A method for early warning and assessment of the status of thermal power equipment

Abstract

The present invention provides a method for early warning and assessment of the state of thermal power equipment, belonging to the technical field of early warning of the state of thermal power equipment. The method includes the following steps: Selecting a set of equipment measurement points as modeling parameters, applying a regression algorithm to construct an early warning regression model, and grouping the measurement points according to the correlation; Analyzing the influence and association relationship of the measurement points to obtain influencing measurement points, affected measurement points, and non-influencing measurement points; Dynamically calibrating the model evaluation value based on the influencing measurement points, affected measurement points, and non-influencing measurement points. The present invention can eliminate the deviation influence of abnormal measurement points on the calculation of the estimated values of other measurement points, correct the evaluation values of normal measurement points of the equipment to the level that conforms to their true state, improve the accuracy of the regression algorithm, and effectively reduce the false alarms caused by the influence of abnormal measurement points.

Description

Technical Field

[0001] The invention belongs to the technical field of thermal power equipment state early warning, and in particular relates to a thermal power equipment state early warning evaluation method. Background Art

[0002] There are many important equipment in thermal power plants. The operating status of these equipment has a huge impact on whether the factory can produce normally. Once the equipment is abnormal or fails, it will cause severe consequences and serious production losses. Therefore, real-time monitoring and timely prediction of the status of key equipment in thermal power plants are particularly important. In recent years, equipment intelligent early warning systems have been increasingly used in thermal power plants, and play a vital role in risk monitoring of thermal power equipment, ensuring the production safety of production personnel to a certain extent and extending the service life of thermal power equipment.

[0003] There are currently two major methods for thermal power equipment status warning: one is the traditional fixed threshold discrimination method, which uses the thresholds of different warning levels of each component set by the manufacturing process to determine whether the various parameters of the production equipment are normal. This method is relatively safe, but it is easy to ignore the characteristics of equipment performance degradation. In addition, different components will have different degradation degrees due to material, usage, and wear. Therefore, this traditional warning method often has more missed alarms in the later stage of equipment use, greatly reducing the accuracy of equipment status warning. The other is to use data mining methods to explore the complex nonlinear influence relationship between various parameters of the equipment, and to perform status analysis by analyzing the deviation between the actual value and the estimated value of the equipment parameter. Its basic principle is to use the selected mathematical modeling method to fully mine the historical operation data of the equipment parameters to establish an efficient and practical model for real-time equipment status evaluation. Whether the model reliably evaluates the real-time status of the equipment is the basis for the success of equipment warning. The model parameters of the data modeling method have a great influence on the model performance. These parameters generally need to be optimized on the basis of the original method embedded parameter optimization strategy to reduce the deviation of the real-time value evaluation of the equipment. When a parameter in the equipment is abnormal, the estimated values ​​of other parameters in the model calculation will deviate to varying degrees due to the nonlinear relationship in the model, making it difficult to truly reflect the operating status of the current equipment components, resulting in certain false alarms and reducing the accuracy of the equipment status warning method.

[0004] In view of this, the present invention provides a thermal power equipment status early warning evaluation method to solve the shortcomings of the existing method and improve the evaluation accuracy of the equipment status early warning algorithm, which is very necessary. Summary of the invention

[0005] To solve the problem of low accuracy of various current regression algorithms in the early warning assessment of equipment status, the present invention provides a method for early warning assessment of thermal power equipment status, which eliminates the deviation influence of abnormal measurement points on the calculation of estimated values of other measurement points, corrects the assessment values of normal measurement points of the equipment to a level that conforms to their true status, improves the accuracy of the regression algorithm, and effectively reduces false alarms caused by the influence of abnormal measurement points.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for early warning assessment of thermal power equipment status, comprising the following steps:

[0008] S1. Select the equipment measurement point set as the modeling parameter, apply the regression algorithm to construct an early warning regression model, and group the measurement points according to the correlation.

[0009] S2. Analyze the influence correlation relationship of the measurement points to obtain the influencing measurement points, the affected measurement points, and the non-influencing measurement points;

[0010] S3. Dynamically calibrate the model assessment value based on the influencing measurement points, the affected measurement points, and the non-influencing measurement points.

