A cloud-edge collaborative monitoring and early warning method, device and medium for a gas generator set

Abstract

This invention discloses a cloud-edge collaborative monitoring and early warning method for gas generator sets, comprising the following steps: S1. Determining the main fault early warning range through preliminary parameters, disassembling the gas generator set equipment for analysis to build an intelligent early warning model, and analyzing the key monitoring parameters related to the early warning model of each device; S2. Making preliminary selections for the early warning range and related parameters of the early warning model, studying the correlation relationship of parameters in each model through correlation algorithms, and extracting the key feature variables required for the early warning model; S3. Cleaning the key feature variables; S4. Optimizing the intelligent early warning model using a method based on algorithm fusion of key feature variables; S5. Setting early warning rules for the optimized early warning model, and then using the early warning model for monitoring and early warning of the gas generator set. This invention improves the accuracy of the intelligent early warning model by establishing an early warning capability based on the fusion of multiple algorithms through deep self-learning of unit parameter data.

Description

Technical Field

[0001] The present invention relates to the field of gas generator set detection and early warning technology, and in particular to a cloud-edge collaborative monitoring and early warning method, equipment and medium for a gas generator set. Background Art

[0002] The production site of heavy-duty gas turbine generator sets still generally relies on the alarm of the unit control system and the experience analysis of the operating personnel to realize the monitoring and diagnosis decision of the equipment. The unit operation and maintenance monitoring decision-making model under the existing model is highly dependent on the overall technical quality of the on-site personnel and is greatly affected by personal subjective factors.

[0003] At the same time, the DCS system can only alarm for a single parameter that exceeds the limit. It has no better analysis means and methods, and is unable to provide corresponding prompts and decision-making guidance in the budding stage of equipment failure. A large amount of operation and fault data has not been effectively mined, resulting in a huge waste of data resources.

[0004] Therefore, with the use of heavy-duty gas turbine generator sets, remote monitoring and diagnosis platforms have gradually been established based on the equipment, such as the STA-RMS platform, iCenter platform, remote monitoring and maintenance support system, etc., which can provide real-time monitoring, remote abnormality warning and other services for gas turbines. However, the accuracy of the relevant intelligent warning and diagnostic work warning of the remote monitoring and diagnosis platform is currently low, which greatly restricts the application of intelligent warning technology.

[0005] To this end, this application specifically proposes a cloud-edge collaborative monitoring and early warning method for gas generator sets to solve the above-mentioned technical problems.

[0006] In addition, it should be noted that the above introduction to the technical background is only for the convenience of a clear and complete description of the technical solutions of this specification and to assist those skilled in the art in understanding the technical solutions of the present application. It cannot be considered that the above technical solutions are well known to those skilled in the art simply because these solutions are described in the background technology section of this specification. Therefore, the description in this section only provides background information related to the disclosure of this specification, and does not mean that the above content is admitted to be the closest prior art. Summary of the invention

[0007] The main purpose of the present invention is to provide a cloud-edge collaborative monitoring and early warning method for a gas generator set to solve the technical problems raised in the background technology.

[0008] The present invention adopts the following technical solutions to solve the above technical problems:

[0009] A cloud-edge collaborative monitoring and early warning method for a gas generator set, wherein a computer device performs the following steps:

[0010] S1. Determine the main fault warning range through preliminary parameters, disassemble and analyze the gas generator set equipment to build an intelligent warning model, and analyze the key monitoring parameters related to the warning model of each equipment in combination with the knowledge of field experts;

[0011] S2. Based on expert knowledge, preliminary selection of warning range and warning model related parameters is carried out, and the correlation relationship of parameters in each model is studied through correlation algorithm to extract key characteristic variables required for the warning model;

[0012] S3. Perform data cleaning operations on key feature variables;

[0013] S4. Use the key characteristic variables after data cleaning to optimize the intelligent early warning model based on the algorithm fusion method;

[0014] S5. Setting warning rules for the optimized warning model. The warning model after the warning rules are set is used for monitoring and warning of gas-fired power generation units.

[0015] Preferably, the specific operation flow of studying the correlation between parameters in each model and extracting key characteristic variables required for the early warning model in step S2 includes:

[0016] S21. After selecting the relevant parameters of the model based on the expert domain knowledge, the initially selected relevant parameters are randomly selected as the initial variables X and Y. Then the Pearson correlation coefficient ρ is:

[0017]

[0018] where cov(X, Y) is the covariance of the initial variables X and Y, σ x , σ y are the standard deviations of the initial variables X and Y respectively;

[0019] At this time, based on the Pearson correlation coefficient ρ x , y The value range is between [-1,1], and |ρ x , y The closer | is to 1, the stronger the linear correlation between variables X and Y is. x , y The closer | is to 0, the weaker the linear correlation between variables X and Y. x , y = 0 means that there is no linear relationship between variables X and Y, ρ x , y When it is a positive number, it means that there is a positive correlation between variables X and Y. x , y When it is a negative number, it means that there is a negative correlation between variables X and Y;

[0020] S22. After repeating the above step S21 for multiple times to select relevant parameters and calculate the linear correlation, the final early warning model parameters are selected as the key feature variables required for the early warning model based on the parameter variables with higher linear correlation.

