Method for monitoring circulation of cold water system in stator winding

Abstract

The invention provides a method for monitoring circulation of a cooling water system in a stator winding. The method comprises the following steps: obtaining a training sample set according to obtained historical outlet water temperature sequence data; training the temperature prediction model by using the training sample set; acquiring current temperature sequence data of the plurality of temperature measuring points in real time, and inputting the current temperature data into the trained temperature prediction model to obtain predicted temperature sequence data; and determining a fault diagnosis result according to the predicted temperature data and a preset temperature threshold. The temperature prediction model is obtained through historical temperature data training, and the temperature prediction model can be utilized to realize temperature prediction in an operation state, so that fault monitoring is realized.

Description

Technical Field

[0001] The present application belongs to the technical field of operation and maintenance management of power industry generator equipment, and specifically, relates to a method for monitoring the flowability of a cooling water system in a stator winding. Background Art

[0002] The stator windings of water-cooled steam turbine generators and their internal cooling water systems play a vital role in power production. The cooling system in the stator windings needs to be kept in good operating condition to ensure the reliability and safety of the generator under high load conditions. If the cooling system fails, it may cause the stator windings to overheat, which in turn may cause equipment damage or shutdown accidents, seriously affecting the stability of power supply.

[0003] Traditionally, the detection of stator winding cooling systems mainly relies on ultrasonic and hot water flow methods during shutdown inspections. Although these methods can evaluate the performance of the cooling system to a certain extent, they have significant limitations. First, they cannot monitor the flow of cooling water in real time when the generator is running, which limits the ability to diagnose dynamic thermal faults. Second, the measurement accuracy is easily restricted by on-site conditions, such as the length limit of the insulating water pipe, resulting in insufficient accuracy of the detection results. In addition, these methods often have difficulty in effectively identifying anomalies in the case of minor blockages or slight deviations in the sensor position.

[0004] In order to improve the accuracy and timeliness of fault warning, discover potential problems in the cooling water system of the stator winding at an early stage, and prevent serious accidents, advanced sensing technology and data analysis methods should be introduced. Some researchers have tried to establish a mechanism-based fault diagnosis model by studying the temperature distribution law, but due to the large number of influencing factors, the model is prone to failure under complex and changeable working conditions. Summary of the invention

[0005] The technical problem solved by the present application is: how to provide a method for diagnosing faults of the stator winding of a generator in operation.

[0006] The present application provides a method for monitoring the flowability of a cooling water system in a stator winding, and the method for monitoring the flowability of a cooling water system in a stator winding comprises:

[0007] A training sample set is obtained according to the acquired historical water outlet temperature series data;

[0008] Using the training sample set to train the temperature prediction model;

[0009] Acquire current temperature sequence data of multiple temperature measurement points in real time, and input the current temperature data into the trained temperature prediction model to obtain predicted temperature sequence data;

[0010] The fault diagnosis result is determined based on the predicted temperature data and the preset temperature threshold.

[0011] Optionally, the method for obtaining a training sample set according to the acquired historical water outlet temperature sequence data includes:

[0012] Perform sliding window extraction on the acquired historical outlet water temperature series data to obtain the original temperature time series data;

[0013] The original temperature time series data is expanded to obtain simulated temperature series data, and the simulated temperature series data is added to the original temperature time series to obtain a training sample set.

[0014] Optionally, the method of training the temperature prediction model using the training sample set includes:

[0015] Pre-training a temperature prediction model using the training sample set;

[0016] The pre-trained temperature prediction model is fine-tuned using the original temperature time series data.

[0017] Optionally, the method of expanding the original temperature time series data to obtain simulated temperature series data is:

[0018] Generate simulated temperature series data using different kernel functions of the Gaussian process.

[0019] Optionally, the temperature prediction model is a prediction model constructed based on Transformer.

[0020] Optionally, the predicted temperature sequence data includes predicted temperatures at multiple time points at multiple temperature measurement points, and the predicted temperature sequence data can be expressed as in represents the predicted temperature of each measuring point at the jth moment, M represents the number of temperature measuring points, and N represents the number of predicted moments.