[0011] Preferably, the step S1 includes the following steps:

[0012] S11. Select the equipment measurement point set and obtain the historical data of the equipment from the database;

[0013] S12. Clean the historical data to obtain training data, and apply the regression algorithm and the training data to construct an early warning regression model;

[0014] S13. Use the early warning regression model to calculate the deviation threshold of each measurement point of the equipment;

[0015] S14. Use the training data to calculate the correlation coefficient matrix and group the measurement points.

[0016] The application of the regression algorithm and the training data to construct an early warning regression model includes:

[0017] Take all the parameters of the training data as the input layer of the early warning regression model, with the number of nodes being the number of equipment measurement points n, and at the same time take all the parameters of the training data as the output layer of the early warning regression model, with the number of nodes being the number of equipment measurement points n;

[0018] The construction process of the early warning regression model specifically includes:

[0019] First, establish a BP neural network structure. The BP neural network structure includes an input layer, three hidden layers, and an output layer, and the number of nodes in each layer is [n, 20, 30, 20, n]. The number of neurons n in the input layer and the output layer is the number of equipment measurement points;

[0020] Then, normalize the training data to obtain normalized training data;

[0021] Finally, use the normalized training data to train the network parameters of the BP neural network. The activation function of each layer of the BP neural network is selected as the non-linear function sigmod function The connection weights and thresholds of neurons in each layer are continuously iteratively optimized through all the data, and the network parameters that minimize the value of the network loss function are obtained as the final result.

[0022] Preferably, the step S13 includes:

[0023] Substitute the training data into the warning regression model for training to obtain training evaluation values;

[0024] Calculate the difference between the training data and the training evaluation values as the training deviation data;

[0025] Select the maximum and minimum values of the training deviation data as the deviation thresholds of the measuring points.

[0026] Preferably, the step S2 includes the following steps:

[0027] S21. Substitute the real-time operation data into the warning regression model to calculate the evaluation value of each measuring point, and calculate the difference between the evaluation value of the measuring point and the real-time operation data of the measuring point to obtain the measuring point deviation;

[0028] S22. Obtain the candidate measuring point set according to the measuring point deviation and the measuring point alarm event conditions;

[0029] S23. Analyze the candidate measuring point set according to the real-time operation data of the measuring point and the measuring point correlation relationship to obtain the influencing measuring points, the affected measuring points, and the non-influencing measuring points.

[0030] Preferably, the step S22 includes:

[0031] Judge whether the measuring point deviation exceeds the deviation threshold of the measuring point;

[0032] If so, determine that the measuring point belongs to the first candidate measuring point set;

[0033] If not, further judge whether the measuring point is in the alarm event. If the measuring point is in the alarm event, determine that the measuring point belongs to the first candidate measuring point set. If the measuring point is not in the alarm event, determine that the measuring point belongs to the second candidate measuring point set;

[0034] Among them, if the cumulative duration of the measuring point deviation exceeding the deviation threshold within 24 hours of the measuring point exceeds 1 hour, it is considered to be in the alarm event, otherwise it is considered not to be in the alarm event.

[0035] Preferably, the step S23 includes:

[0036] Determine whether the real-time operation data of the measurement points in the first candidate measurement point set exceeds the historical maximum or minimum value of the measurement point;

[0037] If so, the measurement point is determined as an influencing measurement point;

[0038] If not, further determine whether the measurement point in the current alarm event exceeds the maximum or minimum value abnormally. If the measurement point in the current alarm event exceeds the maximum or minimum value abnormally, it is determined as an influencing measurement point; otherwise, it is determined as an affected measurement point;

[0039] Judge whether the measurement points in the second candidate measurement point set are strongly correlated with the influencing measurement points according to the correlation coefficient matrix. If they are strongly correlated, the measurement points are determined as influencing measurement points; otherwise, they are determined as non-influencing measurement points.

[0040] Preferably, step S3 includes the following steps:

[0041] S31. Group the influencing measurement point set according to the measurement point grouping result, and calibrate the evaluation value of the corresponding measurement point by reconstructing the model for each group of influencing measurement points;

[0042] S32. Reconstruct the model for the affected measurement point set to calibrate the evaluation value of the corresponding measurement point;

[0043] S33. Do not perform the evaluation value calibration process on the non-influencing measurement points.