[0021] Preferably, the data cleaning operation in step S3 includes:

[0022] S31. For key characteristic variable data, remove abnormal data including abnormal data and fault data in historical data;

[0023] S32. Identify the working conditions of the key characteristic variable data, and set different sampling granularities for different working conditions including shutdown and operation to re-collect data;

[0024] S33. Normalize the measured values ​​of the key characteristic variable data according to their respective extreme values ​​so that the actual measured values ​​are mapped to the interval [0, 1];

[0025] S34. According to the number of parameters required by the model, a multi-dimensional historical data matrix is ​​constructed. For each parameter, the interval [0, 1] is divided into h equal parts, and several observation vectors are found from the set with a step size of 1 / h and added to the training data;

[0026] S35. The same operation as step S34 is performed on the remaining parameters to obtain a processed matrix, and then duplicate removal is performed to obtain a final matrix.

[0027] Preferably, the algorithm fusion method in step S4 includes fusing the nonlinear state assessment algorithm with the neural network algorithm to effectively improve the accuracy of the early warning model. The specific steps of the algorithm fusion include:

[0028] S41. extract corresponding data from the current data and the normal mode of the stored system operation process state, mine the relationship between the observed variables of the system from the historical data, establish a non-parametric qualitative model and calculate the state estimation of the system through the weighted combination of the normal mode of the system state in the non-parametric qualitative model, and the weight is determined by the similarity between the modes;

[0029] S42. Preset BP neural network algorithm and perform algorithm analysis;

[0030] S43. Composition of a mode matrix, a compensation matrix and a compensation observation vector based on multiple interrelated variables at different times during normal operation of the device;

[0031] S44. Determine a similarity coefficient matrix according to the similarity operator of the system when the device is operating normally, the mode matrix and the compensation matrix, determine a compensation prediction vector according to the nonlinear evaluation algorithm, the mode matrix and the compensation observation vector, and then generate a compensation residual vector;

[0032] S45. Using the BP neural network algorithm, based on the similarity coefficient matrix and the compensated residual vector, a residual prediction model is trained, and the residual in the nonlinear evaluation algorithm is compensated based on the output result of the residual prediction model, thereby reducing the data prediction deviation.

[0033] Preferably, the specific operation steps of establishing a non-parametric qualitative model and calculating the state estimation of the system in step S41 include:

[0034] S411. Let X be an observation vector with n dimensions, then the i-th observation vector is X(i) = [x1(i), x2(i)…x n (i)], assuming that the matrix D with m groups is the historical data of normal operation samples, then:

[0035]

[0036] S412. Since the magnitude of the parameters of each device of the power generation equipment is different, calculation errors will occur in the data operation process. In order to reduce the speed reduction and possible errors in matrix operation caused by excessive data magnitude, the sample data and the observation data are normalized, and the evaluation data is denormalized;

[0037] S413. When the system is running normally, for each set of observation data input, the system evaluates each set of input data, extracts the maximum and minimum values ​​in the sample matrix D and the observation data to form a matrix and then normalizes it, extracts the observation data in the normalized matrix and then performs nonlinear state evaluation, combines the evaluation results with the normalized maximum and minimum values ​​and performs denormalization, and finally obtains the correct state evaluation data.

[0038] Preferably, the specific operation steps of evaluating each set of input data in step S413 include:

[0039] L1. Set the observation data matrix input by the system to X obs , the corresponding evaluation data matrix at the same time is X est , the output value of the preset model is a linear combination of all observation vectors in the sample matrix D, then there exists a weight vector W, which is:

[0040] X est =(D T W) T

[0041] The weight vector W = [w1, w2…w m ];

[0042] L2. There is a residual ε between the input observation value and the output evaluation value. To minimize ε, there exists:

[0043]

[0044] L3. Because the sample matrix D cannot be inverted, a model evaluation rule is formulated. The dot product in the formula is replaced by the Euclidean distance, and then:

[0045]

[0046] We can get:

[0047]

[0048] If the model input matrix X obs When the system is in normal working condition, the value should be close to one of the data in the sample matrix D, and the evaluation matrix X est With X obs Close, the residual ε is small;

[0049] If X obs is fault data, and when it exceeds the sample matrix D, the residual ε becomes larger. Therefore, the abnormality of system data can be predicted according to the change of ε, and the model algorithm can be debugged according to ε;

[0050] because is a quantitative quantity, assuming that the quantitative quantity is A, then:

[0051]

[0052] In the formula, A is only determined by the sample matrix D. It only needs to be calculated once, and the calculation result can be brought into the state evaluation of the next set of observation vectors, reducing the calculation time.

