Pumped storage unit multi-source signal abnormal information extraction and fault diagnosis method based on enhanced hierarchical Poincare map index

Abstract

The invention provides a pumped storage unit multi-source signal abnormal information extraction and fault diagnosis method based on an enhanced hierarchical Poincare map index, and relates to the field of intelligent operation and maintenance of a hydroelectric power generation system. A pumped storage unit multi-source vibration signal feature extraction tool fusing multi-frequency nonlinear dynamic indexes and a feature level fusion strategy is developed, a pumped storage unit intelligent fault diagnosis model integrating signal acquisition, feature extraction and mode recognition is constructed, and accurate recognition of the abnormal state of the unit is achieved. Rich state information contained in multi-source vibration signals of the pumped storage unit can be deeply excavated, on the premise that normal production and power generation of the hydropower unit are not affected, efficient recognition of the hydropower station on the early abnormal state of equipment is improved, shutdown and maintenance cost caused by equipment faults is reduced, and the working efficiency of the hydropower station is improved. And an effective technical means is provided for improving the operation reliability of the shaft system and ensuring the safe and stable operation of the hydroelectric generating set.

Description

Technical Field

[0001] The present invention relates to the field of intelligent operation and maintenance of hydropower generation systems, and in particular to a method for extracting abnormal information and diagnosing faults of multi-source signals of a pumped storage unit based on an enhanced hierarchical Poincare map index. Background Art

[0002] As a key area of ​​hydropower development, pumped storage undertakes tasks such as peak regulation, frequency regulation, and phase regulation, which greatly enhances the stability of the new power system based on renewable energy. However, with the commissioning of a large number of pumped storage power plants (PSPPs), some problems have gradually emerged. On the one hand, ultra-high water heads will produce strong hydraulic excitation, and the resulting water hammer pressure will reduce the stability of pumped storage units (PSUs). On the other hand, PSPPs will frequently switch operating conditions according to grid dispatch instructions, causing the units to operate outside the optimal conditions, accelerating their degradation and wear. Therefore, conducting early fault diagnosis of pumped storage units is of great significance to maintaining the safe operation of power stations.

[0003] Generally, the core link in the fault diagnosis of pumped storage units is the feature extraction of signals. The success or failure of the diagnosis depends on whether it can accurately extract features that reflect the abnormal information of rotating machinery. Extracting features from the collected data has become an important research focus. At present, the most commonly used feature extraction methods in the fault diagnosis of pumped storage units are time domain indicators, frequency domain indicators, and time-frequency domain indicators. However, these methods have the following problems: First, traditional time domain and frequency domain indicators are only applicable to smooth and linear signals and cannot effectively extract information from complex nonlinear signals; second, most of the features extracted by deep learning have poor interpretability, and the parameters of these models are difficult to train.

[0004] In recent years, a large number of nonlinear dynamic indicators have been successfully applied to quantify the fault characteristics of rotating machinery signals, proving the effectiveness of these indicators as feature extraction tools for pumped storage units. However, the vibration transmission path of rotating machinery is very complex, and equipment failures may be caused by a variety of factors. This makes it difficult for a single sensor to fully reflect the motion characteristics of rotating machinery, so it is important to combine multiple sensors for early fault diagnosis of units. However, most of the existing multi-sensor information fusion technologies belong to data-level fusion, which has the disadvantages of large amount of information to be processed, poor fault tolerance, and low fusion level, which may lead to low diagnostic accuracy. Summary of the invention

[0005] In response to these problems, the present invention proposes an enhanced hierarchical Poincare map index (EHPPI), which is a nonlinear dynamic index without any adjustable hyperparameters. At the same time, a feature-level fusion strategy is adopted to propose a method for extracting abnormal information and diagnosing faults of multi-source signals of pumped storage units based on the enhanced hierarchical Poincare map index, so as to realize early intelligent diagnosis of typical faults of pumped storage units. In order to reduce the possibility of shutdown of pumped storage units due to faults and significantly improve the safety of equipment, an effective technical means is provided; at the same time, without affecting the daily operation of the power station site, the perception ability of the pumped storage power station for early faults is improved, and the losses caused to equipment and personnel due to faults are reduced.