[0044] Preferably, in step S31, calibrating the evaluation value of the corresponding measurement point by reconstructing the model for each group of influencing measurement points includes:

[0045] S311. Eliminate the influencing measurement points of other groups, and use the training data of this influencing measurement point group, the training data of the affected measurement point group, and the training data of the non-influencing measurement point group to reconstruct the early warning regression model;

[0046] S312. Substitute the real-time operation data of this influencing measurement point group, the real-time operation data of the affected measurement point group, and the real-time operation data of the non-influencing measurement point group into the early warning regression model reconstructed in step S311 to obtain the corrected evaluation value of the influencing measurement point;

[0047] S313. Use the corrected evaluation value of the influencing measurement point to conduct early warning assessment on the thermal power equipment status.

[0048] Preferably, step S32 includes:

[0049] S321. Eliminate all the influencing measurement points, and use the training data of the affected measurement points and the training data of the non-influencing measurement points to reconstruct the early warning regression model;

[0050] S322. Substitute the real-time operation data of the affected measurement points and the real-time operation data of the non-influencing measurement points into the early warning regression model reconstructed in step S321 to obtain the corrected evaluation value of the affected measurement points;

[0051] S323. Use the corrected evaluation value of the affected measurement points to evaluate the status warning of thermal power equipment.

[0052] The beneficial effects of the present invention are as follows. The present invention can dynamically and in real time avoid the interference of abnormal measurement points on other normal measurement points during the evaluation calculation. The evaluation value calibrated by the regression compensation technology can better reflect the actual operating state of the equipment measurement points, and will not weaken the abnormal degree of the abnormal measurement points, greatly improving the accuracy of the regression model in equipment status warning. At the same time, the present invention has good versatility and can improve the accuracy of various regression algorithms. The present invention uses the correlation discrimination method to mine the strength of the correlation relationship between equipment measurement points from historical training data, and realizes the correlation-based model measurement point grouping during the modeling process without the participation of human experience. Therefore, there will be no delay when calibrating the evaluation value of the equipment affected measurement points online, ensuring the real-time warning effect of the model. The present invention has multiple conditions to distinguish the influencing source measurement points and the affected measurement points in real time, and uses light, simple and accurate logical judgment conditions to ensure the real-time performance of the model, providing a prerequisite for online dynamic calibration of equipment data. The present invention uses the evaluation value dynamic calibration method to calibrate the evaluation value of the affected measurement points. The evaluation value dynamic calibration method can stably output non-interfered evaluation values for normal measurement points and has no any fluctuation effect on the evaluation values of abnormal measurement points. The present invention uses data-driven to run the warning regression model. The evaluation value dynamic calibration process is completely based on the essential characteristics of the data, so that the regression model is not restricted by expert experience knowledge and equipment process structure, making the calculation results of the warning regression model more reliable. Brief Description of the Drawings

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0054] Figure 1 It is a flowchart of the method for evaluating the status warning of thermal power equipment in the embodiment of the present invention.

[0055] Figure 2 It is a flowchart of step S2 in the embodiment.

[0056] Figure 3 It is a flowchart of step S3 in the embodiment.

[0057] Figure 4 It is an effect diagram of evaluating and calibrating the vibration parameter in the X direction of the bearing of the primary air fan A equipment in the embodiment.

[0058] Figure 5This is the effect diagram of the calibration of the temperature 1 parameter of the motor coil C phase of a fan A device. DETAILED DESCRIPTION

[0059] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0060] like Figure 1 As shown, an embodiment of the present invention provides a thermal power equipment status early warning evaluation method, comprising the following steps:

[0061] S1. Select the equipment measurement point set as the modeling parameter, apply the regression algorithm to build the early warning regression model, and group the measurement points according to the correlation.