[0053] Preferably, the BP neural network algorithm in step S42 includes:

[0054] S421. After the sample is input, the BP algorithm is used to adjust the weights and thresholds of the network for multiple training, so that the actual output result is as close as possible to the expected result. At this time, the BP neural network setting includes an input layer i, an output layer k, and a hidden layer j;

[0055] S422. Preset ω ij represents the weight between the input layer and the hidden layer, ω jk Represents the weight between the hidden layer and the output layer. According to the input x, the weight ω ij And the threshold a of the hidden layer, it can be concluded that the output of the hidden layer is:

[0056]

[0057] Where, f is the activation function;

[0058] S423. Based on the output H and weight ω jk , the threshold b of the output layer, the output O of the output layer can be obtained k :

[0059]

[0060] Where, f is the activation function;

[0061] S424. During the implementation process, the measured values ​​x1, x2, ..., x n Get the actual estimate through the neural network Compare the estimated value with the actual value y i Compare the residual values When one of the residual values ​​exceeds the threshold, an early warning message will be triggered.

[0062] Preferably, the specific operation steps for setting the warning rules in step S5 include:

[0063] S51. Analyze the alarm limit values ​​of characteristic parameters in different algorithm models of gas turbine generator sets under different operating conditions and boundary conditions, and realize dynamic threshold setting of multiple parameters within the DCS alarm limit;

[0064] S52. Based on the data-driven algorithm, the expert experience of operation and maintenance is converted into parameter alarm rules to detect and identify abnormalities of related parameters;

[0065] S53. A sliding window method is introduced for the residual alarm rule of a single parameter to continuously and in real time detect the changes in the statistical characteristics of the residual and reduce false alarms;

[0066] S54. The sliding window residual statistics method is used to suppress the adverse effects of isolated large residuals by calculating the statistical characteristics of multiple residuals in the sliding window. At this time, within the preset time, the early warning model predicts the residual sequence as: ε GT =[ε1,ε2,…,ε N ,…]

[0067] For this sequence, take a sliding window with a width of N and calculate the mean of the N consecutive residuals in the window. And standard deviation S:

[0068]

[0069] S55. During the implementation process, the model effect is evaluated and, combined with expert domain knowledge, a reasonable sliding window width N is set for each model to ensure that the early warning model can promptly and quickly reflect the continuous changes in the statistical characteristics of the residuals and eliminate the influence of random factors.

[0070] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0071] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0072] It can be seen from the above technical solution that the present invention provides a cloud-edge collaborative monitoring and early warning method for a gas generator set. Compared with the prior art, the present invention has the following advantages:

[0073] 1. The present invention establishes early warning capabilities based on the fusion of multiple algorithms by conducting deep self-learning on the historical measurement parameter data and simulation data of the unit, and can use data means from multiple aspects to improve the accuracy of the intelligent early warning model.

[0074] 2. The present invention combines nonlinear state evaluation with BP neural network for algorithm fusion, which not only optimizes the prediction performance of the model, but also significantly reduces the prediction deviation by setting a reasonable residual compensation mechanism at the algorithm level, thereby enhancing the model's ability to identify abnormal states, and ultimately ensuring timely and accurate warnings.

[0075] 3. The present invention improves data quality and model robustness by implementing detailed data verification and working condition adaptability processing in the data cleaning stage, such as eliminating abnormal data, adjusting the sampling frequency and data normalization according to different working conditions, and ultimately achieving a more stable and reliable monitoring and early warning effect.

[0076] 4. The present invention sets warning rules by optimizing the warning model and sets a reasonable sliding window width for each model. This can ensure that the warning model can promptly and quickly reflect the continuous changes in the statistical characteristics of the residuals while eliminating the influence of random factors, thereby improving the reliability of intelligent status monitoring and warning and reducing the probability of false alarms.

[0077] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become easy to understand through the following description. Of course, it is not necessary to achieve all of the advantages described above simultaneously for any product implementing the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] The drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0079] Figure 1 This is a flow chart of the intelligent early warning technology of the present invention;

[0080] Figure 2 It is a general flow chart of the method steps of the present invention;

[0081] Figure 3 It is a schematic diagram of the monitoring range of the early warning technology of the present invention;

[0082] Figure 4 A schematic diagram of a warning list of the intelligent warning model of the present invention;

[0083] Figure 5 It is a schematic diagram of the data screening process of the present invention;

[0084] Figure 6 It is a path diagram of the algorithm fusion based on nonlinear state evaluation and neural network of the present invention;

[0085] Figure 7 It is a schematic diagram of the state estimation modeling principle of the present invention;

[0086] Figure 8 It is a schematic diagram of the neural network structure of the present invention;

[0087] Fig. 9 It is a schematic diagram of the residual sequence window sliding of the present invention. DETAILED DESCRIPTION

[0088] 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, rather than all of the embodiments. In the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0089] In the embodiment, see Figures 1 to 9 .

[0090] like Figure 1As shown. The cloud-edge collaborative monitoring and early warning method for a gas-fired generator set proposed in an embodiment of the present invention establishes an early warning capability based on the fusion of multiple algorithms by deep self-learning of the historical measurement parameter data and simulation data of the unit, forming a fault intelligent early warning framework for the entire process from parameter selection, working condition identification, data screening, algorithm fusion, alarm rule setting, etc., and uses data means from multiple aspects such as parameters, data, and algorithms to improve the accuracy of the intelligent early warning model.

