CN110763997A - Early fault early warning method for synchronous motor stator - Google Patents
Early fault early warning method for synchronous motor stator Download PDFInfo
- Publication number
- CN110763997A CN110763997A CN201911063909.0A CN201911063909A CN110763997A CN 110763997 A CN110763997 A CN 110763997A CN 201911063909 A CN201911063909 A CN 201911063909A CN 110763997 A CN110763997 A CN 110763997A
- Authority
- CN
- China
- Prior art keywords
- network model
- stator
- fault
- deep
- deep belief
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/34—Testing dynamo-electric machines
- G01R31/343—Testing dynamo-electric machines in operation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses a synchronous motor stator early fault early warning method. The method comprises the following steps: step 1: constructing a Deep Belief Network (DBN) model; step 2: training a deep confidence network model; and step 3: inputting stator side data acquired by an SCADA system during actual operation of a motor into a trained deep confidence network model after normalization processing, outputting a change trend graph of a reconstruction error by a network, judging whether the change trend graph of the reconstruction error is above a set threshold of the reconstruction error, if so, judging that the stator side has a fault, and turning to the step 4; if not, the stator side runs normally, and then the step 5 is carried out; and 4, step 4: subtracting the reconstructed value and the measured value of the state parameter to obtain a residual value of the state parameter, and judging whether the residual value exceeds a set threshold value; and 5: and (6) ending. The early warning method for the stator fault of the synchronous motor provided by the invention realizes a diagnosis method for the stator fault of the synchronous motor and can warn the stator early fault.
Description
Technical Field
The invention relates to the technical field of motors, in particular to a synchronous motor stator early fault early warning method.
Background
With the development of economic society and the rapid increase of electricity consumption demand, the synchronous motor makes great progress in the aspects of cooling mode, material quality and manufacturing process, but the large-capacity motor has a complex structure, high price and wide fault influence range. The stator is one of the key parts of the synchronous motor, and is influenced by environmental factors such as high temperature, humidity and dust, operation factors such as frequent starting, speed regulation and braking, poor heat dissipation, corrosion aging, stress deformation, oxidation and other material factors, so that the stator is easy to break down.
In synchronous machines, common stator faults are: the method comprises the following steps of inter-turn short circuit of the stator winding, inter-phase short circuit of the stator winding, open circuit of the stator winding, blockage of a cooling water path of a stator bar, local overheating of a stator iron core, blockage of a cooling air path and the like. If the fault is not found in the early stage and proper measures are taken for processing, the fault development is accelerated, the fault range is expanded, and the safe and stable operation of the motor is influenced. Therefore, the method has important significance for early warning of stator faults.
The invention provides a fault prediction technology based on a deep belief network, which is applied to the fault prediction technology by combining an artificial intelligence method with big data analysis and solving the problems that the interaction of each fault on a system is difficult to analyze under the condition of a complex fault, the complex characteristics of equipment data are difficult to analyze, the fault is difficult to accurately predict in real time and the like.
Disclosure of Invention
The invention provides a synchronous motor stator early fault early warning method, which realizes a synchronous motor stator fault diagnosis method and can carry out early warning on stator early faults.
In order to achieve the purpose, the invention provides the following scheme:
an early fault early warning method for a synchronous motor stator comprises the following steps:
step 1: determining design parameters of a deep confidence network, and constructing a deep confidence network model;
step 2: training a deep confidence network model:
(1a) collecting stator side data of an SCADA system under the normal operation state of a synchronous motor, dividing the stator side data collected by the SCADA system into a training sample set and a test sample set, and carrying out normalization processing on the obtained training sample set and the test sample set;
(1b) carrying out unsupervised learning on the normalized training sample set, updating network parameters through a contrast divergence algorithm, and training layer by layer to obtain a pre-trained deep belief network model by taking hidden layer node data as the input of a next-stage limited Boltzmann machine after the training is finished by a first limited Boltzmann machine and repeating the steps;
(1c) calling a stator side label sample to perform layer-by-layer parameter fine adjustment from top to bottom through a BP network at the top layer of the pre-trained deep belief network model to obtain a trained deep belief network model;
(1d) and inputting the normalized test sample set into the trained deep confidence network model, outputting the reconstruction error of the deep confidence network model, and outputting a threshold value according to the self-adaptive threshold value principle.
And step 3: inputting stator side data acquired by an SCADA system during actual operation of a motor into a trained deep confidence network model after normalization processing, outputting a change trend graph of a reconstruction error by a network, judging whether the change trend graph of the reconstruction error is above a set threshold of the reconstruction error, if so, judging that the stator side has a fault, needing to give an alarm, and turning to the step 4; if not, the stator side runs normally, and then the step 5 is carried out;
and 4, step 4: subtracting the reconstructed value and the measured value of each state parameter to obtain a residual value of the state parameter, searching a physical quantity of which the residual value exceeds a set threshold value, and analyzing and obtaining a stator fault reason and reporting the stator fault reason by combining the type of the physical quantity;
and 5: and (6) ending.
Optionally, the step 1: determining design parameters of a deep confidence network, and constructing a deep confidence network model, which specifically comprises the following steps:
and taking the learning rate of the deep belief network model as a problem parameter, taking the energy function of the deep belief network model as a target function, performing cyclic iteration on the learning rate of the deep belief network model by adopting a gradient algorithm, finding out the optimal learning rate of the deep belief network model and the network parameter under the condition of the minimum energy value, and taking the optimal learning rate of the deep belief network model and the network parameter under the condition of the minimum energy value as an updating parameter of the deep belief network module.
