CN111310907A - Microwave assembly fault diagnosis method, device and equipment - Google Patents
Microwave assembly fault diagnosis method, device and equipment Download PDFInfo
- Publication number
- CN111310907A CN111310907A CN201811517864.5A CN201811517864A CN111310907A CN 111310907 A CN111310907 A CN 111310907A CN 201811517864 A CN201811517864 A CN 201811517864A CN 111310907 A CN111310907 A CN 111310907A
- Authority
- CN
- China
- Prior art keywords
- fault
- layer
- neural network
- output
- network model
- 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
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
- G06N3/065—Analogue means
Abstract
The invention discloses a microwave assembly fault diagnosis method, a device and equipment, wherein the microwave assembly fault diagnosis method comprises the following steps: acquiring fault characteristic quantities of a microwave assembly, wherein the fault characteristic quantities comprise fault phenomenon characteristic quantities and fault reason characteristic quantities; establishing a back propagation BP neural network model according to the fault characteristic quantity; training the BP neural network model; and diagnosing the fault information to be diagnosed by using the trained BP neural network model. According to the embodiment of the invention, the fault characteristic quantity is extracted from the historical fault case information, so that the whole model has good robustness and fault tolerance, and the adaptability and stability of the model are improved; by utilizing the capability of high fitting nonlinearity of the BP neural network, the coupling relation between the fault information and the fault reason in the microwave assembly fault diagnosis problem is solved, so that the newly generated fault is quickly positioned to the cause of the newly generated fault, and the intelligent diagnosis of the microwave assembly fault facing the automatic test unit is realized.
Description
Technical Field
The invention relates to the field of fault diagnosis, in particular to a method, a device and equipment for fault diagnosis of a microwave assembly.
Background
The fault diagnosis technology is a technology for discovering the abnormal condition of equipment by monitoring the state parameters of the equipment and analyzing and diagnosing the fault reason after the abnormal condition is discovered, and aims to discover the potential fault of a product by means of all current innovative technologies so as to achieve the purpose of preventing quality accidents of the product in the bud. At present, the fault diagnosis technology has been developed into an independent interdisciplinary comprehensive information processing technology, and is a hot research direction in the control field.
In recent years, a BP (Back Propagation) neural network is a widely used inference model at present. The theoretical development of the method is mature, and a plurality of breakthrough progresses are achieved, thereby arousing high attention and great attention in the academic circles at home and abroad. BP neural networks have been used by technicians in various industries as an emerging modeling technique. In the aspect of fault diagnosis, the network model is widely applied. In a traditional test line, the failure problem of a tested microwave component and the reason causing the failure phenomenon are mostly judged manually by a tester according to a test result, and the results are low diagnosis efficiency and the circulation delay of the tested component.
Disclosure of Invention
In order to solve the technical problems, the invention provides a microwave assembly fault diagnosis method, a device and equipment, which solve the problems of low fault diagnosis efficiency and delayed circulation of a tested piece of a microwave assembly in the prior art.
According to an aspect of the present invention, there is provided a microwave component fault diagnosis method, including:
acquiring fault characteristic quantities of a microwave assembly, wherein the fault characteristic quantities comprise fault phenomenon characteristic quantities and fault reason characteristic quantities;
establishing a back propagation BP neural network model according to the fault characteristic quantity;
training the BP neural network model;
and diagnosing the fault information to be diagnosed by using the trained BP neural network model.
Optionally, the step of acquiring the fault characteristic quantity of the microwave assembly includes:
extracting fault characteristic quantity from an original signal of microwave assembly fault detection;
and converting the fault characteristic quantity into a vector form.
Optionally, the step of converting the fault characteristic quantity into a vector form includes:
according to the formula: f (X)i)＝ST(Xi)/SR(Xi) Calculating the ratio of the current value of the to-be-diagnosed state of the fault characteristic quantity to the preset value of the normal state;
wherein, XiFor the fault characteristic quantity, ST(Xi) Is XiOf the state to be diagnosed, SRIs XiA preset value of normal state of (1);
and determining the vector value of the fault characteristic quantity according to the ratio.
Optionally, the step of building a back propagation BP neural network model according to the fault feature quantity includes:
determining the number of input layer nodes and the number of output layer nodes of the BP neural network model according to the fault characteristic quantity;
and determining the number of hidden layer nodes of the BP neural network model according to the number of the input layer nodes and the number of the output layer nodes.
Optionally, the number of nodes in the input layer is the dimension of the fault phenomenon feature quantity, and the number of nodes in the output layer is the dimension of the fault cause feature quantity.
Optionally, the step of determining the number of hidden layer nodes of the BP neural network model according to the number of input layer nodes and the number of output layer nodes includes:
by the formula
Calculating the value range of the number of the hidden layer nodes:
k is the number of hidden layer nodes, n is the number of input layer nodes, l is the number of output layer nodes, and α is a constant of 1-10.
Optionally, the step of building a back propagation BP neural network model according to the fault feature quantity further includes:
an excitation function is determined.
Optionally, the step of training the BP neural network model includes:
setting initial weight and initial threshold of the BP neural network model;
performing forward propagation operation by using the initial weight, the initial threshold and the fault characteristic quantity to respectively obtain an output result of a hidden layer and an output result of an output layer;
calculating an error according to the output result;
and performing reverse feedback operation according to the error to obtain updated weight and threshold.
Optionally, the step of performing forward propagation operation by using the initial weight, the initial threshold, and the fault feature quantity to obtain an output result of the hidden layer and an output result of the output layer respectively includes:
by the formula:
calculating to obtain an output result of the hidden layer;
wherein, bjRepresents the output of the jth neuron of the hidden layer, wijAs weights of the input layer to the hidden layer, aiIs input to the input layer, thetajThe threshold value from the input layer to the hidden layer is N, and the number of the neurons in the layer before the current neuron is N;
by the formula:
calculating an output result of an output layer according to the output result of the hidden layer;
wherein, ctRepresenting the output of the t-th neuron of the output layer, vjtRepresenting the weight, r, from hidden layer to output layertIs the hidden layer to output layer threshold.
