Intelligent metallurgical process virtual simulation method and system based on digital twinning

Abstract

The application relates to the field of metallurgical industry simulation and intelligent control technology, and discloses an intelligent metallurgical process virtual simulation method and system based on digital twinning. The method comprises the following steps: acquiring multi-modal high-dimensional data of a metallurgical process and reducing the dimension to obtain a feature vector set of a hidden layer space; performing disturbance injection simulation by using the feature vector set of the hidden layer space to obtain a multi-path state sequence with time dependence; predicting an abnormal path by using a recurrent neural network to obtain a predicted state evolution trajectory; obtaining multi-path state representation in a physical space by using a preset inverse mapping function; screening to obtain a risk path set; extracting a key evolution node from the set for processing to obtain a virtual simulation scene; and optimizing simulation of preset process adjustment parameters according to the virtual simulation scene to determine an optimized parameter configuration. The method can solve the problem that it is difficult to construct a comprehensive virtual simulation scene which can truly restore the physical production law.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of metallurgical industry simulation and intelligent control technology, and particularly relates to an intelligent metallurgical process virtual simulation method and system based on digital twinning. BACKGROUND

[0002] At present, the metallurgical industrial production process involves high-temperature multiphase reaction, complex mass and heat transfer behavior and extreme working condition environment, which is directly related to the material quality, energy consumption and production safety. With the deep integration of industrial sensor technology and industrial internet, digital twinning technology has been gradually applied to real-time monitoring and virtual simulation of the whole metallurgical process. By constructing a mirror image of the physical entity in the virtual space, process parameter optimization and potential risk prediction are supported.

[0003] In one prior art, the metallurgical state is usually simulated or single path deduced by using physical mechanism model or simple historical trajectory data. Such method has certain reference value under the condition that the working condition is relatively stable. However, the data collected by the metallurgical field sensor has the characteristics of high dimension, high coupling and strong nonlinearity, covering temperature gradient, instantaneous pressure, dynamic concentration of components and other heterogeneous variables. In actual production, the metallurgical state is significantly affected by random disturbances such as raw material composition fluctuation and equipment operation deviation, showing extremely complex state evolution characteristics. When dealing with such high-dimensional characteristics, the prior art often has difficulty in balancing the real-time performance and physical accuracy of the deduction under limited computing resources.

[0004] There is a difficulty in constructing a comprehensive virtual simulation scene that truly restores the physical production law in the prior art. SUMMARY

[0005] The present application provides an intelligent metallurgical process virtual simulation method and system based on digital twinning to solve the difficulty in constructing a comprehensive virtual simulation scene that truly restores the physical production law in the prior art.

[0006] In a first aspect, to solve the above technical problems, the present application provides an intelligent metallurgical process virtual simulation method based on digital twinning, comprising:

[0007] Obtaining multi-modal high-dimensional data of the metallurgical process, using the encoder part in the pre-trained variational autoencoder to reduce the dimension of the high-dimensional data, and obtaining a feature vector set of the hidden layer space;

[0008] According to the feature vector set of the hidden layer space, using a pre-trained time series generative adversarial model to simulate random disturbance injection, and obtaining a multi-path state sequence with time dependence;

[0009] If there is an abnormal path deviating from the preset path deviation threshold in the multi-path state sequence, a preset inverse mapping function constructed by a decoder part in the variational autoencoder is used to map the predicted state evolution trajectory to a physical space to obtain a multi-path state representation in the physical space;

[0010] If there is an abnormal path deviating from the preset path deviation threshold in the multi-path state sequence, a preset inverse mapping function constructed by a decoder part in the variational autoencoder is used to map the predicted state evolution trajectory to a physical space to obtain a multi-path state representation in the physical space;

[0011] If there is an abnormal path deviating from the preset path deviation threshold in the multi-path state sequence, a preset inverse mapping function constructed by a decoder part in the variational autoencoder is used to map the predicted state evolution trajectory to a physical space to obtain a multi-path state representation in the physical space;

[0012] If there is an abnormal path deviating from the preset path deviation threshold in the multi-path state sequence, a preset inverse mapping function constructed by a decoder part in the variational autoencoder is used to map the predicted state evolution trajectory to a physical space to obtain a multi-path state representation in the physical space;

[0013] If there is an abnormal path deviating from the preset path deviation threshold in the multi-path state sequence, a preset inverse mapping function constructed by a decoder part in the variational autoencoder is used to map the predicted state evolution trajectory to a physical space to obtain a multi-path state representation in the physical space.