Optionally, the performing loop iteration on the learning rate of the deep belief network model by using the learning rate of the deep belief network model as a problem parameter and using the energy function of the deep belief network model as an objective function and adopting a gradient algorithm to find out the network parameter of the deep belief network model under the condition that the optimal learning rate and the energy value of the deep belief network model are minimum, and using the network parameter of the deep belief network model under the condition that the optimal learning rate and the energy value of the deep belief network model are minimum as an update parameter of the deep belief network module specifically includes:
according to energy functionObtaining the joint probability distribution of the visible layer nodes and the hidden layer nodes:
performing dimensionality reduction treatment on the joint probability distribution of the visible layer nodes and the hidden layer nodes to obtain partial edge distribution and conditional probability distribution:
the highest probability of occurrence of the state with the lowest energy is obtained according to the statistical conclusion and is obtained by combining probability distribution:
selecting a sigmoid function as an activation function of the formulas (6) and (7) to obtain an update parameter of the deep belief network module:
wherein γ ═ wij,vi,hjDescribing the energy sum of the unit nodes among the layers of the restricted Boltzmann machine by an energy function, η is an initial learning rate, v represents a visible layer, h represents a hidden layer, w represents a visible layerijRepresents the connection weight, v, of the visible layer element i to the hidden layer element jiDenotes a visible layer unit, hjIndicating hidden layer elements, Δ wijRepresenting the weight update criterion, Δ biDenotes the visible layer cell bias, Δ cjIndicating the hidden layer cell bias.
The (1d) inputting the normalized sample set into the trained deep confidence network model, outputting a reconstruction error of the deep confidence network model, and outputting a threshold value according to a self-adaptive principle, specifically comprising:
selecting a small section of data from the initial part of the reconstructed error change trend graph as an initial data frame;
calculating the frame selection data according to an exponential weighted moving average threshold setting method to obtain a fixed threshold which is used as a reconstruction error threshold at the last moment of the data frame;
moving the data frame by frame from the initial position, repeating the above calculation process to obtain the reconstruction error threshold value at any moment, and connecting to form a self-adaptive threshold value graph fitting the change trend of the reconstruction error;
since the frame data range affects the adaptive capacity of the threshold, in practical applications, an appropriate value should be set according to specific situations, and in view of the fact that the initial reconstruction error changes smoothly and fluctuates little, the set threshold at the last time of the initial data frame is also used as the threshold at the previous time.
Compared with the prior art, the technology has the following beneficial effects:
according to the early-stage fault early-warning method for the stator of the synchronous motor, provided by the invention, the problems that the interaction of each fault on a system is difficult to analyze under the condition that a compound fault is difficult to analyze, the complex characteristics of equipment data are difficult to analyze, the fault is difficult to accurately predict in real time and the like exist in the conventional fault prediction method, and the fault prediction technology based on deep learning is provided by applying an artificial intelligence method to the fault prediction technology in combination with big data analysis. Compared with the existing stator fault diagnosis method, the method has the following advantages:
(1) the physical quantity related to the stator is fully considered, and the condition that the predicted value is inaccurate due to inaccurate measurement of a certain variable or interference of a signal is avoided. (2) Although the DBN network still does not fall off the scope of the multi-layer neural network, compared with the traditional decision neural network, the DBN network has the functions of discrimination and generation, can estimate the prior probability and the posterior probability by using the joint probability distribution of input sample data and a label, and is a breakthrough that the traditional model can only estimate the posterior probability. (3) The invention can find the fault before the fault is developed to a serious fault, and plays a role in fault early warning. (4) The invention utilizes SCADA data to predict the fault, does not need to add an additional measuring device in the synchronous motor, has the advantages of simplicity, convenience and feasibility, and provides reliable basis for the operation and maintenance of the unit. (5) Compared with the traditional fixed threshold early warning, the method adopts the self-adaptive threshold early warning, advances the early warning time by 200-2000s, has the characteristic of local compactness, can improve the sensitivity of fault recognition and reduce the occurrence of false alarm.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of an early warning method for stator faults of a synchronous motor according to an embodiment of the present invention;
FIG. 2 is a diagram of an RBM structure model according to an embodiment of the present invention;
FIG. 3 illustrates the encoding and decoding processes of a DBN according to an embodiment of the invention;
FIG. 4 is a diagram of a DBN structure according to an embodiment of the invention;
FIG. 5 is a schematic diagram of a DBN failure prediction model according to an embodiment of the invention;
FIG. 6 is a diagram illustrating the trend of Re variation of a training set under a normal operation condition of a stator according to an embodiment of the present invention;
FIG. 7 is a diagram illustrating the trend of Re variation of a test set under normal operation of a stator according to an embodiment of the present invention;
fig. 8 is a diagram of setting the threshold value of the stator Re based on the control principle in the fault state according to the embodiment of the present invention;
fig. 9 is a diagram illustrating threshold setting of the stator Re based on the adaptive principle in the normal state according to the embodiment of the present invention;
fig. 10 is a diagram of threshold setting of the stator Re based on the adaptive principle in the fault state according to the embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention provides a synchronous motor stator early fault early warning method, which realizes a synchronous motor stator fault diagnosis method and can carry out early warning on stator early faults.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Fig. 1 is a flowchart of an early-stage fault early-warning method for a stator of a synchronous motor according to an embodiment of the present invention, and as shown in fig. 1, the early-stage fault early-warning method for a stator of a synchronous motor based on a deep belief network includes the following steps:
step 1: defining a basic theory of the deep belief network and design parameters influencing the classification effect of the deep belief network, and constructing a deep belief network model;
step 2: training a deep confidence network model:
(1a) collecting stator side data collected by an SCADA system under the normal operation state of a synchronous motor, dividing the stator side data collected by the SCADA system into a training sample set and a test sample set, and carrying out normalization processing on the obtained training sample set and the test sample set;
(1b) carrying out unsupervised learning on the normalized training sample set, updating network parameters through a contrast divergence algorithm, and training layer by layer to obtain a pre-trained deep belief network model by taking hidden layer node data as the input of a next-stage limited Boltzmann machine after the training is finished by a first limited Boltzmann machine and repeating the steps;
(1c) calling a stator side label sample to perform layer-by-layer parameter fine adjustment from top to bottom through a BP network at the top layer of the pre-trained deep belief network model to obtain a trained deep belief network model;
(1d) inputting the normalized test sample set into a trained deep confidence network model, outputting a reconstruction error of the deep confidence network model, and outputting a threshold value according to a self-adaptive threshold value principle;
and step 3: inputting stator side data acquired by an SCADA system during actual operation of a motor into a trained deep confidence network model after normalization processing, outputting a change trend graph of a reconstruction error by a network, judging whether the change trend graph of the reconstruction error is above a set threshold of the reconstruction error, if so, judging that the stator side has a fault, needing to give an alarm, and turning to the step 4; if not, the stator side runs normally, and then the step 5 is carried out;
and 4, step 4: subtracting the reconstructed value and the measured value of each state parameter to obtain a residual value of the state parameter, searching a physical quantity of which the residual value exceeds a set threshold value, and analyzing and obtaining a stator fault reason and reporting the stator fault reason by combining the type of the physical quantity;
and 5: and (6) ending.