Optionally, the step of calculating an error according to the output result comprises:
by the formula:
calculating to obtain the error;
wherein E is an error, dtTo the desired output, ctThe output of the t-th neuron of the output layer is N, and the N is the number of neurons in the layer before the current neuron.
Optionally, performing inverse feedback operation according to the error, and obtaining the updated weight and the threshold includes:
by the formula
Calculating to obtain updated weight;
wherein, wijIs the weight of the input layer to the hidden layer, wij' updated input layer to hidden layer weight, η learning efficiency of the BP neural network model, bjOutput of the jth neuron of the hidden layer, aiIs the input of the input layer; v. ofjtIs the weight from hidden layer to output layer, vjt' updated weight from hidden layer to output layer, et＝dt-ct，dtTo the desired output, ctIs the output of the t-th neuron of the output layer;
by the formula
Calculating to obtain an updated threshold value;
wherein, thetajFor the threshold of the input layer to the hidden layer, θj' is the updated input layer to hidden layer threshold; r istFor a threshold from the hidden layer to the output layer, rt' is the updated hidden layer to output layer threshold.
Optionally, the step of training the BP neural network model further includes:
judging whether the iteration times of the BP neural network model reach preset times or not;
if the preset times are reached, ending the training of the BP neural network model, otherwise, repeating the training process of the BP neural network model.
Optionally, the step of diagnosing the fault information to be diagnosed by using the trained BP neural network model includes:
inputting the fault information to be diagnosed into the trained BP neural network model;
and carrying out fault diagnosis on the fault information to be diagnosed by using the trained BP neural network model.
According to another aspect of the present invention, there is provided a microwave component failure diagnosis apparatus including:
the microwave component fault detection device comprises a first acquisition module, a second acquisition module and a control module, wherein the first acquisition module is used for acquiring fault characteristic quantities of a microwave component, and the fault characteristic quantities comprise fault phenomenon characteristic quantities and fault reason characteristic quantities;
the model establishing module is used for establishing a back propagation BP neural network model according to the fault characteristic quantity;
the model training module is used for training the BP neural network model;
and the fault diagnosis module is used for diagnosing the fault information to be diagnosed by utilizing the trained BP neural network model.
According to still another aspect of the present invention, there is provided a microwave component fault diagnosis apparatus, including a processor, a memory, and a computer program stored on the memory and operable on the processor, wherein the computer program, when executed by the processor, implements the steps of the above-mentioned microwave component fault diagnosis method.
The embodiment of the invention has the beneficial effects that:
in the scheme, the historical fault case information is dispersedly stored in each neuron in the network, so that the network structure can be dynamically adjusted, the whole model has good robustness and fault tolerance, and the adaptability and stability of the model are improved to a certain extent; the nonlinear ability of the BP neural network can be highly fitted to approximate the coupling relation between the fault information and the fault reason in the fault diagnosis problem of the microwave assembly, so that the newly generated fault can be quickly positioned to the cause of the newly generated fault. And the intelligent diagnosis of the microwave assembly fault facing the automatic test unit is realized.
Drawings
FIG. 1 shows a flow chart of a microwave assembly fault diagnosis method of an embodiment of the invention;
FIG. 2 is a diagram illustrating an excitation function according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a BP neural network model according to an embodiment of the present invention;
FIG. 4 shows a flowchart of BP neural network model training according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a fault diagnosis process according to an embodiment of the present invention;
fig. 6 is a block diagram showing a configuration of a microwave module fault diagnosis apparatus according to an embodiment of the present invention.
Detailed Description
Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
As shown in fig. 1, an embodiment of the present invention provides a microwave assembly fault diagnosis method, including:
in the embodiment, the existing fault case data is used as a training set for model training, and identifiable fault information is extracted from the existing fault case data. In an original signal obtained by fault detection of the microwave assembly, a plurality of pieces of information capable of reflecting the fault state of the microwave assembly exist, fault characteristic quantities capable of reflecting the characteristics of the microwave assembly are extracted from the information, the fault characteristic quantities comprise fault phenomenon characteristic quantities and fault reason characteristic quantities, and the extracted fault characteristic quantities reflect regularity and sensitivity in the fault characteristic quantities, namely the fault independence and the fault regularity are better.
and building a BP neural network model according to the extracted fault characteristic quantity, and determining the number of input layer nodes, the number of hidden layer nodes, the number of output layer nodes, an excitation function and the like of the BP neural network model according to the fault characteristic quantity.
training the BP neural network model, inputting the extracted fault characteristic quantity as sample data of the BP neural network model into the network model for operation, and adjusting a numerical matrix, weight and threshold value through multiple iterative operations. And when the preset maximum iteration times are reached or the error of the output fault diagnosis result is reduced to be lower than the expected error, considering that a termination condition is met, and finishing the model training.
And step 14, diagnosing the fault information to be diagnosed by using the trained BP neural network model.
In the embodiment, after the training of the BP neural network model is finished, newly generated fault information to be diagnosed is input into the learned BP neural network model in a vector form, a fault reason corresponding to the fault information is obtained through calculation of the BP neural network model, and is output and displayed in a probability form, so that intelligent fault diagnosis of the microwave component of the automatic test unit is realized.
According to the scheme, historical fault case information is stored in each neuron in the network in a dispersed manner, so that the network structure can be dynamically adjusted, the whole model has good robustness and fault tolerance, and the adaptability and stability of the model are improved to a certain extent; the nonlinear ability of the BP neural network can be highly fitted to approximate the coupling relation between the fault information and the fault reason in the fault diagnosis problem of the microwave assembly, so that the newly generated fault can be quickly positioned to the cause of the newly generated fault. And the intelligent diagnosis of the microwave assembly fault facing the automatic test unit is realized.
Specifically, the step 11 includes: extracting fault characteristic quantity from an original signal of microwave assembly fault detection; and converting the fault characteristic quantity into a vector form.
After the fault characteristic quantity is extracted from the fault information, the fault characteristic quantity needs to be converted into a multidimensional vector, namely, the detected original data is converted into an input vector which can be identified by a network model. The step of converting the fault characteristic quantity into a vector form comprises the following steps:
according to the formula: f (X)i)＝ST(Xi)/SR(Xi) Calculating the ratio of the current value of the to-be-diagnosed state of the fault characteristic quantity to the preset value of the normal state; and determining the vector value of the fault characteristic quantity according to the ratio.
