Power generation equipment vibration and temperature combined monitoring method and related system

Abstract

The invention belongs to the technical field of power generation equipment, discloses a power generation equipment vibration and temperature combined monitoring method and a related system, and overcomes the defect that a traditional system only depends on a single vibration parameter by continuously acquiring real-time state information of the power generation equipment and comprehensively monitoring various parameters such as vibration and temperature. And more comprehensive equipment state evaluation is provided. According to the method, the defects of the power generation equipment can be quickly identified and fed back based on the grouping preprocessing of the preset period, so that the fault discovery and response time is shortened, and the maintenance efficiency is improved. According to the method, anomaly detection is performed on the periodic sequence information by using the pre-constructed signal mutation model, so that potential problems can be identified more accurately, the risks of missing detection and false detection are reduced, and higher detection sensitivity is ensured. According to the method, deep learning and particle filtering algorithms are combined, and the constructed conjoint analysis model can perform deep analysis on abnormal points, so that an operation and maintenance team is helped to make a more scientific decision.

Description

Technical Field

[0001] The present invention belongs to the technical field of power generation equipment, and in particular relates to a method for jointly monitoring vibration and temperature of power generation equipment and a related system. Background Art

[0002] Power generation equipment monitoring is an important means to ensure the safe and stable operation of the power system. By real-time monitoring of the operating status of the equipment, potential problems can be discovered in time, failures can be avoided, and the reliability and service life of the equipment can be improved.

[0003] In a wind farm, by installing a vibration, speed, and temperature three-parameter combined probe to monitor the wind turbine in real time, the accuracy of fault warning can be significantly improved, and downtime and maintenance costs caused by equipment failure can be reduced. Specifically, a temperature-vibration integrated sensor can be used to achieve real-time monitoring of the vibration and temperature information of wind power generation equipment. The temperature-vibration integrated sensor combines vibration and temperature monitoring functions, and can simultaneously obtain the vibration parameters and temperature information of the equipment, providing an accurate reference for equipment fault diagnosis and predictive maintenance.

[0004] Chinese patent CN116877358B discloses a wind turbine vibration monitoring and analysis system, which analyzes and obtains the vibration influence coefficients of blades, nacelles and towers, comprehensively considers the factors affecting the vibration of wind turbines, and corrects the analysis results of the wind turbine vibration; obtains the vibration curves of blades, nacelles and towers, analyzes the vibration proportional coefficients of blades, nacelles and towers, and uses curve images to perform dynamic and trend analysis on vibration data; analyzes the vibration over-limit index of blades, nacelles and towers based on the vibration influence coefficients and vibration proportional coefficients of blades, nacelles and towers; However, the existing system only evaluates the status of the generator set by a single type of vibration parameter of the generator set, which is not good, and cannot timely and effectively feedback the potential risks of the generator set. In response to the above problems, we propose a joint monitoring system and method for vibration and temperature of power generation equipment. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the existing system that the status of the generator set is poorly evaluated by only using a single type of vibration parameter of the generator set and cannot provide timely and effective feedback to the generator set, and to provide a method and related system for jointly monitoring the vibration and temperature of power generation equipment.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for jointly monitoring vibration and temperature of power generation equipment, comprising the following steps: Continuously acquiring real-time status information of power generation equipment, and performing group preprocessing on the real-time status information of power generation equipment based on a preset period to obtain at least one group of period sequence information; Pre-construct a signal mutation model, input the periodic sequence information into the pre-constructed signal mutation model, the signal mutation model performs anomaly detection on the periodic sequence information, and outputs an abnormal point queue; Based on deep learning combined with particle filtering algorithm and outlier queue, a joint analysis model is constructed; The outlier queue is input into the joint analysis model. The joint analysis model conducts joint analysis on the outlier queue, outputs the risk value corresponding to the outlier, compares the risk value with the preset risk threshold, and retains the outliers corresponding to the risk values ​​greater than the preset risk threshold. The abnormal points whose risk values ​​exceed the preset risk threshold are loaded, and the comprehensive risk coefficient of the power generation equipment is calculated based on the iterative method combined with the cascading failure algorithm.

[0007] A further improvement of the present invention is that the real-time status information of the power generation equipment includes an equipment vibration signal, an equipment temperature signal, an equipment speed signal and an equipment environment signal.