Fig. 2 is a model diagram of an RBM structure according to an embodiment of the present invention, and as shown in fig. 2, a Deep Belief Network (DBN) is widely applied to solve a processing problem of a nonlinear relationship of high-dimensional data as one of mainstream models in a Deep learning research field. The whole network is formed by stacking limiting Boltzmann machines serving as basic units from bottom to top, and can be regarded as a tool for extracting the characteristics and the association rules contained in sample data.
A Restricted Boltzmann Machine (RBM) is a random neural network model with a double-layer structure, symmetrical connection and no self-feedback, and comprises m visual nodes which jointly form a visual layer and use a vector vi＝(v1,v2,…,vm) Representing n hidden nodes together forming a hidden layer, using a vector hj＝(h1,h2,…,hn) And (4) showing. The nodes have independence, the nodes of the visible layer are independent from each other, the states of the nodes are only related to the corresponding n hidden layer nodes, the hidden nodes are only related to the corresponding m visible layer nodes, and the structural relation that no connection exists in the layer and the layers are fully connected is met.
The visible layer is used as an input port of data, the hidden layer is used as a feature extractor, each layer of nodes has corresponding offset, and the offset of the visible node is marked as a vector bi＝(b1,b2,…,bm) The offset of an implicit node is denoted as vector cj＝(c1,c2,…,cn) Interlayer connectionThe weighting matrix wijTogether, these parameters were all treated to obey a random minimum of the gaussian distribution prior to training.
The energy function that defines the RBM is:
wherein γ ═ wij,vi,hj}. The function describes the energy sum of RBM interlayer unit nodes, and a gradient method is used for iterative solution;
based on the energy function, the joint probability distribution of the visible layer nodes and the hidden layer nodes is defined as follows:
the edge distribution of a particular variable can be solved by the conditional probability distribution of other variables, so that the edge distribution and the conditional probability distribution can be solved by dimension reduction processing by using the joint probability distribution:
the conclusion in statistical mechanics indicates that the state with the lowest energy has the highest probability of occurring, and this is the target of the actual solution. From the joint probability distribution:
the activation function applied by the two formulas is a sigmoid function, the former formula embodies a feedforward coding process of mapping input to output, and the RBM node realizes the function of an automatic encoder; the latter expression represents the decoding process of back propagation, and the output is the input reconstruction value. Fig. 3 illustrates an encoding and decoding process of a DBN according to an embodiment of the present invention.
The finally obtained RBM parameter updating formula is as follows:
η is an initial learning rate, generally set to 0.1, a gradient is expressed by making a difference between an input and a network reconstruction value, and loop iteration is carried out to lead the difference to tend to be minimum, thereby realizing the update of RBM network parameters gamma and achieving the purpose of leading the distribution of input samples under the RBM representation to approach the actual distribution of synchronous motor stator physical quantity data as much as possible.
Wherein γ ═ wij,vi,hjDescribing the energy sum of the unit nodes among the layers of the restricted Boltzmann machine by an energy function, η is an initial learning rate, v represents a visible layer, h represents a hidden layer, w represents a visible layerijRepresents the connection weight of the visible layer unit i and the hidden layer unit j, v represents the visible layer, h represents the hidden layer, wijRepresents the connection weight, v, of the visible layer element i to the hidden layer element jiDenotes a visible layer unit, hjIndicating hidden layer elements, Δ wijRepresenting the weight update criterion, Δ biIndicating the visible layer cell bias and,Δcjindicating the hidden layer cell bias.
Reconstruction error
When the motor stably runs in a normal state, the corresponding reconstruction error stably changes within an allowable range. When the synchronous motor has a stator side fault, the reconstruction error can have obvious trend change, so that the fault on the stator side can be judged, and the fault occurrence and development can be predicted according to the trend change condition. Therefore, a Reconstruction error (Re) can be defined as an evaluation index, and the change trend of the Reconstruction error (Re) reflects the running state of the stator of the synchronous motor, and the formula is as follows:
Ret＝||xt-Xt||2
in the formula, xtReconstructing a value, X, for an input sample at time ttInput sample data at time t.
