CN118035799B

CN118035799B - Filling mining equipment fault diagnosis method and device based on multidimensional data fusion

Info

Publication number: CN118035799B
Application number: CN202410432046.4A
Authority: CN
Inventors: 马兵; 尹旭; 王野; 张军; 张红静; 王明华
Original assignee: Shandong Energy Shuzhiyun Technology Co ltd
Current assignee: Shandong Energy Shuzhiyun Technology Co ltd
Priority date: 2024-04-11
Filing date: 2024-04-11
Publication date: 2024-07-12
Anticipated expiration: 2044-04-11
Also published as: CN118035799A

Abstract

The invention provides a filling mining equipment fault diagnosis method and device based on multidimensional data fusion, which relate to the technical field of data processing and comprise the following steps: acquiring monitoring data of the target filling and mining equipment in the operation process; the monitoring data comprise state data of the target filling and mining equipment and operation data of the target filling and mining equipment, wherein the operation data are acquired by sensors with multiple dimensions; assembling the monitoring data to generate a vector to be detected; inputting the vector to be detected into a pre-constructed fault diagnosis model, classifying the vector to be detected through the fault diagnosis model, and determining a classification result; the fault diagnosis model is constructed based on an improved quantum coding high-order neural network classifier algorithm, the quantum coding is introduced to capture and represent complex relations in data, and the expression capacity of the model is enhanced through the high-order neural network, so that the fault diagnosis precision can be improved.

Description

Filling mining equipment fault diagnosis method and device based on multidimensional data fusion

Technical Field

The invention relates to the technical field of data processing, in particular to a filling mining equipment fault diagnosis method and device based on multidimensional data fusion.

Background

Filling mining is a key process in coal mine production, and a filling pump is used as core equipment in the process and is responsible for stably and efficiently feeding filling materials into a mine so as to ensure the stability of a mine body. However, in an actual mine production environment, the charge pump is often exposed to various risks of failure due to its continued operation and the effects of external conditions. For example, pumps may suffer from seal failure, bearing wear or overheating due to long-term mechanical friction, material washout, temperature changes, etc. If these problems are not detected and handled in a timely manner, they may lead to more serious equipment failures and even threatens the safety of miners. However, prior art models have limited expressive power: complex relationships in the data are difficult to capture, resulting in models that can be difficult to handle in some complex and nonlinear situations, affecting diagnostic accuracy.

Disclosure of Invention

Accordingly, the invention aims to provide a fault diagnosis method and device for filling mining equipment based on multidimensional data fusion, which can improve the precision of fault diagnosis.

In a first aspect, an embodiment of the present invention provides a method for diagnosing a fault of a filling mining device based on multidimensional data fusion, which is characterized by comprising: acquiring monitoring data of the target filling and mining equipment in the operation process; the monitoring data comprise state data of the target filling and mining equipment and operation data of the target filling and mining equipment, wherein the operation data are acquired by sensors with multiple dimensions; assembling the monitoring data to generate a vector to be detected; inputting the vector to be detected into a pre-constructed fault diagnosis model, classifying the vector to be detected through the fault diagnosis model, and determining a classification result; the fault diagnosis model is constructed based on an improved quantum coding high-order neural network classifier algorithm, and a training set for training the fault diagnosis model comprises a multi-dimensional diagnosis sample, wherein the multi-dimensional diagnosis sample comprises state data of filling and mining equipment, operation data of the filling and mining equipment, acquired by sensors with multiple dimensions, and corresponding sample labels; and carrying out fault diagnosis on the target filling and mining equipment based on the classification result.

In a second aspect, an embodiment of the present invention provides a fault diagnosis device for a filling mining apparatus based on multidimensional data fusion, where the device includes: the data acquisition module is used for acquiring monitoring data of the target filling mining equipment in the running process; the monitoring data comprise state data of the target filling and mining equipment and operation data of the target filling and mining equipment, wherein the operation data are acquired by sensors with multiple dimensions; the data processing module is used for assembling the monitoring data to generate a vector to be detected; the execution module is used for inputting the vector to be detected into a pre-constructed fault diagnosis model, classifying the vector to be detected through the fault diagnosis model, and determining a classification result; the fault diagnosis model is constructed based on an improved quantum coding high-order neural network classifier algorithm, and a training set for training the fault diagnosis model comprises a multi-dimensional diagnosis sample, wherein the multi-dimensional diagnosis sample comprises state data of filling and mining equipment, operation data of the filling and mining equipment, acquired by sensors with multiple dimensions, and corresponding sample labels; and the output module is used for carrying out fault diagnosis on the target filling and mining equipment based on the classification result.

The embodiment of the invention has the following beneficial effects: according to the fault diagnosis method and device for the filling and mining equipment based on multidimensional data fusion, the fault diagnosis is carried out on the monitoring data of the filling and mining equipment in the operation process through the fault diagnosis model constructed based on the improved quantum coding high-order neural network classifier algorithm, wherein the complex relation in the data is captured and represented through introducing quantum coding, and the expression capacity of the model is enhanced through the high-order neural network, so that the fault diagnosis precision can be improved.

Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and drawings.

In order to make the above objects, features and advantages of the present invention more comprehensible, preferred embodiments accompanied with figures are described in detail below.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the present invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.

FIG. 1 is a flow chart of a filling mining equipment fault diagnosis method based on multidimensional data fusion provided by an embodiment of the invention;

FIG. 2 is a flowchart of another method for diagnosing faults of filling and mining equipment based on multidimensional data fusion according to an embodiment of the present invention;

FIG. 3 is a flowchart of another method for diagnosing faults of filling and mining equipment based on multidimensional data fusion according to an embodiment of the present invention;

fig. 4 is a schematic structural diagram of a fault diagnosis device of filling mining equipment based on multidimensional data fusion according to an embodiment of the present invention;

FIG. 5 is a schematic structural diagram of another filling mining equipment fault diagnosis device based on multidimensional data fusion according to an embodiment of the present invention;

fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed Description

For the purposes of clarity, technical solutions, and advantages of the embodiments of the present disclosure, the following description describes embodiments of the present disclosure with specific examples, and other advantages and effects of the present disclosure will be readily apparent to those skilled in the art from the disclosure herein. It will be apparent that the described embodiments are merely some, but not all embodiments of the present disclosure. The disclosure may be embodied or practiced in other different specific embodiments, and details within the subject specification may be modified or changed from various points of view and applications without departing from the spirit of the disclosure. It should be noted that the following embodiments and features in the embodiments may be combined with each other without conflict. All other embodiments, which can be made by one of ordinary skill in the art without inventive effort, based on the embodiments in this disclosure are intended to be within the scope of this disclosure.

It is noted that various aspects of the embodiments are described below within the scope of the following claims. It should be apparent that the aspects described in this disclosure may be embodied in a wide variety of forms and that any specific structure and/or function described in this disclosure is illustrative only. Based on the present disclosure, one skilled in the art will appreciate that one aspect described in this disclosure may be implemented independently of any other aspects, and that two or more of these aspects may be combined in various ways. For example, apparatus may be implemented and/or methods practiced using any number of the aspects set forth in this disclosure. In addition, such apparatus may be implemented and/or such method practiced using other structure and/or functionality in addition to one or more of the aspects set forth in the disclosure.