[0008] A further improvement of the present invention is that the specific method for pre-constructing the signal mutation model is as follows: The Stacking model is used as the initial model of the signal mutation model. The initial model includes an input layer, a base learner, a meta learner, and an output layer. The Isolation Forest algorithm is introduced to improve the base learner in the initial model and freeze the decision tree of the base learner; The LSTM is used to replace the gradient descent optimization method of the meta-learner in the initial model, and the meta-learner is improved and optimized to obtain an improved initial model; The periodic sequence information is divided into a mutation training sample set and a mutation test sample set. Multiple base learners are trained based on the K-fold cross-validation method. The mutation training sample set is divided into training data and test data. Both the training data and the test data contain high-dimensional input features and signal labels. Each group of base learners generates corresponding base prediction results. Obtain at least one set of base prediction results, integrate the base prediction results into a new low-dimensional output feature and signal label composite queue, train a meta-learner using the low-dimensional output feature and signal label composite queue, and generate a meta-prediction result; Use low-dimensional output features for forward propagation and back propagation, calculate the gradient of the LSTM loss function, and use the LSTM model to update the network parameters of the meta-learner until the total loss function of the initial model reaches the preset convergence condition; Obtain a mutation test sample set, and verify whether the meta-prediction result meets the preset model accuracy threshold based on the mutation test sample set; If the preset model accuracy threshold is met, the signal mutation model after training is output.

[0009] A further improvement of the present invention is that, based on the K-fold cross-validation method, multiple base learners are trained, and when the mutation training sample set is divided into training data and test data, the mutation training sample set is evenly divided into K parts, of which K-1 parts are training data and 1 part is test data.

[0010] A further improvement of the present invention is that the periodic sequence information is input into a pre-built signal mutation model, and the signal mutation model performs anomaly detection on the periodic sequence information. The specific method of outputting anomaly point queue is as follows: Obtain periodic sequence information, and the input layer of the signal mutation model performs dimension reduction processing on the periodic sequence information to obtain dimension-reduced sequence data; Load the reduced-dimensionality sequence data, and the base learner calculates the abnormal clustering degree corresponding to the reduced-dimensionality sequence data based on the Isolation Forest algorithm; Obtain the abnormal clustering degree corresponding to the reduced-dimensional sequence data, and the meta-learner corrects the abnormal clustering degree based on the fuzzy c-means clustering algorithm to obtain the clustering correction value; Perform weighted summation on the cluster correction value and the abnormal clustering degree to obtain the weighted abnormal value; Determine whether the weighted outlier exceeds the preset outlier threshold. If so, output the outlier point corresponding to the reduced dimensionality sequence data. If the abnormal threshold is not exceeded, the dimension-reduced sequence data is normal; At least one group of outliers is integrated to obtain an outlier queue.

[0011] A further improvement of the present invention is that the specific method of constructing a joint analysis model based on deep learning combined with a particle filter algorithm and an outlier queue is as follows: Obtain an abnormal point queue, divide the abnormal point queue into an abnormal training sample set and an abnormal verification sample set, and initialize the abnormal index of the abnormal verification sample set; The initial model of the joint analysis model is constructed based on deep learning combined with particle filtering algorithm, wherein the initial model includes a probability distributor, a state transition module and an observer; An optimization signal algorithm is introduced into the state transfer module, and the optimization signal algorithm optimizes the parameters of the state transfer module; Load the abnormal training sample set, and use the probability distributor to extract the characteristic factors of the abnormal training sample set to obtain a characteristic factor extraction set; The state transfer module decomposes the characteristic factor extraction set to obtain a characteristic decomposition set, and determines the state transfer amount of the characteristic decomposition set and the optimal lag time corresponding to the state transfer amount; The state transition amount and the optimal delay corresponding to the state transition amount are loaded, and the observer calculates the forget gate corresponding to the state transition amount; The optimal lag time is updated based on the forget gate corresponding to the state transfer amount, and the risk value corresponding to the abnormal point is calculated by combining the optimal lag time and the state transfer amount; Continuously calculate the initial model loss function value of the joint analysis model, update the Bayesian posterior estimate, and iteratively optimize the joint analysis model; Obtain an abnormal verification sample set, evaluate the performance of the joint analysis model based on the abnormal verification sample set, calculate the group stability index between the predicted risk value and the actual risk value, and output the joint analysis model if the group stability index is less than the preset threshold.

[0012] A further improvement of the present invention is that the outlier queue is input into the joint analysis model, and the joint analysis model conducts joint analysis on the outlier queue, and the specific method of outputting the risk value corresponding to the outlier is as follows: Load the outlier queue, perform feature extraction on the data vectors in the outlier queue, and obtain a feature factor extraction set; Determine an optimal lag corresponding to the feature factor extraction set, and update the optimal lag based on a forget gate corresponding to the state transfer amount; Calculate the risk value corresponding to the abnormal point by combining the optimal lag time and state transfer amount; Output the risk value corresponding to the abnormal point.

