal time point under normal operating conditions, and thus form the normal state region.

[0017] S24. Calculate the Euclidean distance between the charge density mapping neuron and the normal state region at each acquisition time point. α i Let be the weight vector of the i-th neuron in the normal state region, where i is the number of each neuron in the normal state region, i = 1, 2, ... N, and N is the total number of neurons in the normal state region. distance(·) is the distance calculation function.

[0018] S25, Based on the analysis formula The fluctuation coefficient SL1 of the charge density of the corresponding lubricating oil of the power station equipment during the target data collection period is analyzed, where D is the set reference charge density value.

[0019] S26. Based on the calculation method of the change fluctuation coefficient of the charge density of the lubricating oil corresponding to the power station equipment during the target acquisition time period, the change fluctuation coefficients of the charge distribution uniformity, charge change rate and charge fluctuation amplitude of the lubricating oil corresponding to the power station equipment during the target acquisition time period are calculated in the same way.

[0020] S27. Finally, the preliminary assessment coefficient Ψ for the oil deterioration of the corresponding lubricating oil for the power station equipment is identified. The specific identification formula is as follows:

[0021]

[0022] In the formula, SL2, SL3, and SL4 represent the variation coefficients of charge distribution uniformity, charge change rate, and charge fluctuation amplitude of the lubricating oil corresponding to the power station equipment during the target data collection period, respectively; SL1', SL2', SL3', and SL4' are the maximum permissible variation coefficients of charge density, charge distribution uniformity, charge change rate, and charge fluctuation amplitude, respectively; ω1, ω2, ω3, and ω4 are the allocation weight factors corresponding to charge density, charge distribution uniformity, charge change rate, and charge fluctuation amplitude, respectively, reflecting the importance of each charge data in the preliminary assessment of oil deterioration.

[0023] As a preferred technical solution of the present invention: Step S21, training the SOM model until the SOM model converges, specifically includes the following steps:

[0024] S211. Select a two-dimensional grid to represent the SOM model, and randomly initialize the weight vector of each neuron in the SOM model.

[0025] S212. Randomly select a sample x(t) from the standardized dataset as input data into the SOM model. Calculate the distance between the sample and all nodes in the SOM model, and take the node with the smallest distance as the best matching node for sample x(t):

[0026]

[0027] In the formula, the arg min(·) function represents finding the parameter values ​​that minimize a certain function; w j (t) is the weight vector of node j, where j is the number of each node, and BMU is the best matching node for sample x(t);

[0028] S213. Then, the weight vectors of each node in the SOM model are updated sequentially, with the following update formula:

[0029] w j (t+1)=w j (t)+β1(t·β2) bj (t)·[x(t)-w j (t)];

[0030]

[0031] In the formula, w j (t+1) represents the weight vector of node j after one update, β1(t) represents the learning rate, and β2 bj (t) represents the neighborhood function, exp(·) is the exponential function, and γ b γ j Let represent the grid coordinates of the best matching node and node j, respectively, and σ(t) be the neighborhood width;

[0032] S214. Then, repeat the iterative update, gradually reducing the learning rate and neighborhood width until the two-dimensional grid converges. When the two-dimensional grid converges, it means that the SOM model has converged.

[0033] As a preferred embodiment of the present invention: step S23, forming a normal state neuronal region specifically includes the following steps:

[0034] S231. Extract the weight vectors of each neuron in the normal state region from the SOM model and summarize them to form a dataset.

[0035] S232. Determine the number of clusters R using the elbow rule;

[0036] S233. Randomly select R weight vectors as initial cluster centers;

[0037] S234. For each neuron weight vector α i Calculate the Euclidean distance between it and each cluster center;

[0038] S235. Then, based on the Euclidean distance between each neuron weight vector and each cluster center, each neuron weight vector is assigned to the cluster center with the smallest Euclidean distance.

[0039] S236. Iteratively update each cluster center. The specific update steps are: obtain the number of internal neurons; and simultaneously update each cluster center by summing the neuron weight vectors of each cluster center and taking the average value to recalculate the center point of each cluster.

[0040] S237. Repeatedly update each cluster center iteratively until the center point of each cluster center no longer changes or is less than the set threshold, which means that each cluster center has reached convergence.

[0041] S238. After clustering is completed, the cluster center regions of neurons in the normal state region are identified, and they are summarized to obtain the normal state neuron regions.

[0042] As a preferred technical solution of the present invention: In step S3, the logic for identifying whether there is oil deterioration in the lubricating oil circulation system corresponding to the power station equipment is as follows: the preliminary assessment coefficient of oil deterioration of the lubricating oil corresponding to the power station equipment is compared with the preset oil deterioration threshold coefficient. If the preliminary assessment coefficient of oil deterioration of the lubricating oil corresponding to the power station equipment is greater than the preset oil deterioration threshold coefficient, it is determined that there is oil deterioration in the lubricating oil circulation system corresponding to the power station equipment.