Principle of threshold setting
By setting a proper threshold value, and comparing the reconstruction error Re with the threshold value, the operating condition of the equipment is judged according to the variation trend and the degree of Re, and the purpose of fault prediction is achieved. The threshold setting is performed in two steps as follows:
fixed threshold setting based on control principle
An exponential Weighted Moving-Average (EWMA) method is commonly used for statistical data processing, and fully considers the information of all observed values before in the form of setting a weighting coefficient to reflect the recent change trend of the target quantity. And tracking and monitoring the change trend of Re by adopting a control chart based on an EWMA principle, and dividing normal, early warning and warning intervals by setting a control line.
The expression for the EWMA control point value is as follows:
vt＝βRet+(1-β)vt-1
in the formula RetRepresenting the reconstruction error at time t, coefficient β represents the weight coefficient of the EWMA control chart to the historical data, β e (0, 1)]Set β to 0.9. v0Take the mean of the first short time. Finding vtThe standard deviation σ of (a) is as follows:
in the formula, k is the standard deviation of the reconstruction error, n represents the sampling interval, and in conclusion, the threshold function of the EWMA control chart is set as follows:
v＝uvt+zσ
in the formula uvtIs v istConsidering that a certain margin and the sensitivity of a model need to be reserved, the z value can be 4 and 6, the threshold value when the z value is 4 is determined as a warning line for reminding the preliminary occurrence of the stator fault, and the threshold value when the z value is 6 is used as a warning limit for the development of the stator fault to be serious.
Dynamic threshold setting based on adaptive principle
The self-adaptive threshold setting method for fitting the change trend of Re data is adopted, and the specific steps are as follows:
selecting a small section of data from the initial part of the reconstructed error change trend graph as an initial data frame;
calculating the frame selection data according to an EWMA threshold setting method introduced in the previous section to obtain a fixed threshold, and taking the fixed threshold as a Re threshold at the last moment of the data frame;
the data frame is moved from the initial position frame by frame, the calculation process is repeated, the Re threshold value at any time can be obtained, and the Re threshold value are connected to form an adaptive threshold value graph fitting the variation trend of the Re.
Since the data range selected by the frame affects the adaptive capability of the threshold, an appropriate value should be set according to specific situations in practical applications. Since the initial Re portion varies smoothly and the fluctuation is small, the set threshold value at the last time of the initial data frame is also taken as the threshold value at the previous time.
The normal running state and the fault state of the stator can be distinguished according to the change trend graph of the reconstruction error Re and a set threshold, and the synchronous motor stator fault prediction step flow based on the DBN model is as follows:
selecting the SCADA data on the stator side of the synchronous motor in a normal operation state as a training data set, and carrying out normalization processing on the SCADA data in order to eliminate numerical value difference caused by different types and keep the integral structure of the data of the same type unchanged and help the gradient descent algorithm to quickly and accurately converge:
in the formula Tmax，TminRespectively representing the maximum value and the minimum value of the same type data of the whole input sample, and T is the normalization result of the data T.
Carrying out unsupervised learning on a training data set, updating network parameters through a divergence algorithm, after the first RBM is trained, using hidden layer node data as the input of the next-level RBM, and so on, training layer by layer to obtain a complete DBN network model, and finally calling a stator side label sample to carry out layer by layer parameter fine tuning from top to bottom through a BP network at the top layer of the DBN network.
And inputting SCADA data of normal state data of the stator side for testing, reconstructing an input vector by using the trained DBN network to obtain a variation trend graph of a reconstruction error, and verifying the fitting effect of the model.
And inputting stator side data acquired by SCADA (supervisory control and data acquisition) during actual operation of the motor into a trained DBN (database-based network) model, and performing data reconstruction to obtain a change trend graph of a reconstruction error. If the variation trend exceeds the threshold and is kept above the threshold, the fault on the stator side can be judged and an abnormal condition can be alarmed.
If an alarm signal exists, the fault reason can be preliminarily judged according to a residual trend chart of each state variable, wherein the residual refers to the difference between a reconstructed value and an actual value, namely xt-Xt. If the reconstructed residual error of a certain physical quantity exceeds the corresponding threshold value, the fault reason can be further judged.
The implementation of the present invention is illustrated by a 200MW water-hydrogen turbine generator of a thermal power plant. Firstly, SCADA data on a stator side in a certain period of normal operation of a unit are selected to train a DBN (digital-to-analog converter) network, so that a characteristic rule under the normal operation state of the DBN network is mined, and a basis is provided for stator fault prediction of a synchronous motor. Specifically, SCADA state monitoring data on a stator side in a certain period of time when a synchronous motor set of a certain power plant normally operates are selected as total training samples, sampling intervals are 10s, the time duration is 5 hours and 33 minutes, and the total number of the SCADA state monitoring data is 2000 data samples. The SCADA parameter type of the stator side of the synchronous motor selects the three-phase current of the motor stator, the stator voltage, the temperature of a cold air area, the temperature of a hot air area, the temperature of a stator core, the temperature of water inlet, the temperature of water outlet of a stator coil, the interlayer temperature of a stator coil and the like as input variables. Table 1 shows the selected SCADA parameter types on the stator side of the synchronous motor, as shown in table 1:
TABLE 1 Water-Hydrogen turbogenerator stator side principal parameters
Fig. 4 and 5 are DBN models thereof.