[0014] In a second aspect, the present application provides a digital twin intelligent metallurgical process virtual simulation system, comprising:

[0015] A data dimension reduction module is configured to obtain multi-modal high-dimensional data of a metallurgical process, and utilize an encoder part in a pre-trained variational autoencoder to perform dimension reduction processing on the high-dimensional data to obtain a feature vector set in a hidden layer space.

[0016] A multi-path simulation module is configured to utilize a pre-trained time series generative adversarial model to perform random disturbance injection simulation according to the feature vector set in the hidden layer space to obtain a multi-path state sequence with time dependence.

[0017] A state prediction module is configured to, if there is an abnormal path deviating from a preset path deviation threshold in the multi-path state sequence, utilize a pre-trained recurrent neural network to perform time series evolution trend prediction on the abnormal path to obtain a predicted state evolution trajectory.

[0018] An inverse mapping module is configured to utilize a preset inverse mapping function constructed by a decoder part in the variational autoencoder to map the predicted state evolution trajectory to a physical space to obtain a multi-path state representation in the physical space.

[0019] A risk screening module is configured to, according to the multi-path state representation in the physical space, fuse a preset correlation constraint to perform potential risk path screening to obtain a risk path set.

[0020] A scene generation module is configured to extract key evolution nodes from the risk path set and perform data enhancement mapping processing on the key evolution nodes by using a pre-trained node expansion generation adversarial model to obtain a virtual simulation scene.

[0021] A parameter optimization module is configured to perform optimization simulation on preset process adjustment parameters according to the virtual simulation scene to determine an optimized parameter configuration.

[0022] Compared with the prior art, the present application has the following beneficial effects:

[0023] (1) The present application realizes high-precision mapping of a high-dimensional feature space and a physical space by using an encoder and a decoder architecture of a variational autoencoder. The present application performs nonlinear projection on high-dimensional heterogeneous data of a metallurgical process by using a pre-trained variational autoencoder, reduces the original data to a low-dimensional manifold space for deduction, and corrects in the dimension recovery stage by using the hidden layer space manifold distribution characteristics learned by the decoder in advance in cooperation with physical constraint conditions. Compared with a traditional linear dimension reduction algorithm (such as PCA), this mode can effectively capture the complex nonlinear coupling relationship between metallurgical variables, eliminate numerical distortion and logical deviation of high-dimensional features when inversely mapping back to physical coordinates, and ensure the physical consistency of the simulation scene.

[0024] (2) The present application improves the time sequence logic and risk coverage of multi-path deduction by introducing a time step attention mechanism of a time sequence generation adversarial model. The present application injects random disturbance in the hidden layer feature space and simulates multiple potential paths of the evolution of the metallurgical state over time by using a time sequence generation adversarial model. By using the time step attention mechanism built in the model, it is ensured that the generated simulation instantaneous state has strong causal dependence on the time axis, overcoming the sequence disorder or physical mutation problem that is prone to occur in traditional methods when randomly deducing. This process enables the digital twin system to predict more extreme and rare potential risk conditions, and enhances the perception depth of the system to the complex dynamic production environment.

[0025] (3) The present application realizes a logical closed loop from risk perception to precise decision by combining the node expansion enhanced virtual scene with a metallurgical dynamics evolution model. After identifying the key abnormal evolution nodes, the present application performs feature derivation by using a generation adversarial model to construct comprehensive and formatted virtual scene data. By mapping the scene components to a preset metallurgical dynamics evolution model, the key process indicators such as reaction rate fluctuation and slag layer thickness distribution are simulated, and the multi-objective particle swarm optimization algorithm is used to iteratively correct the process parameters. This process realizes the deep transformation from abstract risk data to concrete dynamics simulation, and then to executable parameter configuration, and improves the pertinence of process optimization and the steady-state control level of metallurgical production. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a digital twin intelligent metallurgical process virtual simulation method flowchart provided by the first embodiment of the present application;