It should also be noted that the illustrations provided in the following embodiments merely illustrate the basic concepts of the disclosure by way of illustration, and only the components related to the disclosure are shown in the drawings and are not drawn according to the number, shape and size of the components in actual implementation, and the form, number and proportion of the components in actual implementation may be arbitrarily changed, and the layout of the components may be more complicated. In addition, in the following description, specific details are provided in order to provide a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects may be practiced without these specific details.

The filling mining equipment fault diagnosis method and device based on multidimensional data fusion, provided by the embodiment of the invention, can improve the fault diagnosis precision.

For the sake of understanding the present embodiment, first, a method for diagnosing a fault of a filling and mining device based on multidimensional data fusion disclosed in the present embodiment of the present invention is described in detail, and fig. 1 shows a flowchart of a method for diagnosing a fault of a filling and mining device based on multidimensional data fusion provided in the present embodiment, as shown in fig. 1, where the method includes the following steps:

and step S102, acquiring monitoring data of the target filling and mining equipment in the operation process.

Step S104, assembling the monitoring data to generate a vector to be detected.

Step S106, inputting the vector to be tested into a pre-constructed fault diagnosis model, classifying the vector to be tested through the fault diagnosis model, and determining a classification result.

The embodiment of the invention performs fault diagnosis on the filling mining equipment through a fault diagnosis model, wherein the corresponding fault classification result is determined by collecting monitoring data of the equipment in the operation process, assembling the monitoring data into a vector to be detected and then performing data identification.

Specifically, the monitoring data comprises state data of the target filling and mining equipment and operation data of the target filling and mining equipment, wherein the operation data are acquired by sensors with multiple dimensions; the data of the invention is collected by a plurality of sensors of the device, in one embodiment, a filling pump in the coal mine filling and mining device is selected as a collecting and fault diagnosis object, and the plurality of sensors comprise a pressure sensor, a flow sensor, a temperature sensor, a vibration sensor and the like. Each data record is a vector, consisting of: a ₁,a₂,...,a₁₀.

Wherein each element represents a measured value of a device attribute, and specifically, each attribute is expressed as: a _x: the pressure value of the equipment in operation is obtained by the pressure sensor. a _y: the flow value of the device when in operation is obtained by a flow sensor. a _z: the temperature value of the equipment in operation is obtained by a temperature sensor. a _m: the vibration frequency of the device when in operation is obtained by the vibration sensor. a _n: the current value of the equipment in operation is obtained by a current sensor. a _o: the rotational speed of the device is obtained by a rotational speed sensor. a _p: the working time of the equipment is calculated according to the time difference from the starting to the current starting. a _q: the operating state of the device, obtained by the device state sensor, comprises: normal, standby, failure. a _r: the humidity value of the equipment in operation is obtained by a humidity sensor. a _s: the operating frequency of the device is obtained by a frequency sensor.

Correspondingly, the fault diagnosis model is constructed based on an improved quantum-coded high-order neural network classifier algorithm, and the training set for training the fault diagnosis model comprises a multi-dimensional diagnosis sample, wherein the multi-dimensional diagnosis sample comprises state data of filling and mining equipment, operation data of the filling and mining equipment acquired by a plurality of dimensional sensors and corresponding sample labels. The fault diagnosis model determines a classification result by determining a label corresponding to the vector to be detected.

Further, the embodiment of the invention classifies the vector to be detected through the fault diagnosis model, determines initial output, and combines the initial output with the Bayesian decision theory to obtain a classification result corresponding to the vector to be detected.

Specifically, using bayesian decision theory in combination with the output of the model to determine the failure mode of the device can be expressed as:。

Wherein, Is observation data; is the output of a certain failure mode. Here, bayesian decision theory is used, wherein, Is given dataPosterior probability of a device being in a certain failure mode,Is the likelihood probability that the user will be able to determine,Is the prior probability of being a priori,Is the evidence probability.

And step S108, performing fault diagnosis on the target filling and mining equipment based on the classification result.

According to the filling mining equipment fault diagnosis method based on multidimensional data fusion, fault diagnosis is carried out on monitoring data of the filling mining equipment in the operation process through the fault diagnosis model constructed based on the improved quantum coding high-order neural network classifier algorithm, complex relations in the data are captured and represented through introducing quantum coding, and the expression capacity of the model is enhanced through the high-order neural network, so that the fault diagnosis precision can be improved.

Further, the following technical problems exist in the prior art: (1) data preprocessing is not enough normative: the lack of a unified data preprocessing method may lead to data in different ranges, thereby affecting the training and prediction effects of the model; (2) limited data expansion methods: the prior art, in the face of limited data sets, may not have sufficient means to perform data enhancement, resulting in insufficient generalization capability of the model; (3) traditional optimization methods may be slow: rather than finding the optimal solution as quickly as seismic wave propagation-based methods, longer time and more computing resources may be required for model training.

Based on the method, the embodiment of the invention also provides another filling mining equipment fault diagnosis method based on multidimensional data fusion. Specifically, in the embodiment of the present invention, the monitoring data is identified through a fault diagnosis model, so as to perform fault diagnosis on the equipment, where the model is trained through a training set, and in a specific implementation, fig. 2 shows a flowchart of another fault diagnosis method for filling mining equipment based on multidimensional data fusion, which is provided in the embodiment of the present invention, and a construction step of the training set is described, with reference to fig. 2, and includes the following steps:

step S202, collecting a multi-dimensional diagnosis sample of filling and mining equipment.

And step S204, labeling the sample according to the equipment working state corresponding to the multi-dimensional diagnosis sample, and generating a sample label.

The multi-dimensional diagnosis sample refers to the embodiment, further, the collected data is subjected to data labeling, and the labeling process is completed based on the actual running state of the equipment. When the equipment is in a normal working state, the data label is normal; when the equipment fails, the data label is of a specific failure type, such as 'over-high pressure', 'abnormal flow', and the like.

In one embodiment, one specific data record is ：a₁=12.5,a₂=300,a₃=35.2,a₄=60,a₅=10.5,a₆=1500,a₇=2.5,a₈=" normal ", a ₉=40,a₁₀ =50 Hz. ：a_x：12.5bar;a_y：300L/min;a_z：35.2°C;a_m：60Hz;a_n：10.5A;a_o：1500rpm;a_p：2.5h;a_q： is normal in this data; a _r：40%;a_s: 50Hz. This data record shows that the filling pump, at a specific point in time, has a pressure of 12.5bar, a flow rate of 300L/min, a temperature of 35.2 ℃, a vibration frequency of 60Hz, a current of 10.5A, a rotational speed of 1500rpm, is operating for 2.5 hours, is in a normal state, a humidity of 40% and an operating frequency of 50Hz.

After data acquisition is performed on fault diagnosis of filling and mining equipment, raw data needs to be preprocessed in order to ensure high efficiency and accuracy of an algorithm. The preprocessing aims to eliminate noise, abnormal values, missing values and the like in the data, so that the data is more standardized, and the processing of subsequent steps is convenient. Firstly, each attribute of the data is normalized to ensure that the value of each attribute is within a specific range, and the invention limits the value of the data to be within the [0,1] interval. Setting the value of a certain attribute in the original data asThe maximum value and the minimum value are respectivelyAndThe manner of using Min-Max normalization can be expressed as: . Wherein, Is normalized data.