[0043] As a preferred technical solution of the present invention: in step S4, the dielectric data of the power plant equipment oil at each collection time point includes dielectric constant, dielectric loss factor, dielectric strength and dielectric response time.

[0044] As a preferred technical solution of the present invention: Step S4, the in-depth evaluation of the oil condition of the lubricating oil circulation system corresponding to the power station equipment specifically includes the following steps:

[0045] S41. Extract the dielectric constant, dielectric loss factor, dielectric strength, and dielectric response time of the power plant equipment oil at each collection time point, and preprocess them using a standardized data processing method to obtain the preprocessed dielectric constant, dielectric loss factor, dielectric strength, and dielectric response time of the power plant equipment oil at each collection time point, and label them as δ1(h), δ2(h), δ3(h), and δ4(h), respectively.

[0046] S42. Represent the dielectric properties at each acquisition time point as a vector X(h) = [δ1(h),δ2(h),δ3(h),δ4(h)], and then calculate the Mahalanobis distance MS(h) between the dielectric data of the power plant equipment oil at each acquisition time point and the mean of the normal dielectric data.

[0047] S43. Based on the time series of the collected data, perform statistical analysis to extract the degree of deviation of the current oil product from the normal state and the rate of change of the oil product deterioration trend.

[0048] S44. The final definition of the oil degradation index for the lubricating oil circulation system corresponding to power plant equipment is:

[0049]

[0050] In the formula, θ represents the degree to which the overall oil quality deviates from its normal state, and θ represents the rate of change of the oil quality deterioration trend in the power plant equipment during the target data collection period.

[0051] S45. Compare the oil deterioration index of the lubricating oil circulation system corresponding to the power plant equipment with the set oil deterioration standard range [Q1, Q2], where Q1 is the set lower limit of oil deterioration and Q2 is the set upper limit of oil deterioration.

[0052] like This indicates that the oil condition of the corresponding lubricating oil circulation system of the power station equipment is normal.

[0053] like This indicates that the oil condition of the corresponding lubricating oil circulation system of the power plant equipment is slightly deteriorated;

[0054] like This indicates that the oil condition of the corresponding lubricating oil circulation system of the power plant equipment is severely deteriorated.

[0055] As a preferred technical solution of the present invention: In step S42, calculating the Mahalanobis distance between the dielectric data of the power plant equipment oil at each collection time point and the average normal dielectric data specifically includes the following steps:

[0056] S421. Export the dielectric data of the power station equipment at each normal time point under normal operating conditions from the equipment database, and perform mean calculation on the data to obtain the mean dielectric data of the power station equipment under normal operating conditions. Similarly, construct the dielectric data vector X′=[δ1′,δ2′,δ3′,δ4′] of the power station equipment under normal operating conditions, and construct the normal covariance matrix of the power station equipment under normal operating conditions:

[0057]

[0058] In the formula, δ1′ p δ2′ p δ3′ p δ4′ p Let Var(·) represent the dielectric constant, dielectric loss factor, dielectric strength, and dielectric response time of the power plant equipment at the p-th normal time point under normal operating conditions, respectively. Var(·) represents the variance calculation function, Cov(·) represents the covariance calculation function, and p is the number of each normal time point.

[0059] S422. Calculate the Mahalanobis distance between the dielectric data of the power plant equipment oil at each data collection time point and the mean of the normal dielectric data:

[0060]

[0061] In the formula, T represents the transpose symbol.

[0062] As a preferred technical solution of the present invention: Step S43, extracting the degree of deviation of the current oil product from the normal state and the rate of change of the oil product deterioration trend specifically includes the following steps:

[0063] S431, through calculation formula Calculate the degree to which the current overall oil quality deviates from its normal state.

[0064] S432. Synchronously calculate the rate of change of oil deterioration trend in power plant equipment during the target data collection period:

[0065]

[0066] In the formula, Δt is the preset data acquisition time interval, and t h This represents the time value at the h-th data collection point.

[0067] Compared with existing technologies, this invention has the following advantages: By integrating charge sensors and high-precision dielectric sensors into the lubricating oil circulation system of power plant equipment, and combining this with preset acquisition time intervals to perform real-time data acquisition and status assessment of the lubricating oil, this method has significant benefits and is necessary. On the one hand, the application of charge sensors can efficiently capture changes in charge in the lubricating oil caused by deterioration, preliminarily assess the degree of oil deterioration, and provide a fast and intuitive assessment basis for monitoring the equipment's operating status. On the other hand, the synchronous deployment of high-precision dielectric sensors can further extract changes in key parameters such as dielectric constant and loss factor of the oil, deeply assess the deterioration trend of oil performance, and comprehensively assess the oil status from multiple dimensions. Through multi-parameter fusion analysis, not only is the monitoring accuracy improved, but the initial signs of oil deterioration can also be detected earlier, avoiding equipment failure due to decreased oil performance. In addition, this real-time online monitoring method can overcome the lag of traditional laboratory testing and achieve dynamic tracking of oil status. Therefore, this invention, through the joint monitoring of charge and dielectric parameters, achieves accurate assessment and early warning of lubricating oil deterioration, and has extremely high practicality and promotional value. Attached Figure Description