[0006] In order to realize the above technical features, the object of the present invention is achieved as follows: a method for extracting abnormal information and diagnosing faults of multi-source signals of a pumped storage unit based on an enhanced hierarchical Poincare map index, comprising: Step 1: Obtain vibration signals of the pumped storage unit through vibration sensors and data acquisition instruments; Step 2: Integrate the enhanced hierarchical decomposition method and the Poincaré map index to propose an enhanced hierarchical Poincaré map index; Step 3: Use the enhanced hierarchical Poincare map index to extract the information of vibration signals from different channels, connect all the information in series, and form features that characterize multi-sensor data fusion; Step 4: Establish an intelligent diagnosis model for pumped storage units based on enhanced hierarchical Poincare map indicators and random forests; Step 5: Use the existing vibration data of the pumped storage unit to train the intelligent diagnosis model of the pumped storage unit to obtain a high-precision diagnosis model of the pumped storage unit; Step 6: Use the pumped storage unit test data to test the fault detection effect of the pumped storage unit intelligent diagnosis model to verify the effectiveness of the technology.

[0007] Preferably, the process of acquiring multi-source vibration data is as follows: Acceleration sensors are installed in the horizontal X direction, Y direction and vertical direction of the top cover, and the sensor electrical signals are sent to the data acquisition instrument through hard wiring, and the electrical signals are converted into vibration signals through calibration.

[0008] Preferably, the enhanced hierarchical decomposition method and the Poincaré map index are integrated to propose an enhanced hierarchical Poincaré map index (EHPPI), which specifically includes the following steps: First, using the moving average algorithm and the moving difference algorithm, we obtain Signal components of different frequencies, the process is: The average operator and differential operators for: ; ; in, represents the original signal, represent Middle i Sample point data, N Represents the total number of sample points, and Respectively represent Low-frequency information and high-frequency information; When the number of layers is hour, When it is 0 or 1, the corresponding operator for: ; Conclusion The components of different frequencies are: ; in, Indicates the number of layers The Quantity; Represents the index number of the frequency component, and its expression is: ; in, , ; Then, according to the phase space reconstruction theory, we get two vectors and , and and Using the sample points in as the abscissa and ordinate, draw the corresponding Poincare map: ; in, yes The Sample points; Finally, extracting from different parameters SD 1 and SD 2 as the index at different frequency scales, the complete enhanced hierarchical Poincare map index is obtained, and the process is as follows: Extract from the Poincare map of different signal components SD 1 and SD 2 as indicators of different time and frequency scales: ; ; in, and Indicates signal components SD 1 and SD 2; All frequency scales and Putting it all together, we get EHPPI: ; .

[0009] Preferably, the enhanced hierarchical Poincare map index is used to calculate the indexes of the vibration signals of the top cover in the horizontal X direction, Y direction and vertical direction respectively, and the indexes obtained from the three vibration signals are connected in series to obtain the characteristics characterizing the current operating state of the pumped storage unit: ; in, It is a characteristic that characterizes the current operating status of the pumped storage unit; , , The vibration signals of the top cover in the horizontal X direction, Y direction and vertical direction are extracted by EHPPI respectively. SD 1 indicator; , , The vibration signals of the top cover in the horizontal X direction, Y direction and vertical direction are extracted by EHPPI respectively. SD 2 indicators.

[0010] Preferably, random forest is introduced as an identifier, and an intelligent fault diagnosis model of pumped storage unit based on enhanced hierarchical Poincare map index and random forest is established; in this model, the enhanced hierarchical Poincare map index plays the role of feature extraction, which inputs the comprehensive features extracted from the vibration signals in the horizontal X direction, Y direction and vertical Z direction of the top cover into the random forest.

[0011] Preferably, in step 5, the existing vibration data of the pumped storage unit is used to train the intelligent diagnosis model of the pumped storage unit; wherein, these vibration data are with clear data labels, and the intelligent diagnosis model of the pumped storage unit finds the best mapping relationship between the vibration signal and the state through continuous training and fitting, and obtains a trained high-precision intelligent diagnosis model of the pumped storage unit; Preferably, the existing vibration data of the pumped storage unit is used to train the intelligent diagnosis model of the pumped storage unit to obtain a high-precision diagnosis model of the pumped storage unit; the test data of the pumped storage unit is used to test the fault detection effect of the intelligent diagnosis model of the pumped storage unit, so as to verify the effectiveness of the technology.

