CN117892251B - Rigging forging process parameter monitoring and early warning method and device based on artificial intelligence - Google Patents

Abstract

The invention provides a method and a device for monitoring and early warning of a forging technological parameter of a rigging based on artificial intelligence, which relate to the technical field of data processing. The parameter monitoring model is constructed based on a classifier model constructed by a method of determining anchor points, a feature extraction model constructed by a neural network algorithm based on multi-direction communication simulation particle swarm optimization is used for carrying out feature extraction on a training sample set of a training classifier model, and a data dimension reduction model is used for carrying out dimension reduction on the training sample set. The invention can effectively process and analyze a large number of complex forging process parameters, improves the efficiency and accuracy of data processing, so as to more accurately predict potential risks and anomalies and ensure the safety and efficiency of the forging process.

Description

Rigging forging process parameter monitoring and early warning method and device based on artificial intelligence

Technical Field

The invention relates to the technical field of data processing, in particular to a rigging forging process parameter monitoring and early warning method and device based on artificial intelligence.

Background

Rigging forging is a complex process involving extreme conditions of high temperature, high pressure, etc., and its quality control is critical to the performance and reliability of the product. Conventional forging processes rely on experienced operators and some basic automated equipment to control process parameters such as temperature, pressure, cooling rate and material properties. However, due to the complexity and variability of the forging process, such conventional methods often have difficulty capturing subtle changes in all critical parameters, and thus are not effective in predicting and protecting against potential risks and anomalies.

With the advent of intelligent manufacturing, new requirements and challenges are presented for control and monitoring of forging processes in combination with advanced data analysis and machine learning techniques. In particular in the field of data acquisition and analysis, the limitations of conventional methods are emerging. The prior art often cannot effectively process a large amount of complex data, and particularly has defects in aspects of feature extraction and data dimension reduction. In addition, traditional classifiers tend to be computationally complex and inefficient in the face of high dimensional and nonlinear data. Therefore, a new technical scheme is urgently needed to improve the accuracy and efficiency of data processing and enhance the prediction capability of potential risks and abnormal states in the forging process.

Disclosure of Invention

Therefore, the invention aims to provide the rigging forging process parameter monitoring and early warning method and device based on artificial intelligence, which can more accurately predict the potential risk and abnormality and ensure the safety and efficiency of the forging process.

In a first aspect, an embodiment of the present invention provides a method for monitoring and early warning parameters of a rigging forging process based on artificial intelligence, the method comprising: obtaining forging technological parameters of a target rigging; the forging process parameters comprise parameters for monitoring the target rigging in real time in a rigging forging process; converting forging technological parameters into data vectors, inputting the data vectors into a pre-constructed parameter monitoring model, and outputting parameter monitoring results; the parameter monitoring model is built based on a preset classifier model, a training sample set for training the classifier model is used for carrying out feature extraction through a preset feature extraction model, and dimension reduction is carried out through a preset data dimension reduction model; the feature extraction model is constructed based on a neural network algorithm of multi-directional communication simulation particle swarm optimization, and the classifier model is constructed based on a method for determining anchor points; and predicting the potential risk and abnormal state of the target rigging in the forging process according to the parameter monitoring result.

In a second aspect, an embodiment of the present invention provides an artificial intelligence-based monitoring and early warning device for parameters of a rigging forging process, the device comprising: the data acquisition module is used for acquiring forging process parameters of the target rigging; the forging process parameters comprise parameters for monitoring the target rigging in real time in a rigging forging process; the execution module is used for converting forging process parameters into data vectors, inputting the data vectors into a pre-constructed parameter monitoring model and outputting parameter monitoring results; the parameter monitoring model is built based on a preset classifier model, a training sample set for training the classifier model is extracted through a preset feature extraction model, and dimension reduction is performed through a preset data dimension reduction model; the feature extraction model is constructed based on a neural network algorithm of multi-directional communication simulation particle swarm optimization, and the classifier model is constructed based on a method for determining anchor points; and the output module is used for predicting potential risks and abnormal states of the target rigging in the forging process according to the parameter monitoring result.

The embodiment of the invention has the following beneficial effects: according to the rigging forging process parameter monitoring and early warning device based on artificial intelligence, which is provided by the embodiment of the invention, the forging process parameters are identified through the pre-constructed parameter monitoring model, and the parameter monitoring result is output. The parameter monitoring model is built based on a preset classifier model, and the classifier model is built based on a method for determining anchor points, so that the calculation efficiency and accuracy of the classifier can be improved. The training sample set for training the classifier model is subjected to feature extraction through a preset feature extraction model, and dimension reduction processing is performed through a preset data dimension reduction model; the feature extraction model is constructed based on a neural network algorithm of multi-direction communication simulation particle swarm optimization, the accuracy and optimizing efficiency of feature extraction can be improved, the classifier model is trained after the feature extraction model and the data dimension reduction model are respectively processed, the high-dimension data after feature extraction is changed into a more compact and efficient low-dimension representation after the feature dimension reduction processing, the classification accuracy is improved, the potential risk and abnormality can be predicted more accurately based on the feature extraction model and the data dimension reduction model, and the safety and efficiency of the forging process are ensured. Based on the method, a large number of complex forging process parameters can be effectively processed and analyzed, the efficiency and accuracy of data processing are improved, potential risks and anomalies are predicted more accurately, and the safety and efficiency of the forging process are ensured.

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 method for monitoring and early warning of parameters of a rigging forging process based on artificial intelligence provided by an embodiment of the invention;

FIG. 2 is a flowchart of another method for monitoring and early warning parameters of a rigging forging process based on artificial intelligence according to an embodiment of the present invention;

FIG. 3 is a flowchart of another method for monitoring and early warning parameters of a rigging forging process based on artificial intelligence according to an embodiment of the present invention;

FIG. 4 is a flowchart of another method for monitoring and early warning parameters of a rigging forging process based on artificial intelligence according to an embodiment of the present invention;

FIG. 5 is a schematic structural diagram of a monitoring and early warning device for parameters of a rigging forging process based on artificial intelligence according to an embodiment of the invention;

FIG. 6 is a schematic structural diagram of another monitoring and early warning device for parameters of a rigging forging process based on artificial intelligence according to an embodiment of the present invention;

fig. 7 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 herein may be embodied in a wide variety of forms and that any specific structure and/or function described herein is merely illustrative. Based on the present disclosure, one skilled in the art will appreciate that one aspect described herein may be implemented independently of any other aspect, and that two or more of these aspects may be combined in various ways. For example, an apparatus may be implemented and/or a method practiced using any number of the aspects set forth herein. In addition, such apparatus may be implemented and/or such methods practiced using other structure and/or functionality in addition to one or more of the aspects set forth herein.

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.