[0013] In a second aspect, the present invention provides a combined vibration and temperature monitoring system for power generation equipment, comprising: An information preprocessing module, used to continuously obtain real-time status information of power generation equipment, and perform group preprocessing on the real-time status information of power generation equipment based on a preset period to obtain at least one group of period sequence information; An abnormal point queue output module is used to pre-build a signal mutation model, input the periodic sequence information into the pre-built signal mutation model, and the signal mutation model performs abnormality detection on the periodic sequence information and outputs an abnormal point queue; Joint analysis model building module, which is used to build a joint analysis model based on deep learning combined with particle filtering algorithm and outlier queue; The risk value screening module is used to input the outlier queue into the joint analysis model. The joint analysis model conducts joint analysis on the outlier queue, outputs the risk value corresponding to the outlier, compares the risk value with the preset risk threshold, and retains the outliers corresponding to the risk values ​​greater than the preset risk threshold; The comprehensive risk coefficient acquisition module is used to load abnormal points whose risk values ​​exceed the preset risk threshold, and calculate the comprehensive risk coefficient of the power generation equipment based on the iterative method combined with the cascading failure algorithm.

[0014] In a third aspect, the present invention provides an electronic device including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of a method for jointly monitoring vibration and temperature of power generation equipment when executing the computer program.

[0015] In a fourth aspect, the present invention provides a storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a method for jointly monitoring vibration and temperature of power generation equipment.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention overcomes the shortcomings of traditional systems that rely only on a single vibration parameter by continuously acquiring real-time status information of power generation equipment and comprehensively monitoring multiple parameters such as vibration and temperature, and provides a more comprehensive equipment status assessment. The present invention is based on group preprocessing of preset cycles, which can quickly identify and feedback the shortcomings of power generation equipment, thereby shortening the fault discovery and response time and improving maintenance efficiency. The present invention uses a pre-constructed signal mutation model to perform anomaly detection on periodic sequence information, which can more accurately identify potential problems, reduce the risk of missed detection and false detection, and ensure higher detection sensitivity. The present invention combines deep learning and particle filtering algorithms to construct a joint analysis model that can conduct in-depth analysis of abnormal points and help the operation and maintenance team make more scientific decisions. The present invention can accurately identify high-risk states by comparing the risk value of abnormal points with a preset threshold, take timely measures, and avoid the occurrence of equipment failures. The present invention uses an iterative method and a cascading failure algorithm to calculate the comprehensive risk coefficient of power generation equipment, which helps to predict potential chain failures, thereby improving the overall reliability and safety of the equipment. In summary, the present invention improves the monitoring and maintenance effect of power generation equipment, not only improves the fault warning capability, but also enhances the scientificity and effectiveness of operation and maintenance management.

[0017] Furthermore, the signal mutation model of the present invention uses the Stacking model as the initial model, introduces the IsolationForest algorithm to improve the base learner, and uses LSTM to replace the gradient descent optimization method of the meta-learner to improve and optimize the meta-learner, so that the signal mutation model can solve the problem of large computing resources and time consumption when the Stacking model processes vibration and temperature signals, and also improves the prediction performance and generalization ability of the model.

[0018] Furthermore, the present invention constructs an initial model of the joint analysis model based on deep learning combined with a particle filtering algorithm. At the same time, the joint analysis model introduces an optimized signal algorithm, which optimizes the parameters of the state transfer module, thereby helping the model to better handle uncertainty and noise. When combined with deep learning, the generalization ability of the model can be further enhanced, so that it can maintain good performance on unseen data. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flow chart of the present invention; Figure 2 is a flow chart of Example 1; Figure 3 This is a flow chart for implementing the method for pre-constructing a signal mutation model in Example 2; Figure 4 The signal mutation model in Example 3 performs anomaly detection on periodic sequence information and outputs anomaly point queue implementation flow chart; Figure 5 This is a flowchart for implementing the training method of the joint analysis model in Example 4; Figure 6 This is a flowchart of the implementation of the joint analysis method for the abnormal point queue by the joint analysis model in Example 5; Figure 7 This is a system schematic diagram of Example 6; Figure 8 It is a schematic diagram of the system structure of the present invention; DETAILED DESCRIPTION In order to further understand the content of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the embodiments are only for explaining the present invention and are not intended to limit it.

[0020] See also Figure 1 , a method for jointly monitoring vibration and temperature of power generation equipment, comprising the following steps: S1, continuously acquiring real-time status information of power generation equipment, and performing group preprocessing on the real-time status information of power generation equipment based on a preset period to obtain at least one group of period sequence information.

[0021] S2, pre-constructing a signal mutation model, inputting the periodic sequence information into the pre-constructed signal mutation model, and the signal mutation model performs anomaly detection on the periodic sequence information and outputs an abnormal point queue.

[0022] S3, based on deep learning combined with particle filtering algorithm and outlier queue, builds a joint analysis model.

[0023] S4, input the outlier queue into the joint analysis model, the joint analysis model conducts joint analysis on the outlier queue, outputs the risk value corresponding to the outlier, compares the risk value with the preset risk threshold, and retains the outliers corresponding to the risk values ​​greater than the preset risk threshold.

[0024] S5, loading the abnormal points whose risk values ​​exceed the preset risk threshold, and calculating the comprehensive risk coefficient of the power generation equipment based on the iterative method combined with the cascading failure algorithm.