Wherein, XiFor the fault characteristic quantity, ST(Xi) Is XiOf the state to be diagnosed, SRIs XiA preset value of normal state of (1);
in the embodiment, the current value of the state to be diagnosed of the fault characteristic quantity is compared with the preset normal value, and the ratio is used as the basis for judging whether the fault phenomenon occurs. Optionally, a ratio F (X) of a current value of the to-be-diagnosed state of the fault characteristic quantity to a preset value of the normal statei) Comparing with a preset threshold value, if F (X)i) If the value of the fault characteristic quantity in the corresponding multidimensional vector is larger than the threshold value, the value of the fault characteristic quantity in the corresponding multidimensional vector is '1', the fault is shown to occur, otherwise, the value of the fault characteristic quantity in the corresponding multidimensional vector is '0', the fault is not shown to occur.
Specifically, the step 12 includes:
determining the number of input layer nodes and the number of output layer nodes of the BP neural network model according to the fault characteristic quantity; the number of the nodes of the input layer is the dimension of the fault phenomenon characteristic quantity, and the number of the nodes of the output layer is the dimension of the fault reason characteristic quantity.
And determining the number of hidden layer nodes of the BP neural network model according to the number of the input layer nodes and the number of the output layer nodes. Increasing the number of hidden layer units can reduce network errors and improve accuracy, but can also complicate the network, thereby increasing the training time of the network and causing an "overfitting" phenomenon. This embodiment is illustrated with a layer 3 network as an example, i.e. there is only one hidden layer. A lower error is then obtained by adjusting the number of hidden layer element nodes.
Further, the step of determining the number of hidden layer nodes of the BP neural network model according to the number of input layer nodes and the number of output layer nodes comprises:
by the formula
Calculating the value range of the number of the hidden layer nodes:
k is the number of hidden layer nodes, n is the number of input layer nodes, l is the number of output layer nodes, and α is a constant of 1-10.
In this embodiment, the number of hidden layer nodes has a direct relationship with the requirement for solving the problem and the number of input/output units. If the number of the hidden layer nodes is too small, enough connection weight combination numbers cannot be generated to meet the learning requirement of a plurality of samples; if the number of the hidden layer unit nodes is too large, the generalization capability of the network model after learning becomes poor. Calculating to obtain the value range of the number of hidden layer nodes by the method, then increasing the number of the hidden layer nodes one by one from the minimum value in the value range, comparing the network convergence speed under the condition that the training set is the same, and finally selecting the optimal number of the hidden layer nodes by combining the training error obtained by the training result.
In the above embodiment of the present invention, the step of establishing a back propagation BP neural network model according to the fault feature quantity further includes: an excitation function is determined.
In this embodiment, a coupling phenomenon of a corresponding relationship between a fault phenomenon and a fault reason generally occurs in the problem of fault diagnosis of a microwave component (that is, the coupling phenomenon corresponds to different fault reasons in the case of different fault phenomenon combinations), and it is desirable that a diagnosis result can be located to a faulty module in a probabilistic manner, so for this classification problem, a Sigmoid function with an output range between [0, 1] is selected as an excitation function of a BP neural network model, and a function formula of the Sigmoid function is:
where x is the input to the excitation function, i.e. the fault feature quantity converted into vector form, and e is the natural logarithm.
The Sigmoid function is shown in fig. 2, and has the characteristics of nonlinearity, bounded function values and a value between 0 and 1. Aiming at the problem of microwave component fault diagnosis, the Sigmoid function as a neuron function has the following advantages:
the nonlinearity guarantees that the nonlinear relation between input and output can be expressed when a plurality of neurons cooperate in parallel;
two, the output value between [0, 1] can be directly used on the output layer and can be used for representing the probability of the diagnosis result;
the derivative of the Sigmoid function can be conveniently represented by a function body, and the derivative is easy to obtain f' (x) ═ f (x) × (1-f (x)), so that the method is easy to implement in a neural network using a gradient descent method.
One drawback of Sigmoid function is that when the function value is close to 0 or 1 (close to the boundary), its gradient (absolute value of the derivative value) will be very small, in which case the neural network weight values and thresholds will change slowly. It is therefore desirable to avoid the output of neurons approaching 0 or 1 before the error is small when creating a neural network.
After the number of nodes of the input layer, the number of nodes of the output layer, the number of nodes of the hidden layer and the excitation function of the BP neural network model are determined, the modeling of the BP neural network model is completed, wherein the BP neural network model is shown as a in figure 3, and a in figure 31、a2、a3...anAs input to the model input layer, c1、c2...ctIs the output of the model output layer. And after the modeling of the BP neural network model is completed, training the BP neural network model by using the fault characteristic quantity extracted from the fault information.
Specifically, as shown in fig. 4, the step of training the BP neural network model includes:
As can be seen from the defect of the excitation function Sigmoid function selected in the above embodiment, it is necessary to avoid as much as possible that the output of the next layer of neurons in the forward transfer process is close to 0 or 1 in the initial state during the initialization process. Thus, in this embodiment, the initialized gaussian distribution is set to have a standard deviation of
Where N represents the number of neurons in the layer preceding the neuron. In this case, if the outputs of the neurons in the previous layer are all 1 (extreme state), then the current neuron inputs are summed to mean 0 and the standard deviation is 0
In this state, the probability that the absolute value of the sum of neuron inputs is greater than 3 is extremely low, and therefore, it is possible to effectively prevent the neuron from being saturated due to initialization.