Machine learning is a sub-area of Artificial Intelligence (AI) that gives computer systems the ability to learn empirically and improve their performance without requiring explicit programming. This means that the machine learning model can automatically identify patterns and rules in the data and use this knowledge to make decisions or predictions. The following are some technical data of machine learning relevant to the present invention:

Data driving: machine learning relies on large amounts of data to train a model. The data may be labeled (supervised learning), unlabeled (unsupervised learning), or obtained through interaction with the environment (reinforcement learning).

And (3) model: in machine learning, models are mathematical representations of real world problems that can be learned from data. Common models include decision trees, neural networks, support vector machines, and the like.

Learning algorithm: the learning algorithm defines how the model is adjusted or "learned". The algorithm runs on the data and optimizes according to the behavior of the model, for example, by minimizing the difference between the predicted and actual results.

Evaluation: the performance of the model needs to be measured by some form of evaluation. This typically involves dividing the data into a training set for learning and a test set for evaluating the ability of the model to generalize to unseen data.

Overfitting and underfilling: overfitting means that the model performs well on training data, but cannot be generalized to new data. Under-fitting means that the model does not learn enough features of the data to be able to predict efficiently.

And (3) supervised learning: this type of machine learning involves training a model using a set of labeled samples, i.e., input data paired with expected output. Common supervised learning tasks include classification and regression.

According to the rigging forging process parameter monitoring and early warning method and device based on artificial intelligence, which are provided by the embodiment of the invention, the potential risk and abnormality can be predicted more accurately, and the safety and efficiency of the forging process are ensured.

For the convenience of understanding the present embodiment, first, a detailed description is given of a method for monitoring and early warning parameters of a rigging forging process based on artificial intelligence disclosed in the present embodiment, fig. 1 shows a flowchart of a method for monitoring and early warning parameters of a rigging forging process based on artificial intelligence provided in the present embodiment, and as shown in fig. 1, the method includes the following steps:

step S102, obtaining forging technological parameters of the target rigging.

Step S104, converting the forging process parameters into data vectors, inputting the data vectors into a pre-constructed parameter monitoring model, and outputting parameter monitoring results.

And S106, predicting the potential risk and abnormal state of the target rigging in the forging process according to the parameter monitoring result.

In specific implementation, the forging process parameters of the embodiment of the invention comprise parameters for monitoring the target rigging in real time in a rigging forging process; in one embodiment, including but not limited to temperature, pressure, cooling rate, material properties, and the like. These parameters are input in vector form and can be expressed as: . Further, the forging process parameters are identified through a pre-constructed parameter monitoring model, and a parameter monitoring result is output.

In specific implementation, the parameter monitoring model is trained based on sample labels, the sample labels are used for indicating classification categories of samples, and corresponding parameter monitoring results comprise classification categories corresponding to data vectors. Specifically, the classification category indicated by the parameter monitoring result is determined, and then the state indicated by the classification category is determined as the abnormal state of the target rigging in the forging process. The classification category includes category numbers such as 1, 2, 3; corresponding, 1-normal operating state, 2-abnormal state, 3-potentially risky state. Further, the potential risk and abnormal state of the target rigging in the forging process can be predicted according to the classification, and an early warning result is generated, where in an embodiment, the early warning result includes: 1. abnormal temperature, 2, insufficient pressure, 3, excessive pressure, 4, shape deviation.

The parameter monitoring model is built based on the preset classifier model, and the classifier model is built based on the anchor point determining method, so that the calculation efficiency and accuracy of the classifier can be improved. The training sample set for training the classifier model is subjected to feature extraction through a preset feature extraction model, and dimension reduction processing is performed through a preset data dimension reduction model; the feature extraction model is constructed based on a neural network algorithm of multi-direction communication simulation particle swarm optimization, the accuracy and optimizing efficiency of feature extraction can be improved, the classifier model is trained after the feature extraction model and the data dimension reduction model are respectively processed, the high-dimension data after feature extraction is changed into a more compact and efficient low-dimension representation after the feature dimension reduction processing, the classification accuracy is improved, the potential risk and abnormality can be predicted more accurately based on the feature extraction model and the data dimension reduction model, and the safety and efficiency of the forging process are ensured.

Furthermore, on the basis of the embodiment, the embodiment of the invention also provides another monitoring and early warning method for parameters of the rigging forging process based on artificial intelligence. Further, the prior art may suffer from the following drawbacks: 1. the prior art may lack sufficient data volume, or the data quality is not high, limiting the training and predictive capabilities of the model. 2. Existing methods may have limitations in extracting useful features from complex data, especially when dealing with high-dimensional and non-linear data. 3. In the prior art, feature dimension reduction may be inefficient and non-linear features cannot be effectively handled. 4. Existing classifiers, such as conventional support vector machines, may suffer from high computational complexity and inefficiency in processing large-scale or high-dimensional data.

Correspondingly, the embodiment of the invention provides the following technology for constructing a parameter monitoring model: 1. the generation countermeasure network based on the random gradient Hamiltonian Monte Carlo is utilized, the problem of insufficient training data is effectively solved, and the generalization capability and precision of the model are improved. 2. The neural network algorithm based on multi-direction communication simulation particle swarm optimization is adopted for feature extraction, and the accuracy and the optimizing efficiency of feature extraction are improved. 3. The self-coding neural network based on the local sensitive hash is utilized for feature dimension reduction, the dimension reduction efficiency and precision are improved, and the compact and efficient representation of data is ensured. 4. The anchor point-based support vector machine algorithm is applied, so that the number of support vectors is effectively reduced, and the calculation efficiency and accuracy of the classifier are improved. In specific implementation, the embodiment of the invention performs feature extraction on a pre-built training sample set through a preset feature extraction model, performs dimension reduction through a preset data dimension reduction model, trains a classifier model, and builds a parameter monitoring model according to the trained classifier model.

Firstly, a method for constructing a feature extraction model is described, and in a specific implementation, a preset training sample set is input into a feature extraction module for feature extraction. The invention provides a neural network algorithm for simulating particle swarm optimization based on multi-directional communication, which simulates the searching behavior of the particle swarm, and enhances the information exchange among particles through a multi-directional communication mechanism, thereby improving the optimizing efficiency and accuracy. In addition, each particle not only exchanges information with the global optimal solution, but also shares information with adjacent particles, so that the exploration capability of the algorithm is improved.

Fig. 2 shows a flowchart of another method for monitoring and early warning parameters of a rigging forging process based on artificial intelligence according to an embodiment of the present invention, and mainly describes steps for constructing a feature extraction model with reference to fig. 2, wherein the method includes the following steps:

Step S11, a pre-constructed training sample set and a pre-set neural network are obtained.