[0021] Optionally, the method for determining a fault diagnosis result according to the predicted temperature data and a preset temperature threshold comprises:

[0022] The temperature difference sequence {ΔX1, ΔX2, ..., ΔX N}, for each prediction time j, the maximum temperature difference between the measuring points is:

[0023] Determine the maximum value in the temperature difference series: ΔX max =max({ΔX1,ΔX2,...,ΔX N});

[0024] When the maximum value ΔXmax Greater than the preset temperature threshold ΔX threshold , it is determined that a thermal fault has occurred.

[0025] The present application provides a method for monitoring the flowability of a cooling water system in a stator winding, which has the following technical effects:

[0026] The present application obtains a temperature prediction model by training historical temperature data, and the temperature prediction model can be used to realize temperature prediction in the running state, thereby realizing fault monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A flow chart of a method for monitoring the flowability of a cooling water system in a stator winding according to one or more embodiments;

[0028] Figure 2 is a schematic diagram of a kernel function of a Gaussian process according to one or more embodiments;

[0029] Figure 3 is a schematic diagram of a network structure of a temperature prediction model according to one or more embodiments;

[0030] Figure 4 is a schematic diagram of prediction results of different models according to one or more embodiments;

[0031] Figure 5 Schematic diagram of actual temperature difference and predicted temperature difference distribution according to one or more embodiments. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0033] Before describing the various embodiments of the present application in detail, the technical concept of the present application is first briefly described: the current means of monitoring thermal faults in stator windings cannot be monitored in the operating state, and the measurement accuracy is easily constrained by on-site conditions, resulting in monitoring difficulties. To this end, the present application provides a method for monitoring the fluidity of a cooling water system in a stator winding. Based on a deep learning method, a temperature prediction model is first obtained by training based on historical water outlet temperature sequence data, and then the current temperature sequence data of multiple temperature measurement points are collected in real time and input into the temperature prediction model to obtain predicted temperature sequence data. The fault diagnosis result is determined in combination with a preset temperature threshold. This method can realize fault monitoring in the operating state. The specific principles of the method for monitoring the fluidity of a cooling water system in a stator winding of the present application are described below in combination with more embodiments.

[0034] Specifically, Figure 1As shown, the method for monitoring the flowability of the cooling water system in the stator winding provided in the first embodiment includes the following steps:

[0035] Step S10: obtaining a training sample set according to the acquired historical water outlet temperature sequence data.

[0036] Step S20: training the temperature prediction model using the training sample set.

[0037] Step S30: collect and obtain current temperature sequence data of multiple temperature measurement points in real time, and input the current temperature data into the trained temperature prediction model to obtain predicted temperature sequence data.

[0038] Step S40: determining a fault diagnosis result according to the predicted temperature data and a preset temperature threshold.

[0039] Exemplarily, in order to adopt a model training method similar to natural language, this embodiment first normalizes the data and then discretizes it into a fixed number of discrete values.

[0040] For the time series vector X, the normalization process is as follows:

[0041]

[0042] Among them, X′ is the normalized time series, X min and X max are the minimum and maximum values ​​of X respectively.

[0043] Next, the normalized time series X′ is discretized into a fixed number of discrete values, as shown in the following formula.

[0044]

[0045] Among them, q(x) represents the discretized value, b1, b2, …, b B-1 is the threshold point determined in the discretization process, and B is the number of discrete intervals.

[0046] In one or more embodiments, the method for obtaining a training sample set according to the acquired historical water outlet temperature sequence data in step S10 includes the following sub-steps:

[0047] Step S101: Perform sliding window extraction on the acquired historical water outlet temperature sequence data to obtain the original temperature time series data. Assume that the historical water outlet temperature sequence data is X = {x1, x2, ..., x n}, the size of the sliding window is w, then the original temperature time series data extracted by the sliding window can be expressed as X i ={x i ,x i+1,...,x i+w-1}, where i is the starting index of the sliding window.