[0068] Figure 1 The flowchart is for the power plant equipment condition deterioration early warning method provided by the present invention. Detailed Implementation

[0069] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0070] The implementation logic of this invention is as follows: First, the rapid screening capability of the charge sensor can quickly detect changes in charge distribution in the lubricating oil, providing a preliminary assessment of degradation. The charge sensor is suitable for rapid screening and identifying potential problems. Then, through the in-depth analysis capability of the high-precision dielectric sensor, if the charge sensor detects potential degradation, the high-precision dielectric sensor is used for more precise analysis. The dielectric sensor can provide more detailed dielectric property information, helping to confirm and assess the degree of degradation.

[0071] This sequence helps improve detection efficiency, save resources, and ensure the accuracy of results. Using a dielectric sensor first might add unnecessary complexity and cost, as it is typically used for more in-depth analysis.

[0072] like Figure 1 As shown, a method for early warning of equipment condition deterioration in power plants specifically includes the following steps:

[0073] S1. Integrate a charge sensor and a high-precision dielectric sensor into the lubricating oil circulation system of power station equipment;

[0074] S2. Set the collection time interval, divide the target collection time period into each collection time point according to the preset collection time interval, and then initially collect the charge data of the power station equipment oil at each collection time point, and based on it, identify the preliminary evaluation coefficient of the oil deterioration of the corresponding lubricating oil of the power station equipment.

[0075] The charge data of the power plant equipment oil at each collection time point includes charge density, charge distribution uniformity, charge change rate, and charge fluctuation amplitude. The charge amount at each location point in the power plant equipment lubricating oil circulation system at each collection time point is collected by charge sensors and transmitted as output data to the central processing unit through a standardized data interface. The central processing unit analyzes the charge amount to obtain the charge data of the power plant equipment oil at each collection time point.

[0076] S21. The z-score standardization method is used to preprocess the charge density, charge distribution uniformity, charge change rate and charge fluctuation amplitude of the oil in the power station equipment at each collection time point, and the SOM model is trained at the same time until the SOM model converges.

[0077] Using the SOM model to identify preliminary assessment coefficients for lubricant degradation in power plant equipment offers significant advantages: SOM effectively handles high-dimensional and nonlinear data, reducing complex charge density data to an easily analyzable two-dimensional plane through self-organizing mapping. This method captures underlying patterns and trends in the data, providing an intuitive understanding of the oil's condition. Compared to traditional methods, SOM not only identifies normal and abnormal states but also provides a more detailed assessment of the degree of degradation. It offers greater flexibility and adaptability, enabling earlier detection of potential problems, reduced equipment downtime, extended equipment lifespan, and improved maintenance accuracy and efficiency.

[0078] S211. Select a two-dimensional grid to represent the SOM model, and randomly initialize the weight vector of each neuron in the SOM model.

[0079] S212. Randomly select a sample x(t) from the standardized dataset as input data into the SOM model. Calculate the distance between the sample and all nodes in the SOM model, and take the node with the smallest distance as the best matching node for sample x(t):

[0080]

[0081] In the formula, the argmin(·) function represents finding the parameter values ​​that minimize a given function; w j (t) is the weight vector of node j, where j is the number of each node, and BMU is the best matching node for sample x(t);

[0082] Updating the weight vector after finding the neuron with the smallest distance is to make the neuron and its neighborhood closer to the input data, so that the network can adaptively capture the distribution characteristics of the data, gradually adapt the network to the distribution of the input data, and maintain the topological structure of the input data.

[0083] By updating the weight vector, SOM can adjust the position of neurons during training to better represent the feature distribution of the input data, thereby improving the network's mapping accuracy and classification ability. This update mechanism ensures that SOM can effectively capture the topological structure of the data and effectively perform data clustering and visualization, while gradually forming a low-dimensional representation of the data, mapping similar data points to adjacent neurons, forming meaningful clusters and pattern recognition.

[0084] S213. Then, the weight vectors of each node in the SOM model are updated sequentially, with the following update formula:

[0085] w j (t+1)=w j (t)+β1(t·β2) bj (t)·[x(t)-wj (t)];

[0086]

[0087] In the formula, w j (t+1) represents the weight vector of node j after one update, β1(t) represents the learning rate, and β2 bj (t) represents the neighborhood function, exp(·) is the exponential function, and γ b γ j Let represent the grid coordinates of the best matching node and node j, respectively, and σ(t) be the neighborhood width;

[0088] S214. Then, repeat the iterative update, gradually reducing the learning rate and neighborhood width until the two-dimensional grid converges. When the two-dimensional grid converges, it means that the SOM model has converged.