[0091] like Figure 1 and Figure 2 As shown, the monitoring and early warning method specifically performs the following steps through a computer device:

[0092] S1. Determine the main fault warning range through preliminary parameters, disassemble and analyze the gas generator set equipment to build an intelligent warning model, and analyze the key monitoring parameters related to the warning model of each equipment in combination with the knowledge of field experts;

[0093] Since the heavy-duty gas turbine is a complex system, its structure needs to be decomposed, and then preliminary parameter selection and mechanism analysis are carried out to provide a basis for the establishment of the early warning model. The intelligent early warning model includes mainstream E-class and F-class gas generator sets. The development and implementation process of the intelligent early warning technology first disassembles and analyzes the equipment based on the knowledge of domain experts:

[0094] Take a certain F-class gas-steam combined cycle generator set as an example, refer to Figure 3 , the unit includes a gas turbine, a waste heat boiler, a steam turbine, a generator and other main equipment and related auxiliary systems;

[0095] Among them, the gas turbine includes main system components such as compressor, combustion chamber, turbine, etc. When the unit is running, the gas turbine compressor compresses the ambient air and mixes it with natural gas to burn and drive the turbine to do work. The high-temperature exhaust gas of the gas turbine continues to enter the waste heat boiler to heat the boiler water to produce high-temperature and high-pressure steam and then drive the steam turbine power generation equipment. According to the equipment level, the gas-steam combined cycle unit can be subdivided into main equipment such as gas turbine, waste heat boiler, steam turbine, generator and important auxiliary systems; the gas turbine body can be further decomposed into key systems such as intake and exhaust system, compressor system, combustion chamber system, turbine system, gas turbine shaft system, etc.; the waste heat boiler mainly includes high-pressure system, medium-pressure system, low-pressure system, flue gas system and other main systems; the steam turbine mainly includes the steam turbine shaft;

[0096] refer to Figure 4 According to the hierarchical division of gas-fired power generation units, a list of main intelligent early warning models is formulated. After determining the scope of the main fault early warning, it is necessary to analyze the key monitoring parameters related to the early warning models of each equipment in combination with the knowledge of field experts. The equipment and monitoring parameter analysis is carried out by taking the early warning model of the gas turbine as an example:

[0097] (1) Compressor system

[0098] The compressor is driven by a turbine. Its main performance parameters include real-time monitoring parameters and index calculation parameters such as speed, flow, air flow, pressure ratio and efficiency. After analysis based on expert domain knowledge, the important parameters for compressor system status monitoring are initially obtained as follows:

[0099]

[0100] (2) Turbine system

[0101] The main function of the gas turbine turbine is to output mechanical work. Part of the output mechanical work is used to drive the compressor and auxiliary machines, and part of it is output to drive power generation. After disassembly and analysis based on expert domain knowledge, the important parameters for gas turbine status monitoring are as follows:

[0102]

[0103]

[0104] (3) Combustion chamber

[0105] The combustion chamber is composed of fuel nozzle, guide sleeve, flame tube, cross-flame tube, transition section and other components. After disassembly and analysis based on expert domain knowledge, the important parameters for monitoring the combustion chamber of the gas turbine are as follows:

[0106]

[0107] S2. Based on expert knowledge, preliminary selection is made for the warning range and related parameters of the warning model. Then, the correlation algorithm is used to further refine the correlation between the parameters in each model and extract the key characteristic variables required for the warning model. The specific operation process at this time includes:

[0108] S21. After selecting the relevant parameters of the model based on the expert domain knowledge, the initially selected relevant parameters are randomly selected as the initial variables X and Y. Then the Pearson correlation coefficient ρ is:

[0109]

[0110] where cov(X, Y) is the covariance of the initial variables X and Y, σ x , σ y are the standard deviations of the initial variables X and Y respectively;

[0111] At this time, based on the Pearson correlation coefficient ρ x , y The value range is between [-1,1], and |ρ x , yThe closer | is to 1, the stronger the linear correlation between variables X and Y is. x , y The closer | is to 0, the weaker the linear correlation between variables X and Y. x , y = 0 means that there is no linear relationship between variables X and Y, ρ x , y When it is a positive number, it means that there is a positive correlation between variables X and Y. x , y When it is a negative number, it means that there is a negative correlation between variables X and Y;

[0112] S22. After repeating the above step S21 for multiple times to select relevant parameters to calculate the linear correlation, the final warning model parameters are selected as the key feature variables required for the warning model based on the parameter variables with higher linear correlation;

[0113] S3. Perform data cleaning operations on key feature variables. The data cleaning operations include:

[0114] S31. For key characteristic variable data, remove abnormal data including abnormal data and fault data in historical data;

[0115] S32. In order to obtain as much historical data as possible under different working conditions, reduce redundant data and increase the amount of historical data, the working condition of key characteristic variable data is judged, and different sampling granularities are set for different working conditions including shutdown and operation to re-collect data;

[0116] At this time, by implementing detailed data verification and working condition adaptability processing in the data cleaning stage, such as eliminating abnormal data, adjusting the sampling frequency and data normalization according to different working conditions, it plays a role in improving data quality and model robustness, and ultimately achieves a more stable and reliable monitoring and early warning effect;