42, performing forward propagation operation by using the initial weight, the initial threshold and the fault characteristic quantity to respectively obtain an output result of a hidden layer and an output result of an output layer;
specifically, the step of obtaining the output result of the hidden layer and the output result of the output layer by the forward propagation operation includes:
by the formula:
calculating to obtain an output result of the hidden layer;
wherein, bjRepresents the output of the jth neuron of the hidden layer, wijAs weights of the input layer to the hidden layer, aiIs input to the input layer, thetajThe threshold value from the input layer to the hidden layer is N, and the number of the neurons in the layer before the current neuron is N;
by the formula:
calculating an output result of an output layer according to the output result of the hidden layer;
wherein, ctRepresenting the output of the t-th neuron of the output layer, vjtRepresenting the weight, r, from hidden layer to output layertIs the hidden layer to output layer threshold.
specifically, the step of calculating an error from the output result includes:
by the formula:
calculating to obtain the error;
wherein E is an error, dtTo the desired output, ctThe output of the t-th neuron of the output layer is N, and the N is the number of neurons in the layer before the current neuron. Note dt-ct＝etThen the error can be expressed as:
and 44, performing reverse feedback operation according to the error to obtain updated weight and threshold. The goal of error back-propagation is to minimize the error function.
Specifically, the step of performing inverse feedback operation according to the error to obtain updated weight and threshold includes:
by the formula
Calculating to obtain updated weight;
wherein, wijIs the weight of the input layer to the hidden layer, wij' updated input layer to hidden layer weight, η learning efficiency of the BP neural network model, bjOutput of the jth neuron of the hidden layer, aiIs the input of the input layer; v. ofjtIs the weight from hidden layer to output layer, vjt' updated weight from hidden layer to output layer, et＝dt-ct，dtTo the desired output, ctIs the output of the t-th neuron of the output layer;
the weight updating formula is calculated according to a gradient descent method, that is, the weight from the input layer to the hidden layer is updated as follows:
wherein the content of the first and second substances,
the weight from the hidden layer to the output layer is updated as:
by the formula
Calculating to obtain an updated threshold value;
wherein, thetajFor the threshold of the input layer to the hidden layer, θj' is the updated input layer to hidden layer threshold; r istFor a threshold from the hidden layer to the output layer, rt' is the updated hidden layer to output layer threshold.
Wherein, the threshold value from the input layer to the hidden layer is updated as follows:
wherein the content of the first and second substances,
the hidden layer to output layer threshold is updated as:
the update formula of the threshold is: r ist′＝rt+ηet。
According to the updating method of the weight and the threshold, the BP neural network model is subjected to repeated iterative computation, so that the weight parameter and the threshold which can enable the calculation result of the BP neural network model to be most accurate are obtained.
Optionally, the step of training the BP neural network model further includes:
judging whether the iteration times of the BP neural network model reach preset times or not; if the preset times are reached, ending the training of the BP neural network model, otherwise, repeating the training process of the BP neural network model.
In this embodiment, whether the termination condition of the algorithm is satisfied can be determined according to the iteration number of the BP neural network model, forward propagation operation can be performed according to the updated weight and the updated threshold value, the outputs of the hidden layer and the output layer are calculated, an error is further calculated, and whether the error of the output fault diagnosis result is smaller than an expected error is determined, so that whether the network model training is ended is determined.
In the above embodiment of the present invention, the step of diagnosing the fault information to be diagnosed by using the trained BP neural network model includes:
inputting the fault information to be diagnosed into the trained BP neural network model;
and carrying out fault diagnosis on the fault information to be diagnosed by using the trained BP neural network model.
In the embodiment, newly generated fault information to be diagnosed is converted into a form of most vectors, the most vectors are input into a trained BP neural network model, an operation result is output through a calculation output layer, the three maximum positions in the output multidimensional vectors are converted into corresponding fault reasons, and the numerical result represents the probability of determining the fault reasons.
The specific fault diagnosis process is as shown in fig. 5, obtaining newly generated fault phenomenon information to be diagnosed, converting the fault phenomenon information into an input vector, performing forward propagation operation to obtain output of a hidden layer and output of an output layer, then performing reverse feedback operation to obtain updated weight and threshold, performing error calculation by using the updated weight and threshold to judge whether the error is smaller than a preset allowable error, if so, obtaining output vector of the output layer, and selecting three of the output vectors with the largest probability to convert into fault cause information, thereby obtaining a diagnosis result of fault diagnosis.
According to the embodiment, a network model is determined according to the problem-oriented requirements and the vector data structure, and the existing fault case data is extracted to obtain identifiable fault information and is led into the network model for training and learning. After training is finished, newly generated fault information is input into the learned network model, the fault reason corresponding to the newly generated fault information can be obtained through calculation, and the fault reason is output and displayed in a probability mode, so that intelligent fault diagnosis of the microwave assembly of the automatic test unit is realized. The intelligent diagnosis of the microwave assembly fault facing the automatic test unit is realized; historical fault case information is dispersedly stored in each neuron in the network, and the network structure can be dynamically adjusted, so that the whole model has good robustness and fault tolerance, and the adaptability and stability of the model are improved to a certain extent; the nonlinear ability of the BP neural network can be highly fitted to approximate the coupling relation between the fault information and the fault reason in the fault diagnosis problem of the microwave assembly, so that the newly generated fault can be quickly positioned to the cause of the newly generated fault.
According to the scheme, the intelligent fault diagnosis problem of the microwave assembly of the automatic test unit is analyzed, a fault information identification and fault information standardization method is provided, an intelligent fault diagnosis method based on a neural network is researched, the test efficiency can be improved, and the real-time, high-efficiency, fine and intelligent test judgment processing is realized.
As shown in fig. 6, an embodiment of the present invention provides a microwave component fault diagnosis apparatus, including:
the first acquiring module 61 is configured to acquire fault characteristic quantities of the microwave assembly, where the fault characteristic quantities include fault phenomenon characteristic quantities and fault cause characteristic quantities;
in the embodiment, the existing fault case data is used as a training set for model training, and identifiable fault information is extracted from the existing fault case data. In an original signal obtained by fault detection of the microwave assembly, a plurality of pieces of information capable of reflecting the fault state of the microwave assembly exist, fault characteristic quantities capable of reflecting the characteristics of the microwave assembly are extracted from the information, the fault characteristic quantities comprise fault phenomenon characteristic quantities and fault reason characteristic quantities, and the extracted fault characteristic quantities reflect regularity and sensitivity in the fault characteristic quantities, namely the fault independence and the fault regularity are better.