[0062] The step S1 specifically includes the following steps:

[0063] S11. Select a set of equipment measurement points and obtain the historical data of the equipment from the PI database;

[0064] Taking the primary fan A equipment of a thermal power plant in the north as an example, all the data of the equipment's measurement point parameter set from January 2020 to December 2020 were obtained from the PI database. According to the start and stop condition parameters of the equipment, the shutdown data in all the data were identified and eliminated, and the processed data were used as the historical data of the primary fan A equipment. The equipment regression model mainly uses 27 measurement point sets of the primary fan, including vibration measurement points, pressure measurement points, current measurement points, and temperature measurement points, as model modeling parameters, with a data sampling interval of 1 minute.

[0065] The data files of primary fan A equipment are saved in m×n matrix format, including m time points and n measurement points of the equipment. The specific format is as follows:

[0066]

[0067] S12. Perform data cleaning on historical data to obtain training data, and apply regression algorithm and training data to build an early warning regression model.

[0068] In order to eliminate abnormal data from historical data and obtain healthy operation data, it is necessary to clean the historical data of the primary fan A device obtained in step S11, and screen out the historical operation data of the device in the normal state as the training data for constructing the regression model. Among them, in the trend chart, if phenomena such as the data trend exceeding the protection set value, violent fluctuations, or no fluctuations in the numerical value occur, this paragraph of data will be regarded as abnormal data and eliminated. The remaining data after screening can be used as the training data for regression modeling.

[0069] Suppose the training data T after cleaning the historical data F of the device at n measurement points contains k moments, then this training data file should be in the matrix form of k×n, which can be expressed as:

[0070]

[0071] At the same time, calculate the maximum and minimum values of each measurement point of the training data T and store them in the MaxMin array. The form size of the MaxMin array is 2×n.

[0072] Taking the BP neural network regression algorithm as an example to introduce the construction process of the early warning regression model. In order to better explore the non-linear regression relationship of each measurement point parameter of the primary fan A device, use the training data T to construct a 5-layer BP neural network regression model. All parameters of the training data T are used as both the input layer of the regression model, with the number of nodes being the number of device measurement points n, and at the same time, all parameters of the training data T are also used as the output layer of the model, with the number of nodes being the number of device measurement points n, and three hidden layers are built in the middle. Use the BP neural network regression model to calculate the evaluation value of the historical data. Subtract the historical evaluation value from the historical data to obtain the historical deviation data. Use the BP neural network regression model to calculate the evaluation value of the training data. Subtract the evaluation value of the training data from the training data to obtain the training deviation data.

[0073] The construction process of the early warning regression model specifically includes:

[0074] First, establish the BP neural network structure. The BP neural network structure includes an input layer, three hidden layers, and an output layer. The number of nodes in each layer is [n, 20, 30, 20, n]. The number of neurons n in the input layer and the output layer is the number of measurement points of the primary fan A;

[0075] Then, perform normalization processing on the training data T to obtain the normalized training data T a_norm , and map the normalized training data to the interval [0, 1] using the following formula:

[0076]

[0077] Among them, T a_norm [i, j] is the normalized value of parameter j at time i, and T a[i, j] is the training value of parameter j at time i, T a_max_j is the maximum value on parameter j, T a_min_j is the minimum value on parameter j.

[0078] Finally, use the normalized training data T a_norm to train the network parameters of the BP neural network. All the parameters of the training data serve as both the input nodes and the output nodes of the BP neural network. The activation function of each layer of the BP neural network selects the non-linear function sigmod function The connection weights and thresholds of each layer of neurons are continuously iteratively optimized through all the data, and the network parameters that minimize the value of the network loss function are obtained as the final result.

[0079] The training process of the BP neural network regression model is as follows:

[0080] Step 1, the data signal is input from the input layer and propagated layer by layer through the hidden layer to the output layer. During this process, it is necessary to calculate the weights between neurons and the activation function in the neurons, and finally the calculated value of the neural network can be output from the output layer.

[0081] Taking a certain node j in the hidden layer or output layer as an example, the input S of node j j is the weighted cumulative output value of the i neurons in the previous layer, that is: S j = ∑w ij y i , where w ij are the respective weights from i to j, and y i is the output value of i.

[0082] Assuming the threshold of node j is θ j , then the actual input of node j is: u ij = Σw ij y i - θ j ; at the same time, the output of node j is y i = f(u j ) = f(S j - θ j ), where f(u j ) is the activation function sigmod function.