Firstly, a training sample set is obtained and is used as input data to be input into a preset neural network for training, and the input data in the embodiment of the invention is parameters of a rigging forging process, and the embodiment can be referred to specifically. According to the embodiment of the invention, the initial sample is acquired in advance, and the training sample set is constructed after the initial sample is marked. The data labeling mode of the embodiment of the invention comprises the following steps: and manually marking by a professional technician to generate a sample label. The labeling category comprises normal operation state, abnormal state and potential risk state. And constructing an initial sample set according to the sample label and the initial sample. The data source of the invention is derived from various parameters in the process of the rigging forging technology, and relates to temperature, pressure, time and material properties. These data are collected from the sensor and monitoring system in a data format that is structured vector data, each data point representing an instantaneous state of the process. Further, the acquisition mode comprises real-time monitoring and historical record analysis. Real-time monitoring relies on a sensor array to continuously track process parameter changes; historian analysis provides additional depth and context to the dataset by analyzing past production data.

In one embodiment, the attributes of the data include, are set upIs a data vector for a rigging forging process, wherein: /(I): Forging temperature, which represents a temperature value during forging; /(I): A pressure index representing a pressure level at the time of forging; /(I): Duration, which refers to the length of time during forging; /(I): The hardness of the material reflects the hardness characteristics of the material used; /(I): The cooling rate represents the cooling rate of the material after forging; /(I): Chemical composition ratio, indicating the ratio of each element in the material; /(I): Shape deviation, namely measuring the deviation degree of the shape of the forged product from an ideal standard; /(I): Surface roughness, reflecting the smoothness of the product surface; /(I): Acoustic emission signals, monitoring acoustic wave characteristics in the forging process; /(I): Vibration data for analyzing the device operating state and process stability. It should be emphasized that in practical applications, the attributes of the data may be far more than the number of attributes listed in this embodiment, and may typically reach tens or hundreds.

It can be understood that in the task of the invention, the acquisition and labeling of the training data are time-consuming and labor-consuming, and the insufficient training samples easily cause poor generalization capability of the model, and simultaneously influence the precision of the model. The invention provides a random gradient Hamiltonian Monte Carlo-based generation countermeasure network algorithm to solve the problems that training data are insufficient and high-cost data are acquired, so that generalization capability and precision of a model are improved. Specifically, the embodiment of the invention also uses a preset generation countermeasure network to carry out sample expansion on the initial sample set so as to obtain an expanded sample; evaluating the expansion sample, and performing iterative optimization on the generated countermeasure network according to the evaluation result; and constructing a training sample set according to the current expansion sample and the initial sample set until the expansion sample meets the preset evaluation requirement. The extended samples of the embodiment of the invention are generated by a generator by using a random gradient hamilton Monte Carlo method. Further, conventional data expansion techniques such as rotation, flipping, scaling, noise injection, etc. may also be used.

Specifically, the generating countermeasure network structure based on the random gradient Hamiltonian Monte Carlo in the embodiment of the invention comprises two parts: a generator and a arbiter. The generator is responsible for generating new data samples, and the arbiter judges whether the samples come from the real data set, namely: the generator (G) can more effectively explore a data space by using a random gradient Hamiltonian Monte Carlo method to generate synthetic data with more similar real data distribution, and the introduction of the random gradient Hamiltonian Monte Carlo allows the generator to consider uncertainty in the data generation process, so that the data diversity is enhanced. The aim of the arbiter (D) is to distinguish between real data and data produced by the generator, in which the arbiter is improved by the Laplace approximation, increasing the sensitivity to fine features, so that the quality of the produced data can be evaluated more accurately.

The inputs to the generation countermeasure network based on the random gradient hamilton monte carlo are: raw data, i.e., vector data or discrete data collected during the rig forging process, includes, but is not limited to, temperature, pressure, time, material properties, and the like. The data format is a structured data vector, which in one embodiment may be represented asIncluding forging temperature, pressure index, duration, etc.

The output of the generation countermeasure network based on the random gradient hamilton monte carlo is as follows: the data is augmented, i.e., by generating a forging process parameter dataset that is enhanced against the network algorithm, to improve the training effect and prediction accuracy of the subsequent model. Specifically, the training process of generating the countermeasure network based on the random gradient Hamiltonian Monte Carlo is as follows:

1) And determining potential energy functions, momentums and energy matrixes corresponding to the initial sample set based on preset generator parameters.

2) An initial extended sample is generated from the potential energy function, momentum, and energy matrix.

First, initializing: setting initial parameters of generator and discriminator, specifically, initializing generatorSum discriminator/>In the form of parameters, the generator parameter initialization and the arbiter parameter initialization can be expressed as:

further, random gradient hamilton monte carlo samples are generated, and initial extended samples are generated: the generator generates data samples using a random gradient hamiltonian monte carlo method, and in particular, according to the hamiltonian dynamics principle, can be expressed as:

Wherein, As a Hamiltonian energy function,/>Is a potential energy function,/>Is momentum,/>Is a quality matrix. The quality matrix is preset by human beings, and in one embodiment, the quality matrix is an identity matrix.

Further, the manner of gradient update can be expressed as:

Wherein, Is the step size,/>Is the coefficient of friction,/>Is a standard normal distribution. /(I)Is a noise injection function. In one embodiment, step size/>Is set to 0.01. Coefficient of friction/>Set to 0.1.

Further, potential energy functionThe calculation of (2) can be expressed as:

Wherein, Representing a given parameter/>Time data/>Conditional probability of/>Is an a priori distribution of parameters.

Further, momentumThe update style of (c) can be expressed as:

Wherein, Is a momentum decay parameter,/>Is learning rate,/>Is the noise covariance. Momentum/>The Hamiltonian energy function is influenced, and the Hamiltonian energy function is used in generating a contrast network, and the Hamiltonian energy function influences the movement of a generator in a data space by describing energy distribution, and the generator generates samples similar to real data distribution as much as possible by minimizing the Hamiltonian energy function, so that the quality of generated data is improved.

Further, noise injection functionAdaptive noise injection mechanisms are employed to enhance the diversity and quality of the generated data. Specifically, the calculation manner of the noise injection function may be expressed as:

Wherein, Is at the/>Noise intensity of step,/>Is a standard normal distribution.

Further, noise intensityThe adjustment of (2) can be expressed as:

Wherein, Is the initial noise intensity,/>Is the attenuation coefficient,/>Is at the/>Step data diversity metric, in one embodiment, a Kullback-Leibler divergence between the generated data and the real data is used.

3) And judging the initial expansion sample by using a Laplace approximation discriminator to obtain a judging result.

Specifically, the discriminator of the embodiment of the invention applies the Laplace approximation to optimize the judgment accuracy. Specifically, first, a posterior probability approximation is calculated, which can be expressed as:

Wherein, Is a data sample,/>And/>The mean vector and covariance matrix, respectively.

Further, covariance matrixThe calculation of (2) can be expressed as:

Wherein, Is at the mean/>Potential energy function at/>Is a hessian matrix of (c).

4) And updating parameters of the generator according to the judging result, and performing countermeasure training on the judging device and the generator.