The model establishing module 62 is configured to establish a back propagation BP neural network model according to the fault feature quantity;
and building a BP neural network model according to the extracted fault characteristic quantity, and determining the number of input layer nodes, the number of hidden layer nodes, the number of output layer nodes, an excitation function and the like of the BP neural network model according to the fault characteristic quantity.
A model training module 63, configured to train the BP neural network model;
training the BP neural network model, inputting the extracted fault characteristic quantity as sample data of the BP neural network model into the network model for operation, and adjusting a numerical matrix, weight and threshold value through multiple iterative operations. And when the preset maximum iteration times are reached or the error of the output fault diagnosis result is reduced to be lower than the expected error, considering that a termination condition is met, and finishing the model training.
And the fault diagnosis module 64 is configured to diagnose the fault information to be diagnosed by using the trained BP neural network model.
In the embodiment, after the training of the BP neural network model is finished, newly generated fault information to be diagnosed is input into the learned BP neural network model in a vector form, a fault reason corresponding to the fault information is obtained through calculation of the BP neural network model, and is output and displayed in a probability form, so that intelligent fault diagnosis of the microwave component of the automatic test unit is realized.
According to the scheme, historical fault case information is stored in each neuron in the network in a dispersed manner, so that the network structure can be dynamically adjusted, the whole model has good robustness and fault tolerance, and the adaptability and stability of the model are improved to a certain extent; the nonlinear ability of the BP neural network can be highly fitted to approximate the coupling relation between the fault information and the fault reason in the fault diagnosis problem of the microwave assembly, so that the newly generated fault can be quickly positioned to the cause of the newly generated fault. And the intelligent diagnosis of the microwave assembly fault facing the automatic test unit is realized.
In the above embodiment of the present invention, the first obtaining module 61 includes:
the extracting unit is used for extracting fault characteristic quantity from an original signal of the microwave assembly fault detection;
and the conversion unit is used for converting the fault characteristic quantity into a vector form.
After the fault characteristic quantity is extracted from the fault information, the fault characteristic quantity needs to be converted into a multidimensional vector, namely, the detected original data is converted into an input vector which can be identified by a network model. The conversion unit is specifically configured to:
according to the formula: f (X)i)＝ST(Xi)/SR(Xi) Calculating the ratio of the current value of the to-be-diagnosed state of the fault characteristic quantity to the preset value of the normal state; and determining the vector value of the fault characteristic quantity according to the ratio.
Wherein, XiFor the fault characteristic quantity, ST(Xi) Is XiOf the state to be diagnosed, SRIs XiA preset value of normal state of (1);
in the embodiment, the current value of the state to be diagnosed of the fault characteristic quantity is compared with the preset normal value, and the ratio is used as the basis for judging whether the fault phenomenon occurs. Optionally, a ratio F (X) of a current value of the to-be-diagnosed state of the fault characteristic quantity to a preset value of the normal statei) Comparing with a preset threshold value, if F (X)i) If the value of the fault characteristic quantity in the corresponding multidimensional vector is larger than the threshold value, the value of the fault characteristic quantity in the corresponding multidimensional vector is '1', the fault is shown to occur, otherwise, the value of the fault characteristic quantity in the corresponding multidimensional vector is '0', the fault is not shown to occur.
In the above embodiment of the present invention, the model building module 62 includes:
the first determining unit is used for determining the number of input layer nodes and the number of output layer nodes of the BP neural network model according to the fault characteristic quantity; the number of the nodes of the input layer is the dimension of the fault phenomenon characteristic quantity, and the number of the nodes of the output layer is the dimension of the fault reason characteristic quantity.
And the second determining unit is used for determining the number of hidden layer nodes of the BP neural network model according to the number of the input layer nodes and the number of the output layer nodes. Increasing the number of hidden layer units can reduce network errors and improve accuracy, but can also complicate the network, thereby increasing the training time of the network and causing an "overfitting" phenomenon. This embodiment is illustrated with a layer 3 network as an example, i.e. there is only one hidden layer. A lower error is then obtained by adjusting the number of hidden layer element nodes.
Specifically, the second determining unit is specifically configured to:
by the formula
Calculating the value range of the number of the hidden layer nodes:
k is the number of hidden layer nodes, n is the number of input layer nodes, l is the number of output layer nodes, and α is a constant of 1-10.
In this embodiment, the number of hidden layer nodes has a direct relationship with the requirement for solving the problem and the number of input/output units. If the number of the hidden layer nodes is too small, enough connection weight combination numbers cannot be generated to meet the learning requirement of a plurality of samples; if the number of the hidden layer unit nodes is too large, the generalization capability of the network model after learning becomes poor. Calculating to obtain the value range of the number of hidden layer nodes by the method, then increasing the number of the hidden layer nodes one by one from the minimum value in the value range, comparing the network convergence speed under the condition that the training set is the same, and finally selecting the optimal number of the hidden layer nodes by combining the training error obtained by the training result.
In the above embodiment of the present invention, the model building module 62 further includes:
a third determination unit for determining the excitation function.
In this embodiment, a coupling phenomenon of a corresponding relationship between a fault phenomenon and a fault reason generally occurs in the problem of fault diagnosis of a microwave component (that is, the coupling phenomenon corresponds to different fault reasons in the case of different fault phenomenon combinations), and it is desirable that a diagnosis result can be located to a faulty module in a probabilistic manner, so for this classification problem, a Sigmoid function with an output range between [0, 1] is selected as an excitation function of a BP neural network model, and a function formula of the Sigmoid function is:
where x is the input to the excitation function, i.e. the fault feature quantity converted into vector form, and e is the natural logarithm.
The Sigmoid function is shown in fig. 2, and has the characteristics of nonlinearity, bounded function values and a value between 0 and 1. Aiming at the problem of microwave component fault diagnosis, the Sigmoid function as a neuron function has the following advantages:
the nonlinearity guarantees that the nonlinear relation between input and output can be expressed when a plurality of neurons cooperate in parallel;
two, the output value between [0, 1] can be directly used on the output layer and can be used for representing the probability of the diagnosis result;
the derivative of the Sigmoid function can be conveniently represented by a function body, and the derivative is easy to obtain f' (x) ═ f (x) × (1-f (x)), so that the method is easy to implement in a neural network using a gradient descent method.