[0025] See also Figure 8, power generation equipment vibration and temperature joint monitoring system, including: An information preprocessing module, used to continuously obtain real-time status information of power generation equipment, and perform group preprocessing on the real-time status information of power generation equipment based on a preset period to obtain at least one group of period sequence information; An abnormal point queue output module is used to pre-build a signal mutation model, input the periodic sequence information into the pre-built signal mutation model, and the signal mutation model performs abnormality detection on the periodic sequence information and outputs an abnormal point queue; Joint analysis model building module, which is used to build a joint analysis model based on deep learning combined with particle filtering algorithm and outlier queue; The risk value screening module is used to input the outlier queue into the joint analysis model. The joint analysis model conducts joint analysis on the outlier queue, outputs the risk value corresponding to the outlier, compares the risk value with the preset risk threshold, and retains the outliers corresponding to the risk values ​​greater than the preset risk threshold; The comprehensive risk coefficient acquisition module is used to load abnormal points whose risk values ​​exceed the preset risk threshold, and calculate the comprehensive risk coefficient of the power generation equipment based on the iterative method combined with the cascading failure algorithm.

[0026] Embodiment 1: The embodiment of the present invention provides a method for jointly monitoring vibration and temperature of power generation equipment. Figure 2 The schematic diagram of the implementation process of the method for joint monitoring of vibration and temperature of power generation equipment is shown. The method for joint monitoring of vibration and temperature of power generation equipment specifically includes: Step S10, obtaining real-time status information of the power generation equipment, the real-time status information including equipment vibration signal, equipment temperature signal, equipment speed signal, and equipment environment signal, and grouping and preprocessing the real-time status information of the power generation equipment based on a set period to obtain at least one set of period sequence information; It should be noted that the power generation equipment includes but is not limited to thermal power generation sets, hydropower generation sets, wind power generation sets, and solar power generation sets. The wind power generation sets are composed of multiple module structures such as wind turbines, blades, hubs, gear boxes, asynchronous generators, transformers, sensors, etc. Among them, when the real-time status information of the power generation equipment is grouped and preprocessed based on the set period, the set period can be 8 hours, 24 hours, 48 ​​hours, 7 days, 15 days, 30 days or 60 days, and the method of grouping the real-time status information of the power generation equipment can be data cleaning and denoising preprocessing.

[0027] Step S20, loading the periodic sequence information, pre-building a signal mutation model, taking the periodic sequence information as input, executing the signal mutation model, the signal mutation model performs anomaly detection on the periodic sequence information, and outputs an abnormal point queue; Step S30, obtain the outlier queue, input the outlier queue into the trained joint analysis model, the joint analysis model conducts joint analysis on the outlier queue, outputs the risk value corresponding to the outlier, and determines whether the risk value exceeds the preset risk threshold. If it is greater than the preset risk threshold, the risk value corresponding to the current outlier is retained.

[0028] In this embodiment, the risk threshold can be set to 0.3-0.35. If the risk value exceeds the preset risk threshold, the warning information is pushed to the background in real time based on the MODBUS protocol.

[0029] Step S40, loading the abnormal points whose risk values ​​exceed the preset risk threshold, and calculating the comprehensive risk coefficient of the power generation equipment based on the iterative method combined with the cascading failure algorithm.

[0030] In this embodiment, the comprehensive risk coefficient can be calculated by multiplying and summing the risk value of each power generation equipment and the preset weight of the abnormal point of the power generation equipment, thereby obtaining the comprehensive risk coefficient, which reflects the overall risk level faced by the equipment.

[0031] In the embodiment of the present invention, a method for joint monitoring of vibration and temperature of power generation equipment is provided. By pre-building a signal mutation model and a joint analysis model, accurate detection of abnormal points is achieved, and vibration and temperature signals from different sensors are integrated to jointly evaluate the risk value of power generation equipment. The current state and future trend of the equipment can be more comprehensively evaluated. By effectively utilizing the advantages of these two models, the safe and stable operation of power generation equipment can be better guaranteed, and energy utilization efficiency can be improved. The problem that the existing system only evaluates the state of the generator set through a single type of vibration parameter of the generator set is poor, and the potential risks of the generator set cannot be fed back in a timely and effective manner is overcome.

[0032] Embodiment 2: The embodiment of the present invention provides a method for pre-constructing a signal mutation model. Figure 3 A schematic diagram of a method for pre-building a signal mutation model is shown. The method for pre-building a signal mutation model specifically includes: Step S101, using the Stacking model as the initial model of the signal mutation model, the initial model includes an input layer, a base learner, a meta learner and an output layer; It should be noted that the core idea of ​​the Stacking model is to train a meta-learner to make the final prediction by taking the outputs of multiple base learners as input features. This method can effectively combine the advantages of different models, thereby improving the generalization ability of the entire system. In the embodiment of the present invention, using the Stacking model as the initial model of the signal mutation model has significant advantages for abnormal signal detection.