[0012] The present invention has the following beneficial effects: 1. The present invention discloses a method for extracting abnormal information of multi-source signals of pumped storage units based on enhanced hierarchical Poincare map indicators, which can retain the characteristic information of multi-source signals to the greatest extent. This method overcomes the problems of large amount of information processing, poor fault tolerance and low fusion level in data-level fusion strategy when extracting multi-source signal features, and provides a new technical means for multi-sensor state fusion identification of hydropower units.

[0013] 2. The intelligent fault diagnosis method for pumped storage units based on enhanced hierarchical Poincare map indicators and random forests disclosed in the present invention can directly use multi-source vibration signals to directly determine the operating status of the unit. It is simple to operate, does not increase the burden of on-site operation, and has good engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0015] Figure 1 It is a schematic diagram of the process of the present invention.

[0016] Figure 2 This is a structural block diagram of the enhanced hierarchical Poincare map index of the present invention.

[0017] Figure 3 It is a signal acquisition flow chart of the present invention.

[0018] Figure 4 (a)(b)(c)(d)(e) are the on-site pictures of equipment failure of a pumped storage unit.

[0019] Figure 5 (a)(b)(c)(d)(e) are the waveform diagrams of the vibration signal of a pumped storage unit failure.

[0020] Figure 6 The present invention extracts a high-dimensional feature distribution map of the pumped storage unit.

[0021] Figure 7 The present invention extracts a low-dimensional feature distribution map of the pumped storage unit.

[0022] Figure 8 This is a diagram showing the recognition results of the method of the present invention on the test set samples.

[0023] Fig. 9 This is a diagram showing the recognition results of the method of the present invention on the test set samples. DETAILED DESCRIPTION

[0024] The present invention is further described below in conjunction with embodiments. It is necessary to point out that the following embodiments are only used to further illustrate the present invention and cannot be understood as limiting the scope of protection of the present invention. Some non-essential improvements and adjustments made by technical personnel in the field based on the above invention content still fall within the scope of protection of the present invention.

[0025] Embodiment 1: See also Figure 1 , a method for extracting abnormal information and diagnosing faults of multi-source signals of pumped storage units based on enhanced hierarchical Poincare map index, comprising: Step 1: Install acceleration sensors in the horizontal X direction, Y direction and vertical direction of the top cover, and send the sensor electrical signals to the data acquisition instrument through hard wiring, and convert the electrical signals into vibration signals through calibration.

[0026] Step 2: The enhanced hierarchical decomposition method and the Poincaré map index are integrated to propose the enhanced hierarchical Poincaré map index (EHPPI). The EHPPI specifically includes the following steps: First, using the moving average algorithm and the moving difference algorithm, we obtain Signal components of different frequencies, the process is: The average operator and differential operators for: ; ; in, represents the original signal, represent Middle i Sample point data, N Represents the total number of sample points, and Respectively represent Low-frequency information and high-frequency information; When the number of layers is hour, When it is 0 or 1, the corresponding operator for: ; Conclusion The components of different frequencies are: ; in, Indicates the number of layers The Quantity; Represents the index number of the frequency component, and its expression is: ; in, , ; Then, according to the phase space reconstruction theory, we get two vectors and , and and Using the sample points in as the abscissa and ordinate, draw the corresponding Poincare map: ; in, yes The Sample points; Finally, extracting from different parameters SD 1 and SD 2 as the index at different frequency scales, the complete enhanced hierarchical Poincare map index is obtained, and the process is as follows: Extract from the Poincare map of different signal components SD 1 and SD 2 as indicators of different time and frequency scales: ; ; in, and Indicates signal components SD 1 and SD 2; All frequency scales and Putting it all together, we get EHPPI: ; .

[0027] Step 3: Use the enhanced hierarchical Poincare map index to calculate the indexes of the vibration signals in the horizontal X direction, Y direction and vertical direction of the top cover respectively, and connect the indexes obtained from these three vibration signals in series to obtain the characteristics that characterize the current operating status of the pumped storage unit: ; in, It is a characteristic that characterizes the current operating status of the pumped storage unit; , , The vibration signals of the top cover in the horizontal X direction, Y direction and vertical direction are extracted by EHPPI respectively. SD 1 indicator; , , The vibration signals of the top cover in the horizontal X direction, Y direction and vertical direction are extracted by EHPPI respectively. SD 2 indicators.