One drawback of Sigmoid function is that when the function value is close to 0 or 1 (close to the boundary), its gradient (absolute value of the derivative value) will be very small, in which case the neural network weight values and thresholds will change slowly. It is therefore desirable to avoid the output of neurons approaching 0 or 1 before the error is small when creating a neural network.
After the number of nodes of the input layer, the number of nodes of the output layer, the number of nodes of the hidden layer and the excitation function of the BP neural network model are determined, the modeling of the BP neural network model is completed, wherein the BP neural network model is shown as a in figure 3, and a in figure 31、a2、a3...anAs input to the model input layer, c1、c2...ctIs the output of the model output layer. And after the modeling of the BP neural network model is completed, training the BP neural network model by using the fault characteristic quantity extracted from the fault information.
Specifically, the model training module 63 includes:
the parameter setting unit is used for setting the initial weight and the initial threshold of the BP neural network model; namely, initializing a neural network model, and randomly setting initial weights and threshold values.
As can be seen from the defect of the excitation function Sigmoid function selected in the above embodiment, it is necessary to avoid as much as possible that the output of the next layer of neurons in the forward transfer process is close to 0 or 1 in the initial state during the initialization process. Thus, in this embodiment, the initialized gaussian distribution is set to have a standard deviation of
Where N represents the number of neurons in the layer preceding the neuron. In this case, if the outputs of the preceding layer neurons are all 1 (extreme state), then the current neuron inputs sum to a mean value of 0, labeledThe tolerance is
In this state, the probability that the absolute value of the sum of neuron inputs is greater than 3 is extremely low, and therefore, it is possible to effectively prevent the neuron from being saturated due to initialization.
The first calculation unit is used for performing forward propagation operation by using the initial weight, the initial threshold and the fault characteristic quantity to respectively obtain an output result of a hidden layer and an output result of an output layer;
specifically, the first computing unit is specifically configured to:
by the formula:
calculating to obtain an output result of the hidden layer;
wherein, bjRepresents the output of the jth neuron of the hidden layer, wijAs weights of the input layer to the hidden layer, aiIs input to the input layer, thetajThe threshold value from the input layer to the hidden layer is N, and the number of the neurons in the layer before the current neuron is N;
by the formula:
calculating an output result of an output layer according to the output result of the hidden layer;
wherein, ctRepresenting the output of the t-th neuron of the output layer, vjtRepresenting the weight, r, from hidden layer to output layertIs the hidden layer to output layer threshold.
A second calculation unit for calculating an error according to the output result;
specifically, the second computing unit is specifically configured to:
by the formula:
calculating to obtain the error;
wherein E is an error, dtTo the desired output, ctThe output of the t-th neuron of the output layer is N, and the N is the number of neurons in the layer before the current neuron. Note dt-ct＝etThen the error can be expressed as:
and the third calculating unit is used for carrying out reverse feedback operation according to the error to obtain the updated weight and the threshold value. The goal of error back-propagation is to minimize the error function.
Specifically, the third computing unit is specifically configured to:
by the formula
Calculating to obtain updated weight;
wherein, wijIs the weight of the input layer to the hidden layer, wij' updated input layer to hidden layer weight, η learning efficiency of the BP neural network model, bjOutput of the jth neuron of the hidden layer, aiIs the input of the input layer; v. ofjtIs the weight from hidden layer to output layer, vjt' updated weight from hidden layer to output layer, et＝dt-ct，dtTo the desired output, ctIs the output of the t-th neuron of the output layer;
the weight updating formula is calculated according to a gradient descent method, that is, the weight from the input layer to the hidden layer is updated as follows:
wherein the content of the first and second substances,
the update formula of the weight is:
the weight from the hidden layer to the output layer is updated as:
by the formula
Calculating to obtain an updated threshold value;
wherein, thetajFor the threshold of the input layer to the hidden layer, θj' is the updated input layer to hidden layer threshold; r istFor a threshold from the hidden layer to the output layer, rt' is the updated hidden layer to output layer threshold.
Wherein, the threshold value from the input layer to the hidden layer is updated as follows:
wherein the content of the first and second substances,
the hidden layer to output layer threshold is updated as:
the update formula of the threshold is: r ist′＝rt+ηet。
According to the updating method of the weight and the threshold, the BP neural network model is subjected to repeated iterative computation, so that the weight parameter and the threshold which can enable the calculation result of the BP neural network model to be most accurate are obtained.
In the above embodiment of the present invention, the model training module 63 further includes:
the judging unit is used for judging whether the iteration times of the BP neural network model reach preset times or not;
and the control unit is used for ending the training of the BP neural network model if the preset times are reached, and otherwise, repeating the training process of the BP neural network model.
In this embodiment, whether the termination condition of the algorithm is satisfied can be determined according to the iteration number of the BP neural network model, forward propagation operation can be performed according to the updated weight and the updated threshold value, the outputs of the hidden layer and the output layer are calculated, an error is further calculated, and whether the error of the output fault diagnosis result is smaller than an expected error is determined, so that whether the network model training is ended is determined.
In the above embodiment of the present invention, the fault diagnosis module includes:
the input unit is used for inputting the fault information to be diagnosed into the trained BP neural network model;
and the diagnosis unit is used for carrying out fault diagnosis on the fault information to be diagnosed by utilizing the trained BP neural network model.
In the embodiment, newly generated fault information to be diagnosed is converted into a form of most vectors, the most vectors are input into a trained BP neural network model, an operation result is output through a calculation output layer, the three maximum positions in the output multidimensional vectors are converted into corresponding fault reasons, and the numerical result represents the probability of determining the fault reasons.
The specific fault diagnosis process is as shown in fig. 5, obtaining newly generated fault phenomenon information to be diagnosed, converting the fault phenomenon information into an input vector, performing forward propagation operation to obtain output of a hidden layer and output of an output layer, then performing reverse feedback operation to obtain updated weight and threshold, performing error calculation by using the updated weight and threshold to judge whether the error is smaller than a preset allowable error, if so, obtaining output vector of the output layer, and selecting three of the output vectors with the largest probability to convert into fault cause information, thereby obtaining a diagnosis result of fault diagnosis.