[0083] Step 2, the network performs error backpropagation. First, calculate the error between the output of the output layer and the expected output of the network, compare the error with the learning accuracy set by the network. If the error is greater than the learning accuracy, calculate the partial derivatives of the error with respect to the weights and thresholds in the neural network, and adjust the weights and thresholds between neurons according to the gradient descent method.

[0084] These two processes run continuously in a loop until the termination conditions such as the number of iterations being greater than the set maximum number of iterations or the error being less than the set learning accuracy are met, then the BP neural network regression model BPModel is successfully trained.

[0085] S13. Calculate the deviation threshold for each measuring point of the device using the warning regression model, specifically including:

[0086] Substitute the training data into the warning regression model for training to obtain the training evaluation value;

[0087] Calculate the difference between the training data and the training evaluation value as the training deviation data;

[0088] Select the maximum and minimum values of the training deviation data as the deviation threshold for the measuring point.

[0089] This step mainly subtracts the training evaluation value from the training data to obtain the training deviation data, and uses the maximum and minimum values of the training deviation data as the deviation threshold for the measuring point; substitute the normalized training data T a_norm into the BP neural network regression model for training to obtain the training evaluation value T fore , calculate the difference between the training data T and the training evaluation value T fore as the training deviation data R ta , the training deviation data R ta is a data matrix with the form size of k×n. By calculating the maximum and minimum values of each column of data in R ta to generate the measuring point deviation threshold array RE with the form size of 2×n.

[0090] S14. Calculate the correlation coefficient matrix using the training data and group the measuring points;

[0091] Use the training data T to calculate the correlation coefficient matrix Corr to measure the strength of the correlation between pairs of measuring points. Initialize the classification, with each measuring point as a group. Then, according to the measuring point group similarity formula, perform clustering of similar measuring point groups. Through multiple rounds of polling, merge the measuring point groups with high similarity pairwise until the similarity between each measuring point group is greater than the similarity threshold CT, where the similarity threshold CT is 0.7. The formula for calculating the similarity between pairs of measuring point groups is as follows:

[0092]

[0093] Among them, ρ xy is the similarity coefficient between measuring point x and measuring point y, m is the number of measuring points in group 1, and n is the number of measuring points in group 2. After the similarity-based measuring point grouping, 27 measuring points can be divided into 3 measuring point groups. The first group is mainly vibration measuring points, the second group is mainly pressure measuring points, and the third group is mainly temperature measuring points.

[0094] S2. Analyze the influence correlation relationship of the measurement points to obtain the influencing measurement points, the affected measurement points, and the non - affected measurement points. As Figure 2 shown, it specifically includes the following steps:

[0095] Take the failure data of the primary fan A equipment from January 2021 to July 2021 as the real - time operation data.

[0096] S21. Substitute the real - time operation data into the early - warning regression model to calculate the evaluation value of each measurement point, and subtract the evaluation value of the measurement point from the real - time operation data of the measurement point to obtain the measurement point deviation.

[0097] Substitute the real - time operation data of the primary fan A equipment from January 2021 to July 2021 into the BP neural network regression model to calculate the evaluation value. Normalize the real - time operation data RT i to obtain RT nomi . Substitute RT nomi into the trained BP neural network regression model BPModel to first obtain the evaluation data RF i of each parameter, and then calculate its deviation L i , that is: L i = RT i - RF i .

[0098] S22. Obtain the candidate measurement point set according to the measurement point deviation and the measurement point alarm event condition, specifically including:[[]]

[0099] Judge whether the measurement point deviation exceeds the deviation threshold of the measurement point;

[0100] If so, determine that the measurement point belongs to the first candidate measurement point set;

[0101] If not, further judge whether the measurement point is in the alarm event. If the measurement point is in the alarm event, determine that the measurement point belongs to the first candidate measurement point set. If the measurement point is not in the alarm event, determine that the measurement point belongs to the second candidate measurement point set;

[0102] Among them, if the cumulative duration of the measurement point deviation exceeding the deviation threshold within 24 hours exceeds 1 hour, it is considered to be in the alarm event, otherwise it is considered not to be in the alarm event.