Further, the manner of parameter update can be expressed as:

Wherein, Is a loss function,/>Is learning rate,/>Is a regularization coefficient,/>Is a label. In one embodiment, the learning rate/>Is set to 0.001, regularization coefficient/>Set to 0.05.

Further, the generator and the arbiter are also counter-trained, with parameters being continuously adjusted to improve performance. Specifically, the loss calculation mode of the generator can be expressed as:

Wherein, Is noise extracted from the a priori distribution.

Further, the loss calculation mode of the discriminator can be expressed as:

Wherein, Is a true data sample,/>Is the corresponding tag.

Further, the method comprises the steps of,The calculation of (2) can be expressed as:

wherein the log likelihood is approximated as a square loss, making the training more stable.

Further, the arbiter losesThe way in which the cross entropy terms of (a) are calculated can be expressed as:

5) Until the loss of the generator converges, and the loss of the arbiter converges, determining the current initial extended sample as the extended sample corresponding to the initial sample set.

In specific implementation, the quality of the generated data and the accuracy of the discriminant are evaluated, and iterative optimization is performed according to the evaluation result. Specifically, the manner of quality assessment can be expressed as:

Wherein, Is an evaluation function, in one embodiment, calculates the distance between the distribution of the generated data and the true data distribution.

Further, the manner of iterative optimization can be expressed as:

Wherein, Is the learning rate of iterative optimization. In one embodiment, the learning rate/>, is iteratively optimizedSet/>0.002.

Further, a quality assessment functionThe calculation of (2) can be expressed as:

When the actual forging process data is processed, the expanded data can cover more possible operation and environment change conditions, so that the early warning system can more accurately predict the potential risks and anomalies, and the safety and the efficiency of the forging process are ensured.

Further, after the corresponding training sample set is constructed by the method, the neural network is trained, and the feature extraction model is constructed, so that the generalization capability and precision of the model are improved.

Step S12, initializing a particle swarm of the neural network and setting a global optimal solution.

Step S13, inputting the training sample set into the neural network, updating the particle positions of the neural network based on the global optimal solution and a preset multi-directional communication mechanism, and determining the particle optimal value of each dimension so as to train the neural network.

First, initializing: randomly initializing a particle swarm and the position thereof (namely, the neural network parameters of the feature extraction), and setting a globally optimal solution. Specifically, the particle group size is set asEach particle/>(Wherein/>) The location in the neural network parameter space is denoted/>The globally optimal solution is/>The manner of initialization can be expressed as:

Wherein, Is a fitness function. /(I)Representing all parameter range spaces,/>As a random function, in one embodiment, particle population size/>Set to 50.

Further, the particle positions of the neural network are updated and the particle optimum value for each dimension is determined. Specifically, a preset learning factor is obtained, and the particle position of the neural network is updated according to the learning factor, the information of the adjacent particles and the global optimal solution.

The embodiment of the invention updates through multi-directional communication: each particle updates its own location based on its own experience, the experience of neighboring particles, and the globally optimal solution. Specifically, for each particleThe way to update the information whose location considers neighboring particles and the global optimal solution can be expressed as:

Wherein, And/>Is a learning factor, and controls the influence degree of the global optimal solution and the adjacent optimal solution on the particle position; Is the optimal location of the adjacent particles.

Further, the learning factor is dynamically adjusted according to the historical information of the particles and the importance of the adjacent particle information, and can be expressed as follows:

Wherein, Is an adjustment factor,/>And/>The fitness of the current particle and the neighboring particle, respectively.

Wherein the importance is embodied in the learning factorsAnd/>These learning factors are based on the history information of the particles themselves (fitness/>) And information of neighboring particles (fitness/>) To adjust. /(I)Representing the impact weight of the particle's own history information at the time of location update, which is dynamically adjusted by comparing the difference in fitness of the particle itself with the optimal fitness of neighboring particles. When the fitness of the particle itself is much greater than the optimal fitness of the neighboring particles (/ >)），/>The value of (2) is close to 0, so that/>Approaching 1 means that the particle will update its location based mainly on its own history. /(I)Represents the impact weight of the neighbor information at the time of location update by subtracting/>, from 1Is obtained by the method. When/>Larger, i.e. when the history of the particles themselves is considered more important,/>Will be correspondingly smaller, indicating that the influence of the neighboring particle information is smaller. Conversely, when the fitness of the particle itself is not as good as the optimal fitness of the neighboring particles,/>Will decrease,/>Then the increase indicates that the neighboring particle information occupies a greater proportion in the particle location update. Thus, the importance is passed/>And/>The dynamic adjustment of the particle position updating strategy reflects the relative importance of the history information of the particle and the adjacent particle information in the current particle position updating strategy.

Further, according to the embodiment of the invention, the update factor is determined according to the distance between the current position of the particle and the target position; and updating the position of the particle in the current dimension based on the updating factor, and determining the optimal value of the particle in the current dimension.

Specifically, the coordinate ascent update is performed: the position of the particle is updated separately in each dimension, finding the optimal value for that dimension. Specifically, in each dimensionIn the above, the way to update the position of the particle alone to find the optimal value for this dimension can be expressed as:

wherein rand (0, 1) is a random number between 0 and 1. Is an update factor,/>Is/>And (5) an optimal solution of the dimension.

Further, the update factor depends on the distance between the current position of the particle and the target position, and can be expressed as:

Wherein, Is a scaling factor. In one embodiment, the scaling factor/>Set to 0.05.

Further, neural network training is performed: the neural network is trained by forward and backward propagation using the updated particle positions (neural network parameters). Specifically, the manner in which the forward propagation and the backward propagation of the neural network are performed can be expressed as:

Wherein, For the output of neural networks,/>Is a neural network model,/>For the error between the model output and the target value,/>As a loss function,/>For/>Fitness of individual particles.

Further, in the neural network model training process, the dynamically adjusted learning rate is usedTo update the weights, which can be expressed as:

Wherein, Is weight/>Is a learning rate of the first learning device; /(I)Is a dynamic adjustment factor for adjusting the magnitude of the learning rate; /(I)Is weight/>Is a gradient of (2); /(I)Is a weight in the neural network; /(I)Is the updated weight.

Further, in the learning rate dynamic adjustment, the weight contribution evaluation is performed first, that is, in each iteration, the contribution of each weight to the model performance is evaluated, which is achieved by calculating the gradient size of the weight, that is, the larger the absolute value of the weight gradient, the higher the contribution thereof. Based on the contribution degree of the weights, the learning rate of the corresponding weights is dynamically adjusted, and can be expressed as follows:

Wherein, Is the weight/>Original learning rate of/>Is an adjustment factor,/>Is the weight/>Is a gradient absolute value of (c).

And S14, carrying out fitness evaluation on the trained neural network, and updating the global optimal solution and the individual optimal solution according to the fitness evaluation result.

And S15, constructing a feature extraction model based on the current neural network until a preset iteration condition is met.