[0027] Figure 2 is a digital twin intelligent metallurgical process virtual simulation system structure diagram provided by the second embodiment of the present application. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0029] With reference to Figure 1 , the first embodiment of the present application provides a digital twin intelligent metallurgical process virtual simulation method, comprising the following steps:

[0030] S11, obtaining multi-modal high-dimensional data of a metallurgical process, using an encoder part in a pre-trained variational autoencoder to perform dimension reduction processing on the high-dimensional data to obtain a feature vector set of a hidden layer space;

[0031] S12, according to the feature vector set of the hidden layer space, using a pre-trained time series generation adversarial model to perform random disturbance injection simulation to obtain a multi-path state sequence with time dependence;

[0032] S13, if there is an abnormal path deviating from a preset path deviation threshold in the multi-path state sequence, using a pre-trained recurrent neural network to perform time series evolution trend prediction on the abnormal path to obtain a predicted state evolution trajectory;

[0033] S14, using a preset inverse mapping function constructed by a decoder part in the variational autoencoder to map the predicted state evolution trajectory to a physical space to obtain a multi-path state representation in the physical space;

[0034] S15, according to the multi-path state representation in the physical space, fusing a preset correlation constraint to perform potential risk path screening to obtain a risk path set;

[0035] S16, extracting a key evolution node from the risk path set and using a pre-trained node expansion generation adversarial model to perform data enhancement mapping processing on the key evolution node to obtain a virtual simulation scene;

[0036] S17, according to the virtual simulation scene, performing optimization simulation on a preset process adjustment parameter to determine an optimized parameter configuration.

[0037] In step S11, multi-modal high-dimensional data of a metallurgical process is acquired, and the high-dimensional data is reduced in dimension by using an encoder part in a pre-trained variational autoencoder to obtain a feature vector set of a hidden layer space, including:

[0038] The multi-modal high-dimensional data of the metallurgical process is acquired, and an initial high-dimensional matrix is constructed using the multi-modal high-dimensional data.

[0039] The initial high-dimensional matrix is input into the encoder part, and multi-modal mixed convolution mapping processing is performed by using the encoder part to project the initial high-dimensional matrix to a low-dimensional manifold space.

[0040] The hidden variable distribution parameters of the low-dimensional manifold space are extracted, and determined as the feature vector set of the hidden layer space.

[0041] In an implementation manner, the multi-modal high-dimensional data is acquired by a multi-source sensor cluster. The multi-source sensor cluster includes but is not limited to a platinum-rhodium thermocouple rigidly installed at different depths of an inner wall of a furnace lining, a capacitive pressure transmitter installed in a dust removal pipeline at the top of the furnace, an electromagnetic flowmeter for monitoring an oxygen lance system, and an infrared thermal imager installed at an observation window of the metallurgical furnace. The multi-modal high-dimensional data includes real-time collected smelting temperature data, furnace pressure data, oxygen supply flow data, and infrared thermal image data representing the spatial distribution of the surface temperature field of the molten pool in the furnace.

[0042] In an implementation manner, the initial high-dimensional matrix is constructed by using the multi-modal high-dimensional data. First, the original electric signals fed back by each sensor are subjected to synchronous sampling processing, and the sampling frequency is set to 50 Hz. The value is selected according to the characteristic frequency of the main physical quantity fluctuation in the electric arc furnace steelmaking. Then, each variable sequence collected synchronously is subjected to standardization processing. The specific processing logic is that, for each dimension variable, the arithmetic mean and the standard deviation in a preset historical observation period are calculated. The preset historical observation period is preferably the real-time monitoring data in the last 24 hours or the historical data of the last 100 complete metallurgical furnace times. The original observation value is subtracted from the arithmetic mean, and the obtained difference value is divided by the standard deviation to output the dimensionless standardized characteristic value. For the infrared thermal image data, it is converted into a corresponding gray matrix component, and is aligned and superimposed with the standardized electric signal features according to the sampling time stamp, so as to construct the initial high-dimensional matrix containing time axis information, spatial field information and multi-dimensional process components.