[0117] S33. Normalize the measured values ​​of the key characteristic variable data according to their respective extreme values ​​so that the actual measured values ​​are mapped to the interval [0, 1];

[0118] S34. According to the number of parameters required by the model, a multi-dimensional historical data matrix is ​​constructed. For each parameter, the interval [0, 1] is divided into h equal parts, and several observation vectors are found from the set with a step size of 1 / h and added to the training data;

[0119] S35. The same operation as step S34 is performed on the remaining parameters to obtain the processed matrix, and then deduplication is performed to obtain the final matrix;

[0120] S4. Use the key feature variables after data cleaning to filter and use. At this time, you can refer to step S3 for reference Figure 5 As shown in the figure, the intelligent early warning model is optimized based on the algorithm fusion method. Since the traditional early warning technology is based on the current single algorithm model, there is a problem of large prediction error. Therefore, the algorithm fusion method at this time includes fusing the nonlinear state assessment algorithm with the neural network algorithm to effectively improve the accuracy of the early warning model. At this time, the fusion optimization path is as follows Figure 6 As shown in Figure 2, the specific steps of algorithm fusion include:

[0121] S41. Extract corresponding data from the normal mode of the current data and the stored system operation process state, dig out the relationship between the various observed variables of the system from the historical data, and establish a non-parametric qualitative model based on the nonlinear state estimation method of the similarity principle in the prior art. This method estimates the system operation state by comparing the similarity between the current data and the stored system operation process state, that is, monitoring the similarity between these multi-dimensional signals. By extracting the normal mode of the system state, digging out the relationship between the various observed variables of the system from the historical data, a non-parametric qualitative model is established;

[0122] Then, the system state estimation is calculated by weighted combination of normal modes of system state in nonparametric qualitative model. The weight is determined by the similarity between modes. The principle of state estimation modeling is as follows: Figure 7 As shown;

[0123] At this time, the specific steps to establish a non-parametric qualitative model and calculate the state estimate of the system include:

[0124] S411. Let X be an observation vector with n dimensions, then the i-th observation vector is X(i) = [x1(i), x2(i)…x n (i)], assuming that the matrix D with m groups is the historical data of normal operation samples, then:

[0125]

[0126] S412. Since the magnitude of the parameters of each device of the power generation equipment is different, calculation errors will occur in the data operation process. In order to reduce the speed reduction and possible errors in the matrix operation caused by the excessive magnitude of the data, the sample data and the observation data are normalized, and the evaluation data is denormalized. In actual operation, the sensor needs to collect a large amount of data. There may be empty data in the sample D due to the different data collection frequencies of each device and the interference of the sensor. Therefore, it is necessary to clean the data of D;

[0127] S413. When the system is running normally, when each set of observation data is input, the system evaluates each set of input data, extracts the maximum and minimum values ​​in the sample matrix D and the observation data to form a matrix and then normalizes it, extracts the observation data in the normalized matrix and then performs nonlinear state evaluation, combines the evaluation results with the normalized maximum and minimum values ​​and performs denormalization, and finally obtains the correct state evaluation data;

[0128] Furthermore, the specific steps for evaluating each set of input data include:

[0129] L1. Set the observation data matrix input by the system to X obs , the corresponding evaluation data matrix at the same time is X est , the output value of the preset model is a linear combination of all observation vectors in the sample matrix D, then there is a weight vector W, which is:

[0130] X est =(D T W) T

[0131] The weight vector W = [w1, w2…w m ];

[0132] L2. There is a residual ε between the input observation value and the output evaluation value. To minimize ε, there exists:

[0133]

[0134] L3. Because the sample matrix D cannot be inverted, a model evaluation rule is formulated. The dot product in the formula is replaced by the Euclidean distance, and then:

[0135]

[0136] We can get:

[0137]

[0138] If the model input matrix X obs When the system is in normal working condition, the value should be close to one of the data in the sample matrix D, and the evaluation matrix X est With X obs is closer, and the residual ε is smaller;

[0139] If X obs is fault data, and when it exceeds the sample matrix D, the residual ε becomes larger. Therefore, the system data abnormality can be predicted according to the change of ε, and the model algorithm can be debugged according to ε;

[0140] because is a quantity, let the quantity be A, then:

[0141]

[0142] In the formula, A is only determined by the sample matrix D, and only needs to be calculated once, and the calculation result can be brought into the state evaluation of the next set of observation vectors, reducing the calculation time;

[0143] S42. Preset the BP neural network algorithm and perform algorithm analysis. The BP neural network algorithm here is as follows: Figure 8 After the sample is input, the BP algorithm is used to adjust and train the network weights and thresholds multiple times to make the actual output results as close as possible to the expected results, including:

[0144] S421. After the sample is input, the BP algorithm is used to adjust the weights and thresholds of the network for multiple training, so that the actual output result is as close as possible to the expected result. At this time, the BP neural network setting includes an input layer i, an output layer k, and a hidden layer j;