And calculating fitness (such as an error function) according to the output of the neural network and the target value, and updating the global optimal solution and the individual optimal solution. Specifically, the manner in which fitness is evaluated and global and individual optimal solutions are updated can be expressed as:

Further, the fitness function is the inverse of the error function, the error function is the mean square error, and the calculation mode of the fitness function can be expressed as:

Wherein, Is the mean square error between the model output and the target value.

Further, the iterative loop: the steps are repeated until the maximum iteration number is met, and the maximum iteration number in the embodiment of the invention can be preset manually. In one embodiment, the maximum number of iterations is 1000.

Further, the trained neural network may be used to perform feature extraction on the rig forging process parameter data, which may be expressed as:

Wherein, Is the position of the particle at the end of the training, i.e. the neural network parameters. The output of the neural network is a feature extraction result for the rig forging process parameters. By applying a multidirectional communication simulation particle swarm optimization algorithm, the extracted characteristics more accurately reflect the key parameter changes in the forging process, and potential quality problems can be early warned in advance.

Further, a method for constructing the data dimension reduction model is described. In specific implementation, the embodiment of the invention reduces the dimension of the data after the feature extraction through the constructed data dimension reduction model. The data dimension reduction model can be built through a preset training sample set, and when the data dimension reduction model is specifically implemented, after the feature extraction model is trained through the training sample set built through the embodiment, the output of the feature extraction model is used as the training sample set of the data dimension reduction model, so that key parameter changes in the forging process are reflected more accurately.

The invention provides a self-coding neural network algorithm based on local sensitive hash, which utilizes local sensitive hash technology and compressed sensing to improve the efficiency and precision of feature dimension reduction so as to realize efficient and accurate feature dimension reduction. The self-encoder of the present invention comprises two main parts: an encoder and a decoder. The encoder maps the input data to a low-dimensional feature space through a series of layers, and the decoder restores the low-dimensional features to an output that approximates the original data. In this process, a locality sensitive hash is used at the output layer of the encoder to ensure that similar feature vectors are mapped to similar hash codes.

Referring to fig. 3, a flowchart of another method for monitoring and early warning parameters of a rigging forging process based on artificial intelligence according to an embodiment of the invention is shown, and as shown in fig. 3, the method includes the following steps:

Step S21, a preset training sample set is obtained, and normalization processing is carried out on the training sample set.

In a specific implementation, the input data of the embodiment of the present invention is: feature vector after feature extraction. First, data normalization is performed: and carrying out normalization processing on the data after the feature extraction. In one embodiment, a set of rigged forging process parameter data is provided as/>Each data point/>Is a high-dimensional vector, and the normalization can be expressed as: /(I); Wherein/>Is normalized data point,/>Is the mean value of the data after feature extraction,/>Is the standard deviation of the data after feature extraction.

Further, the average valueThe calculation of (2) can be expressed as: /(I); Wherein/>Is the original data point of the data,Is the total number of data points.

Standard deviation ofThe calculation of (2) can be expressed as: /(I); Wherein/>The square of the difference between each data point and the mean is represented, and the standard deviation is obtained by summing and averaging.

Step S22, the training sample set subjected to normalization processing is input into a preset self-encoder, local sensitive hash mapping is carried out on the training sample set, and corresponding low-dimensional hash codes are determined.

Locality sensitive hash mapping: applying a locality sensitive hashing technique at the output layer of the encoder, mapping data to a low-dimensional hash code, can be expressed as: ; wherein/> Is corresponding to input/>Hash code of/>Representing a locally sensitive hash function. Further, the locality sensitive hash function maps data points to a lower dimensional hash space by a set of random projections, the data points being assigned to different hash values based on the results of the projections.

In one embodiment, a device is providedIs one/>Data points of a dimension are mapped to one/>Hash code of dimension/>Wherein. The locality sensitive hash function can be expressed as:

Wherein, Is one/>Each element of which is randomly chosen from a gaussian distribution probability distribution, this matrix being used to transform the original/>Dimension data point projection to a lower/>Dimensional space. /(I)Is one/>The random vector of dimension is used for introducing random offset in the hash function and enhancing the diversity of the hash. /(I)The function is a signed function, returning 1 for each element if the element is greater than 0; otherwise, return to-1.

Step S23, extracting and compressing key features from the low-dimensional hash codes through a compressed sensing technology.

In specific implementation, key features are extracted and compressed from the hash codes by a compressed sensing technology, and the method can be realized by solving the following optimization problems:

Wherein, Is a compressed feature representation,/>Is a compressed matrix,/>Is a regularization parameter,/>Is the L1 norm for promoting sparsity.

Further, the matrix is compressedThe configuration of (a) can be expressed as:

Wherein, Representing one/>Random matrix generation function of size,/>Is the feature dimension after compression,/>Is the dimension of the original data. This random matrix is typically generated by a gaussian distribution. In one embodiment, the feature extracted data dimension/>Setting the compressed characteristic dimension/>。

And S24, reconstructing the compressed characteristics into an approximate representation of the original data by using a preset decoder to obtain an initial coding result.

Using a decoder to reconstruct the compressed features into an approximate representation of the original data, the decoding process can be expressed as: ; wherein/> Is a reconstructed data point,/>Representing the decoding function.

Step S25, determining a loss function corresponding to the initial coding result, and optimizing parameters of the self-encoder according to the loss function.

And S26, constructing a data dimension reduction model based on the self-encoder until the loss function converges.

Further, by optimizing the parameters of the self-encoder by back-propagation and gradient descent algorithms, the loss function can be expressed as:

Wherein, Is the reconstruction error over the entire training set.

Further, the calculation mode of the gradient descent algorithm can be expressed as:

Wherein, Representing model parameters,/>Is a weight adjustment factor,/>Is learning rate,/>Is a loss function/>With respect toIs a gradient of (a). In one embodiment, the learning rate/>Is set to 0.01.

Further, the weight adjustment factorThe update speed of each weight is dynamically adjusted by the following calculation method:

Wherein, And/>Is a superparameter controlling the weight to adjust the intensity and sensitivity,/>Is a hyperbolic tangent function for ensuring a smooth variation of the adjustment factor. /(I)The absolute value of the weight is represented for evaluating the importance of the weight. In one embodiment,/>And/>Set to 0.01 and 0.1, respectively.

And (3) data output: and (5) carrying out low-dimensional representation of the technological parameters of the rigging forging after the characteristic dimension reduction treatment. Through data dimension reduction, the efficiency of data processing is improved, and the accuracy of the model in the aspects of predicting and warning potential production problems is enhanced; the high-dimensional data after feature extraction is subjected to feature dimension reduction processing and then becomes a more compact and efficient low-dimensional representation, so that the classification accuracy is improved.