[0043] It should be noted that the initial high-dimensional matrix is input into the encoder part in this embodiment, and a multimodal mixed convolution mapping process is performed using the encoder part. The encoder part is constructed as a mixed convolutional neural network architecture, which includes a first-order convolution branch for extracting time-series correlation features of process variables (with a preset convolution kernel size of and a step size of 1), and a second-order convolution branch for extracting spatial feature distribution of furnace temperature field in infrared thermal images (with a preset convolution kernel size of and a step size of 1). A rectified linear unit (ReLU) is nested between the convolution layers as a nonlinear activation function. In this embodiment, the time-series features and spatial field features are nonlinearly projected through a multimodal feature fusion layer to eliminate data redundancy, thereby compressing and projecting the high-dimensional original information to a low-dimensional manifold space. This process effectively preserves the core manifold structure representing thermodynamic equilibrium and chemical reaction rate.

[0044] In the multimodal mixed convolution mapping process, after the time-series convolution branch and the spatial convolution branch perform feature extraction, the output feature maps are spliced in the channel dimension to form a fusion feature map. The fusion feature map is then integrated and reduced in the channel dimension through a 1x1 convolution layer, and then input into the subsequent fully connected layer, which ensures the deep fusion of time-series signals and spatial image information at the feature level.

[0045] It should be noted that the hidden variable distribution parameters of the low-dimensional manifold space are extracted in this embodiment. Specifically, the end of the encoder part is connected with two parallel fully connected layers, which are the mean mapping layer and the variance mapping layer. In this embodiment, the projected features are simultaneously input into the above two layers, and the linear transformation matrix outputs the mean vector and the logarithmic variance vector , which are determined as the feature vector set of the hidden layer space. In this embodiment, the dimension of the feature vector set is preset to 256 dimensions, which is determined based on a compression ratio of 10% to 20% of the original high-dimensional data features, aiming to ensure the condensation of essential features while retaining sufficient expression ability, and to provide sufficient feature support for the restoration of the three-dimensional tensor in the subsequent steps.

[0046] It should be noted that the pre-trained variational autoencoder is trained using image-signal samples containing historical abnormal working conditions in the offline stage. In this embodiment, a composite loss function composed of a reconstruction loss term and a divergence constraint term is constructed as the optimization objective. In this embodiment, the adaptive moment estimation (Adam) algorithm is used to perform iterative solution, which minimizes the reconstruction error, so that the model training is sufficient and there is no significant risk of gradient disappearance, ensuring that the dimensionality reduction mapping process can accurately preserve the metallurgical holographic state.

[0047] In step S12, according to the feature vector set of the hidden layer space, a pre-trained time sequence generation adversarial model is used for random disturbance injection simulation to obtain a time-dependent multi-path state sequence, including:

[0048] A preset random disturbance factor and a preset random noise vector are called, and the random disturbance factor is superimposed into the random noise vector to synthesize a disturbance input vector set;

[0049] The feature vector set of the hidden layer space and the disturbance input vector set are synchronously input into the time sequence generation adversarial model;

[0050] The time step attention mechanism built in the time sequence generation adversarial model is used to perform time-dependent mapping to obtain a set of mutually related simulated instantaneous states;

[0051] According to the generated time sequence, time sequence connection processing is performed on the data points in the set of simulated instantaneous states to construct the multi-path state sequence.

[0052] In an implementation manner, the embodiment uses a pre-trained time sequence generation adversarial model for random disturbance injection simulation. The time sequence generation adversarial model is constructed as a sequence inference architecture based on a conditional generative adversarial network (CGAN), which includes a generator network and a discriminator network. The generator network uses a gated recurrent unit (GRU) as a core time sequence processing unit, and has an attention calculation layer nested at each output time step, which is used to capture the long-range time sequence dependence of the metallurgical state.

[0053] In an implementation manner, the embodiment calls a preset random noise vector . The random noise vector follows a standard normal distribution with a mean of 0 and a variance of 1 , and the dimension is preset to 100 dimensions. The dimension is selected based on providing sufficient random entropy space to cover complex process fluctuations. The preset random disturbance factor represents uncontrollable physical fluctuations in the metallurgical process, such as batch differences of raw materials or equipment wear. The value range is determined by statistical analysis. The deviation sequences of the oxygen lance height, bottom blowing flow rate and raw material composition in the historical process log of the metallurgical site are collected; the standard deviation multiples of each deviation sequence relative to the mean value are calculated; in this embodiment, the value range of the random disturbance factor is set to between -0.15 and 0.15, which is consistent with the scale of the normalized data in S11, ensuring the physical reasonableness of the disturbance injection. In this embodiment, an algebraic addition operation is performed to superimpose the random disturbance factor into the components of the random noise vector , thereby synthesizing the disturbance input vector set.