[0145] S422. Preset ω ij represents the weight between the input layer and the hidden layer, ω jk Represents the weight between the hidden layer and the output layer. According to the input x, the weight ω ij And the threshold a of the hidden layer, it can be concluded that the output of the hidden layer is:

[0146]

[0147] Among them, f is the activation function, which is used to convert the input value into an output value within a limited range through function conversion;

[0148] S423. Based on the output H and weight ω jk , the threshold b of the output layer, the output O of the output layer can be obtained k :

[0149]

[0150] Among them, f is the activation function, which is used to convert the input value into an output value within a limited range through function conversion;

[0151] S424. During the implementation process, the measured values ​​x1, x2, ..., x n Get the actual estimate through the neural network Compare the estimated value with the actual value y i Compare the residual values When one of the residual values ​​exceeds the threshold, an early warning message will be triggered;

[0152] S43. Through algorithm analysis, it can be known that the nonlinear state assessment algorithm predicts and alarms the system operation status by comparing the similarity between current data and historical data (that is, monitoring the similarity between these multidimensional signals). However, in the implementation process, on the one hand, due to the limitations of the nonlinear state assessment algorithm itself, the large amount of historical data will cause the historical data training process to be too slow, or even unable to calculate the results; on the other hand, when the amount of historical data is certain, it is difficult to enumerate all boundary environmental conditions and operating conditions of power generation equipment. Therefore, the early warning model based on the nonlinear assessment algorithm still has the problem of increased prediction residuals due to the deviation between current data and historical data, and then the early warning model will be falsely reported and the accuracy will be reduced. Therefore, based on multiple interrelated variables at different times when the equipment is operating normally, a pattern matrix, a compensation matrix and a compensation observation vector are formed;

[0153] S44. Determine the similarity coefficient matrix according to the similarity operator, pattern matrix and compensation matrix of the system when the device is operating normally, determine the compensation prediction vector according to the nonlinear evaluation algorithm, the pattern matrix and the compensation observation vector, and then generate the compensation residual vector;

[0154] S45. Using the BP neural network algorithm, based on the similarity coefficient matrix and the compensation residual vector, a residual prediction model is trained, and the residual in the nonlinear evaluation algorithm is compensated based on the output result of the residual prediction model, thereby reducing the data prediction deviation and ultimately improving the model accuracy;

[0155] In summary, by combining nonlinear state evaluation with BP neural network for algorithm fusion, not only the prediction performance of the model is optimized, but also a reasonable residual compensation mechanism is set at the algorithm level, which can significantly reduce the prediction deviation, thereby enhancing the model's ability to identify abnormal states, and ultimately ensuring that the early warning is timely and accurate;

[0156] S5. Setting warning rules for the optimized warning model. The warning model after the warning rules are set is used for monitoring and warning of gas-fired generator sets. The specific steps for setting the warning rules include:

[0157] S51. Analyze the alarm limit values ​​of characteristic parameters in different algorithm models of gas turbine generator sets under different operating conditions and boundary conditions, and realize dynamic threshold setting of multiple parameters within the DCS alarm limit;

[0158] S52. Based on the data-driven algorithm, the expert experience of operation and maintenance is converted into parameter alarm rules to detect and identify abnormalities of related parameters;

[0159] S53. A sliding window method is introduced for the residual alarm rule of a single parameter to continuously and in real time detect the changes in the statistical characteristics of the residual and reduce false alarms;

[0160] S54. Reference Fig. 9 , the sliding window residual statistics method is used to suppress the adverse effects of isolated large residuals by calculating the statistical characteristics of multiple residuals in the sliding window. At this time, within the preset time, the early warning model predicts the residual sequence as: ε GT =[ε1,ε2,…,ε N ,…]

[0161] For this sequence, take a sliding window with a width of N and calculate the mean of the N consecutive residuals in the window. And standard deviation S:

[0162]

[0163] S55. During the implementation process, the model effect is evaluated, and a reasonable sliding window width N is set for each model in combination with expert domain knowledge, so as to ensure that the early warning model can promptly and quickly reflect the continuous changes in the statistical characteristics of the residuals, eliminate the influence of random factors, improve the reliability of intelligent state monitoring early warning, and reduce the probability of false alarms;

[0164] At this time, the warning rules are set by optimizing the warning model and setting a reasonable sliding window width for each model. This can ensure that the warning model can promptly and quickly reflect the continuous changes in the statistical characteristics of the residuals while eliminating the influence of random factors, thereby improving the reliability of intelligent status monitoring and warning and reducing the probability of false alarms.

[0165] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0166] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0167] In another embodiment provided in the present application, a computer program product comprising instructions is also provided, which, when executed on a computer, enables the computer to execute the cloud-edge collaborative monitoring and early warning method for any gas generator set in the above-mentioned embodiments.

[0168] It is understandable that the system provided by the embodiment of the present invention corresponds to the method provided by the embodiment of the present invention, and the explanation, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the above method.

[0169] The embodiment of the present application also provides an electronic device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus.

[0170] Memory, used to store computer programs;

[0171] The processor is used to implement the cloud-edge collaborative monitoring and early warning method of the above-mentioned gas generator set when executing the program stored in the memory.