When the synchronous motor stably runs in a normal state, the state monitoring data of the stator side SCADA system fluctuates in a reasonable range along with time, and the state monitoring data is maintained in a dynamic balance state. And extracting the related SCADA data of the stator side of the synchronous motor as an input sample of the training model, and learning the characteristic rule of the stator side data in a normal state in a mode of iteratively updating the model parameters to be optimal.
After the sample is input, if the network model has excellent performance and high accuracy, the difference between the reconstructed value and the input value is theoretically small. However, when a stator side of the synchronous motor breaks down, the dynamic balance relation among the SCADA data is broken, if the data in the abnormal state is reconstructed by using a normal DBN network, the reconstruction error can have obvious trend change, the stator side can be judged to break down, and the development of the fault can be predicted according to the trend change condition.
Fig. 6 is a graph of the Re variation trend of the training set in the normal operation state of the stator in the embodiment of the present invention, as shown in fig. 6, a fixed threshold is set according to the EWMA principle, and a simulation curve indicates that the Re variation trend dynamically and steadily varies within the early warning threshold range in the time range from 0s to t1, which indicates that the stator of the synchronous generator works well in this time period. However, at the time t1, Re reaches the boundary of the early warning threshold, and then exceeds the threshold limit range and is always in a fluctuation rising state, so that the stator can be judged to be in an abnormal condition, an early warning signal is sent out, and closer monitoring and trend analysis are performed on the stator. When the gradual fault exceeds the alarm threshold value when the gradual fault is developed to t2, and the variation trend of Re is unchanged and the gradient is accelerated, the stator fault can be judged, an alarm signal is sent, and meanwhile, shutdown inspection and maintenance are required. With adaptive threshold setting for the same fault, as shown in fig. 7, before the stator fault occurs, i.e., before time t3, the Re trend dynamically changes smoothly within the adaptive threshold range, and after time t3, the Re trend exceeds the set threshold and always assumes an ascending situation. Based on the method, a fault early warning signal can be sent out, so that the purpose of predicting the stator fault is achieved by detecting the occurrence of early faults. Comparing fig. 6 and 7, it can be seen that, compared with the early warning effect of the fixed threshold, the adaptive threshold advances 2070 seconds in the early warning time, and has the characteristic of local compactness, which can improve the sensitivity of identifying the fault and reduce the occurrence of false alarm.
The invention provides a synchronous motor stator early fault early warning method based on a deep belief network, which solves the technical problems that the prior art has defects in the aspects of fault monitoring timeliness, fault threshold determination, early fault early warning and the like. At present, in order to monitor the operation condition of power generation equipment in real time, a supervisory control and Data Acquisition (SCADA) system is provided for a power plant, so that various operation Data of a synchronous generator are recorded in detail, and a reliable basis is provided for monitoring and maintaining the state of a unit. Aiming at the defects of the conventional fault prediction method in the aspects of processing complex faults, analyzing fault characteristics, finding slight faults, predicting faults in real time and the like, a deep learning algorithm can be introduced, the fault characteristics are extracted by using equipment state monitoring data, and a fault early warning model is constructed to realize the online dynamic prediction of the stator faults. Meanwhile, the SCADA system reserves a large amount of information rich in mining values, such as historical alarms, historical faults and the like, is favorable for fully exerting the modeling capacity of deep learning on big data, is easy to construct a fault prediction model based on the deep learning, and provides auxiliary support for daily fault processing means. The invention provides a fault prediction technology based on deep learning, which is applied to the fault prediction technology by combining an artificial intelligence method with big data analysis and solves the problems that the interaction of each fault on a system is difficult to analyze under the condition of a complex fault, the complex characteristics of equipment data are difficult to analyze, the fault is difficult to accurately predict in real time and the like. Compared with the existing stator fault diagnosis method, the method has the following advantages:
(1) the physical quantity related to the stator is fully considered, and the condition that the predicted value is inaccurate due to inaccurate measurement of a certain variable or interference of a signal is avoided. (2) Although the DBN network still does not fall off the scope of the multi-layer neural network, compared with the traditional decision neural network, the DBN network has the functions of discrimination and generation, can estimate the prior probability and the posterior probability by using the joint probability distribution of input sample data and a label, and is a breakthrough that the traditional model can only estimate the posterior probability. (3) The invention can find the fault before the fault is developed to a serious fault, and plays a role in fault early warning. (4) The invention utilizes SCADA data to predict the fault, does not need to add an additional measuring device in the synchronous motor, has the advantages of simplicity, convenience and feasibility, and provides reliable basis for the operation and maintenance of the unit. (5) Compared with the traditional fixed threshold early warning, the method adopts the self-adaptive threshold early warning, advances the early warning time by 200-2000s, has the characteristic of local compactness, can improve the sensitivity of fault recognition and reduce the occurrence of false alarm. A synchronous motor stator winding fault diagnosis method based on a deep belief network utilizes the advantages of deep learning in the aspect of historical data non-visual relevance mining to complete model training through a large number of training samples, and achieves fault early warning and fault diagnosis in a mode of distinguishing difference of Re trend graphs in a fault state and a normal state by taking a reconstruction error Re as a fault prediction index. The invention provides a synchronous motor stator early fault early warning method based on a deep belief network, which is used for realizing a synchronous motor stator fault diagnosis method and can be used for early warning stator early faults.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (4)
1. The early warning method for the stator fault of the synchronous motor is characterized by comprising the following steps of:
step 1: determining design parameters of a deep confidence network, and constructing a deep confidence network model;
step 2: training a deep confidence network model:
(1a) collecting stator side data of an SCADA system under the normal operation state of a synchronous motor, dividing the stator side data collected by the SCADA system into a training sample set and a test sample set, and carrying out normalization processing on the obtained training sample set and the test sample set;
(1b) carrying out unsupervised learning on the normalized training sample set, updating network parameters through a divergence algorithm, taking hidden layer node data of a first Restricted Boltzmann Machine (RBM) as the input of the next Restricted Boltzmann Machine after the training is finished, and training layer by layer in the same way to obtain a pre-trained deep belief network model;
(1c) calling a stator side label sample to perform layer-by-layer parameter fine adjustment from top to bottom through a BP network at the top layer of the pre-trained deep belief network model to obtain a trained deep belief network model;
(1d) inputting the normalized test sample set into a trained deep confidence network model, outputting a reconstruction error of the deep confidence network model, and outputting a threshold value according to a self-adaptive threshold value principle;
and step 3: inputting stator side data acquired by an SCADA system during actual operation of a motor into a trained deep confidence network model after normalization processing, outputting a change trend graph of a reconstruction error by a network, judging whether the change trend graph of the reconstruction error is above a set threshold of the reconstruction error, if so, judging that the stator side has a fault, needing to give an alarm, and turning to the step 4; if not, the stator side runs normally, and then the step 5 is carried out;
and 4, step 4: subtracting the reconstructed value and the measured value of each state parameter to obtain a residual value of the state parameter, searching a physical quantity of which the residual value exceeds a set threshold value, and analyzing and obtaining a stator fault reason and reporting the stator fault reason by combining the type of the physical quantity;
and 5: and (6) ending.