[0033] Step S102, introducing the Isolation Forest algorithm to improve the base learner and freezing the decision tree of the base learner; Step S103, using LSTM to replace the gradient descent optimization method of the meta-learner, improving and optimizing the meta-learner, and obtaining an improved initial model; In this embodiment, the signal mutation model uses the Stacking model as the initial model, and introduces the IsolationForest algorithm to improve the base learner. LSTM is used to replace the gradient descent optimization method of the meta-learner to improve and optimize the meta-learner. This enables the signal mutation model to solve the problem of large computing resources and time consumption when the Stacking model processes vibration and temperature signals, and also improves the prediction performance and generalization ability of the model.

[0034] Step S104, loading the periodic sequence information, dividing the periodic sequence information into a mutation training sample set and a mutation test sample set, training multiple base learners based on the K-fold cross-validation method, dividing the mutation training sample set into K parts on average, where K-1 parts are training data and 1 part is test data, and both the training data and the test data contain high-dimensional input features and signal labels, and each group of base learners generates a corresponding base prediction result; Step S105, obtaining at least one set of base prediction results, integrating the base prediction results into a new low-dimensional output feature and signal label composite queue, using the low-dimensional output feature and signal label composite queue to train a meta-learner to generate a meta-prediction result; Step S106, using the low-dimensional output features to perform forward propagation and back propagation, calculating the gradient of the LSTM loss function, and using the LSTM model to update the network parameters of the meta-learner until the total loss function of the initial model reaches a preset convergence condition; Step S107, obtaining a mutation test sample set, and verifying whether the meta-prediction result meets a preset model accuracy threshold based on the mutation test sample set; Step S108: If the model meets the preset model accuracy threshold, the trained signal mutation model is output. It should be noted that the preset model accuracy threshold can be set to 0.9-0.96.

[0035] If the preset model accuracy threshold is not met, the process returns to step S104 to continue iterative training of the initial model.

[0036] Embodiment 3: The embodiment of the present invention provides a method for detecting anomalies of periodic sequence information using a signal mutation model and outputting anomaly point queues. Figure 4The present invention shows a schematic diagram of a method for implementing anomaly detection of periodic sequence information by a signal mutation model and outputting anomaly point queues. The method for implementing anomaly detection of periodic sequence information by a signal mutation model and outputting anomaly point queues specifically includes: Step S201, obtaining periodic sequence information, and the input layer of the signal mutation model performs dimension reduction processing on the periodic sequence information to obtain dimension-reduced sequence data; It should be noted that when the input layer of the signal mutation model performs dimensionality reduction processing on the periodic sequence information, principal component analysis and linear analysis can be used to reduce the dimensionality of the periodic sequence information. By reducing the dimensionality of the periodic sequence information, the degree of data fusion can be improved, thereby ensuring the generalization ability and robustness of the model.

[0037] Step S202, loading the reduced-dimensionality sequence data, and the base learner calculates the abnormal clustering degree corresponding to the reduced-dimensionality sequence data based on the Isolation Forest algorithm; Among them, the abnormal clustering degree is calculated by the following formula: (1) in, is the abnormal clustering degree, Input vector representation for dimensionality-reduced sequence data, is the weight coefficient corresponding to the dimension-reduced sequence data, is the deviation coefficient of the path length in the Isolation Forest algorithm. In this embodiment, the deviation coefficient may be 0.02-0.05. is the time corresponding to the dimension-reduced sequence data, It is the cluster center point vector representation of the dimension-reduced sequence data; (2) in, The dimension reduction dimension of the reduced sequence data can be 1-8. is the dataset size for dimensionality-reduced sequence data.

[0038] Step S203, obtaining the abnormal clustering degree corresponding to the dimension-reduced sequence data, and the meta-learner corrects the abnormal clustering degree based on the fuzzy c-means clustering algorithm to obtain a cluster correction value; The cluster correction value is calculated by the following formula: (3) in, is the cluster correction value, is the cluster center point vector representation of the dimensionality reduction sequence data, is the Euclidean distance between the abnormal clustering degree and the cluster center of the reduced dimensionality sequence data, is the number of cluster centers, is the clustering fuzziness; (4) Step S204, performing weighted summation on the cluster correction value and the abnormal clustering degree to obtain a weighted abnormal value; (5) in, is the sampling interval of the dimension-reduced sequence data; Step S205, determining whether the weighted abnormal value exceeds a preset abnormal threshold; Step S206: if exceeded, output the abnormal point corresponding to the reduced dimensionality sequence data; If the abnormal threshold is not exceeded, the dimension-reduced sequence data is normal; Step S207: Integrate at least one group of outliers to obtain an outlier queue.