[0172] The communication bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0173] The communication interface is used for communication between the above electronic device and other devices.

[0174] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0175] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, and discrete hardware components.

[0176] It should also be noted that electronic devices also include terminal devices, which can also be called terminals, user equipment (UE), mobile stations (MS), mobile terminals (MT), etc. Terminal devices can be mobile phones, smart TVs, wearable devices, tablet computers (Pad), computers with wireless transceiver functions, virtual reality (VR) terminal devices, augmented reality (AR) terminal devices, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in remote medical surgery, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, wireless terminals in smart homes, etc. The embodiments of the present application do not limit the specific technology and specific device form adopted by the terminal devices.

[0177] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.

[0178] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

[0179] In addition, it should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative position relationship, movement status, etc. between the components in a certain specific posture. If the specific posture changes, the directional indication will also change accordingly.

[0180] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the meaning of "and / or" appearing in the full text includes three parallel schemes. Taking "A and / or B" as an example, it includes scheme A, or scheme B, or schemes that A and B meet at the same time. In addition, in the embodiments of the present invention, "multiple" refers to more than two. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary technicians in the field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

Claims

1. A cloud-edge collaborative monitoring and early warning method for a gas generator set, characterized in that: Follow these steps on your computer device: S1. Determine the main fault warning range through preliminary parameters, disassemble and analyze the gas generator set equipment to build an intelligent warning model, and analyze the key monitoring parameters related to the warning model of each equipment in combination with the knowledge of field experts; S2. Based on expert knowledge, preliminary selection of warning range and warning model related parameters is carried out, and the correlation relationship of parameters in each model is studied through correlation algorithm to extract key characteristic variables required for the warning model; S3. Perform data cleaning operations on key feature variables; S4. Use the key characteristic variables after data cleaning to optimize the intelligent early warning model based on the algorithm fusion method; S5. Setting warning rules for the optimized warning model. The warning model after the warning rules are set is used for monitoring and warning of gas-fired power generation units.

2. The cloud-edge collaborative monitoring and early warning method for a gas generator set according to claim 1, characterized in that: The specific operation process of studying the correlation between parameters in each model and extracting the key characteristic variables required for the early warning model in step S2 includes: S21. After selecting the relevant parameters of the model based on the expert domain knowledge, the initially selected relevant parameters are randomly selected as the initial variables X and Y. Then the Pearson correlation coefficient ρ is: where cov(X, Y) is the covariance of the initial variables X and Y, σ x , σ y are the standard deviations of the initial variables X and Y respectively; At this time, based on the Pearson correlation coefficient ρ x , y The value range is between [-1,1], and |ρ x , y The closer | is to 1, the stronger the linear correlation between variables X and Y is. x , y The closer | is to 0, the weaker the linear correlation between variables X and Y. x , y = 0 means that there is no linear relationship between variables X and Y, ρ x , y When it is a positive number, it means that there is a positive correlation between variables X and Y. x , y When it is a negative number, it means that there is a negative correlation between variables X and Y; S22. After repeating the above step S21 for multiple times to select relevant parameters and calculate the linear correlation, the final early warning model parameters are selected as the key feature variables required for the early warning model based on the parameter variables with higher linear correlation.

3. The cloud-edge collaborative monitoring and early warning method for a gas generator set according to claim 1, characterized in that: The data cleaning operation in step S3 includes: S31. For key characteristic variable data, remove abnormal data including abnormal data and fault data in historical data; S32. Identify the working conditions of the key characteristic variable data, and set different sampling granularities for different working conditions including shutdown and operation to re-collect data; S33. Normalize the measured values ​​of the key characteristic variable data according to their respective extreme values ​​so that the actual measured values ​​are mapped to the interval [0, 1]; S34. According to the number of parameters required by the model, a multi-dimensional historical data matrix is ​​constructed. For each parameter, the interval [0, 1] is divided into h equal parts, and several observation vectors are found from the set with a step size of 1 / h and added to the training data; S35. The same operation as step S34 is performed on the remaining parameters to obtain a processed matrix, and then duplicate removal is performed to obtain a final matrix.

4. The cloud-edge collaborative monitoring and early warning method for a gas generator set according to claim 1, characterized in that: The algorithm fusion method in step S4 includes fusing the nonlinear state assessment algorithm with the neural network algorithm to effectively improve the accuracy of the early warning model. The specific steps of the algorithm fusion include: S41. extract corresponding data from the current data and the normal mode of the stored system operation process state, mine the relationship between the observed variables of the system from the historical data, establish a non-parametric qualitative model and calculate the state estimation of the system through the weighted combination of the normal mode of the system state in the non-parametric qualitative model, and the weight is determined by the similarity between the modes; S42. Preset BP neural network algorithm and perform algorithm analysis; S43. Composition of a mode matrix, a compensation matrix and a compensation observation vector based on multiple interrelated variables at different times during normal operation of the device; S44. Determine a similarity coefficient matrix according to the similarity operator of the system when the device is operating normally, the mode matrix and the compensation matrix, determine a compensation prediction vector according to the nonlinear evaluation algorithm, the mode matrix and the compensation observation vector, and then generate a compensation residual vector; S45. Using the BP neural network algorithm, based on the similarity coefficient matrix and the compensated residual vector, a residual prediction model is trained, and the residual in the nonlinear evaluation algorithm is compensated based on the output result of the residual prediction model, thereby reducing the data prediction deviation.