[0103] This step mainly prepares for the next step of identifying different types of measurement points. Obtain the first candidate measurement point set and the second candidate measurement point set from all the selected measurement points. The first candidate measurement point set is mainly used to further calculate the influencing measurement points and the affected measurement points, and the second candidate measurement point set is mainly used to further calculate the influencing measurement points and the non - affected measurement points. Among them, the alarm rule formed by the measurement point alarm event is that there is an alarm situation where the cumulative duration of the measurement point exceeds the deviation threshold by more than 1 hour within 24 hours.

[0104] S23. Analyze the candidate measurement point set based on the real-time operation data of the measurement points and the relevant relationships of the measurement points to obtain the influencing measurement points, the affected measurement points, and the non-influencing measurement points, specifically including:

[0105] Judge whether the real-time operation data of the measurement points in the first candidate measurement point set exceeds the historical maximum or minimum value of the measurement point;

[0106] If so, the measurement point is determined to be an influencing measurement point;

[0107] If not, further judge whether the measurement point in the current alarm event exceeds the maximum or minimum value abnormally. If the measurement point in the current alarm event exceeds the maximum or minimum value abnormally, it is determined to be an influencing measurement point, otherwise it is determined to be an affected measurement point;

[0108] Judge whether the measurement points in the second candidate measurement point set are strongly correlated with the influencing measurement points according to the correlation coefficient matrix. If they are strongly correlated, the measurement points are determined to be influencing measurement points, otherwise they are determined to be non-influencing measurement points.

[0109] Among them, if the similarity coefficient between the measurement points in the second candidate measurement point set and the influencing measurement points exceeds the strong correlation threshold, it is determined that the two are strongly correlated, otherwise they are not strongly correlated.

[0110] S3. Dynamically calibrate the model evaluation value based on the influencing measurement points, the affected measurement points, and the non-influencing measurement points, as Figure 3 shown, specifically including the following steps:

[0111] S31. Group the influencing measurement point set according to the measurement point grouping result, and calibrate the evaluation value of the corresponding measurement point by reconstructing the model for each group of influencing measurement points;

[0112] Among them, the influencing measurement point set is grouped according to the correlation grouping result in the training stage.

[0113] After grouping, calibrating the evaluation value of the corresponding measurement point by reconstructing the model for each group of influencing measurement points specifically includes:

[0114] S311. Eliminate the influencing measurement points of other groups, and use the training data of this influencing measurement point group, the training data of the affected measurement point group, and the training data of the non-influencing measurement point group to reconstruct the early warning regression model;

[0115] S312. Substitute the real-time operation data of this influencing measurement point group, the real-time operation data of the affected measurement point group, and the real-time operation data of the non-influencing measurement point group into the early warning regression model reconstructed in step S311 to obtain the corrected evaluation value of the influencing measurement point;

[0116] S313. Use the corrected evaluation value of the influencing measurement point to evaluate the state of the thermal power equipment for early warning.

[0117] Among them, reconstructing the early warning regression model means re-training the model according to the method in step S12.

[0118] For example, in the verification example of the primary fan A equipment, after the influence correlation analysis, the vibration measurement point in the X direction of the bearing of the primary fan A equipment belongs to the affected measurement point. The evaluation of the calibration effect is as follows Figure 4 shown. The value of the vibration measurement point in the X direction of the bearing of the primary fan A equipment gradually increases, and the state gradually deteriorates. From the evaluation effect, the deviation between the real-time value and the evaluation value of this measurement point is getting larger and larger, which is in line with the actual situation of the equipment.

[0119] S32. Reconstruct the model for the set of affected measurement points to calibrate the evaluation values of the corresponding measurement points, specifically including:

[0120] S321. Eliminate all affected measurement points, and use the training data of the affected measurement points and the training data of the non-affected measurement points to reconstruct the early warning regression model;

[0121] S322. Substitute the real-time operation data of the affected measurement points and the real-time operation data of the non-affected measurement points into the early warning regression model reconstructed in step S321 to obtain the corrected evaluation values of the affected measurement points;

[0122] S323. Use the corrected evaluation values of the affected measurement points to conduct early warning assessment on the state of thermal power equipment;