[0028] Step 4: Introduce random forest as a recognizer, and establish an intelligent fault diagnosis model for pumped storage units based on enhanced hierarchical Poincare map indicators and random forests; in this model, the enhanced hierarchical Poincare map indicator plays the role of feature extraction, which inputs the comprehensive features extracted from the vibration signals in the horizontal X direction, Y direction, and vertical Z direction of the top cover into the random forest. Random forest plays the role of an automatic recognizer, and its working principle is: First, we use the bootstrap sampling technique to randomly select L samples to create a sub-training set. The process is repeated T times, generate T In each sub-training set, about 2 / 3 of the samples are called in-bag samples, which are used to train the CART model, and the remaining 1 / 3 of the samples are called out-of-bag samples, which are reserved for evaluating the classification error rate of the generated tree model.

[0029] Then, a training CART model is constructed on each unique in-bag dataset without any pruning. Note that during model creation, the features of each node are randomly selected from the available features.

[0030] Finally, the test data is passed to the trained CART model to generate T The final prediction result is determined by majority voting, i.e. T The prediction with the highest frequency among the results is taken as the final result.

[0031] Step 5: Use the existing vibration data of the pumped storage unit to train the intelligent diagnosis model of the pumped storage unit; wherein, these vibration data have clear data labels, and the intelligent diagnosis model of the pumped storage unit finds the best mapping relationship between the vibration signal and the state through continuous training and fitting, and obtains a trained high-precision intelligent diagnosis model of the pumped storage unit.

[0032] Step 6: Use the intelligent diagnosis model of the pumped storage unit to extract features and recognize patterns of the vibration signals in the test data set, give the operating status of these unknown samples, and use indicators such as accuracy, precision, recall, and F1-score to evaluate the performance of the intelligent diagnosis model.

[0033] Example: The main body of the micro pumped storage power station used in this embodiment consists of an upstream reservoir, a piping system, a regulating room, a downstream reservoir, a monitoring system, a measurement system, a water supply system and an excitation system. Detailed information about these systems. The two units are arranged symmetrically and share the same piping system. Unit #1 is a double-fed variable-speed pump-turbine and has been in operation for 3 years. Figure 3 The data collection process of the present invention is shown. Figure 3 As shown, the acceleration sensors installed in the horizontal X direction, Y direction and vertical direction of the top cover of the pumped storage unit are used to sense the change of the vibration of the hydropower unit over time, and the current signal is connected to the data acquisition instrument through hard wiring to be calibrated and converted into a vibration signal; During the three-year operation of the micro-hydropower plant, Unit #1 suffered from multiple faults such as vortex eccentricity, slight main shaft friction, bearing wear and guide vane misalignment due to water hammer effect, high water impurity content and poor component manufacturing quality. These mechanical failures led to the unplanned shutdown of Unit #1 and the inability to connect to the grid. Specific on-site pictures can be seen Figure 4 From the historical database, the present invention collected 50 groups of data on five operating states, including normal, vortex belt eccentricity, slight main shaft friction, bearing wear and guide vane misalignment. Each group of signals contained vibration signals of three channels in the horizontal X direction, Y direction and vertical direction of the top cover. The sampling frequency of each group of signals was 2048 Hz and the duration was 1 s.

[0034] Figure 5 The waveform diagram of the vibration signal under these states is shown, from which it can be seen that the signal amplitude under normal and bearing wear states is smaller than the signal amplitude under the other three states. The reason for this change is that normal and bearing wear indicate healthy or sub-healthy states, while the other three states indicate states that seriously endanger the safety of the unit. At the same time, we also observed amplitude changes between vibration signals of different channels. This change is due to the different performance of the pumped storage unit when various faults occur, resulting in different information transmission modes. These transmission relationships help us capture the effective information contained in the signal. However, the waveforms of these signals are too complex, and there is a certain amount of environmental noise. Therefore, it is extremely difficult to judge the working state of the pumped storage unit based solely on the differences in amplitude and waveform. Therefore, it is necessary to apply the method involved in the present invention in this case.