5. The cloud-edge collaborative monitoring and early warning method for a gas generator set according to claim 4, characterized in that: The specific operation steps of establishing a non-parametric qualitative model and calculating the state estimation of the system in step S41 include: S411. Let X be an observation vector with n dimensions, then the i-th observation vector is i(i) = [x1(i), x2(i)…x n (i)], assuming that the matrix D with m groups is the historical data of normal operation samples, then: S412. Since the magnitude of the parameters of each device of the power generation equipment is different, calculation errors will occur in the data operation process. In order to reduce the speed reduction and possible errors in matrix operation caused by excessive data magnitude, the sample data and the observation data are normalized, and the evaluation data is denormalized; S413. When the system is running normally, for each set of observation data input, the system evaluates each set of input data, extracts the maximum and minimum values ​​in the sample matrix D and the observation data to form a matrix and then normalizes it, extracts the observation data in the normalized matrix and then performs nonlinear state evaluation, combines the evaluation results with the normalized maximum and minimum values ​​and performs denormalization, and finally obtains the correct state evaluation data.

6. The cloud-edge collaborative monitoring and early warning method for a gas generator set according to claim 5, characterized in that: The specific operation steps of evaluating each set of input data in step S413 include: L1. Set the observation data matrix input by the system to X obs , the corresponding evaluation data matrix at the same time is X est , the output value of the preset model is a linear combination of all observation vectors in the sample matrix D, then there exists a weight vector W, which is: X est =(D T W) T The weight vector W = [w1, w2…w m ]; L2. There is a residual ε between the input observation value and the output evaluation value. To minimize ε, there exists: L3. Because the sample matrix D cannot be inverted, a model evaluation rule is formulated. The dot product in the formula is replaced by the Euclidean distance, and then: We can get: If the model input matrix X obs When the system is in normal working condition, the value should be close to one of the data in the sample matrix D, and the evaluation matrix X est With X obs Close, the residual ε is small; If X obs is fault data, and when it exceeds the sample matrix D, the residual ε becomes larger. Therefore, the abnormality of system data can be predicted according to the change of ε, and the model algorithm can be debugged according to ε; because is a quantity, let the quantity be A, then: In the formula, A is only determined by the sample matrix D. It only needs to be calculated once, and the calculation result can be brought into the state evaluation of the next set of observation vectors, reducing the calculation time.

7. The cloud-edge collaborative monitoring and early warning method for a gas generator set according to claim 4, characterized in that: The BP neural network algorithm in step S42 includes: S421. After the sample is input, the BP algorithm is used to adjust the weights and thresholds of the network for multiple training, so that the actual output result is as close as possible to the expected result. At this time, the BP neural network setting includes an input layer i, an output layer k, and a hidden layer j; S422. Preset ω ij represents the weight between the input layer and the hidden layer, ω jk Represents the weight between the hidden layer and the output layer. According to the input x, the weight ω ij And the threshold a of the hidden layer, it can be concluded that the output of the hidden layer is: Where, f is the activation function; S423. Based on the output H and weight ω jk , the threshold b of the output layer, the output O of the output layer can be obtained k : Where, f is the activation function; S424. During the implementation process, the measured values ​​x1, x2, ..., x n Get the actual estimate through the neural network Compare the estimated value with the actual value y o Compare the residual values When one of the residual values ​​exceeds the threshold, an early warning message will be triggered.

8. The cloud-edge collaborative monitoring and early warning method for a gas generator set according to claim 1, characterized in that: The specific operation steps for setting the warning rules in step S5 include: S51. Analyze the alarm limit values ​​of characteristic parameters in different algorithm models of gas turbine generator sets under different operating conditions and boundary conditions, and realize dynamic threshold setting of multiple parameters within the DCS alarm limit; S52. Based on the data-driven algorithm, the expert experience of operation and maintenance is converted into parameter alarm rules to detect and identify abnormalities of related parameters; S53. A sliding window method is introduced for the residual alarm rule of a single parameter to continuously and in real time detect the changes in the statistical characteristics of the residual and reduce false alarms; S54. The sliding window residual statistics method is used to suppress the adverse effects of isolated large residuals by calculating the statistical characteristics of multiple residuals in the sliding window. At this time, within the preset time, the early warning model predicts the residual sequence as: ε GT =[ε1,ε2,…,ε N ,…] For this sequence, take a sliding window with a width of N and calculate the mean of the N consecutive residuals in the window. And standard deviation S: S55. During the implementation process, the model effect is evaluated and, combined with expert domain knowledge, a reasonable sliding window width N is set for each model to ensure that the early warning model can promptly and quickly reflect the continuous changes in the statistical characteristics of the residuals and eliminate the influence of random factors.

9. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 8.

10. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 8.

Images