[0048] Step S102: Expand the original temperature time series data to obtain simulated temperature series data, add the simulated temperature series data to the original temperature time series, and obtain a training sample set. Exemplarily, different kernel functions of the Gaussian process are used to generate simulated temperature series data. This can alleviate the problem of fewer fault operation data samples and greatly expand the training sample space.

[0049] Compared with the huge number of parameters of the Transformer model, the sample size of the time series data of the outlet water temperature of the cooling water system in the stator winding is still very insufficient. Although traditional data enhancement methods, such as adding noise and masking, can alleviate overfitting to a certain extent, increasing the number of data sets is still the best choice. Time series are usually split into components such as trends, seasonality, and random errors, and for some more complex time series, the splitting granularity will be more complex. In order to generate a large amount of time series data with diverse patterns, this application uses Gaussian processes to sample with different kernel functions to generate different time series patterns.

[0050] Gaussian process (GP) is a nonparametric Bayesian method for defining the distribution of a function. A Gaussian process is defined by its mean function m(t) and covariance function (or kernel function) k(t, t′), expressed as:

[0051]

[0052] Where m(t) is the expected value of the Gaussian process at point t, and κ(t,t′) is the covariance between points t and t′. By selecting different kernel functions, different time series patterns can be captured, such as Figure 2 shown.

[0053] The linear kernel is defined as κ(t,t′)=σ 2 (tc)(t′-c), suitable for capturing time series with linear trends.

[0054] The RBF kernel (radial basis function kernel) is defined as Good for capturing smooth local changes.

[0055] The periodic core is defined as Good for capturing seasonal or cyclical patterns.

[0056] The combination of these kernel functions can generate synthetic time series with complex patterns. These time series are then combined with the existing temperature time series to randomly draw Time series of a certain length I are sampled from existing and generated datasets, and their weighted combination is taken as the new series:

[0057]

[0058] in represents the i-th time series, λ i represents the weight of the i-th time series.

[0059] In one or more embodiments, the temperature prediction model in step S20 is a prediction model constructed based on Transformer. Figure 3 The encoder-decoder structure shown in the figure, where the encoder input is the input sequence, that is, the historical temperature sequence, the decoder input is the biased output sequence, that is, the temperature sequence to be predicted, and the output is the probability distribution of the next moment. In the inference process, it is in the form of autoregression, and the result of each step of calculation will be further placed in the biased output sequence of the decoder input link.

[0060] Exemplarily, in step S20, the method of training the temperature prediction model using the training sample set includes: pre-training the temperature prediction model using the training sample set; and fine-tuning the pre-trained temperature prediction model using the original temperature time series data. The initial learning rate used for fine-tuning is 0.001, and decays to 0.95 times the previous value every 10 rounds.

[0061] When the current temperature sequence data of multiple temperature measuring points are collected in real time in step S20, the temperature measuring points should be evenly distributed on the outlet pipe of the cold water system of the stator winding, and the collection cycle is 1 minute.

[0062] For example, the historical water outlet temperature sequence data has a total of 42 measurement points, of which measurement point 17 has a thermal fault, and the data sampling period is 1 minute. The data is segmented into sliding windows, with a window length of 100 minutes and a step length of 20 minutes.

[0063] For example, the predicted temperature sequence data obtained in step S30 includes predicted temperatures at multiple time points of multiple temperature measurement points. The predicted temperature sequence data can be expressed as in represents the predicted temperature of each measuring point at the jth moment, M represents the number of temperature measuring points, and N represents the number of predicted moments.

[0064] Furthermore, the method for determining the fault diagnosis result according to the predicted temperature data and the preset temperature threshold in step S40 includes:

[0065] Step S401: Calculate the temperature difference sequence {ΔX1, ΔX2, ..., ΔXN}, where for each prediction time j, the maximum temperature difference between the measurement points is:

[0066]

[0067] Step S402: Determine the maximum value in the temperature difference sequence: ΔX max =max({ΔX1,ΔX2,...,ΔX N});

[0068] Step S403: When the maximum value ΔX max Greater than the preset temperature threshold ΔX threshold When a thermal fault occurs, the system issues a fault warning. Otherwise, the system continues to monitor the real-time temperature data and repeats the above steps.