Further, aiming at the missing value in the data, the processing mode is to adopt mean value substitution. Setting data attributesThere is a deletion of the mean value of。

The mean substitution approach can be expressed as: ; wherein, Is the filled data.

Further, an operation of removing outliers, which are data points that deviate far from the normal range, is performed, and these values may be due to sensor failure or other abnormal operation, and need to be detected and removed for failure diagnosis. Let the ith data point of attribute bz be bz _i, its mean value beStandard deviation isSetting a threshold valueTypically taken as 3, when the data point meets the following conditions, it is considered an outlier and removed:。

further, some noisy data is smoothed using a moving average or the like. Set the use length as Is subjected to moving average of windows of the data attribute ofThe moving average formula is: . Wherein, Is smoothed data.

Step S206, constructing an initial sample set based on the sample label and the multi-dimensional diagnostic sample.

Further, an initial sample set is constructed from the samples and sample tags.

It will be appreciated that the collection, labeling and preprocessing of data is time-consuming and labor-consuming, and that insufficient training samples can easily result in difficult model convergence and poor model generalization capability. The invention provides an improved generation countermeasure network algorithm for sample generation, thereby realizing data expansion.

The generation countermeasure network consists of two networks: a generator (G) and a discriminator (D). The generator attempts to generate dummy data similar to the real data, and the arbiter attempts to distinguish between the real and dummy data. The two networks are opposed by each other and the final generator is able to generate dummy data very close to real data. Because equipment data in a coal mine filling mining task may have the defect of certain specific modes, the invention introduces participation limit to ensure that the generated data is more diversified and avoid generating false data which deviates from the real data distribution too much. Further, the embodiment of the invention aims to adjust the quantum state to cooperatively optimize the generator by introducing the quantum state modulation operation, and based on the quantum state modulation operation, data which is highly similar to real data can be generated.

The training flow of the improved generation countermeasure network algorithm is as follows:

Step S208, randomly sampling from the initial sample set to obtain sampling samples, and distributing sample weights to the sampling samples according to data distribution corresponding to the sampling samples.

First, data sampling and weight distribution are performed: a batch of data is randomly sampled from the real data. Based on the distribution of the data, each data point is assigned a weight such that a minority class of data has a higher weight. Specifically, let the real data be X, where each a _i represents a data point, the data weight PW (i.e., the sample weight) may be represented as: pw= { PW ₁,pw₂,…,pw_i,…,pw_n }. Wherein the weight w _i is proportional to the inverse frequency of the class of data points a _i. That is, a minority class of data points may be more weighted and may be expressed as:。

Wherein, For the frequency of occurrence of the data samples pa _i,Indicating the frequency value corresponding to the sample with the largest frequency of occurrence among all the data samples, pw _i is the weight of the data sample pa _i,Is a non-uniformity coefficientThe coefficient is calculated based on the number of data points of each class, and can be expressed as: . Wherein, Is the number of total data points that are to be counted,Is the firstThe number of data points of the class.

Step S210, performing data expansion on the initial sample set through a preset sample expansion algorithm to generate an expansion sample.

In specific implementation, the sample expansion algorithm comprises a generation antagonism network algorithm, and the embodiment of the invention performs data expansion by generating the antagonism network algorithm, generates samples by the generation antagonism network algorithm, and performs sample discrimination by the discriminator. First, a generator is initializedDistinguishing deviceIs a weight of (2). At the same time, a threshold parameter is set, deciding when to introduce engagement limits and when to adjust the preferences of the generator. And setting super parameters of the quasi-Newton method, such as memory size, step size and the like.

Further, a permit a leave data is generated using generator G. The quality of these dummy data is evaluated by the discriminator D and a quality score is assigned to each dummy data. Specifically, the output of generator G may represent: . Where z is the random noise, and where, Is a set of parameters of the generator and,Is the generated data point.

Further, the discriminator D outputs a scoreIndicating its judgmentIs the probability of real data. Also, the output of the arbiter D may be expressed asWherein, the method comprises the steps of, wherein,Is the parameter set of the arbiter.

Step S212, updating the sample weight based on the mass fraction of the extended sample, and optimizing the sample extension algorithm.

After the above-described expanded samples are obtained, the discriminator D is trained using real data and dummy data. Meanwhile, the data weight is updated according to the quality score of the false data, and the learning direction of the generator is optimized. When the method is specifically implemented, the expanded sample generated by the generator is judged by a discriminator for generating an countermeasure network algorithm, and a judging result is determined; calculating a loss function of the discriminator based on the discrimination result; the sample weights pw _i are updated based on the loss function.

Specifically, the loss function L _D of the arbiter is represented by:

Where L _D is the loss function that is the arbiter.

Further, according to L _D, the weight of D is updated. pa _i is a sample of real data input to the generation countermeasure network.

Further, for the generator, the adjustment of the weights is based on the feedback of the arbiter. That is, the generator wants to make it more difficult for the arbiter to distinguish between the false data it generates.

The loss function of the generator L _G is: Wherein L _G is the loss function of the generator.

Further, the weight update of the arbiter can be performed by the following gradient descent formula: Wherein, the method comprises the steps of, wherein, Is the learning rate.Is a set of parameters for the generator; is a parameter set of the arbiter. Further, after each time of generating the extended sample, a sample weight is allocated to the sample according to the data distribution of the current sample, so as to update the sample weight.

Meanwhile, the embodiment of the invention optimizes the sample expansion algorithm, which comprises the following steps: 1-calculating the proportion of each type of sample in the expanded samples, and adjusting the parameters of the generator according to the proportion; 2, based on the discrimination result, carrying out joint optimization on the parameters of the generator and the parameters of the discriminator; and 3, optimizing the learning strategies of the generator and the discriminator by using a preset meta learning method.

1) The parameters of the generator G are updated using quasi-newton methods, making the dummy data it generates more difficult to distinguish by D. In this process, the generator's preferences are adjusted according to the data weights, making it more prone to generating minority classes of data.

Specifically, using a quasi-newton method, performing iterative optimization on G, and setting the weight of the generator as θ, where the iteration of the quasi-newton method is as follows: Wherein, the method comprises the steps of, wherein, Is the inverse of the Hessian matrix,Is the step size of the step,Is a loss functionWith respect toIs a gradient of (a).

Further, participation evaluation and dynamic adjustment are also performed:

The proportion of each type of data in the generated data is calculated. Further, it is judged whether a predetermined threshold is satisfied or not, in comparison with the true data distribution. If the amount of generation of a certain class of data exceeds or falls below a threshold, the parameters of the generator G are dynamically adjusted to change its generation preferences.

Specifically, the difference between the category distribution of the generated data and the category distribution of the real data, in which the engagement degree P is calculated, can be expressed as: Wherein, the method comprises the steps of, wherein, For the frequency of occurrence of the real data samples pa _i input to the generation of the challenge network,The frequency of occurrence of the data samples c _i generated for the generator.

When P exceeds a set threshold, parameters of the generator are adjusted to change its generation preferences.

2) The joint tuning is based on the joint loss function for parameter updating, and the joint loss function is based on preset quantum state calculation.

Specifically, the weight of the generator G is fine-tuned based on the feedback of the arbiter D. At the same time, the feedback of the generator G is used to fine tune the weight of the arbiter D, ensuring the balance between the two.

Specifically, in order to ensure balance between the two, let us say that the joint loss function L _GD is expressed by the following equation: Wherein, the method comprises the steps of, wherein, Is a balance coefficient, and L _GD is a loss function that is a combination of the generator and the arbiter.