Further, a method for constructing a classifier model is described, wherein the embodiment of the invention inputs the dimension reduced data into the classifier to output a parameter monitoring result. The classifier model of the embodiment of the invention is trained through the preset training sample set, and after the training sample set is constructed through the embodiment, the feature extraction model is used for extracting the features, and the data dimension reduction model is used for carrying out dimension reduction and then training the classifier model, so that the data processing efficiency is improved, and the accuracy of the model in the aspects of predicting and warning potential production problems is enhanced. The high-dimensional data after feature extraction is subjected to feature dimension reduction processing and then becomes a more compact and efficient low-dimensional representation, so that the classification accuracy is improved.

Aiming at the problems of low computational complexity and low efficiency of the traditional support vector machine algorithm when facing large-scale or high-dimensional data, the invention provides the support vector machine algorithm based on the anchor point method, which determines the anchor point by evaluating the association degree of data points and decision boundaries, thereby optimizing the selection of support vectors, effectively reducing the number of the support vectors and simultaneously maintaining the accuracy of a classifier. In addition, the improvement of the combined kernel skills allows the algorithm to process nonlinear data more flexibly, and by using the adaptive kernel function, the algorithm can dynamically adjust parameters of the kernel function according to the characteristics of the data, so that better classification effect is realized under different data distribution.

Referring to fig. 4, a flowchart of another method for monitoring and early warning parameters of a rigging forging process based on artificial intelligence according to an embodiment of the invention is shown, and as shown in fig. 4, the method includes the following steps:

Step S31, a preset training sample set and a preset support vector machine are obtained.

Step S32, inputting the training sample set into a support vector machine, determining an anchor point and a kernel function corresponding to the training sample set, and training the support vector machine by using the anchor point and the kernel function.

Specifically, the input data in the embodiment of the invention is a reduced-dimension data setAnd inputting the data into a preset support vector machine for training. The embodiment of the invention trains the support vector machine by determining the anchor point and the kernel function corresponding to the training sample set.

1) Anchor point selection: the embodiment of the invention obtains the decision boundary of the support vector machine; carrying out relevance evaluation on the training sample set according to the decision boundary to obtain a relevance result; judging a relevance result based on a preset relevance threshold value, and determining a sample of which the relevance result meets the relevance threshold value as an anchor point.

Specifically, the embodiment of the invention selects the anchor point by a method based on the association degree of the data point and the decision boundary, and sets the data set after dimension reductionComprises/>Samples of/>Representing the feature vector after dimension reduction,/>Is the corresponding tag. The process of selecting the anchor point is completed by evaluating the association degree of the data point and the decision boundary, and a threshold value/> issetFor each data point/>Calculate its distance from decision boundary/>And selecting a distance less than/>As anchor points, can be expressed as:

Wherein, Is a distance threshold. In one embodiment, a distance threshold/>, is setFor/>Then only the distance from the decision boundary is smaller than/>The data points are selected as anchor points/>。

Further, the method comprises the steps of,The calculation of (2) can be expressed as:

Wherein, Is a weight vector,/>Is an offset term,/>Representation/>Is a norm of (c).

2) Adaptive adjustment of kernel functions: according to the embodiment of the invention, the parameters of the target vector machine corresponding to any two data points in the training sample set are determined; and generating a kernel function corresponding to the data point based on the target vector machine parameters.

Specifically, parameters of the kernel function are dynamically adjusted according to the characteristics of the data, and the kernel function form and parameters which are most suitable for the current data set are selected by analyzing the distribution characteristics and the nonlinear relation of the data. Specifically, a kernel functionIs obtained based on data characteristics, set/>，/>For any two data points, the kernel function can be expressed as:

Wherein, Is a parameter automatically adjusted according to the data, and is determined by minimizing cross-validation errors, and can be expressed as follows: /(I)(Cross validation error/>))。

3) Model training: training of the model is performed using the selected anchor points and kernel functions. Specifically, the goal of the support vector machine is to find a decision boundary, which can be achieved by the following optimization problem:

The constraint conditions are as follows:

Wherein, Is a weight vector,/>Is an offset term,/>Is a regularization parameter,/>Is a relaxation variable. In one embodiment, regularization parameters/>Is set as/>。

And step S33, evaluating the trained support vector machine through a preset verification method to obtain a verification result.

And step S34, adjusting model parameters of the support vector machine according to the verification result until a preset verification requirement is met, and constructing a classifier model based on the current support vector machine.

Further, model verification and optimization are performed: and evaluating and optimizing the model through cross verification, and adjusting model parameters according to a verification result so as to achieve the optimal prediction effect. Specifically, a verification set is setAccuracy/>, on verification set, of calculation modelPerformance metrics, which can be expressed as:

Wherein, Is an indication function for judging whether the model prediction is accurate.

And (3) outputting: the prediction indicates whether a potential forging problem exists, and in one embodiment, the output categories include temperature anomalies, pressure starvation, and the like.

In summary, another monitoring and early warning method for parameters of a rigging forging process based on artificial intelligence provided by the embodiment of the invention has the following innovation and technical effects:

1. The generation countermeasure network based on the random gradient Hamiltonian Monte Carlo effectively solves the problems of insufficient training data and high-cost data acquisition by generating new data samples and evaluating the sample quality through a discriminator, and enhances the generalization capability and precision of the model.

2. The neural network for multi-direction communication simulation particle swarm optimization enhances the optimization efficiency and accuracy of the neural network in the feature extraction process by simulating the searching behavior of the particle swarm and a multi-direction communication mechanism.

3. The self-coding neural network based on the local sensitive hash is combined with compressed sensing through the local sensitive hash technology, so that the efficiency and the precision of feature dimension reduction are improved, and efficient and accurate feature dimension reduction processing is realized.

4. The support vector machine algorithm based on the anchor point method determines the anchor point by evaluating the association degree of the data point and the decision boundary, optimizes the selection of the support vectors, effectively reduces the number of the support vectors, and simultaneously maintains the accuracy of the classifier.

In summary, the invention can effectively process and analyze a large number of complex forging process parameters, and improve the efficiency and accuracy of data processing. The invention significantly improves the adaptability and generalization of the model to new data by generating the countermeasure network and other advanced algorithms. Through particle swarm optimization and a local sensitive hash technology, the method can extract key features more accurately, effectively reduce the dimension of data and facilitate subsequent analysis and prediction. The invention can more accurately predict the potential risk and abnormality and ensure the safety and efficiency of the forging process.

On the basis of the above embodiment, the embodiment of the present invention further provides a device for monitoring and early warning of a parameter of a forging process of a rigging based on artificial intelligence, and fig. 5 shows a schematic structural diagram of the device for monitoring and early warning of a parameter of a forging process of a rigging based on artificial intelligence provided by the embodiment of the present invention, as shown in fig. 5, the device includes: the data acquisition module 100 is used for acquiring forging process parameters of the target rigging; the forging process parameters comprise parameters for monitoring the target rigging in real time in a rigging forging process; the execution module 200 is used for converting forging process parameters into data vectors, inputting the data vectors into a pre-constructed parameter monitoring model and outputting parameter monitoring results; the parameter monitoring model is built based on a preset classifier model, a training sample set for training the classifier model is extracted through a preset feature extraction model, and dimension reduction is performed through a preset data dimension reduction model; the feature extraction model is constructed based on a neural network algorithm of multi-directional communication simulation particle swarm optimization, and the classifier model is constructed based on a method for determining anchor points; and the output module 300 is used for predicting potential risks and abnormal states of the target rigging in the forging process according to the parameter monitoring result.