[0069] For example, the preset temperature threshold ΔX threshold Set to 7°C and perform liquidity monitoring on the existing test data set. The temperature difference at multiple measuring points is compared with the threshold. If it is greater than the threshold, the system issues a fault warning. Otherwise, the system continues to monitor the real-time temperature data and repeats the above steps.

[0070] Exemplarily, in order to verify the prediction accuracy of the temperature prediction model of this embodiment, different models are compared, and the comparison models are a model based on the ARIMA algorithm and a model based on the LSTM algorithm. Figure 4 The single prediction examples of different models on the same data are shown. It can be seen that the temperature prediction model of this embodiment (based on the Transformer algorithm) has a stronger ability to express data and a higher degree of fit. The recognition accuracy of the operating state reaches 94.5%, and it has a high precision and recall rate, ensuring a low false alarm rate and missed alarm rate. Figure 5 As shown, the distribution of the real temperature difference and the predicted temperature difference obtained in this embodiment is shown. It can be seen that the real temperature difference and the predicted temperature difference are generally linearly correlated, that is, the predicted temperature difference threshold can replace the real temperature difference threshold as a basis for judgment.

[0071] The specific implementation methods of the present application are described in detail above. Although some embodiments have been shown and described, those skilled in the art should understand that these embodiments can be modified and improved without departing from the principles and spirit of the present application whose scope is defined by the claims and their equivalents. These modifications and improvements should also be within the scope of protection of the present application.

Claims

1. A method for monitoring the flowability of a cooling water system in a stator winding, characterized in that: The method for monitoring the flowability of the cooling water system in the stator winding comprises: A training sample set is obtained according to the acquired historical water outlet temperature series data; Using the training sample set to train the temperature prediction model; Acquire current temperature sequence data of multiple temperature measurement points in real time, and input the current temperature data into the trained temperature prediction model to obtain predicted temperature sequence data; The fault diagnosis result is determined based on the predicted temperature data and the preset temperature threshold.

2. The method for monitoring the flowability of the cooling water system in the stator winding according to claim 1, characterized in that: The method for obtaining a training sample set based on the acquired historical water outlet temperature sequence data includes: Perform sliding window extraction on the acquired historical outlet water temperature series data to obtain the original temperature time series data; The original temperature time series data is expanded to obtain simulated temperature series data, and the simulated temperature series data is added to the original temperature time series to obtain a training sample set.

3. The method for monitoring the flowability of the cooling water system in the stator winding according to claim 2, characterized in that: The method for training the temperature prediction model using the training sample set includes: Pre-training a temperature prediction model using the training sample set; The pre-trained temperature prediction model is fine-tuned using the original temperature time series data.

4. The method for monitoring the flowability of the cooling water system in the stator winding according to claim 1, characterized in that: The method of expanding the original temperature time series data to obtain simulated temperature series data is: Generate simulated temperature series data using different kernel functions of the Gaussian process.

5. The method for monitoring the flowability of the cooling water system in the stator winding according to claim 1, characterized in that: The temperature prediction model is a prediction model built based on Transformer.

6. The method for monitoring the flowability of the cooling water system in the stator winding according to claim 1, characterized in that: The predicted temperature sequence data includes predicted temperatures at multiple time points at multiple temperature measurement points. The predicted temperature sequence data can be expressed as in represents the predicted temperature of each measuring point at the jth moment, M represents the number of temperature measuring points, and N represents the number of predicted moments.

7. The method for monitoring the flowability of the cooling water system in the stator winding according to claim 6, characterized in that: The method for determining a fault diagnosis result according to predicted temperature data and a preset temperature threshold comprises: The temperature difference sequence {ΔX1, ΔX2, ..., ΔX N }, for each prediction time j, the maximum temperature difference between the measuring points is: Determine the maximum value in the temperature difference series: ΔX max =max({ΔX1,ΔX2,...,ΔX N }); When the maximum value ΔX max Greater than the preset temperature threshold ΔX threshold , it is determined that a thermal fault has occurred.

Images