As a synergy loss function, the effect is to use quantum states to help optimize the parameters of the generator G when using the generation of the antagonistic network enhancement data. I.e. for each input sampleDefine its quantum state asCan be expressed as: Where pa _i,j is the j-th feature of sample pa _i, Is the ground state corresponding to the feature.

Further, a quantum state modulation operation is introducedWherein, the method comprises the steps of, wherein,Is a parameter set of a modulation operation whose purpose is to adjust the quantum states to co-optimize the generator, based on which data can be generated that is highly similar to real data, which can be expressed as:。

Then, the synergy loss function The difference between the dummy data generated by the generator and the data after quantum state modulation can be expressed as:。

3) Meta learning and self-adaptive adjustment:

The learning strategies of the generator and the arbiter are optimized using a meta learning method. If the data quality is not significantly improved in successive rounds, the learning rate or other super-parameters are automatically adjusted. Specifically, meta learning is performed using a model-agnostic meta learning algorithm (Model Agnostic META LEARNING, abbreviated as MAML), which can be simplified by the following formula: Wherein, the method comprises the steps of, wherein, Is the weight after the optimization of the weight,Representing the MAML algorithm.

Further, the key to MAML is to perform two gradient updates, which can be expressed as:。

for the outer loop, the update formula is: Wherein, the method comprises the steps of, wherein, Is the weight after the update of the weight,Is thatThe updated learning rate is used to determine the learning rate,Is thatIs used to update the gradient.

Step S214, until the mass fraction of the extended sample meets the preset requirement, a training set is constructed based on the extended sample and the initial sample set.

In a specific implementation, the improvement amplitude of successive rounds is evaluated to determine if the algorithm has converged. If the improvement amplitude is smaller than a preset threshold value, an early-stopping strategy is started, and training is finished.

Further, the trained generator G is returned. And generating a new sample by using the trained generator, and adding the new sample into the initial data set to construct a training set.

In one embodiment, the improved generation counter network algorithm has a training iteration number of 5000, 500 pieces of data in the original data set, wherein for 1 piece of original data, the data has 8192 sampling points. The data is further transformed by fourier transformation, and the improved generation by training is performed against the signal image of the network algorithm for sample generation of the generated data. The generated data is further transformed by fourier transformation, based on which the generated data has extremely high similarity to the original data, both in the time domain and in the frequency domain.

Further, on the basis of the above embodiment, the fault diagnosis model is constructed based on an improved quantum-coded high-order neural network classifier algorithm, wherein the embodiment of the invention inputs the data after feature extraction into the classifier for classification. The invention provides a high-order neural network classifier algorithm based on improved quantum coding, which utilizes the combination of a high-order neural network based on improved quantum coding and an echo state network to construct a deep learning model for diagnosing equipment faults. The improved quantum coding provides a brand new parameter coding mode for the model of the invention, the mode can capture and represent complex relations in data, and the higher order neural network enhances the expression capacity of the model by carrying out high-order nonlinear mapping on input data. Meanwhile, the echo state network has excellent performance in time sequence analysis and dynamic system modeling, and the echo state network is used for processing information in a time dimension, so that the model has stronger sensitivity to the running state and fault mode of equipment. Fig. 3 shows another method for diagnosing faults of filling and mining equipment based on multidimensional data fusion, which is provided by the embodiment of the invention, as shown in fig. 3, and comprises the following steps:

Step S302, a pre-constructed training set is obtained.

And S304, carrying out feature extraction on the training set through a pre-constructed feature extraction model, and determining target feature parameters.

The feature extraction model is constructed through the preset training set, and in one embodiment, the feature extraction model can be constructed through the training set constructed through the embodiment.

Some schemes use neural networks for feature extraction, and in some neural network structures, problems of gradient disappearance, gradient explosion or sinking into a locally optimal solution may be encountered, which affects the training stability and the performance of the model. The invention provides a neural network optimization algorithm based on seismic wave propagation, which is inspired by a seismic wave propagation principle, and when an earthquake occurs, seismic waves can propagate along the fastest paths, and the paths are influenced by characteristics (such as density, elasticity and the like) of underground media. The invention regards the loss curved surface of the neural network as the terrain, and regards the parameter updating path as the seismic wave, and attempts to simulate the propagation of the seismic wave in the complex terrain so as to find the fastest optimizing path.

In specific implementation, the embodiment of the invention constructs a feature extraction model through the following steps:

1) A pre-constructed training set is obtained.

2) Inputting the training set into a preset neural network, and calculating an initial loss function.

First, the neural network parameters are initialized and an initial loss function of the neural network is calculated.

Specifically, the weights (w) and biases (b) of the neural network are both randomly initialized from within a predefined range. Defining L as a loss function, w as a weight matrix, b as a bias vector, E as seismic wave energy, alpha as a learning rate, G as a gradient, M _i as a dielectric characteristic of an ith layer loss curved surface, and R as a reflection coefficient.

3) A multi-layer loss topography is constructed based on the gradient of the initial loss function, and medium characteristics are assigned to each layer of loss topography.

Calculating the gradient of the current loss function:

the gradient of the loss function is calculated from the current weight (w) and bias (b) configuration. In particular, for any weight in the neural network Bias and method of making sameThe gradient calculation method of the loss function L can be expressed as follows:；。

Further, simulating the propagation of seismic waves in the subsurface may be considered as a detection of a lost surface. To make the model more stable in the optimization process, a small disturbance is added to the loss surface after each update, which can be expressed as: ； . Wherein, Is a small constant; l is the value of the loss function; to a small disturbance of the loss surface.

Constructing a multilayer loss topography: wherein, simulate a plurality of subsurface layers to each layer loses the curved surface and distributes different medium characteristic. Specifically, the subsurface layer defined for each gradient G simulates its dielectric propertiesThe manner of (a) can be expressed as: Wherein, the method comprises the steps of, wherein, Representing the maximum absolute value of the gradient,Is the derivative of the neuron activation function.

4) Based on the medium characteristics, the propagation and reflection of the seismic wave in the lost terrain are simulated, and the path and energy of the seismic wave are recorded.

Specifically, an initial seismic wave is first transmitted: the simulated seismic wave is transmitted at the current parameter location, i.e., its initial energy E ₀ is set.

Further, seismic wave propagation and reflection are simulated: based on the characteristics of the medium, the propagation and reflection of seismic waves between the loss curved surface and the underground layer are simulated. Specifically, as seismic waves propagate between media, their energy is partially reflected.

The reflection coefficient R is defined by the following formula: Wherein, the method comprises the steps of, wherein, Is the dielectric property of the i-th layer. The reflection energy of the seismic wave between the ith layer and the (i+1) th layer is as follows: ; further, the energy to continue to propagate is: . Further, the path and energy of the seismic wave after each reflection are recorded, and the energy attenuation of the seismic wave, i.e., the trend of change in energy E, is taken into account.

5) And determining a parameter updating path from the paths according to the energy, updating network parameters of the neural network based on the parameter updating path, and constructing a feature extraction model.

Determining a parameter updating path:

determining energy corresponding to the frequency based on a preset relational expression by calculating the frequency in the neural network optimization process; carrying out propagation of seismic waves according to the frequency, and calculating an updating loss function; determining an adaptability function corresponding to the updated loss function, and determining a target loss function according to the adaptability function; and determining the energy corresponding to the target loss function as target energy, and determining the path corresponding to the target energy as a parameter updating path.