2. The early warning method for the stator fault of the synchronous motor according to claim 1, wherein the step 1: determining design parameters of a deep confidence network, and constructing a deep confidence network model, which specifically comprises the following steps:
and taking the learning rate of the deep belief network model as a problem parameter, taking the energy function of the deep belief network model as a target function, performing cyclic iteration on the learning rate of the deep belief network model by adopting a gradient algorithm, finding out the optimal learning rate of the deep belief network model and the network parameter under the condition of the minimum energy value, and taking the optimal learning rate of the deep belief network model and the network parameter under the condition of the minimum energy value as an updating parameter of the deep belief network module.
3. The early warning method for the stator fault of the synchronous motor according to claim 2, wherein the learning rate of the deep belief network model is used as a problem parameter, the energy function of the deep belief network model is used as an objective function, a gradient algorithm is adopted to carry out cyclic iteration on the learning rate of the deep belief network model, a network parameter under the condition that the optimal learning rate and the energy value of the deep belief network model are minimum is found out, and a network parameter under the condition that the optimal learning rate and the energy value of the deep belief network model are minimum is used as an update parameter of the deep belief network module, and the method specifically comprises the following steps:
according to energy function
Obtaining the joint probability distribution of the visible layer nodes and the hidden layer nodes:
performing dimensionality reduction treatment on the joint probability distribution of the visible layer nodes and the hidden layer nodes to obtain partial edge distribution and conditional probability distribution:
the highest probability of occurrence of the state with the lowest energy is obtained according to the statistical conclusion and is obtained by combining probability distribution:
selecting a sigmoid function as an activation function of the formulas (6) and (7) to obtain an update parameter of the deep belief network module:
wherein γ ═ wij,vi,hjDescribing the energy sum of the unit nodes among the layers of the restricted Boltzmann machine by an energy function, η is an initial learning rate, v represents a visible layer, h represents a hidden layer, w represents a visible layerijRepresents the connection weight, v, of the visible layer element i to the hidden layer element jiDenotes a visible layer unit, hjDenotes a hidden layer unit, γ denotes γ ═ wij,vi,hj}，ΔwijRepresenting the weight update criterion, Δ biDenotes the visible layer cell bias, Δ cjIndicating the hidden layer cell bias.
4. The early warning method for the stator fault of the synchronous motor according to claim 1, wherein (1d) the normalized sample set is input into a trained deep confidence network model, a reconstruction error of the deep confidence network model is output, and a threshold value is output according to an adaptive principle, and specifically comprises:
selecting a small section of data from the initial part of the reconstructed error change trend graph as an initial data frame;
calculating the frame selection data according to an exponential weighted moving average threshold setting method to obtain a fixed threshold which is used as a reconstruction error threshold at the last moment of the data frame;
moving the data frame by frame from the initial position, repeating the above calculation process to obtain the reconstruction error threshold value at any moment, and connecting to form a self-adaptive threshold value graph fitting the change trend of the reconstruction error;
since the frame data range affects the adaptive capacity of the threshold, in practical applications, an appropriate value should be set according to specific situations, and in view of the fact that the initial reconstruction error changes smoothly and fluctuates little, the set threshold at the last time of the initial data frame is also used as the threshold at the previous time.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911063909.0A CN110763997A (en) | 2019-11-04 | 2019-11-04 | Early fault early warning method for synchronous motor stator |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911063909.0A CN110763997A (en) | 2019-11-04 | 2019-11-04 | Early fault early warning method for synchronous motor stator |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110763997A true CN110763997A (en) | 2020-02-07 |
Family
ID=69335465
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911063909.0A Pending CN110763997A (en) | 2019-11-04 | 2019-11-04 | Early fault early warning method for synchronous motor stator |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110763997A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111273635A (en) * | 2020-01-19 | 2020-06-12 | 浙江大学 | Unknown anomaly detection method for industrial control equipment |