[0089] S22. Map the charge density of the preprocessed power plant equipment oil at each sampling time point to the trained SOM model to obtain the weight vector of the neurons in the SOM model that maps the charge density of the preprocessed power plant equipment oil at each sampling time point. This weight vector is then denoted as α. h h is the number of each collection time point, and the value of h ranges from 1 to H, where H is the total number of collection time points;

[0090] S23. Export the charge data of the power plant equipment at each normal time point under normal operating conditions from the equipment database. This gives the charge density of the power plant equipment at each normal time point under normal operating conditions. Similarly, input this data into the SOM model to obtain the weight vector of the charge density mapping neuron of the power plant equipment at each normal time point under normal operating conditions, and thus form the normal state region.

[0091] S231. Extract the weight vectors of each neuron in the normal state region from the SOM model and summarize them to form a dataset.

[0092] S232. Determine the number of clusters R using the elbow rule;

[0093] S233. Randomly select R weight vectors as initial cluster centers;

[0094] S234. For each neuron weight vector α i Calculate the Euclidean distance between it and each cluster center;

[0095] S235. Then, based on the Euclidean distance between each neuron weight vector and each cluster center, each neuron weight vector is assigned to the cluster center with the smallest Euclidean distance.

[0096] S236. Iteratively update each cluster center. The specific update steps are: obtain the number of internal neurons; and simultaneously update each cluster center by summing the neuron weight vectors of each cluster center and taking the average value to recalculate the center point of each cluster.

[0097] S237. Repeatedly update each cluster center iteratively until the center point of each cluster center no longer changes or is less than the set threshold, which means that each cluster center has reached convergence.

[0098] S238. After clustering is completed, the cluster center regions of neurons in the normal state region are identified, and they are summarized to obtain the normal state neuron regions.

[0099] S24. Calculate the Euclidean distance between the charge density mapping neuron and the normal state region at each acquisition time point. α i Let be the weight vector of the i-th neuron in the normal state region, where i is the number of each neuron in the normal state region, i = 1, 2, ... N, and N is the total number of neurons in the normal state region. distance(·) is the distance calculation function.

[0100] S25, Based on the analysis formula The fluctuation coefficient SL1 of the charge density of the corresponding lubricating oil of the power station equipment during the target data collection period is analyzed, where D is the set reference charge density value.

[0101] S26. Based on the calculation method of the change fluctuation coefficient of the charge density of the lubricating oil corresponding to the power station equipment during the target acquisition time period, the change fluctuation coefficients of the charge distribution uniformity, charge change rate and charge fluctuation amplitude of the lubricating oil corresponding to the power station equipment during the target acquisition time period are calculated in the same way.

[0102] The core logic of this method involves subtracting a set baseline charge density value from the Euclidean distance between the charge density mapping neurons at each sampling time point and the normal state region, and then dividing the result by the set baseline charge density value. This process quantifies the deviation between the current state and the normal state into a relative change (i.e., a change fluctuation coefficient) through normalization. Specifically, the Euclidean distance is used to measure the overall difference between the charge density characteristics of the current sampling point and the normal state, while the baseline charge density value serves as a reference standard for the normal state. Subtracting the baseline value yields the absolute amount of deviation, and dividing by the baseline value achieves normalization, converting the deviation into a percentage change relative to the normal state. This allows a dimensionless relative value (fluctuation coefficient) to reflect the dynamic trend of charge density changes, eliminating the influence of dimensions and clearly revealing the relative magnitude of the deviation, facilitating quantitative analysis and comparison of oil deterioration trends.

[0103] Similarly, based on the calculation method of the change fluctuation coefficient of the charge density of the lubricating oil corresponding to the power station equipment during the target acquisition time period, the change fluctuation coefficients of the charge distribution uniformity, charge change rate and charge fluctuation amplitude of the lubricating oil corresponding to the power station equipment during the target acquisition time period are calculated.

[0104] S27. Finally, the preliminary assessment coefficient Ψ for the oil deterioration of the corresponding lubricating oil for the power station equipment is identified. The specific identification formula is as follows:

[0105]

[0106] In the formula, SL2, SL3, and SL4 represent the variation coefficients of charge distribution uniformity, charge change rate, and charge fluctuation amplitude of the lubricating oil corresponding to the power station equipment during the target data collection period, respectively; SL1', SL2', SL3', and SL4' are the maximum permissible variation coefficients of charge density, charge distribution uniformity, charge change rate, and charge fluctuation amplitude, respectively; ω1, ω2, ω3, and ω4 are the allocation weight factors corresponding to charge density, charge distribution uniformity, charge change rate, and charge fluctuation amplitude, respectively, reflecting the importance of each charge data in the preliminary assessment of oil deterioration.

[0107] A preliminary assessment coefficient for oil degradation is calculated using charge density, charge distribution uniformity, charge change rate, and the fluctuation coefficient of charge fluctuation amplitude. These electrical properties directly reflect the physical and chemical changes in lubricating oil. Changes in charge density can reveal an increase in contaminants or impurities in the oil; charge distribution uniformity indicates the homogeneity and stability within the oil; the charge change rate reflects the oil's reaction rate under different operating conditions; and the fluctuation coefficient of charge fluctuation amplitude reveals the oil's stability under dynamic conditions. Combining these indicators and their fluctuation coefficients provides a comprehensive perspective for assessing the degree of oil degradation, thereby helping to predict equipment maintenance needs, avoid potential failures, and ensure reliable equipment operation.