[0123] For example, in the verification example of the primary fan A equipment, after the influence correlation analysis, the temperature measurement point 1 of phase C of the motor coil of the primary fan A equipment belongs to the affected measurement point. The value of the temperature measurement point 1 of phase C of the motor coil of the primary fan A equipment is within the normal range during the verification period, and there is no abnormal state. The parameter of the temperature measurement point 1 of phase C of the motor coil of the primary fan A equipment is affected by the abnormal state parameter of the vibration in the X direction of the bearing of the primary fan A equipment, and the deviation between its regression value and the actual value is very large. After the model measurement point influence correlation analysis and the dynamic calibration application of the model evaluation value of the present invention, the evaluation and calibration effect of this measurement point is as follows Figure 5 shown. The parameter of the temperature measurement point 1 of phase C of the motor coil of the primary fan A equipment without abnormal state is no longer affected by the abnormal state parameter of the vibration in the X direction of the bearing of the primary fan A equipment, and the evaluation value is close to the actual value. Only 11.5 days of data deviate greatly within 6 months, and the effective rate of evaluation and calibration reaches 94.5%.

[0124] S33. Do not perform evaluation value calibration processing on non-affected measurement points.

[0125] For the non-affected measurement points determined by applying the influence correlation analysis technology, their evaluation values are no longer subjected to evaluation and calibration processing, and the original evaluation value RF in step S21 is directly used.

[0126] Through the technical verification of the primary air fan A equipment this time, it can be seen that the evaluation effect of the early warning state of the invention in thermal power equipment is that when the X-direction vibration measurement point of the bearing of the primary air fan A equipment is in a deteriorated abnormal state, the other measurement points of the early warning regression model of the equipment are no longer affected by the abnormal measurement point, and no false alarm events will occur. The evaluation effect of the invention on thermal power equipment conforms to the actual situation of the equipment at that time, and can effectively improve the accuracy of the early warning regression model.

[0127] Although the present invention has been described in detail by referring to the accompanying drawings and in combination with the preferred embodiments, the present invention is not limited thereto. Without departing from the spirit and essence of the present invention, those of ordinary skill in the art can make various equivalent modifications or substitutions to the embodiments of the present invention, and all such modifications or substitutions should be within the scope of the present invention. / Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, and all should be covered within the protection scope of the present invention.

Claims

1. A method for early warning and assessment of the status of thermal power equipment, characterized in that, It includes the following steps: S1. Select the set of equipment measurement points as modeling parameters, apply the regression algorithm to construct an early warning regression model, and group the measurement points according to the correlation; S2. Analyze the influence correlation relationship of the measurement points to obtain influencing measurement points, affected measurement points, and non-influencing measurement points; S3. Dynamically calibrate the model evaluation values based on the influencing measurement points, affected measurement points, and non-influencing measurement points, including: S31. Group the influencing measurement point set according to the measurement point grouping result, and reconstruct the model for each group of influencing measurement points to calibrate the evaluation value of the corresponding measurement point; S32. Reconstruct the model for the affected measurement point set to calibrate the evaluation value of the corresponding measurement point; S33. Do not perform evaluation value calibration processing on the non-influencing measurement points.

2. The method for early warning and assessment of the state of thermal power equipment according to claim 1, characterized in that , the step S1 includes the following steps: S11. Select the set of equipment measurement points and obtain the historical data of the equipment from the database; S12. Clean the historical data to obtain training data, and apply the regression algorithm and the training data to construct an early warning regression model; S13. Use the early warning regression model to calculate the deviation threshold of each measurement point of the equipment; S14. Calculate the correlation coefficient matrix using the training data and group the measurement points.

3. The method for early warning and evaluation of the thermal power equipment status according to claim 2, characterized in that, Applying the regression algorithm and the training data to construct an early warning regression model includes: Take all the parameters of the training data as the input layer of the early warning regression model, with the number of nodes being the number of equipment measurement points n. At the same time, take all the parameters of the training data as the output layer of the early warning regression model, with the number of nodes being the number of equipment measurement points n; The construction process of the early warning regression model specifically includes: First, establish a BP neural network structure. The BP neural network structure includes an input layer, three hidden layers, and an output layer, with the number of nodes in each layer being [n, 20, 30, 20, n]. The number of neurons n in the input layer and the output layer is the number of equipment measurement points; Then, perform normalization processing on the training data to obtain normalized training data; Finally, the BP neural network is trained with the normalized training data for the network parameters. The activation function of each layer of the BP neural network selects the nonlinear sigmod function. The connection weights and thresholds of neurons in each layer are continuously iteratively optimized through all the data, and the network parameters that minimize the value of the network loss function are obtained as the final result.