| CN111722145A (en) * | 2020-06-12 | 2020-09-29 | 华北电力大学（保定） | Method for diagnosing slight fault of turn-to-turn short circuit of excitation winding of synchronous motor |
| CN112727704A (en) * | 2020-12-15 | 2021-04-30 | 北京天泽智云科技有限公司 | Method and system for monitoring corrosion of leading edge of blade |
| CN112766702A (en) * | 2021-01-13 | 2021-05-07 | 广东能源集团科学技术研究院有限公司 | Distributed power station fault analysis method and system based on deep belief network |
| CN113171059A (en) * | 2021-04-25 | 2021-07-27 | 首都医科大学附属北京天坛医院 | Postoperative END risk early warning of multi-modal monitoring information and related equipment |
| CN113191219A (en) * | 2021-04-15 | 2021-07-30 | 华能威宁风力发电有限公司 | Fan bearing fault self-adaptive identification method |
| CN114089221A (en) * | 2021-11-23 | 2022-02-25 | 北京市汉华环球科技发展有限责任公司 | Vibrating mirror stator winding loop open-circuit fault detection method and protection module |
| CN116845936A (en) * | 2023-09-01 | 2023-10-03 | 华夏天信智能物联股份有限公司 | Intelligent control method of flywheel energy storage device |
| CN117034020A (en) * | 2023-10-09 | 2023-11-10 | 贵州大学 | Unmanned aerial vehicle sensor zero sample fault detection method based on CVAE-GAN model |
| CN117407824A (en) * | 2023-12-14 | 2024-01-16 | 四川蜀能电科能源技术有限公司 | Health detection method, equipment and medium of power time synchronization device |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2000043882A1 (en) * | 1999-01-20 | 2000-07-27 | Fujitsu Limited | Database switching apparatus |
| CN105973594A (en) * | 2016-04-25 | 2016-09-28 | 西北工业大学 | Rolling bearing fault prediction method based on continuous deep belief network |
| CN106769048A (en) * | 2017-01-17 | 2017-05-31 | 苏州大学 | Self adaptation depth confidence network Method for Bearing Fault Diagnosis based on Nesterov momentum methods |
| CN107725283A (en) * | 2017-09-19 | 2018-02-23 | 江苏方天电力技术有限公司 | A kind of fan trouble detection method based on depth belief network model |
| CN108304927A (en) * | 2018-01-25 | 2018-07-20 | 清华大学 | Bearing fault modality diagnostic method and system based on deep learning |
| CN108732528A (en) * | 2018-05-28 | 2018-11-02 | 国网福建省电力有限公司电力科学研究院 | A kind of digitalized electrical energy meter method for diagnosing faults based on depth confidence network |
| CN109086817A (en) * | 2018-07-25 | 2018-12-25 | 西安工程大学 | A kind of Fault Diagnosis for HV Circuit Breakers method based on deepness belief network |
| CN110084148A (en) * | 2019-04-09 | 2019-08-02 | 东南大学 | A kind of Mechanical Failure of HV Circuit Breaker diagnostic method |
| CN110297178A (en) * | 2018-05-11 | 2019-10-01 | 宫文峰 | Diesel generating set fault diagnosis and detection device and method based on deep learning |
-
2019
- 2019-11-04 CN CN201911063909.0A patent/CN110763997A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2000043882A1 (en) * | 1999-01-20 | 2000-07-27 | Fujitsu Limited | Database switching apparatus |
| CN105973594A (en) * | 2016-04-25 | 2016-09-28 | 西北工业大学 | Rolling bearing fault prediction method based on continuous deep belief network |
| CN106769048A (en) * | 2017-01-17 | 2017-05-31 | 苏州大学 | Self adaptation depth confidence network Method for Bearing Fault Diagnosis based on Nesterov momentum methods |
| CN107725283A (en) * | 2017-09-19 | 2018-02-23 | 江苏方天电力技术有限公司 | A kind of fan trouble detection method based on depth belief network model |
| CN108304927A (en) * | 2018-01-25 | 2018-07-20 | 清华大学 | Bearing fault modality diagnostic method and system based on deep learning |
| CN110297178A (en) * | 2018-05-11 | 2019-10-01 | 宫文峰 | Diesel generating set fault diagnosis and detection device and method based on deep learning |
| CN108732528A (en) * | 2018-05-28 | 2018-11-02 | 国网福建省电力有限公司电力科学研究院 | A kind of digitalized electrical energy meter method for diagnosing faults based on depth confidence network |
| CN109086817A (en) * | 2018-07-25 | 2018-12-25 | 西安工程大学 | A kind of Fault Diagnosis for HV Circuit Breakers method based on deepness belief network |
| CN110084148A (en) * | 2019-04-09 | 2019-08-02 | 东南大学 | A kind of Mechanical Failure of HV Circuit Breaker diagnostic method |
Non-Patent Citations (5)
| Title |
|---|
| JIHONGYAN: "Rotor unbalance fault diagnosis using DBN based on multi-source heterogeneous information fusion", 《PROCEDIA MANUFACTURING》 * |
| 刘辉海: "基于深度学习方法的风电机组驱动链故障预测研究", 《中国优秀硕士学位论文全文数据库 工程科技II辑》 * |
| 杨立敏: "《大数据背景下数据挖掘机处理分析》", 31 January 2019 * |
| 赵洪山: "基于堆叠自编码网络的风电机组发电机状态监测与故障诊断", 《电力系统自动化》 * |
| 赵洪山: "基于性能改善深度置信网络的风电机组主轴承状态分析", 《电力自动化设备》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111273635B (en) * | 2020-01-19 | 2021-02-12 | 浙江大学 | Unknown anomaly detection method for industrial control equipment |
| CN111273635A (en) * | 2020-01-19 | 2020-06-12 | 浙江大学 | Unknown anomaly detection method for industrial control equipment |
| CN111722145A (en) * | 2020-06-12 | 2020-09-29 | 华北电力大学（保定） | Method for diagnosing slight fault of turn-to-turn short circuit of excitation winding of synchronous motor |