[0108] Algorithm convergence means that each cluster centroid no longer changes significantly. There are two common criteria for this:

[0109] Suppose that after the f-th iteration, each cluster center point is updated to... Updated after the (f+1)th iteration to So:

[0110] The first condition for judgment is that the change is 0: if This indicates that the centroid of the c-th cluster has been completely stabilized; c is the cluster number, and f is the iteration number.

[0111] The second judgment condition is that the change is less than the set value: This indicates that the change in the center point of the c-th cluster is very slight, and it can be considered that the cluster has converged. ψ is the set threshold.

[0112] S3. Based on the preliminary assessment coefficient of the oil deterioration of the lubricating oil corresponding to the power station equipment, identify whether there is oil deterioration in the lubricating oil circulation system corresponding to the power station equipment. If so, proceed to step S4; otherwise, exit the method loop.

[0113] The preliminary assessment coefficient of oil deterioration of the lubricating oil corresponding to the power plant equipment is compared with the preset oil deterioration threshold coefficient. If the preliminary assessment coefficient of oil deterioration of the lubricating oil corresponding to the power plant equipment is greater than the preset oil deterioration threshold coefficient, it is determined that there is oil deterioration in the lubricating oil circulation system corresponding to the power plant equipment.

[0114] S4. Based on the deployed high-precision dielectric sensors, the dielectric data of the power station equipment oil at each collection time point are extracted synchronously to deeply evaluate the oil status of the corresponding lubricating oil circulation system of the power station equipment and send corresponding early warning signals.

[0115] The dielectric data of the oil in the power plant equipment at each time point of collection includes dielectric constant, dielectric loss factor, dielectric strength, and dielectric response time.

[0116] S41. Extract the dielectric constant, dielectric loss factor, dielectric strength, and dielectric response time of the power plant equipment oil at each collection time point, and preprocess them using a standardized data processing method to obtain the preprocessed dielectric constant, dielectric loss factor, dielectric strength, and dielectric response time of the power plant equipment oil at each collection time point, and label them as δ1(h), δ2(h), δ3(h), and δ4(h), respectively.

[0117] S42. Represent the dielectric properties at each acquisition time point as a vector X(h) = [δ1(h),δ2(h),δ3(h),δ4(h)], and then calculate the Mahalanobis distance MS(h) between the dielectric data of the power plant equipment oil at each acquisition time point and the mean of the normal dielectric data.

[0118] S421. Export the dielectric data of the power station equipment at each normal time point under normal operating conditions from the equipment database, and perform mean calculation on the data to obtain the mean dielectric data of the power station equipment under normal operating conditions. Similarly, construct the dielectric data vector X′=[δ1′,δ2′,δ3′,δ4′] of the power station equipment under normal operating conditions, and construct the normal covariance matrix of the power station equipment under normal operating conditions:

[0119]

[0120] In the formula, δ1′ p δ2′ p δ3′p δ4′ p Let Var(·) represent the dielectric constant, dielectric loss factor, dielectric strength, and dielectric response time of the power plant equipment at the p-th normal time point under normal operating conditions, respectively. Var(·) represents the variance calculation function, Cov(·) represents the covariance calculation function, and p is the number of each normal time point.

[0121] S422. Calculate the Mahalanobis distance between the dielectric data of the power plant equipment oil at each data collection time point and the mean of the normal dielectric data:

[0122]

[0123] In the formula, T represents the transpose symbol.

[0124] Mahalanobis distance is a metric based on multivariate distributions that comprehensively considers the correlation between variables, rather than just the independent bias of a single variable. By standardizing the parameter vector, the influence of different parameter dimensions and scales can be eliminated, making the differences between parameters comparable. Furthermore, by combining the dielectric data vector and covariance matrix of power plant equipment under normal operating conditions, Mahalanobis distance can reflect the current state's position within the normal distribution and distinguish outlier data points or trends. Compared to simple Euclidean distance, Mahalanobis distance adjusts the weights of different parameters through the inverse of the covariance matrix, giving more attention to parameters with smaller fluctuations but higher sensitivity in anomaly detection. In addition, Mahalanobis distance is statistically significant and can be combined with thresholds (such as the critical value of the chi-square distribution) to determine whether the current state significantly deviates from the normal range.

[0125] S43. Based on the time series of the collected data, perform statistical analysis to extract the degree of deviation of the current oil product from the normal state and the rate of change of the oil product deterioration trend.

[0126] S431, through calculation formula Calculate the degree to which the current overall oil quality deviates from its normal state.