[0035] EHPPI is used to extract the vibration signal characteristics of the three channels of the top cover of the pumped storage unit in the horizontal X direction, Y direction and vertical direction, while RF is used to automatically identify these abnormal conditions. Figure 6 The vibration signal characteristics of the pumped storage unit under different working conditions are shown. Figure 6As shown in the figure, the feature differences between signals in the same state are not large, and only individual features fluctuate, indicating that the features extracted by EHPPI have a certain stability. From the distribution of features between signals in different states, there are great differences between them. Among them, the feature values ​​of guide vane misalignment and slight friction of the main shaft are significantly higher than those of the other three states, while the feature values ​​of vortex belt eccentricity are slightly higher than those of normal and bearing wear, and show an obvious positive correlation with the signal amplitude. Based on these feature results, we can distinguish guide vane misalignment, slight friction of the main shaft, and vortex belt eccentricity from the features very well, but it is very difficult to completely distinguish normal and bearing wear. This is mainly because bearing wear represents an undesirable state, and its main impact is to increase the temperature of the equipment rather than increase vibration.

[0036] The T-distributed stochastic neighbor embedding algorithm (TSNE) is used to reduce the dimension of the obtained high-dimensional features. The specific feature distribution of the obtained low-dimensional features can be seen Figure 7 .like Figure 7 As shown, guide vane misalignment and slight friction with the main shaft are easily distinguished, while sporadic bearing wear is mixed with normal wear, which is consistent with the above analysis. In general, the features extracted by the EHPPI algorithm involved in the present invention have a high degree of discrimination.

[0037] Subsequently, in order to verify the effectiveness of the intelligent fault diagnosis model of pumped storage units based on enhanced hierarchical Poincare map index and random forest involved in the present invention, the present invention divides the data set into a training set and a test set equally. The former is used to train the RF model, and the latter is used to evaluate the recognition effect of the model. Figure 8 The figure shows the diagnostic results of the proposed pumped storage unit intelligent fault diagnosis model after training and applying it to the training set. From the figure, we can see that all the states of the pumped storage unit are accurately identified, which shows that the intelligent diagnosis model has been well trained. The trained high-precision pumped storage unit diagnosis model is applied to the test data set with unknown sample labels. The results are shown in Figure 2. Fig. 9 As shown. Fig. 9 The results shown show that only one bearing wear is in normal state, which is consistent with the feature distribution analysis. Despite this misclassification, the sub-healthy state of bearing wear does not directly endanger the normal operation of the equipment because it can be quickly resolved through temperature monitoring. In addition, the present invention introduces four indicators to quantify the recognition performance of different models: accuracy rate (AR), precision rate (PR), recall rate (RR) and F1-score. The specific formula is as follows: ; ; ; ; In the formula, TP , TN , FP and FN Denotes true positive, true negative, false positive, and false negative.

[0038] Fig. 9 The indicators of all states are displayed. It can be seen from the figure that the diagnostic indicators of all states are above 96%, which illustrates the superiority of the multi-source signal abnormal information extraction and fault diagnosis method of the pumped storage unit based on the enhanced hierarchical Poincare map indicator involved in the present invention.

[0039] The present invention is the embodiment of the present invention as described above. The above embodiments and the specific parameters in the embodiments are only for the purpose of clearly describing the invention verification process, and are not used to limit the patent protection scope of the present invention. The patent protection scope of the present invention is still subject to its claims. Any equivalent structural changes made by using the description and drawings of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for extracting abnormal information and diagnosing faults of multi-source signals of pumped storage units based on enhanced hierarchical Poincare map indicators, characterized in that: include: Step 1: Obtain vibration signals of the pumped storage unit through vibration sensors and data acquisition instruments; Step 2: Integrate the enhanced hierarchical decomposition method and the Poincaré map index to propose an enhanced hierarchical Poincaré map index; Step 3: Use the enhanced hierarchical Poincare map index to extract the information of vibration signals from different channels, connect all the information in series, and form features that characterize multi-sensor data fusion; Step 4: Establish an intelligent diagnosis model for pumped storage units based on enhanced hierarchical Poincare map indicators and random forests; Step 5: Use the existing vibration data of the pumped storage unit to train the intelligent diagnosis model of the pumped storage unit to obtain a high-precision diagnosis model of the pumped storage unit; Step 6: Use the pumped storage unit test data to test the fault detection effect of the pumped storage unit intelligent diagnosis model to verify the effectiveness of the technology.