| CN112727704A (en) * | 2020-12-15 | 2021-04-30 | 北京天泽智云科技有限公司 | Method and system for monitoring corrosion of leading edge of blade |
| CN112727704B (en) * | 2020-12-15 | 2021-11-30 | 北京天泽智云科技有限公司 | Method and system for monitoring corrosion of leading edge of blade |
| CN112766702A (en) * | 2021-01-13 | 2021-05-07 | 广东能源集团科学技术研究院有限公司 | Distributed power station fault analysis method and system based on deep belief network |
| CN113191219B (en) * | 2021-04-15 | 2022-11-01 | 华能威宁风力发电有限公司 | Fan bearing fault self-adaptive identification method |
| CN113191219A (en) * | 2021-04-15 | 2021-07-30 | 华能威宁风力发电有限公司 | Fan bearing fault self-adaptive identification method |
| CN113171059A (en) * | 2021-04-25 | 2021-07-27 | 首都医科大学附属北京天坛医院 | Postoperative END risk early warning of multi-modal monitoring information and related equipment |
| CN113171059B (en) * | 2021-04-25 | 2022-03-01 | 首都医科大学附属北京天坛医院 | Postoperative END risk early warning of multi-modal monitoring information and related equipment |
| CN114089221A (en) * | 2021-11-23 | 2022-02-25 | 北京市汉华环球科技发展有限责任公司 | Vibrating mirror stator winding loop open-circuit fault detection method and protection module |
| CN114089221B (en) * | 2021-11-23 | 2024-03-15 | 北京市汉华环球科技发展有限责任公司 | Method for detecting open-circuit fault of vibrating mirror stator winding loop and protection module |
| CN116845936A (en) * | 2023-09-01 | 2023-10-03 | 华夏天信智能物联股份有限公司 | Intelligent control method of flywheel energy storage device |
| CN116845936B (en) * | 2023-09-01 | 2023-11-21 | 华夏天信智能物联股份有限公司 | Intelligent control method of flywheel energy storage device |
| CN117034020A (en) * | 2023-10-09 | 2023-11-10 | 贵州大学 | Unmanned aerial vehicle sensor zero sample fault detection method based on CVAE-GAN model |
| CN117034020B (en) * | 2023-10-09 | 2024-01-09 | 贵州大学 | Unmanned aerial vehicle sensor zero sample fault detection method based on CVAE-GAN model |
| CN117407824A (en) * | 2023-12-14 | 2024-01-16 | 四川蜀能电科能源技术有限公司 | Health detection method, equipment and medium of power time synchronization device |
| CN117407824B (en) * | 2023-12-14 | 2024-02-27 | 四川蜀能电科能源技术有限公司 | Health detection method, equipment and medium of power time synchronization device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110763997A (en) | Early fault early warning method for synchronous motor stator | |
| Arunthavanathan et al. | A deep learning model for process fault prognosis | |
| JP7242101B2 (en) | Engine surge failure prediction system and method based on fusion neural network model | |
| Soualhi et al. | Prognosis of bearing failures using hidden Markov models and the adaptive neuro-fuzzy inference system | |
| Liu et al. | Diagnosis and prognosis of degradation process via hidden semi-Markov model | |
| JP2020119605A (en) | Abnormality detection system, abnormality detection method, abnormality detection program, and method for generating learned model | |
| US11840998B2 (en) | Hydraulic turbine cavitation acoustic signal identification method based on big data machine learning | |
| CN111722145B (en) | Synchronous motor excitation winding turn-to-turn short circuit mild fault diagnosis method | |
| CN111178553A (en) | Industrial equipment health trend analysis method and system based on ARIMA and LSTM algorithms | |
| CN111337244B (en) | Method and device for monitoring and diagnosing faults of input shaft of fan gearbox | |
| CN112069727B (en) | Intelligent transient stability evaluation system and method with high reliability for power system | |
| CN110969185A (en) | Equipment abnormal state detection method based on data reconstruction | |
| Hu et al. | A hybrid model for predicting the degradation trend of hydropower units based on deep learning | |
| CN114912640A (en) | Method and system for detecting abnormal mode of generator set based on deep learning | |
| Cheng et al. | Predicting the remaining useful life of rolling element bearings using locally linear fusion regression | |
| CN113470694A (en) | Remote listening monitoring method, device and system for hydraulic turbine set | |
| Xu et al. | Anomaly detection with gru based bi-autoencoder for industrial multimode process | |
| Abdelli et al. | Machine learning based data driven diagnostic and prognostic approach for laser reliability enhancement | |
| Melendez et al. | Self-supervised Multi-stage Estimation of Remaining Useful Life for Electric Drive Units | |
| CN115423370A (en) | Relay protection equipment health state assessment method and device | |
| CN114298413A (en) | Hydroelectric generating set runout trend prediction method | |
| CN114819261A (en) | Fault prediction method and system based on real-time running state of equipment | |
| Zhang et al. | Distributed quality-related process monitoring framework using parallel DVIB-VAE-mRMR for large-scale processes | |
| CN117034157B (en) | Hydropower equipment fault identification method and system combining multimodal operation data | |
| Soualhi et al. | Remaining useful life estimation of turbofan engines using adaptive fault detection learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200207 |
|
| RJ01 | Rejection of invention patent application after publication |