The rigging forging process parameter monitoring and early warning device based on the artificial intelligence provided by the embodiment of the invention has the same technical characteristics as the rigging forging process parameter monitoring and early warning method based on the artificial intelligence provided by the embodiment, so that the same technical problems can be solved, and the same technical effects can be achieved.

On the basis of the embodiment, the embodiment of the invention also provides another rope forging process parameter monitoring and early warning device based on artificial intelligence, and fig. 6 shows a schematic structural diagram of the rope forging process parameter monitoring and early warning device based on artificial intelligence, and as shown in fig. 6, the parameter monitoring result comprises classification categories corresponding to data vectors; the output module 300 is further configured to determine a classification category indicated by the parameter monitoring result; and determining the state indicated by the classification category as an abnormal state of the target rigging in the forging process.

The device further comprises a model construction module 400, configured to obtain a preset training sample set and a preset support vector machine; inputting the training sample set into a support vector machine, determining an anchor point and a kernel function corresponding to the training sample set, and training the support vector machine by using the anchor point and the kernel function; evaluating the trained support vector machine through a preset verification method to obtain a verification result; and adjusting model parameters of the support vector machine according to the verification result until a preset verification requirement is met, and building a classifier model based on the current support vector machine.

The model building module 400 is configured to obtain a decision boundary of a support vector machine; carrying out relevance evaluation on the training sample set according to the decision boundary to obtain a relevance result; judging a relevance result based on a preset relevance threshold value, and determining a sample of which the relevance result meets the relevance threshold value as an anchor point; the step of determining the kernel function corresponding to the training sample set comprises the following steps: determining target vector machine parameters corresponding to any two data points in the training sample set; and generating a kernel function corresponding to the data point based on the target vector machine parameters.

The model construction module 400 is further configured to obtain a pre-constructed training sample set and a preset neural network; initializing a particle swarm of a neural network and setting a global optimal solution; inputting a training sample set into a neural network, updating the particle positions of the neural network based on a global optimal solution and a preset multi-directional communication mechanism, and determining the particle optimal value of each dimension so as to train the neural network; performing fitness evaluation on the trained neural network, and updating a global optimal solution and an individual optimal solution according to a fitness evaluation result; and constructing a feature extraction model based on the current neural network until a preset iteration condition is met.

The model building module 400 is further configured to obtain a preset learning factor, and update a particle position of the neural network according to the learning factor, information of adjacent particles, and a global optimal solution; wherein the learning factor is dynamically adjusted according to the historical information of the particles and the importance of the adjacent particle information.

The model building module 400 is further configured to determine an update factor according to a distance between a current position of the particle and the target position; and updating the position of the particle in the current dimension based on the updating factor, and determining the optimal value of the particle in the current dimension.

The model building module 400 is further configured to obtain a preset training sample set, and normalize the training sample set; inputting the normalized training sample set into a preset self-encoder, performing local sensitive hash mapping on the training sample set, and determining a corresponding low-dimensional hash code; extracting and compressing key features from the low-dimensional hash codes by a compressed sensing technology; reconstructing the compressed features into an approximate representation of the original data by using a preset decoder to obtain an initial coding result; determining a loss function corresponding to the initial coding result, and optimizing parameters of the self-encoder according to the loss function; until the loss function converges, a data dimension reduction model is constructed based on the self-encoder.

The model building module 400 is further configured to obtain an initial sample acquired in advance; labeling the initial sample to generate a sample label; constructing an initial sample set according to the sample label and the initial sample, and performing sample expansion on the initial sample set by using a preset generation countermeasure network to obtain an expanded sample; wherein the extended samples are generated by a generator using a random gradient hamilton monte carlo method; evaluating the expansion sample, and performing iterative optimization on the generated countermeasure network according to the evaluation result; and constructing a training sample set according to the current expansion sample and the initial sample set until the expansion sample meets the preset evaluation requirement.

The model building module 400 is further configured to determine a potential energy function, momentum and energy matrix corresponding to the initial sample set based on parameters of a preset generator; generating an initial expansion sample according to the potential energy function, the momentum and the energy matrix; judging the initial expansion sample by using a Laplace approximation discriminator to obtain a judging result; updating parameters of the generator according to the discrimination result, and performing countermeasure training on the discriminator and the generator; until the loss of the generator converges, and the loss of the arbiter converges, determining the current initial extended sample as the extended sample corresponding to the initial sample set.

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 any one of the figures 1 to 4. 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 any of the above-mentioned figures 1 to 4. The embodiment of the present invention further provides a schematic structural diagram of an electronic device, as shown in fig. 7, where the electronic device includes a processor 71 and a memory 70, where the memory 70 stores computer executable instructions that can be executed by the processor 71, and the processor 71 executes the computer executable instructions to implement the method shown in any of the foregoing fig. 1 to 4. In the embodiment shown in fig. 7, the electronic device further comprises a bus 72 and a communication interface 73, wherein the processor 71, the communication interface 73 and the memory 70 are connected by the bus 72.

The memory 70 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 the at least one other network element is achieved via at least one communication interface 73 (which may be wired or wireless), which may use the internet, a wide area network, a local network, a metropolitan area network, etc. The Bus 72 may be an ISA (Industry Standard Architecture ) Bus, a PCI (PERIPHERAL COMPONENT INTERCONNECT, peripheral component interconnect standard) Bus, or EISA (Extended Industry Standard Architecture ) Bus, etc., or an AMBA (Advanced Microcontroller Bus Architecture, standard for on-chip buses) Bus, where AMBA defines three buses, including an APB (ADVANCED PERIPHERAL Bus) Bus, an AHB (ADVANCED HIGH-performance Bus) Bus, and a AXI (Advanced eXtensible Interface) Bus. The bus 72 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. 7, but not only one bus or type of bus.