According to the scheme, the intelligent fault diagnosis problem of the microwave assembly of the automatic test unit is analyzed, a fault information identification and fault information standardization method is provided, an intelligent fault diagnosis method based on a neural network is researched, the test efficiency can be improved, and the real-time, high-efficiency, fine and intelligent test judgment processing is realized.
It should be noted that the apparatus is an apparatus corresponding to the individual recommendation method, and all implementation manners in the method embodiments are applicable to the embodiment of the apparatus, and the same technical effect can be achieved.
The embodiment of the present invention further provides a microwave component fault diagnosis device, which includes a processor, a memory, and a computer program stored on the memory and operable on the processor, and when the computer program is executed by the processor, the steps of the microwave component fault diagnosis method described above are implemented.
According to the embodiment of the invention, the historical fault case information is dispersedly stored in each neuron in the network, so that the network structure can be dynamically adjusted, the whole model has good robustness and fault tolerance, and the adaptability and stability of the model are improved to a certain extent; the nonlinear ability of the BP neural network can be highly fitted to approximate the coupling relation between the fault information and the fault reason in the fault diagnosis problem of the microwave assembly, so that the newly generated fault can be quickly positioned to the cause of the newly generated fault. And the intelligent diagnosis of the microwave assembly fault facing the automatic test unit is realized.
While the preferred embodiments of the present invention have been described, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention as defined in the following claims.
Claims (15)
1. A method of fault diagnosis for a microwave assembly, comprising:
acquiring fault characteristic quantities of a microwave assembly, wherein the fault characteristic quantities comprise fault phenomenon characteristic quantities and fault reason characteristic quantities;
establishing a back propagation BP neural network model according to the fault characteristic quantity;
training the BP neural network model;
and diagnosing the fault information to be diagnosed by using the trained BP neural network model.
2. The method of claim 1, wherein the step of obtaining the fault characteristic quantity of the microwave assembly comprises:
extracting fault characteristic quantity from an original signal of microwave assembly fault detection;
and converting the fault characteristic quantity into a vector form.
3. The microwave module fault diagnosis method according to claim 2, wherein the step of converting the fault characteristic quantity into a vector form comprises:
according to the formula: f (X)i)＝ST(Xi)/SR(Xi) Calculating the ratio of the current value of the to-be-diagnosed state of the fault characteristic quantity to the preset value of the normal state;
wherein, XiFor the fault characteristic quantity, ST(Xi) Is XiOf the state to be diagnosed, SRIs XiA preset value of normal state of (1);
and determining the vector value of the fault characteristic quantity according to the ratio.
4. The microwave component fault diagnosis method according to claim 1, wherein the step of building a back propagation BP neural network model based on the fault feature quantity comprises:
determining the number of input layer nodes and the number of output layer nodes of the BP neural network model according to the fault characteristic quantity;
and determining the number of hidden layer nodes of the BP neural network model according to the number of the input layer nodes and the number of the output layer nodes.
5. The microwave module fault diagnosis method according to claim 4, wherein the number of the input layer nodes is the number of dimensions of the fault phenomenon feature quantity, and the number of the output layer nodes is the number of dimensions of the fault cause feature quantity.
6. The microwave component fault diagnosis method according to claim 4, wherein the step of determining the number of hidden layer nodes of the BP neural network model according to the number of input layer nodes and the number of output layer nodes comprises:
by the formula
Calculating the value range of the number of the hidden layer nodes:
k is the number of hidden layer nodes, n is the number of input layer nodes, l is the number of output layer nodes, and α is a constant of 1-10.
7. The microwave component fault diagnosis method according to claim 1, wherein the step of building a back propagation BP neural network model based on the fault feature quantity further comprises:
an excitation function is determined.
8. The microwave component fault diagnosis method according to claim 1, wherein the step of training the BP neural network model comprises:
setting initial weight and initial threshold of the BP neural network model;
performing forward propagation operation by using the initial weight, the initial threshold and the fault characteristic quantity to respectively obtain an output result of a hidden layer and an output result of an output layer;
calculating an error according to the output result;
and performing reverse feedback operation according to the error to obtain updated weight and threshold.
9. The microwave component fault diagnosis method according to claim 8, wherein the step of performing forward propagation operation using the initial weight, the initial threshold value and the fault feature quantity to obtain the output result of the hidden layer and the output result of the output layer respectively comprises:
by the formula:
calculating to obtain an output result of the hidden layer;
wherein, bjRepresents the output of the jth neuron of the hidden layer, wijAs weights of the input layer to the hidden layer, aiIs input to the input layer, thetajThe threshold value from the input layer to the hidden layer is N, and the number of the neurons in the layer before the current neuron is N;
by the formula:
calculating an output result of an output layer according to the output result of the hidden layer;
wherein, ctRepresenting the output of the t-th neuron of the output layer, vjtRepresenting the weight, r, from hidden layer to output layertIs the hidden layer to output layer threshold.
10. The microwave assembly fault diagnosis method according to claim 8, wherein the step of calculating an error based on the output result comprises:
by the formula:
calculating to obtain the error;
wherein E is an error, dtTo the desired output, ctIs the output of the t-th neuron of the output layer, and N is the spirit of the previous layer of the current neuronThe number of warp elements.
11. The microwave assembly fault diagnosis method according to claim 8, wherein the step of performing an inverse feedback operation according to the error to obtain updated weights and thresholds comprises:
by the formula
Calculating to obtain updated weight;
wherein, wijIs the weight of the input layer to the hidden layer, wij' updated input layer to hidden layer weight, η learning efficiency of the BP neural network model, bjOutput of the jth neuron of the hidden layer, aiIs the input of the input layer; v. ofjtIs the weight from hidden layer to output layer, vjt' updated weight from hidden layer to output layer, et＝dt-ct，dtTo the desired output, ctIs the output of the t-th neuron of the output layer;
by the formula
Calculating to obtain an updated threshold value;
wherein, thetajFor the threshold of the input layer to the hidden layer, θj' is the updated input layer to hidden layer threshold; r istFor a threshold from the hidden layer to the output layer, rt' is the updated hidden layer to output layer threshold.