[0127] Mean Mahalanobis distance The Mahalanobis distance can represent the degree to which the overall oil quality deviates from its normal state because it comprehensively measures the difference between the current state and the normal state by combining the covariance relationships of multiple parameters (such as dielectric constant, dielectric loss factor, breakdown voltage, polarization time, etc.). Compared to the deviation of a single parameter, the Mahalanobis distance not only considers the changes of each parameter itself, but also reflects the changes in the correlation between them. The average value of these distances can be used to intuitively assess the degree of deviation between the current overall state of the oil and the normal state. The larger the value, the more significant the deviation, indicating a higher degree of oil deterioration.

[0128] S432. Synchronously calculate the rate of change of oil deterioration trend in power plant equipment during the target data collection period:

[0129]

[0130] In the formula, Δt is the preset data acquisition time interval, and t h This represents the time value at the h-th data collection point.

[0131] In the above calculation formula (t) h -Δt) is used to center the time data by mean, thereby eliminating the influence of time series offset on the slope calculation. This centering process can improve the numerical stability of the calculation and ensure that the result only reflects the trend change and is not affected by the absolute value of the time point.

[0132] S44. The final definition of the oil degradation index for the lubricating oil circulation system corresponding to power plant equipment is:

[0133]

[0134] In the formula, θ represents the degree to which the overall oil quality deviates from its normal state, and θ represents the rate of change of the oil quality deterioration trend in the power plant equipment during the target data collection period.

[0135] The oil degradation index is calculated based on the degree to which the overall oil deviates from its normal state and its rate of degradation. Its core idea is to decompose the degree of oil degradation into two key dimensions: the degree of deviation from the current state (static indicator) and the rate of change of the degradation trend (dynamic indicator). This method is primarily based on the physical and chemical processes of oil degradation. In actual operation, lubricating oil degradation is usually accompanied by a gradual deviation of performance parameters from the normal range. This deviation is the result of accumulated factors such as the equipment operating environment, load conditions, and time. Simultaneously, the degradation rate reflects the degree to which the oil is affected by external factors (such as temperature shock, contaminant intrusion, and mechanical friction) in a short period. By combining these two dimensions, the current health status of the oil can be assessed more comprehensively; it not only reflects the current health status of the oil but also predicts future degradation trends, thus providing a scientific basis for operation and maintenance decisions. For example, when the degradation index is high and the degradation rate is accelerating, early warnings can be issued and targeted measures (such as replacing the lubricating oil or adjusting operating parameters) can be taken to prevent equipment failure due to lubrication failure. In addition, this calculation method is highly adaptable and can adjust the weights or parameter ranges according to different equipment or operating conditions, thereby improving the accuracy and practicality of the evaluation.

[0136] From a necessity standpoint, the operation of modern industrial equipment is highly dependent on lubricating oil, and oil deterioration is often an early warning sign of equipment failure. Relying solely on a single static or dynamic indicator may lead to overlooking potential risks or overreacting. However, by introducing a comprehensive indicator such as a deterioration index, the degree of risk can be assessed more scientifically, providing a reliable data foundation for equipment condition monitoring and lifespan management, ultimately improving the safety and economic efficiency of equipment operation.

[0137] S45. Compare the oil deterioration index of the lubricating oil circulation system corresponding to the power plant equipment with the set oil deterioration standard range [Q1, Q2], where Q1 is the set lower limit of oil deterioration and Q2 is the set upper limit of oil deterioration.

[0138] like This indicates that the oil condition of the corresponding lubricating oil circulation system of the power station equipment is normal.

[0139] like This indicates that the oil condition of the corresponding lubricating oil circulation system of the power plant equipment is slightly deteriorated;

[0140] like This indicates that the oil condition of the corresponding lubricating oil circulation system of the power plant equipment is severely deteriorated.

[0141] If the oil condition of the corresponding lubricating oil circulation system of the power station equipment is identified as normal, no warning signal will be issued.

[0142] If the oil condition of the lubricating oil circulation system corresponding to the power station equipment is identified as slightly deteriorated, a warning signal will be issued to the outside world.

[0143] If the oil condition of the corresponding lubricating oil circulation system of the power station equipment is identified as severely deteriorated, warning signals will be continuously issued through various warning methods.

[0144] Various forms of early warning include, but are not limited to, the following:

[0145] Audible and visual alarm: buzzer and flashing light alert;

[0146] Mobile device notifications: Send alerts to users via SMS, email, or a dedicated application;

[0147] Network alert: Sends alert information to networked security systems or control centers.

[0148] All data in this invention have been normalized to become dimensionless values.

[0149] The technical solution of the present invention has been described in conjunction with the specific experimental procedures shown in the accompanying drawings. However, the scope of protection of the present invention is not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions resulting from such changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for early warning of equipment condition deterioration in power plants, characterized in that, Includes the following steps: S1. Integrate a charge sensor and a high-precision dielectric sensor into the lubricating oil circulation system of power station equipment; S2. Set the collection time interval, divide the target collection time period into each collection time point according to the preset collection time interval, and then initially collect the charge data of the power station equipment oil at each collection time point, and based on it, identify the preliminary evaluation coefficient of the oil deterioration of the corresponding lubricating oil of the power station equipment. S3. Based on the preliminary assessment coefficient of the oil deterioration of the lubricating oil corresponding to the power station equipment, identify whether there is oil deterioration in the lubricating oil circulation system corresponding to the power station equipment. If so, proceed to step S4; otherwise, exit the method loop. S4. Based on the deployed high-precision dielectric sensors, the dielectric data of the power station equipment oil at each collection time point are extracted synchronously to deeply evaluate the oil status of the corresponding lubricating oil circulation system of the power station equipment and send corresponding early warning signals.