[0039] In an embodiment of the present invention, a signal mutation model is provided to perform anomaly detection on periodic sequence information and output an abnormal point queue. By performing anomaly detection on periodic sequence information, preprocessing of the periodic sequence information is achieved, and the model calculation complexity of the joint analysis model is reduced, thereby improving the data processing capability of the joint analysis model.

[0040] Embodiment 4: The embodiment of the present invention provides a training method for a joint analysis model. Figure 5 The figure shows a flowchart of a training method for a joint analysis model. The training method for a joint analysis model specifically includes: Step S301, obtaining an abnormal point queue, dividing the abnormal point queue into an abnormal training sample set and an abnormal verification sample set, and initializing the abnormal index of the abnormal verification sample set; In this embodiment, the allocation ratio of the abnormal training sample set to the abnormal verification sample set may be 3:1 or 4:1.

[0041] Step S302, constructing an initial model of the joint analysis model based on deep learning combined with a particle filter algorithm, wherein the initial model includes a probability distributor, a state transition module, and an observer; Step S303, introducing an optimization signal algorithm into the state transfer module, and optimizing the state transfer module parameters by the optimization signal algorithm; Step S304, loading the abnormal training sample set, and extracting characteristic factors from the abnormal training sample set by the probability distributor to obtain a characteristic factor extraction set; Step S305, the state transfer module decomposes the characteristic factor extraction set to obtain a characteristic decomposition set, and determines the state transfer amount of the characteristic decomposition set and the optimal lag time corresponding to the state transfer amount; Step S306, loading the state transition amount and the optimal delay time corresponding to the state transition amount, and the observer calculates the forget gate corresponding to the state transition amount; Among them, the forget gate corresponding to the state transfer amount is calculated by the following formula: (6) in, is the activation function of the observer, is the input weight matrix of the forget gate, is the circulant weight matrix, is the bias vector, and is the state transition amount, is the optimal delay corresponding to the state transfer amount; Step S307, updating the optimal lag time based on the forget gate corresponding to the state transition amount, and calculating the risk value corresponding to the abnormal point in combination with the optimal lag time and the state transition amount; Step S308, continuously calculating the initial model loss function value of the joint analysis model, updating the Bayesian posterior estimation, and iteratively optimizing the joint analysis model; Step S309, obtaining an abnormal verification sample set, evaluating the performance of the joint analysis model based on the abnormal verification sample set, calculating the group stability index between the predicted risk value and the true risk value, and outputting the joint analysis model if the group stability index is less than 0.1; If the population stability index is greater than 0.1, return to step S304.

[0042] In an embodiment of the present invention, an initial model of the joint analysis model is constructed based on deep learning combined with a particle filtering algorithm. At the same time, the joint analysis model introduces an optimized signal algorithm, which optimizes the parameters of the state transfer module, thereby helping the model to better handle uncertainty and noise. When combined with deep learning, the generalization ability of the model can be further enhanced, so that it can maintain good performance on unseen data.

[0043] Embodiment 5: The embodiment of the present invention provides a method for joint analysis of abnormal point queues using a joint analysis model. Figure 6 The figure shows a schematic diagram of the implementation process of the joint analysis model for the joint analysis method of the abnormal point queue, and the method of the joint analysis model for the joint analysis of the abnormal point queue specifically includes: Step S401, loading an outlier queue, performing feature extraction on data vectors in the outlier queue, and obtaining a feature factor extraction set; Among them, when extracting features, the feature extraction function is expressed as: (7) in, is the data vector input representation in the outlier queue, is the feature extraction coefficient. In this embodiment, the feature extraction coefficient may be 0.1-0.5. is the total number of characteristic factors, which can be 1-8; Step S402, determining the optimal lag time corresponding to the feature factor extraction set, and updating the optimal lag time based on the forget gate corresponding to the state transfer amount; The optimal lag time is calculated by the following formula: (8) in, is the output vector representation of the optimal delay, is the proportional control factor of the observer based on Gaussian distribution, and is the state transition amount, is the vibration signal component of the state transition quantity, It is the temperature signal component of the state transition quantity.

[0044] Step S403, calculating the risk value corresponding to the abnormal point by combining the optimal lag time and the state transfer amount; The risk value is calculated by the following formula: (9) in, is the risk value, is the state transfer loss coefficient, which is set based on the IMOMFO algorithm; Step S404, outputting the risk value corresponding to the abnormal point, and determining whether the risk value exceeds a preset risk threshold; Step S405: If the risk value is greater than the preset risk threshold, the risk value corresponding to the current abnormal point is retained.

[0045] Step S406: If it is less than the preset risk threshold, the current abnormal point is determined to be a potential risk abnormal point, and the current abnormal point is marked. If it is still less than the preset risk threshold after three set periods of continuous marking, the current abnormal point is discarded.