12. The microwave component fault diagnosis method according to claim 1, wherein the step of training the BP neural network model further comprises:
judging whether the iteration times of the BP neural network model reach preset times or not;
if the preset times are reached, ending the training of the BP neural network model, otherwise, repeating the training process of the BP neural network model.
13. The microwave assembly fault diagnosis method according to claim 1, wherein the step of diagnosing the fault information to be diagnosed by using the trained BP neural network model comprises:
inputting the fault information to be diagnosed into the trained BP neural network model;
and carrying out fault diagnosis on the fault information to be diagnosed by using the trained BP neural network model.
14. A microwave module fault diagnosis apparatus, comprising:
the microwave component fault detection device comprises a first acquisition module, a second acquisition module and a control module, wherein the first acquisition module is used for acquiring fault characteristic quantities of a microwave component, and the fault characteristic quantities comprise fault phenomenon characteristic quantities and fault reason characteristic quantities;
the model establishing module is used for establishing a back propagation BP neural network model according to the fault characteristic quantity;
the model training module is used for training the BP neural network model;
and the fault diagnosis module is used for diagnosing the fault information to be diagnosed by utilizing the trained BP neural network model.
15. A microwave component fault diagnosis device, characterized by comprising a processor, a memory and a computer program stored on the memory and executable on the processor, the computer program, when executed by the processor, implementing the steps of the microwave component fault diagnosis method according to any one of claims 1 to 13.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811517864.5A CN111310907A (en) | 2018-12-12 | 2018-12-12 | Microwave assembly fault diagnosis method, device and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811517864.5A CN111310907A (en) | 2018-12-12 | 2018-12-12 | Microwave assembly fault diagnosis method, device and equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111310907A true CN111310907A (en) | 2020-06-19 |
Family
ID=71159594
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811517864.5A Pending CN111310907A (en) | 2018-12-12 | 2018-12-12 | Microwave assembly fault diagnosis method, device and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111310907A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111830934A (en) * | 2020-07-17 | 2020-10-27 | 中车大连电力牵引研发中心有限公司 | Electric locomotive fault source positioning method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103364683A (en) * | 2013-07-01 | 2013-10-23 | 东南大学 | Method for detecting open-circuit fault of inverter circuit |
| CN107480787A (en) * | 2017-08-10 | 2017-12-15 | 西安建筑科技大学 | A kind of opencut gyratory crusher method for diagnosing faults based on BP neural network |
| US20170364800A1 (en) * | 2016-06-16 | 2017-12-21 | Qatar University | Method and apparatus for performing motor-fault detection via convolutional neural networks |
-
2018
- 2018-12-12 CN CN201811517864.5A patent/CN111310907A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103364683A (en) * | 2013-07-01 | 2013-10-23 | 东南大学 | Method for detecting open-circuit fault of inverter circuit |
| US20170364800A1 (en) * | 2016-06-16 | 2017-12-21 | Qatar University | Method and apparatus for performing motor-fault detection via convolutional neural networks |
| CN107480787A (en) * | 2017-08-10 | 2017-12-15 | 西安建筑科技大学 | A kind of opencut gyratory crusher method for diagnosing faults based on BP neural network |
Non-Patent Citations (3)
| Title |
|---|
| ZHAOYUXIA517: "BP神经网络，BP推导过程，反向传播算法，误差反向传播，梯度下降，权值阈值更新推导，隐含层权重更新公式", pages 1 - 3, Retrieved from the Internet <URL:《http://blog.csdn.net/zhaoyuxia517/article/details/78116683》> * |
| 张绪锦 等: "基于BP神经网络的故障诊断方法", 《系统工程理论与实践》, no. 6, pages 61 - 66 * |
| 易守平 等: "基于进化神经网络的滤波器电路故障诊断方法", pages 54 - 56 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111830934A (en) * | 2020-07-17 | 2020-10-27 | 中车大连电力牵引研发中心有限公司 | Electric locomotive fault source positioning method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109359385B (en) | Training method and device for service quality assessment model | |
| Xie et al. | Time series prediction based on NARX neural networks: An advanced approach | |
| CN111860982A (en) | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU | |
| CN108960303B (en) | Unmanned aerial vehicle flight data anomaly detection method based on LSTM | |
| CN109141847B (en) | Aircraft system fault diagnosis method based on MSCNN deep learning | |
| CN108958217A (en) | A kind of CAN bus message method for detecting abnormality based on deep learning | |
| CN110335168B (en) | Method and system for optimizing power utilization information acquisition terminal fault prediction model based on GRU | |
| CN106656357B (en) | Power frequency communication channel state evaluation system and method | |
| CN109829514A (en) | A kind of network inbreak detection method, device, computer equipment and storage medium | |
| CN109815855B (en) | Electronic equipment automatic test method and system based on machine learning | |
| CN116089870A (en) | Industrial equipment fault prediction method and device based on meta-learning under small sample condition | |
| CN112763967A (en) | BiGRU-based intelligent electric meter metering module fault prediction and diagnosis method | |
| CN111310722A (en) | Power equipment image fault identification method based on improved neural network | |
| CN111444947A (en) | Bridge settlement intelligent detection system | |
| CN117056865B (en) | Method and device for diagnosing operation faults of machine pump equipment based on feature fusion | |
| CN114189047A (en) | False data detection and correction method for active power distribution network state estimation | |
| CN111310907A (en) | Microwave assembly fault diagnosis method, device and equipment | |
| CN111079348B (en) | Method and device for detecting slowly-varying signal | |
| CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
| CN116094758B (en) | Large-scale network flow acquisition method and system | |
| CN111858343A (en) | Countermeasure sample generation method based on attack capability | |
| CN114387524B (en) | Image identification method and system for small sample learning based on multilevel second-order representation | |
| CN114565051A (en) | Test method of product classification model based on neuron influence degree | |
| CN115099135A (en) | Improved artificial neural network multi-type operation power consumption prediction method | |
| CN114970674A (en) | Time sequence data concept drift adaptation method based on relevance alignment |
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 |