Specifically, a path with minimum energy attenuation is selected as a parameter updating path according to propagation and reflection conditions of seismic waves in each underground layer.

Further, in order to define an fitness index, i.e. fitness function, that measures the "quality" of the neural network parametersCan be expressed as: ; where L is the current loss function value, L ₀ is a reference loss value (e.g., the loss value at initialization or the loss value of the previous round), Is a parameter for adjusting the steepness of the fitness function curve. The fitness function is characterized in that when L is close to or lower than L ₀, F is close to 1, which means that the network performs well; when L is significantly higher than L ₀, F approaches 0, meaning that the network performs poorly.The larger the fitness function the more sensitive the response.

Further, the frequency response is adjusted to take into account the propagation of the seismic wave in the earth's crust, and in addition to its energy propagation and attenuation, the frequency of the seismic wave is also a very important characteristic. In the optimization of neural networks, the response of such frequencies is simulated, which is used as another modulation means to guide the updating of parameters.

Specifically, let theIs a frequency response function, where f is the "frequency" in the neural network optimization process. The "frequency" here may be the rate of model parameter update, the fluctuation rate of the loss function value, etc. Specifically defined as: ; wherein, Is the loss function at timeIs used for the control of the degree of variation of (c),Is the corresponding time interval. Further, considering that high frequency seismic waves are more easily absorbed by the crust in real seismic propagation, the present invention expects to simulate this phenomenon in network optimization, which can be expressed as: ; wherein, Is a constant that determines the maximum magnitude of the response; is a regularization term that prevents zero denominator from being zero when f is zero.

Further, the present invention contemplates simulating the attenuation of high frequency seismic waves in the earth's crust, thus selecting the form of reciprocal square. That is, in order to obtain R (f), the physical meaning of frequency is considered, and the derivation is performed from the viewpoint of energy. In physics, the relationship between energy and frequency can be expressed as: ; where E is energy and hs is Planck constant.

In the present invention, to obtain a relationship inversely proportional to frequency, the relationship is expressed as: ; neglecting the constants, there are: 。

updating network parameters of the neural network:

The weights (w) and offsets (b) are updated according to the selected optimal path. The parameter update strategy simulates the propagation of seismic waves on the loss surface.

Further, to simulate the continuous propagation of seismic waves, a motion term v is introduced for the weights and biases, which can be expressed by the following equation:

；

。

Wherein, Is a momentum coefficient, and is usually valued between [0, 1); w and b are the weights and bias parameters of the neural network; And Is the loss gradient in weight and bias; And Momentum items updated for pre-update parameter weights and offsets; And Updated motion terms are weighted and biased for the updated parameters.

Further, the update formula of the weights and biases is:

；

。

Wherein, For energy below a certain threshold value,Is the learning rate.

In the parameter optimization process, the learning rate is also adjusted, specifically, the energy attenuation of the seismic waves is simulated, and the learning rate is gradually reduced. The energy attenuation of the earthquake wave is simulated, the learning rate is gradually reduced, and the updating mode of the learning rate can be expressed as follows:。

further, to simulate real seismic wave propagation, the seismic wave energy E should not be constant at all times. When the seismic wave continuously propagates on the loss curved surface, the energy of the seismic wave gradually attenuates, and the seismic wave can be expressed as: ; wherein, Is the attenuation coefficient, typically less than 1.Is the learning rate.

Further, to prevent E from decaying to 0 or too much, a limit is added above and below it, which can be expressed as: ; wherein, AndIs the minimum and maximum of energy.

Further, if the change in the loss function value is below a preset threshold, or a preset maximum number of iterations is reached, the algorithm stops.

Further, the current optimal weight (w) and bias (b) configuration is returned. After iteration is completed, the completion of training of the feature extraction model is indicated, and the target feature parameters are output through the current feature extraction model.

And step S306, performing data processing on the target characteristic parameters through a preset middle layer neural network to generate a global characteristic representation.

Specifically, the whole classifier is provided with a plurality of neural network layers, wherein the middle layer neural network and the depth multi-layer perceptron are used for describing, in the embodiment of the invention, the target characteristic parameters are processed through a plurality of processing layers (namely, the middle layer neural network), and in the embodiment of the invention, the middle layer neural network comprises a deep layer network, a recursion back-up state network and an adaptive weight adjustment layer. The network parameters respectively corresponding to the middle layer neural network and the depth multi-layer perceptron are initialized through an improved quantum coding mode.

According to the embodiment of the invention, the target characteristic parameters are sequentially input into a deep network, a recursion lift-state network and a self-adaptive weight adjustment layer to respectively obtain advanced characteristics, recursion output and adjustment layer output; based on the complexity corresponding to the advanced features, the recursion output and the adjustment layer output, respectively, dynamically adjusting the quantum states of the deep network, the recursion lifting state network and the self-adaptive weight adjustment layer; and according to the feedback of the quantum state dynamic adjustment, the network parameters of the subsequent neural network layer are adjusted; and fusing the advanced features, the recursion output and the adjustment layer output to obtain the global feature representation.

1. Specifically, the weights and biases of multiple layers of the model are initialized using the improved quantum encoding scheme can be expressed as:； . Where Q () represents a quantum encoding function, AndIs the initial weight and bias.

Further, for the quantum coding function, the invention adopts a mapping function of a high-dimensional Hilbert space to complete quantum coding, and can be expressed as follows: ; wherein, AndIs a fundamental function of the Hilbert space,Representing the tensor product of the quantum states.

2. Depth feature extraction:

Data first flows through multiple deep networks, each layer using high-order nonlinear transformations to transform raw features into advanced features. Specifically, the conversion of the original features using a high-order nonlinear transformation can be expressed as: ; wherein, A non-linear activation function is used,Is a weight matrix for feature extraction.

3. Quantum state dynamic adjustment:

the output of each layer is sent to a quantum state dynamic adjustment module, and the quantum state is dynamically adjusted according to the complexity of the data, which can be expressed as: ; wherein, Is another nonlinear activation function, used to adjust the quantum state of a feature,Is the weight matrix for that layer.

4. Recursive echo state network processing:

the quantum state dynamically tuned features are fed into a recursive echo state network. In which not only the current state of the data is utilized, but also the states of the previous moments are recursively considered. Specifically, considering the recursive feature, there are: ; wherein, The time step is indicated as such,Is an activation function of the echo state network,Is a matrix of weights that are to be used,Is the output of the last time step.

5. An adaptive weight adjustment layer:

the output of the recursive echo state network is used as an input to this layer, and then the weights of the subsequent neural network layers are automatically adjusted based on the feedback of the quantum state dynamic adjustment. Specifically, the output of this layer can be expressed as: ; wherein W _aw is a weight adjustment matrix based on quantum state dynamic adjustment feedback.

6. Multi-layer feature fusion:

fusing features obtained from all intermediate layers together yields a global feature representation, which can be expressed as: ; wherein, Representing the fusion operation of the features.

Step S308, inputting the global feature representation into a preset depth multi-layer perceptron to perform classification training.

Based on the global features, the data is sent to a deep multi-layer perceptron for final classification training, which can be expressed as: ; wherein, Is the activation function of the final classification layer,Is the weight matrix for that layer.