[0054] It should be noted that the embodiment takes the feature vector set of the hidden layer space as a conditional input to limit the generated evolution trajectory to be always in the neighborhood of the current metallurgical thermodynamic state, preventing the generation of invalid paths that deviate from the actual production background.

[0055] It is worth noting that the embodiment utilizes the time step attention mechanism built-in the time sequence generative adversarial model to perform time sequence dependency mapping, obtaining a set of interrelated simulated instantaneous states. The time step attention mechanism realizes cross-time step fusion of information by calculating the correlation weight between the hidden state of the current deduction time and the hidden state sequence of the historical time. Specifically, the embodiment calculates the attention score using a scaled dot product function, and obtains the weight distribution through normalization processing, guiding the generator to output state points that are logically causally continuous with the historical trend. The time step attention mechanism adopts scaled dot product attention; at each GRU time step of the generator, the hidden state of the current GRU is taken as the query vector, and the sequence of GRU hidden states of all previous time steps is taken as the key vector and the value vector to calculate the attention weight; the weighted sum of the context vector will be fused with the input of the current GRU to determine the output of the time step. This mechanism enables the generator to dynamically focus on the most critical information in the historical state for the current deduction.

[0056] In an implementation manner, the embodiment performs time sequence connection processing on the data points in the set of simulated instantaneous states in the generated time sequence to construct the multi-path state sequence. The embodiment arranges the continuously generated instantaneous state points in a vertical direction according to the timestamp index to form multiple trajectory sequences representing different perturbation evolution directions. Among them, is a preset time step sequence length.

[0057] It should be noted that the time sequence generative adversarial model is trained in the offline stage. The embodiment constructs a composite loss function composed of an adversarial loss term, a time consistency loss term, and a physical manifold constraint term. The physical manifold constraint term is realized by calculating the Euclidean distance between the simulated instantaneous state generated by the generator and its projection on the low-dimensional feature manifold, aiming to constrain the generated trajectory to strictly follow the metallurgical physical law. The embodiment determines the training convergence standard by the performance driving method. The fluctuation rate of the discriminator loss is counted for 100 consecutive batches, and if the fluctuation rate is less than a preset stability threshold (for example ), it is determined that the discriminator and the generator reach Nash equilibrium, the model converges and the training is terminated. This process ensures that the virtual trajectory generated by the system has statistical diversity and can cover more extreme and rare potential risk working conditions.

[0058] In step S13, if there is an abnormal path deviating from the preset path deviation threshold in the multi-path state sequence, the time sequence evolution trend of the abnormal path is predicted by using the pre-trained recurrent neural network, and a predicted state evolution trajectory is obtained, including:

[0059] The state vector in the multi-path state sequence is extracted, and the Euclidean distance between the state vector and the preset standard reference vector is calculated to obtain a path deviation numerical set;

[0060] If the numerical value in the path deviation numerical set exceeds the preset path deviation threshold, the corresponding continuous abnormal node segment in the multi-path state sequence is intercepted, and the abnormal path data to be processed is determined;

[0061] The abnormal path data to be processed is input into the pre-trained recurrent neural network to perform feature hidden layer extraction processing, and a hidden layer state vector is obtained;

[0062] The hidden layer state vector is used for multi-round recursive iteration operation, and a continuous evolution prediction value is calculated;

[0063] According to the continuous evolution prediction value, sequence reorganization processing is performed to obtain the predicted state evolution trajectory.

[0064] In an implementation manner, the state vector in the multi-path state sequence is extracted. The state vector is the hidden layer feature generated by S12 and located in the low-dimensional manifold space. The Euclidean distance between the state vector and the preset standard reference vector is calculated.

[0065] In an implementation manner, the preset standard reference vector is determined by statistical analysis method. The specific determination method is to retrieve the preset historical production database, filter out the high-quality metallurgical batches with a yield greater than 98.5% and energy consumption indicators in the preset low consumption interval; extract the historical hidden layer feature sequence corresponding to the high-quality metallurgical batches, calculate the arithmetic mean of all sample points in each dimension, and construct the preset standard reference vector. The arithmetic square root of the sum of squares of the difference between the current state vector and the preset standard reference vector in each dimension is calculated to output a scalar value representing the degree of deviation of the current evolution state from the ideal working condition. All time steps in the multi-path state sequence are traversed to obtain the path deviation numerical set composed of distance values at each time.