2. According to claim 1, a method for extracting abnormal information and diagnosing faults of multi-source signals of pumped storage units based on enhanced hierarchical Poincare map indicators is characterized in that: The process of acquiring multi-source vibration data: Acceleration sensors are installed in the horizontal X direction, Y direction and vertical direction of the top cover, and the sensor electrical signals are sent to the data acquisition instrument through hard wiring, and the electrical signals are converted into vibration signals through calibration.

3. According to claim 2, a method for extracting abnormal information and diagnosing faults of multi-source signals of pumped storage units based on enhanced hierarchical Poincare map indicators is characterized in that: The Enhanced Hierarchical Poincare Index (EHPPI) consists of the following steps: First, using the moving average algorithm and the moving difference algorithm, we obtain Signal components of different frequencies, the process is: The average operator and differential operators for: ； ； in, represents the original signal, represent Middle i Sample point data, N Represents the total number of sample points, and Respectively represent Low-frequency information and high-frequency information; When the number of layers is hour, When it is 0 or 1, the corresponding operator for: ； Conclusion The components of different frequencies are: ； in, Indicates the number of layers The Quantity; Represents the index number of the frequency component, and its expression is: ； in, , ; Then, according to the phase space reconstruction theory, we get two vectors and , and and Using the sample points in as the abscissa and ordinate, draw the corresponding Poincare map: ； in, yes The Sample points; Finally, extracting from different parameters SD 1 and SD 2 as the index at different frequency scales, the complete enhanced hierarchical Poincare map index is obtained, and the process is as follows: Extract from the Poincare map of different signal components SD 1 and SD 2 as indicators of different time and frequency scales: ； ； in, and Indicates signal components SD 1 and SD 2; All frequency scales and Putting it all together, we get EHPPI: ； 。 4. According to claim 3, a method for extracting abnormal information and diagnosing faults of multi-source signals of pumped storage units based on enhanced hierarchical Poincare map indicators is characterized in that: The enhanced hierarchical Poincare map index is used to calculate the indexes of the vibration signals of the top cover in the horizontal X direction, Y direction and vertical direction respectively, and the indexes obtained from these three vibration signals are connected in series to obtain the characteristics that characterize the current operating status of the pumped storage unit: ; in, It is a characteristic that characterizes the current operating status of the pumped storage unit; , , The vibration signals of the top cover in the horizontal X direction, Y direction and vertical direction are extracted by EHPPI respectively. SD 1 indicator; , , The vibration signals of the top cover in the horizontal X direction, Y direction and vertical direction are extracted by EHPPI respectively. SD 2 indicators.

5. According to claim 4, a method for extracting abnormal information and diagnosing faults of multi-source signals of pumped storage units based on enhanced hierarchical Poincare map indicators is characterized in that: Random forest is introduced as a recognizer, and an intelligent fault diagnosis model of pumped storage unit based on enhanced hierarchical Poincare map index and random forest is established; in this model, the enhanced hierarchical Poincare map index plays the role of feature extraction, which inputs the comprehensive features extracted from the vibration signals in the horizontal X direction, Y direction and vertical Z direction of the top cover into the random forest.

6. According to claim 5, a method for extracting abnormal information and diagnosing faults of multi-source signals of pumped storage units based on enhanced hierarchical Poincare map indicators is characterized in that: In step 5, the existing vibration data of the pumped storage unit is used to train the intelligent diagnosis model of the pumped storage unit; wherein, these vibration data have clear data labels, and the intelligent diagnosis model of the pumped storage unit finds the best mapping relationship between the vibration signal and the state through continuous training and fitting, and obtains a trained high-precision intelligent diagnosis model of the pumped storage unit; The existing vibration data of pumped storage units is used to train the intelligent diagnosis model of pumped storage units to obtain a high-precision diagnosis model of pumped storage units. The test data of pumped storage units is used to test the fault detection effect of the intelligent diagnosis model of pumped storage units to verify the effectiveness of the technology.

Images