[0046] Embodiment 6: See also Figure 7 As shown, the present invention also provides an electronic device 100 for a method for jointly monitoring vibration and temperature of power generation equipment; the electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on the at least one processor 102, and at least one communication bus 104.

[0047] The memory 101 can be used to store the computer program 103, and the processor 102 implements the steps of the method for joint monitoring vibration and temperature of power generation equipment described in Example 1 by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101. The memory 101 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data (such as audio data) created according to the use of the electronic device 100, etc. In addition, the memory 101 can include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other non-volatile solid-state storage devices.

[0048] The at least one processor 102 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor 102 may be a microprocessor or any conventional processor, etc. The processor 102 is the control center of the electronic device 100, and uses various interfaces and lines to connect various parts of the entire electronic device 100.

[0049] The memory 101 in the electronic device 100 stores a plurality of instructions to implement a method for jointly monitoring vibration and temperature of a power generation device. The processor 102 may execute the plurality of instructions to implement: Continuously acquiring real-time status information of power generation equipment, and performing group preprocessing on the real-time status information of power generation equipment based on a preset period to obtain at least one group of period sequence information; Pre-construct a signal mutation model, input the periodic sequence information into the pre-constructed signal mutation model, the signal mutation model performs anomaly detection on the periodic sequence information, and outputs an abnormal point queue; Based on deep learning combined with particle filtering algorithm and outlier queue, a joint analysis model is constructed; The outlier queue is input into the joint analysis model. The joint analysis model conducts joint analysis on the outlier queue, outputs the risk value corresponding to the outlier, compares the risk value with the preset risk threshold, and retains the outliers corresponding to the risk values ​​greater than the preset risk threshold. The abnormal points whose risk values ​​exceed the preset risk threshold are loaded, and the comprehensive risk coefficient of the power generation equipment is calculated based on the iterative method combined with the cascading failure algorithm.

[0050] Example 5 If the module / unit integrated in the electronic device 100 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory and read-only memory (ROM, Read-Only Memory).

[0051] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0052] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0053] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0054] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0055] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for jointly monitoring vibration and temperature of power generation equipment, characterized in that: The following steps are involved: Continuously acquiring real-time status information of power generation equipment, and performing group preprocessing on the real-time status information of power generation equipment based on a preset period to obtain at least one group of period sequence information; Pre-construct a signal mutation model, input the periodic sequence information into the pre-constructed signal mutation model, the signal mutation model performs anomaly detection on the periodic sequence information, and outputs an abnormal point queue; Based on deep learning combined with particle filtering algorithm and outlier queue, a joint analysis model is constructed; The outlier queue is input into the joint analysis model. The joint analysis model conducts joint analysis on the outlier queue, outputs the risk value corresponding to the outlier, compares the risk value with the preset risk threshold, and retains the outliers corresponding to the risk values ​​greater than the preset risk threshold. The abnormal points whose risk values ​​exceed the preset risk threshold are loaded, and the comprehensive risk coefficient of the power generation equipment is calculated based on the iterative method combined with the cascading failure algorithm.

2. The method for joint monitoring of vibration and temperature of power generation equipment according to claim 1, characterized in that: The real-time status information of the power generation equipment includes equipment vibration signals, equipment temperature signals, equipment speed signals and equipment environment signals.

3. The method for joint monitoring of vibration and temperature of power generation equipment according to claim 1, characterized in that: The specific method of pre-building a signal mutation model is as follows: The Stacking model is used as the initial model of the signal mutation model. The initial model includes an input layer, a base learner, a meta learner, and an output layer. The Isolation Forest algorithm is introduced to improve the base learner in the initial model and freeze the decision tree of the base learner; The LSTM is used to replace the gradient descent optimization method of the meta-learner in the initial model, and the meta-learner is improved and optimized to obtain an improved initial model; The periodic sequence information is divided into a mutation training sample set and a mutation test sample set. Multiple base learners are trained based on the K-fold cross-validation method. The mutation training sample set is divided into training data and test data. Both the training data and the test data contain high-dimensional input features and signal labels. Each group of base learners generates corresponding base prediction results. Obtain at least one set of base prediction results, integrate the base prediction results into a new low-dimensional output feature and signal label composite queue, train a meta-learner using the low-dimensional output feature and signal label composite queue, and generate a meta-prediction result; Use low-dimensional output features for forward propagation and back propagation, calculate the gradient of the LSTM loss function, and use the LSTM model to update the network parameters of the meta-learner until the total loss function of the initial model reaches the preset convergence condition; Obtain a mutation test sample set, and verify whether the meta-prediction result meets the preset model accuracy threshold based on the mutation test sample set; If the preset model accuracy threshold is met, the signal mutation model after training is output.

4. The method for joint monitoring of vibration and temperature of power generation equipment according to claim 3, characterized in that: Based on the K-fold cross-validation method, multiple base learners are trained. When the mutation training sample set is divided into training data and test data, the mutation training sample set is evenly divided into K parts, of which K-1 parts are training data and 1 part is test data.