The processor 71 may be an integrated circuit chip with signal processing capabilities. In implementation, the steps of the above method may be performed by integrated logic circuitry in hardware or instructions in software in the processor 71. The processor 71 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 71 reads the information in the memory and in combination with its hardware performs the method as shown in any of the foregoing figures 1 to 4. The embodiment of the application provides a method and a device for monitoring and early warning parameters of a rigging forging process based on artificial intelligence, 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 (8)

1. An artificial intelligence-based monitoring and early warning method for parameters of a rigging forging process is characterized by comprising the following steps:

Obtaining forging technological parameters of a target rigging; the forging process parameters comprise parameters for monitoring the target rigging in real time in a rigging forging process;

Converting the forging process parameters into data vectors, inputting the data vectors into a pre-constructed parameter monitoring model, and outputting parameter monitoring results; the parameter monitoring model is constructed based on a preset classifier model, a training sample set for training the classifier model is used for carrying out feature extraction through a preset feature extraction model, and dimension reduction processing is carried out through a preset data dimension reduction model; the feature extraction model is constructed based on a neural network algorithm of multi-directional communication simulation particle swarm optimization, and the classifier model is constructed based on a method of determining anchor points;

Predicting potential risks and abnormal states of the target rigging in the forging process according to the parameter monitoring result;

The training method of the classifier model comprises the following steps:

acquiring a preset training sample set and a preset support vector machine;

Inputting the training sample set into the support vector machine, determining an anchor point and a kernel function corresponding to the training sample set, and training the support vector machine by using the anchor point and the kernel function;

evaluating the trained support vector machine through a preset verification method to obtain a verification result;

Adjusting model parameters of the support vector machine according to the verification result until a preset verification requirement is met, and building a classifier model based on the current support vector machine;

The step of determining the anchor point corresponding to the training sample set comprises the following steps:

acquiring a decision boundary of the support vector machine;

Performing relevance evaluation on the training sample set according to the decision boundary to obtain a relevance result;

judging the association result based on a preset association threshold, and determining a sample of which the association result meets the association threshold as an anchor point;

The step of determining the kernel function corresponding to the training sample set comprises the following steps:

determining target vector machine parameters corresponding to any two data points in the training sample set;

and generating a kernel function corresponding to the data point based on the target vector machine parameters.

2. The method of claim 1, wherein the parameter monitoring result comprises a classification category corresponding to the data vector;

and predicting potential risks and abnormal states of the target rigging in the forging process according to the parameter monitoring result, wherein the steps comprise:

Determining the classification category indicated by the parameter monitoring result;

and determining the state indicated by the classification category as an abnormal state of the target rigging in the forging process.

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

Acquiring a pre-constructed training sample set and a pre-set neural network;

Initializing a particle swarm of the neural network and setting a global optimal solution;

Inputting the training sample set into the neural network, updating the particle positions of the neural network based on the global optimal solution and a preset multi-directional communication mechanism, and determining the particle optimal value of each dimension so as to train the neural network;

Performing fitness evaluation on the trained neural network, and updating the global optimal solution and the individual optimal solution according to a fitness evaluation result;

And constructing a feature extraction model based on the current neural network until a preset iteration condition is met.

4. A method according to claim 3, wherein the step of updating the particle location of the neural network based on the globally optimal solution and a preset multi-directional communication mechanism comprises:

Acquiring a preset learning factor, and updating the particle position of the neural network according to the learning factor, the information of the adjacent particles and the global optimal solution; wherein the learning factors are dynamically adjusted according to the historical information of the particles and the importance of the adjacent particle information;

A step of determining an optimal value of particles for each dimension, comprising:

determining an update factor according to the distance between the current position of the particle and the target position;

And updating the position of the particle in the current dimension based on the updating factor, and determining the optimal value of the particle in the current dimension.

5. The method of claim 1, wherein the method for constructing the data dimension reduction model comprises:

acquiring a preset training sample set, and carrying out normalization processing on the training sample set;

inputting a normalized training sample set into a preset self-encoder, performing local sensitive hash mapping on the training sample set, and determining a corresponding low-dimensional hash code;

Extracting and compressing key features from the low-dimensional hash code by a compressed sensing technology;

reconstructing the compressed features into an approximate representation of the original data by using a preset decoder to obtain an initial coding result;

Determining a loss function corresponding to the initial coding result, and optimizing parameters of the self-encoder according to the loss function;

Until the loss function converges, a data dimension reduction model is constructed based on the self-encoder.

6. The method according to claim 1, wherein the method further comprises:

Acquiring a pre-acquired initial sample;

Labeling the initial sample to generate a sample label;

constructing an initial sample set according to the sample label and the initial sample, and performing sample expansion on the initial sample set by using a preset generation countermeasure network to obtain an expanded sample; wherein the extended samples are generated by a generator using a random gradient hamilton monte carlo method;

evaluating the expansion sample, and performing iterative optimization on the generated countermeasure network according to an evaluation result;

And constructing a training sample set according to the current expansion sample and the initial sample set until the expansion sample meets the preset evaluation requirement.

7. The method of claim 6, wherein the step of sample expanding the initial sample set using a predetermined generation countermeasure network to obtain an expanded sample comprises:

determining potential energy functions, momentums and energy matrixes corresponding to the initial sample set based on preset generator parameters;

generating an initial expansion sample according to the potential energy function, the momentum and the energy matrix;

judging the initial expansion sample by using a Laplace approximation discriminator to obtain a judging result;

updating the parameters of the generator according to the discrimination result, and performing countermeasure training on the discriminator and the generator;

Until the loss of the generator converges, and the loss of the discriminator converges, determining the current initial extended sample as the extended sample corresponding to the initial sample set.

8. An artificial intelligence-based rigging forging process parameter monitoring and early warning device is characterized in that the device comprises:

the data acquisition module is used for acquiring forging process parameters of the target rigging; the forging process parameters comprise parameters for monitoring the target rigging in real time in a rigging forging process;

The execution module is used for converting the forging process parameters into data vectors, inputting the data vectors into a pre-constructed parameter monitoring model and outputting parameter monitoring results; the parameter monitoring model is built based on a preset classifier model, a training sample set for training the classifier model is extracted through a preset feature extraction model, and dimension reduction is performed through a preset data dimension reduction model; the feature extraction model is constructed based on a neural network algorithm of multi-directional communication simulation particle swarm optimization, and the classifier model is constructed based on a method of determining anchor points;

the output module is used for predicting potential risks and abnormal states of the target rigging in the forging process according to the parameter monitoring result;

The device further comprises a model construction module, a model analysis module and a model analysis module, wherein the model construction module is used for acquiring a preset training sample set and a preset support vector machine; inputting the training sample set into the support vector machine, determining an anchor point and a kernel function corresponding to the training sample set, and training the support vector machine by using the anchor point and the kernel function; evaluating the trained support vector machine through a preset verification method to obtain a verification result; adjusting model parameters of the support vector machine according to the verification result until a preset verification requirement is met, and building a classifier model based on the current support vector machine;

The model construction module is also used for acquiring a decision boundary of the support vector machine; performing relevance evaluation on the training sample set according to the decision boundary to obtain a relevance result; judging the association result based on a preset association threshold, and determining a sample of which the association result meets the association threshold as an anchor point; the step of determining the kernel function corresponding to the training sample set comprises the following steps: determining target vector machine parameters corresponding to any two data points in the training sample set; and generating a kernel function corresponding to the data point based on the target vector machine parameters.