Furthermore, the embodiment of the invention also carries out the training of the classifier (namely the depth multi-layer perceptron) by combining the Bayesian decision theory.

Wherein, using bayesian decision theory in combination with the output of the model to determine the failure mode of the device can be expressed as: ; wherein, Is observation data; is the output of a certain failure mode. Here, bayesian decision theory is used, wherein, Is given dataPosterior probability of a device being in a certain failure mode,Is the likelihood probability that the user will be able to determine,Is the prior probability of being a priori,Is the evidence probability. Further, based on the output result, the classifier is optimized until a preset requirement is met, such as a maximum number of iterations is reached or the loss function falls to a set threshold.

And step S310, optimizing network parameters of the intermediate neural network and network parameters of the deep multi-layer perceptron by using a quantum optimization technology and Monte Carlo simulation.

Besides the traditional gradient descent method, the invention introduces a quantum optimization technology and Monte Carlo simulation to more efficiently find the optimal solution. Specifically, the model is optimized by using a quantum optimization technology and Monte Carlo simulation. In quantum optimization, a gradient descent method of quantum version is adopted, and can be expressed as follows: ; wherein, Is the rate of learning to be performed,Is based on the gradient computation of the quantum states,Is a loss function.AndIs the value of the weight parameter of the neural network in the new iteration and the old iteration.

For the monte carlo simulation, the acceptance probability is defined as: ; wherein, To accept the probability of a new solution in the monte carlo simulation.

Further, likelihood probability, prior probability and evidence probability in the Bayesian decision theory are defined, wherein the likelihood probability is the probability of generating observation data under a certain fault mode given by a model, namely:

; wherein, AndIs the mean and standard deviation of the data for a given failure mode.

Further, a priori probabilitiesIs the probability of a device being in a certain failure mode, typically based on historical data or expert experience. Evidence probabilityIs the probability of observing a certain data, and can be obtained by normalization.

And step S312, constructing a fault diagnosis model based on the depth multi-layer perceptron until a preset optimization threshold is met.

After model training is completed, fault diagnosis is carried out by using the trained model.

According to the method for diagnosing the faults of the filling mining equipment based on multidimensional data fusion, an improved generation countermeasure network is adopted to enhance data, so that data which is highly similar to real data can be generated even under the condition of a limited data set, and the generalization capability of a model is improved. And provides a neural network optimization algorithm based on seismic wave propagation: the optimal neural network parameter optimization path is found by simulating the propagation of seismic waves in complex terrain. And, design a high order neural network classifier based on improved quantum coding: quantum encoding is introduced to capture and represent complex relationships in the data and to enhance the expressive power of the model through higher order neural networks.

Based on the above, the embodiment of the invention can realize high-precision diagnosis: through the innovative technology, the invention can more accurately diagnose the equipment faults, thereby improving the safety and stability of the coal mine equipment. And has strong data expansion capability: even under the condition of limited data, enough samples can be generated, and the training stability and accuracy of the model are ensured. Faster model training is also possible: the optimization algorithm based on seismic wave propagation can find the optimal solution more quickly, and the training time of the model is shortened. In addition, has high sensitivity and response capability: the method combines the quantum coding, the high-order neural network and the model of the echo state network, can quickly respond to the tiny change and the abnormal state of the equipment, and ensures the real-time monitoring of the coal mine equipment.

Further, on the basis of the above method embodiment, the embodiment of the present invention further provides a fault diagnosis device for a filling and mining apparatus based on multidimensional data fusion, and fig. 4 shows a schematic structural diagram of the fault diagnosis device for a filling and mining apparatus based on multidimensional data fusion provided by the embodiment of the present invention, as shown in fig. 4, where the device includes: the data acquisition module 100 is used for acquiring monitoring data of the target filling and mining equipment in the running process; the monitoring data comprise state data of the target filling and mining equipment and operation data of the target filling and mining equipment, wherein the operation data are acquired by sensors with multiple dimensions; the data processing module 200 is used for assembling the monitoring data to generate a vector to be detected; the execution module 300 is configured to input a vector to be tested into a pre-constructed fault diagnosis model, classify the vector to be tested through the fault diagnosis model, and determine a classification result; the fault diagnosis model is constructed based on an improved quantum coding high-order neural network classifier algorithm, and a training set for training the fault diagnosis model comprises a multi-dimensional diagnosis sample, wherein the multi-dimensional diagnosis sample comprises state data of filling and mining equipment, operation data of the filling and mining equipment, acquired by sensors with multiple dimensions, and corresponding sample labels; and an output module 400 for performing fault diagnosis on the target filling and mining equipment based on the classification result.

The filling and mining equipment fault diagnosis device based on multidimensional data fusion provided by the embodiment of the invention has the same technical characteristics as the filling and mining equipment fault diagnosis method based on multidimensional data fusion provided by the embodiment, so that the same technical problems can be solved, and the same technical effects can be achieved.

Further, on the basis of the foregoing embodiment, the embodiment of the present invention further provides another filling and mining equipment fault diagnosis device based on multidimensional data fusion, and fig. 5 shows a schematic structural diagram of another filling and mining equipment fault diagnosis device based on multidimensional data fusion provided by the embodiment of the present invention, where the execution module 300 is further configured to obtain a pre-constructed training set; extracting features of the training set through a pre-constructed feature extraction model, and determining target feature parameters; performing data processing on the target characteristic parameters through a preset middle-layer neural network to generate a global characteristic representation; inputting the global feature representation into a preset depth multi-layer perceptron to perform classification training; initializing network parameters respectively corresponding to the middle layer neural network and the depth multi-layer perceptron through an improved quantum coding mode; optimizing network parameters of the intermediate neural network and network parameters of the deep multi-layer perceptron by using a quantum optimization technology and Monte Carlo simulation; and constructing a fault diagnosis model based on the depth multi-layer perceptron until a preset optimization threshold is met.

Further, the middle layer neural network comprises a deep layer network, a recursion back-rising state network and an adaptive weight adjustment layer; the execution module 300 is further configured to sequentially input the target feature parameters into a deep network, a recursive rising state network, and an adaptive weight adjustment layer, to obtain an advanced feature, a recursive output, and an adjustment layer output, respectively; based on the complexity corresponding to the advanced features, the recursion output and the adjustment layer output, respectively, dynamically adjusting the quantum states of the deep network, the recursion lifting state network and the self-adaptive weight adjustment layer; and according to the feedback of the quantum state dynamic adjustment, the network parameters of the subsequent neural network layer are adjusted; and fusing the advanced features, the recursion output and the adjustment layer output to obtain the global feature representation.

The apparatus further includes a construction module 500 for collecting a multi-dimensional diagnostic sample of the filling mining device; sample labeling is carried out according to the equipment working state corresponding to the multi-dimensional diagnosis sample, and a sample label is generated; constructing an initial sample set based on the sample tag and the multi-dimensional diagnostic sample; randomly sampling from an initial sample set to obtain a sampling sample, and distributing sample weight to the sampling sample according to data distribution corresponding to the sampling sample; performing data expansion on the initial sample set through a preset sample expansion algorithm to generate an expansion sample; updating the sample weight based on the mass fraction of the expanded sample, and optimizing a sample expansion algorithm; until the mass fraction of the expanded sample meets the preset requirement, a training set is constructed based on the expanded sample and the initial sample set.