2. The method according to claim 1, characterized in that: In step S2, the charge data of the power plant equipment oil at each collection time point includes charge density, charge distribution uniformity, charge change rate, and charge fluctuation amplitude.

3. The method according to claim 1, characterized in that: Step S2, identifying the preliminary assessment coefficient of oil degradation for the corresponding lubricating oil of the power station equipment, specifically includes the following steps: S21. The z-score standardization method is used to preprocess the charge density, charge distribution uniformity, charge change rate and charge fluctuation amplitude of the oil in the power station equipment at each collection time point, and the SOM model is trained at the same time until the SOM model converges. S22. Map the charge density of the preprocessed power plant equipment oil at each sampling time point to the trained SOM model to obtain the weight vector of the neurons in the SOM model that maps the charge density of the preprocessed power plant equipment oil at each sampling time point. This weight vector is then denoted as α. h h is the number of each collection time point, and the value of h ranges from 1 to H, where H is the total number of collection time points; S23. Export the charge data of the power plant equipment at each normal time point under normal operating conditions from the equipment database. This gives the charge density of the power plant equipment at each normal time point under normal operating conditions. Similarly, input this data into the SOM model to obtain the weight vector of the charge density mapping neuron of the power plant equipment at each normal time point under normal operating conditions, and thus form the normal state region. S24. Calculate the Euclidean distance between the charge density mapping neuron and the normal state region at each acquisition time point. α i Let be the weight vector of the i-th neuron in the normal state region, where i is the number of each neuron in the normal state region, i = 1, 2, ..., N; N is the total number of neurons in the normal state region, and distance(·) is the distance calculation function; S25, Based on the analysis formula The fluctuation coefficient SL1 of the charge density of the corresponding lubricating oil of the power station equipment during the target data collection period is analyzed, where D is the set reference charge density value. S26. Based on the calculation method of the change fluctuation coefficient of the charge density of the lubricating oil corresponding to the power station equipment during the target acquisition time period, the change fluctuation coefficients of the charge distribution uniformity, charge change rate and charge fluctuation amplitude of the lubricating oil corresponding to the power station equipment during the target acquisition time period are calculated in the same way. S27. Finally, the preliminary assessment coefficient Ψ for the oil deterioration of the corresponding lubricating oil for the power station equipment is identified. The specific identification formula is as follows: In the formula, SL2, SL3, and SL4 represent the variation coefficients of charge distribution uniformity, charge change rate, and charge fluctuation amplitude of the lubricating oil corresponding to the power station equipment during the target data collection period, respectively; SL1', SL2', SL3', and SL4' are the maximum permissible variation coefficients of charge density, charge distribution uniformity, charge change rate, and charge fluctuation amplitude, respectively; ω1, ω2, ω3, and ω4 are the allocation weight factors corresponding to charge density, charge distribution uniformity, charge change rate, and charge fluctuation amplitude, respectively, reflecting the importance of each charge data in the preliminary assessment of oil deterioration.

4. The method according to claim 3, characterized in that: Step S21 involves training the SOM model until it converges, specifically including the following steps: S211. Select a two-dimensional grid to represent the SOM model, and randomly initialize the weight vector of each neuron in the SOM model. S212. Randomly select a sample x(t) from the standardized dataset as input data into the SOM model. Calculate the distance between the sample and all nodes in the SOM model, and take the node with the smallest distance as the best matching node for sample x(t): In the formula, the argmin(·) function represents finding the parameter values ​​that minimize a given function; w j (t) is the weight vector of node j, where j is the number of each node, and BMU is the best matching node for sample x(t); S213. Then, the weight vectors of each node in the SOM model are updated sequentially, with the following update formula: In the formula, w j (t+1) represents the weight vector of node j after one update, β1(t) represents the learning rate, and β2 bj (t) represents the neighborhood function, exp(·) is the exponential function, and γ b γ j Let represent the grid coordinates of the best matching node and node j, respectively, and σ(t) be the neighborhood width; S214. Then, repeat the iterative update, gradually reducing the learning rate and neighborhood width until the two-dimensional grid converges. When the two-dimensional grid converges, it means that the SOM model has converged.