5. The method for joint monitoring of vibration and temperature of power generation equipment according to claim 3, characterized in that: The periodic sequence information is input into the pre-built signal mutation model. The signal mutation model performs anomaly detection on the periodic sequence information. The specific method of outputting the abnormal point queue is as follows: Obtain periodic sequence information, and the input layer of the signal mutation model performs dimension reduction processing on the periodic sequence information to obtain dimension-reduced sequence data; Load the reduced-dimensionality sequence data, and the base learner calculates the abnormal clustering degree corresponding to the reduced-dimensionality sequence data based on the Isolation Forest algorithm; Obtain the abnormal clustering degree corresponding to the dimension-reduced sequence data, and the meta-learner corrects the abnormal clustering degree based on the fuzzy c-means clustering algorithm to obtain the clustering correction value; Perform weighted summation on the cluster correction value and the abnormal clustering degree to obtain the weighted abnormal value; Determine whether the weighted outlier exceeds the preset outlier threshold. If so, output the outlier point corresponding to the reduced dimensionality sequence data. If the abnormal threshold is not exceeded, the dimension-reduced sequence data is normal; At least one group of outliers is integrated to obtain an outlier queue.

6. The method for joint monitoring of vibration and temperature of power generation equipment according to claim 1, characterized in that: The specific method of building a joint analysis model based on deep learning combined with particle filtering algorithm and outlier queue is as follows: Obtain an abnormal point queue, divide the abnormal point queue into an abnormal training sample set and an abnormal verification sample set, and initialize the abnormal index of the abnormal verification sample set; The initial model of the joint analysis model is constructed based on deep learning combined with particle filtering algorithm, wherein the initial model includes a probability distributor, a state transition module and an observer; An optimization signal algorithm is introduced into the state transfer module, and the optimization signal algorithm optimizes the parameters of the state transfer module; Load the abnormal training sample set, and use the probability distributor to extract the characteristic factors of the abnormal training sample set to obtain a characteristic factor extraction set; The state transfer module decomposes the characteristic factor extraction set to obtain a characteristic decomposition set, and determines the state transfer amount of the characteristic decomposition set and the optimal lag time corresponding to the state transfer amount; The state transition amount and the optimal delay corresponding to the state transition amount are loaded, and the observer calculates the forget gate corresponding to the state transition amount; The optimal lag time is updated based on the forget gate corresponding to the state transfer amount, and the risk value corresponding to the abnormal point is calculated by combining the optimal lag time and the state transfer amount; Continuously calculate the initial model loss function value of the joint analysis model, update the Bayesian posterior estimate, and iteratively optimize the joint analysis model; Obtain an abnormal verification sample set, evaluate the performance of the joint analysis model based on the abnormal verification sample set, calculate the group stability index between the predicted risk value and the actual risk value, and output the joint analysis model if the group stability index is less than the preset threshold.

7. The method for joint monitoring of vibration and temperature of power generation equipment according to claim 1, characterized in that: The outlier queue is input into the joint analysis model. The joint analysis model conducts joint analysis on the outlier queue and outputs the risk value corresponding to the outlier as follows: Load the outlier queue, perform feature extraction on the data vectors in the outlier queue, and obtain a feature factor extraction set; Determine an optimal lag corresponding to the feature factor extraction set, and update the optimal lag based on a forget gate corresponding to the state transfer amount; Calculate the risk value corresponding to the abnormal point by combining the optimal lag time and state transfer amount; Output the risk value corresponding to the abnormal point.

8. The power generation equipment vibration and temperature joint monitoring system is characterized by: include: An information preprocessing module, used to continuously obtain real-time status information of power generation equipment, and perform group preprocessing on the real-time status information of power generation equipment based on a preset period to obtain at least one group of period sequence information; An abnormal point queue output module is used to pre-build a signal mutation model, input the periodic sequence information into the pre-built signal mutation model, and the signal mutation model performs abnormality detection on the periodic sequence information and outputs an abnormal point queue; Joint analysis model building module, which is used to build a joint analysis model based on deep learning combined with particle filtering algorithm and outlier queue; The risk value screening module is used to input the outlier queue into the joint analysis model. The joint analysis model conducts joint analysis on the outlier queue, outputs the risk value corresponding to the outlier, compares the risk value with the preset risk threshold, and retains the outliers corresponding to the risk values ​​greater than the preset risk threshold; The comprehensive risk coefficient acquisition module is used to load abnormal points whose risk values ​​exceed the preset risk threshold, and calculate the comprehensive risk coefficient of the power generation equipment based on the iterative method combined with the cascading failure algorithm.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for jointly monitoring vibration and temperature of power generation equipment according to any one of claims 1 to 7 are implemented.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for jointly monitoring vibration and temperature of power generation equipment according to any one of claims 1 to 7 are implemented.

Images