The building block 500 is further configured to generate a counter network algorithm by using a sample expansion algorithm; generating a countermeasure network algorithm, generating samples through a generator, and discriminating the samples through a discriminator; a step of updating the sample weights based on the mass fraction of the expanded samples, comprising: judging the expansion sample generated by the generator through a discriminator for generating an countermeasure network algorithm, and determining a judging result; calculating a loss function of the discriminator based on the discrimination result; and updating the sample weight according to the loss function.

The above construction module 500 is further configured to calculate a proportion of each type of sample in the extended samples, and adjust parameters of the generator according to the proportion; wherein parameters of the generator are updated using quasi-newton methods; based on the discrimination result, the parameters of the generator and the parameters of the discriminator are subjected to joint optimization; the method comprises the steps of performing parameter updating on the basis of joint optimization and joint loss function, and calculating the joint loss function on the basis of a preset quantum state; and optimizing the learning strategies of the generator and the discriminator by using a preset meta learning method.

The execution module 300 is further configured to obtain a pre-constructed training set; inputting the training set into a preset neural network, and calculating an initial loss function; constructing a multi-layer loss topography based on the gradient of the initial loss function, and distributing medium characteristics for each layer of loss topography; based on medium characteristics, simulating the propagation and reflection of seismic waves on lost terrains, and recording the paths and energy of the seismic waves; and determining a parameter updating path from the paths according to the energy, updating network parameters of the neural network based on the parameter updating path, and constructing a feature extraction model.

The execution module 300 is further configured to calculate a frequency in the neural network optimization process, and determine energy corresponding to the frequency based on a preset relational expression; carrying out propagation of seismic waves according to the frequency, and calculating an updating loss function; determining an adaptability function corresponding to the updated loss function, and determining a target loss function according to the adaptability function; and determining the energy corresponding to the target loss function as target energy, and determining the path corresponding to the target energy as a parameter updating path.

The output module 400 is further configured to classify the vector to be tested by using a fault diagnosis model, and determine an initial output; and combining the initial output with a Bayesian decision theory to obtain a classification result corresponding to the vector to be detected.

The embodiment of the invention also provides an electronic device, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the computer program to realize the steps of the method shown in the figures 1 to 3. The embodiments of the present invention also provide a computer readable storage medium having a computer program stored thereon, which when executed by a processor performs the steps of the method shown in fig. 1 to 3 described above.

The embodiment of the present invention further provides a schematic structural diagram of an electronic device, as shown in fig. 6, where the electronic device includes a processor 61 and a memory 60, where the memory 60 stores computer executable instructions that can be executed by the processor 61, and the processor 61 executes the computer executable instructions to implement the methods shown in fig. 1 to 3. In the embodiment shown in fig. 6, the electronic device further comprises a bus 62 and a communication interface 63, wherein the processor 61, the communication interface 63 and the memory 60 are connected by means of the bus 62.

The memory 60 may include a high-speed random access memory (RAM, random Access Memory), and may further include a non-volatile memory (non-volatile memory), such as at least one magnetic disk memory. The communication connection between the system network element and at least one other network element is achieved via at least one communication interface 63 (which may be wired or wireless), and may use the internet, a wide area network, a local network, a metropolitan area network, etc. Bus 62 may be an ISA (Industry Standard Architecture ) Bus, PCI (PERIPHERAL COMPONENT INTERCONNECT, peripheral component interconnect standard) Bus, EISA (Extended Industry Standard Architecture ) Bus, etc., or AMBA (Advanced Microcontroller Bus Architecture, standard for on-chip buses) Bus, where AMBA defines three types of buses, including an APB (ADVANCED PERIPHERAL Bus) Bus, an AHB (ADVANCED HIGH-performance Bus) Bus, and a AXI (Advanced eXtensible Interface) Bus. The bus 62 may be classified as an address bus, a data bus, a control bus, or the like. For ease of illustration, only one bi-directional arrow is shown in FIG. 6, but not only one bus or type of bus.

The processor 61 may be an integrated circuit chip with signal processing capabilities. In implementation, the steps of the above method may be performed by integrated logic circuits of hardware in the processor 61 or by instructions in the form of software. The processor 61 may be a general-purpose processor, including a central processing unit (Central Processing Unit, CPU for short), a network processor (Network Processor, NP for short), and the like; but may also be a digital signal Processor (DIGITAL SIGNAL Processor, DSP), application Specific Integrated Circuit (ASIC), field-Programmable gate array (FPGA) or other Programmable logic device, discrete gate or transistor logic device, discrete hardware components. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the method disclosed in connection with the embodiments of the present application may be embodied directly in the execution of a hardware decoding processor, or in the execution of a combination of hardware and software modules in a decoding processor. The software modules may be located in a random access memory, flash memory, read only memory, programmable read only memory, or electrically erasable programmable memory, registers, etc. as well known in the art. The storage medium is located in a memory and the processor 61 reads the information in the memory and in combination with its hardware performs the method shown in any of the foregoing figures 1 to 3.

The embodiment of the invention provides a method and a device for diagnosing faults of filling and mining equipment based on multidimensional data fusion, which comprises a computer readable storage medium storing program codes, wherein the program codes comprise instructions for executing the method described in the embodiment of the method, and specific implementation can be seen in the embodiment of the method and is not repeated herein. It will be clear to those skilled in the art that, for convenience and brevity of description, reference may be made to the corresponding process in the foregoing method embodiment for the specific working process of the above-described system, which is not described herein again. In addition, in the description of embodiments of the present invention, unless explicitly stated and limited otherwise, the terms "mounted," "connected," and "connected" are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. The specific meaning of the above terms in the present invention will be understood by those skilled in the art in specific cases.

The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

In the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. are based on the directions or positional relationships shown in the drawings, are merely for convenience of describing the present invention and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and thus should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.

Finally, it should be noted that: the above examples are only specific embodiments of the present invention for illustrating the technical solution of the present invention, but not for limiting the scope of the present invention, and although the present invention has been described in detail with reference to the foregoing examples, it will be understood by those skilled in the art that the present invention is not limited thereto: any person skilled in the art may modify or easily conceive of the technical solution described in the foregoing embodiments, or perform equivalent substitution of some of the technical features, while remaining within the technical scope of the present disclosure; such modifications, changes or substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention, and are intended to be included in the scope of the present invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.