5. The method according to claim 3, characterized in that: In step S23, forming a normal neuronal region specifically includes the following steps: S231. Extract the weight vectors of each neuron in the normal state region from the SOM model and summarize them to form a dataset. S232. Determine the number of clusters R using the elbow rule; S233. Randomly select R weight vectors as initial cluster centers; S234. For each neuron weight vector α i Calculate the Euclidean distance between it and each cluster center; S235. Then, based on the Euclidean distance between each neuron weight vector and each cluster center, each neuron weight vector is assigned to the cluster center with the smallest Euclidean distance. S236. Iteratively update each cluster center. The specific update steps are: obtain the number of internal neurons; and simultaneously update each cluster center by summing the neuron weight vectors of each cluster center and taking the average value to recalculate the center point of each cluster. S237. Repeatedly update each cluster center iteratively until the center point of each cluster center no longer changes or is less than the set threshold, which means that each cluster center has reached convergence. S238. After clustering is completed, the cluster center regions of neurons in the normal state region are identified, and they are summarized to obtain the normal state neuron regions.

6. The method according to claim 1, characterized in that: In step S3, the logic for identifying whether there is oil deterioration in the lubricating oil circulation system corresponding to the power station equipment is as follows: compare the preliminary assessment coefficient of oil deterioration of the lubricating oil corresponding to the power station equipment with the preset oil deterioration threshold coefficient. If the preliminary assessment coefficient of oil deterioration of the lubricating oil corresponding to the power station equipment is greater than the preset oil deterioration threshold coefficient, it is determined that there is oil deterioration in the lubricating oil circulation system corresponding to the power station equipment.

7. The method according to claim 1, characterized in that: In step S4, the dielectric data of the power plant equipment oil at each collection time point includes dielectric constant, dielectric loss factor, dielectric strength, and dielectric response time.

8. The method according to claim 1, characterized in that: Step S4, the in-depth assessment of the oil condition of the corresponding lubricating oil circulation system of the power plant equipment, specifically includes the following steps: S41. Extract the dielectric constant, dielectric loss factor, dielectric strength, and dielectric response time of the power plant equipment oil at each collection time point, and preprocess them using a standardized data processing method to obtain the preprocessed dielectric constant, dielectric loss factor, dielectric strength, and dielectric response time of the power plant equipment oil at each collection time point, and label them as δ1(h), δ2(h), δ3(h), and δ4(h), respectively. S42. Represent the dielectric properties at each acquisition time point as a vector X(h) = [δ1(h),δ2(h),δ3(h),δ4(h)], and then calculate the Mahalanobis distance MS(h) between the dielectric data of the power plant equipment oil at each acquisition time point and the mean of the normal dielectric data. S43. Based on the time series of the collected data, perform statistical analysis to extract the degree of deviation of the current oil product from the normal state and the rate of change of the oil product deterioration trend. S44. The final definition of the oil degradation index for the lubricating oil circulation system corresponding to power plant equipment is: In the formula, θ represents the degree to which the overall oil quality deviates from its normal state, and θ represents the rate of change of the oil quality deterioration trend in the power plant equipment during the target data collection period. S45. Compare the oil deterioration index of the lubricating oil circulation system corresponding to the power plant equipment with the set oil deterioration standard range [Q1, Q2], where Q1 is the set lower limit of oil deterioration and Q2 is the set upper limit of oil deterioration. like This indicates that the oil condition of the corresponding lubricating oil circulation system of the power station equipment is normal. like This indicates that the oil condition of the corresponding lubricating oil circulation system of the power plant equipment is slightly deteriorated; like This indicates that the oil condition of the corresponding lubricating oil circulation system of the power plant equipment is severely deteriorated.

9. The method according to claim 8, characterized in that: In step S42, calculating the Mahalanobis distance between the dielectric data of the power plant equipment oil at each data collection time point and the mean of the normal dielectric data specifically includes the following steps: S421. Export the dielectric data of the power station equipment at each normal time point under normal operating conditions from the equipment database, and perform mean calculation on the data to obtain the mean dielectric data of the power station equipment under normal operating conditions. Similarly, construct the dielectric data vector X′=[δ1′,δ2′,δ3′,δ4′] of the power station equipment under normal operating conditions, and construct the normal covariance matrix of the power station equipment under normal operating conditions: In the formula, δ1′ p δ2′ p δ3′ p δ4′ p Let Var(·) represent the dielectric constant, dielectric loss factor, dielectric strength, and dielectric response time of the power plant equipment at the p-th normal time point under normal operating conditions, respectively. Var(·) represents the variance calculation function, Cov(·) represents the covariance calculation function, and p is the number of each normal time point. S422. Calculate the Mahalanobis distance between the dielectric data of the power plant equipment oil at each data collection time point and the mean of the normal dielectric data: In the formula, T represents the transpose symbol.

10. The method according to claim 8, characterized in that: Step S43, extracting the degree to which the current oil deviates from its normal state and the rate of change of the oil deterioration trend, specifically includes the following steps: S431, through calculation formula Calculate the degree to which the current overall oil quality deviates from its normal state. S432. Synchronously calculate the rate of change of oil deterioration trend in power plant equipment during the target data collection period: In the formula, Δt is the preset data acquisition time interval, and t h This represents the time value at the h-th data collection point.

Images