4. The method for warning and evaluating the state of thermal power equipment according to claim 2 or 3, characterized in that , the step S13 includes: Substitute the training data into the early warning regression model for training to obtain training evaluation values; Calculate the difference between the training data and the training evaluation values as the training deviation data; Select the maximum and minimum values of the training deviation data as the deviation threshold of the measurement point.

5. The method for warning and evaluating the state of thermal power equipment according to claim 4, characterized in that, The step S2 includes the following steps: S21. Substitute the real-time operation data into the early warning regression model to calculate the evaluation value of each measurement point, and calculate the difference between the evaluation value of the measurement point and the real-time operation data of the measurement point to obtain the measurement point deviation; S22. Obtain the candidate measurement point set according to the measurement point deviation and the measurement point alarm event condition; S23. Analyze the candidate measurement point set based on the real-time operation data of the measurement point and the measurement point correlation relationship to obtain influencing measurement points, affected measurement points, and non-influencing measurement points.

6. The method for warning and evaluating the state of thermal power equipment according to claim 5, wherein The step S22 includes: Judge whether the measurement point deviation exceeds the deviation threshold of the measurement point; If so, determine that the measurement point belongs to the first candidate measurement point set; If not, further judge whether the measurement point is in an alarm event. If the measurement point is in an alarm event, determine that the measurement point belongs to the first candidate measurement point set. If the measurement point is not in an alarm event, determine that the measurement point belongs to the second candidate measurement point set; If the cumulative duration of the measurement point deviation exceeding the deviation threshold within 24 hours exceeds 1 hour, it is considered to be in an alarm event, otherwise it is considered not to be in an alarm event.

7. The method for warning and evaluating the state of thermal power equipment according to claim 6, wherein The step S23 includes: Determine whether the real-time operation data of the measuring points in the first candidate measuring point set exceeds the historical maximum and minimum values of the measuring points; If so, the measuring point is determined as an influencing measuring point; If not, further determine whether the measuring point in the current alarm event exceeds the maximum and minimum value abnormally. If the measuring point in the current alarm event exceeds the maximum and minimum value abnormally, it is determined as an influencing measuring point, otherwise it is determined as an affected measuring point; Judge whether the measuring points in the second candidate measuring point set are strongly correlated with the influencing measuring points according to the correlation coefficient matrix. If they are strongly correlated, the measuring points are determined as influencing measuring points, otherwise they are determined as non-influencing measuring points.

8. The method for early warning and assessment of the state of thermal power equipment according to claim 1, characterized in that In step S31, the evaluation values of the corresponding measuring points are calibrated by reconstructing the model for each group of influencing measuring points, including: S311. Eliminate the influencing measuring points of other groups, and reconstruct the early warning regression model using the training data of this group of influencing measuring points, the training data of the affected measuring points group, and the training data of the non-influencing measuring points group; S312. Substitute the real-time operation data of this group of influencing measuring points, the real-time operation data of the affected measuring points group, and the real-time operation data of the non-influencing measuring points group into the early warning regression model reconstructed in step S311 to obtain the corrected evaluation value of the influencing measuring points; S313. Use the corrected evaluation value of the influencing measuring points to conduct early warning evaluation on the state of thermal power equipment.

9. The method for early warning and evaluation of the thermal power equipment status according to claim 8, characterized in that The said step S32 includes: S321. Eliminate all influencing measuring points, and reconstruct the early warning regression model using the training data of the affected measuring points and the training data of the non-influencing measuring points; S322. Substitute the real-time operation data of the affected measuring points and the real-time operation data of the non-influencing measuring points into the early warning regression model reconstructed in step S321 to obtain the corrected evaluation value of the affected measuring points; S323. Use the corrected evaluation value of the affected measuring points to conduct early warning evaluation on the state of thermal power equipment.

Images