Claims

1. A filling mining equipment fault diagnosis method based on multidimensional data fusion is characterized by comprising the following steps:

Acquiring monitoring data of the target filling and mining equipment in the operation process; the monitoring data comprise state data of the target filling and mining equipment and operation data of the target filling and mining equipment, wherein the state data of the target filling and mining equipment and the operation data of the target filling and mining equipment are acquired by sensors with multiple dimensions;

assembling the monitoring data to generate a vector to be detected;

Inputting the vector to be detected into a pre-constructed fault diagnosis model, classifying the vector to be detected through the fault diagnosis model, and determining a classification result; the fault diagnosis model is constructed based on an improved quantum coding high-order neural network classifier algorithm, and a training set for training the fault diagnosis model comprises a multi-dimensional diagnosis sample, wherein the multi-dimensional diagnosis sample comprises state data of filling and mining equipment, operation data of the filling and mining equipment, acquired by sensors with multiple dimensions, and corresponding sample labels;

performing fault diagnosis on the target filling and mining equipment based on the classification result;

Wherein, the step of constructing the fault diagnosis model based on the improved quantum coding high-order neural network classifier algorithm comprises the following steps:

acquiring a pre-constructed training set;

performing feature extraction on the training set through a pre-constructed feature extraction model, and determining target feature parameters;

Performing data processing on the target characteristic parameters through a preset middle-layer neural network to generate a global characteristic representation;

inputting the global feature representation into a preset depth multi-layer perceptron to perform classification training; the network parameters respectively corresponding to the middle layer neural network and the depth multi-layer perceptron are initialized through an improved quantum coding mode;

Optimizing network parameters of the middle layer neural network and network parameters of the depth multilayer perceptron by using a quantum optimization technology and Monte Carlo simulation;

until a preset optimization threshold is met, constructing a fault diagnosis model based on the depth multi-layer perceptron;

the middle layer neural network comprises a deep layer network, a recursion lifting state network and a self-adaptive weight adjusting layer;

The step of generating a global feature representation by performing data processing on the target feature parameters through a preset middle layer neural network comprises the following steps:

Sequentially inputting the target characteristic parameters into a deep network, a recursion lifting state network and a self-adaptive weight adjustment layer to respectively obtain advanced characteristics, recursion output and adjustment layer output;

based on the complexity corresponding to the advanced features, the recursion output and the adjustment layer output, respectively dynamically adjusting the quantum states of the deep network, the recursion back-rising state network and the self-adaptive weight adjustment layer; and according to the feedback of the quantum state dynamic adjustment, the network parameters of the subsequent neural network layer are adjusted;

And fusing the advanced features, the recursion output and the adjustment layer output to obtain global feature representation.

2. The method according to claim 1, wherein the training set constructing method comprises:

collecting a multi-dimensional diagnosis sample of filling mining equipment;

Sample labeling is carried out according to the equipment working state corresponding to the multi-dimensional diagnosis sample, and a sample label is generated;

constructing an initial sample set based on the sample tag and the multi-dimensional diagnostic sample;

Randomly sampling from the initial sample set to obtain a sampling sample, and distributing sample weight to the sampling sample according to data distribution corresponding to the sampling sample;

performing data expansion on the initial sample set through a preset sample expansion algorithm to generate an expansion sample;

updating the sample weight based on the mass fraction of the extended sample, and optimizing the sample extension algorithm;

Until the mass fraction of the extended sample meets the preset requirement, a training set is constructed based on the extended sample and the initial sample set.

3. The method of claim 2, wherein the sample expansion algorithm comprises generating a antagonism network algorithm; the generating countermeasure network algorithm generates samples through a generator and judges the samples through a judging device;

a step of updating the sample weights based on the mass scores of the extended samples, comprising:

Judging the expansion sample generated by the generator through a discriminator for generating an countermeasure network algorithm, and determining a judging result;

calculating a loss function of the discriminator based on the discrimination result;

Updating the sample weight based on the loss function.

4. A method according to claim 3, wherein the step of optimizing the sample expansion algorithm comprises:

Calculating the proportion of each type of sample in the extended samples, and adjusting the parameters of the generator according to the proportion; wherein parameters of the generator are updated using quasi-newton methods;

based on the discrimination result, carrying out joint optimization on the parameters of the generator and the parameters of the discriminator; the joint optimization is based on joint loss function to update parameters, and the joint loss function is based on preset quantum state calculation;

and optimizing the learning strategies of the generator and the discriminator by using a preset meta learning method.

5. The method according to claim 1, wherein the method for constructing the feature extraction model includes:

acquiring a pre-constructed training set;

Inputting the training set into a preset neural network, and calculating an initial loss function;

constructing a plurality of layers of loss topography based on the gradient of the initial loss function, and distributing medium characteristics for each layer of loss topography;

simulating the propagation and reflection of a seismic wave on the lost terrain based on the medium characteristics, and recording the path and energy of the seismic wave;

and determining a parameter updating path from the paths according to the energy, updating network parameters of the neural network based on the parameter updating path, and constructing a feature extraction model.

6. The method of claim 5, wherein the step of determining a parameter update path from the path based on the energy comprises:

calculating the frequency in the neural network optimization process, and determining the energy corresponding to the frequency based on a preset relational expression;

according to the frequency, carrying out the propagation of the seismic wave, and calculating an updating loss function;

determining a fitness function corresponding to the updated loss function, and determining a target loss function according to the fitness function;

and determining the energy corresponding to the target loss function as target energy, and determining a path corresponding to the target energy as a parameter updating path.

7. The method according to claim 1, wherein the step of inputting the vector to be measured into a pre-constructed fault diagnosis model, classifying the vector to be measured by the fault diagnosis model, and determining a classification result includes:

classifying the vector to be detected through the fault diagnosis model, and determining initial output;

and combining the initial output with a Bayesian decision theory to obtain a classification result corresponding to the vector to be detected.

8. A filling and mining equipment fault diagnosis device based on multidimensional data fusion, which is characterized by comprising:

The data acquisition module is used for acquiring monitoring data of the target filling mining equipment in the running process; the monitoring data comprise state data of the target filling and mining equipment and operation data of the target filling and mining equipment, wherein the state data of the target filling and mining equipment and the operation data of the target filling and mining equipment are acquired by sensors with multiple dimensions;

the data processing module is used for assembling the monitoring data to generate a vector to be detected;

The execution module is used for inputting the vector to be detected into a pre-constructed fault diagnosis model, classifying the vector to be detected through the fault diagnosis model, and determining a classification result; the fault diagnosis model is constructed based on an improved quantum coding high-order neural network classifier algorithm, and a training set for training the fault diagnosis model comprises a multi-dimensional diagnosis sample, wherein the multi-dimensional diagnosis sample comprises state data of filling and mining equipment, operation data of the filling and mining equipment, acquired by sensors with multiple dimensions, and corresponding sample labels;

The output module is used for carrying out fault diagnosis on the target filling and mining equipment based on the classification result;

The execution module is also used for acquiring a pre-constructed training set; performing feature extraction on the training set through a pre-constructed feature extraction model, and determining target feature parameters; performing data processing on the target characteristic parameters through a preset middle-layer neural network to generate a global characteristic representation; inputting the global feature representation into a preset depth multi-layer perceptron to perform classification training; the network parameters respectively corresponding to the middle layer neural network and the depth multi-layer perceptron are initialized through an improved quantum coding mode; optimizing network parameters of the middle layer neural network and network parameters of the depth multilayer perceptron by using a quantum optimization technology and Monte Carlo simulation; until a preset optimization threshold is met, constructing a fault diagnosis model based on the depth multi-layer perceptron;

The middle layer neural network comprises a deep layer network, a recursion lifting state network and a self-adaptive weight adjusting layer; the execution module is further used for sequentially inputting the target characteristic parameters into a deep network, a recursion lifting state network and a self-adaptive weight adjustment layer to respectively obtain advanced characteristics, recursion output and adjustment layer output; based on the complexity corresponding to the advanced features, the recursion output and the adjustment layer output, respectively dynamically adjusting the quantum states of the deep network, the recursion back-rising state network and the self-adaptive weight adjustment layer; and according to the feedback of the quantum state dynamic adjustment, the network parameters of the subsequent neural network layer are adjusted; and fusing the advanced features, the recursion output and the adjustment layer output to obtain global feature representation.