[0066] It should be noted that the embodiment judges whether the numerical value in the path deviation numerical set exceeds the preset path deviation threshold. The preset path deviation threshold is determined by performance driving method. In the model verification stage, the distance fluctuation distribution under the normal metallurgical working condition is counted, and the mean and standard deviation are calculated. According to the performance driving method, the preset path deviation threshold is determined as the sum of the mean and a certain percentage of the standard deviation. The percentage is determined according to the actual situation. In principle, the preset path deviation threshold is set to the sum of the mean value and 3 times the standard deviation. If the distance value in a certain time period continuously exceeds the threshold, the embodiment intercepts the corresponding continuous abnormal node segment in the multi-path state sequence and determines it as abnormal path data to be processed. In this way, the system can automatically eliminate normal random noise and accurately capture abnormal disturbances with instability trends.

[0067] It is worth noting that the embodiment inputs the abnormal path data to be processed into the pre-trained recurrent neural network to perform feature hidden layer extraction processing and obtain a hidden layer state vector. The pre-trained recurrent neural network adopts a long short-term memory (LSTM) architecture, which includes an input gate, a forget gate, and an output gate, and can effectively process metallurgical evolution features with long-range dependencies.

[0068] In one implementation, the embodiment performs multiple rounds of recursive iteration operations using the hidden layer state vector to calculate a continuous evolution prediction value. The embodiment takes the starting offset time of the abnormal path data to be processed as the initial value of iteration, combines the disturbance features at the current time, and performs the following trend deduction operation: using the hidden layer state vector at the previous time and the predicted output to perform nonlinear mapping to output the prediction value at the current time and the updated hidden layer state vector. The embodiment performs multiple loop iterations until the predicted time step covers the length of the continuous abnormal node segment. Through this deduction rather than forced correction, the embodiment completely retains the subsequent development trend of the abnormal offset under the inertia of physical laws.

[0069] It is worth noting that the embodiment performs sequence reorganization processing according to the continuous evolution prediction value to obtain the predicted state evolution trajectory. The embodiment replaces the abnormal part in the original sequence with the generated continuous evolution prediction value, and performs smooth transition processing on the boundaries of the replacement points using a cubic spline interpolation algorithm to ensure that the generated trajectory has temporal continuity in the hidden layer space, providing a highly reliable deduction basis for the restoration of the physical space in the subsequent steps.

[0070] It is worth noting that the pre-trained recurrent neural network is trained offline. The embodiment uses more than 5000 hours of historical metallurgical process records as a training data set, covering a variety of typical working condition fluctuation scenarios. The embodiment uses the root mean square error (RMSE) as the optimization target, minimizes the residual error between the predicted feature vector and the real historical feature sequence, and uses the root mean square propagation (RMSprop) optimizer to perform parameter fitting. The embodiment continuously performs weight updates until the prediction accuracy tolerance on the validation set reaches the preset accuracy standard, ensuring that the trend prediction result has extremely high deduction reliability in logic.

[0071] In step S14, the predicted state evolution trajectory is mapped to the physical space by using a preset inverse mapping function constructed by the decoder part of the variational autoencoder, to obtain a multi-path state representation in the physical space, including:

[0072] Component dimension disassembly is performed on the predicted state evolution trajectory, to extract a trajectory feature vector sequence;

[0073] The trajectory feature vector sequence is input into a preset inverse mapping function constructed by the decoder part, and dimension recovery operation is performed by using the hidden layer space manifold distribution features learned by the decoder part in advance, to obtain an initial physical space coordinate set;

[0074] The initial physical space coordinate set is subjected to iterative correction processing by using a preset physical constraint condition, to obtain a corrected physical space coordinate;

[0075] The time step information and physical components in the corrected physical space coordinate are integrated and mapped to synthesize the multi-path state representation in the physical space.

[0076] In an implementation manner, component dimension disassembly is performed on the predicted state evolution trajectory by the embodiment. The predicted state evolution trajectory is a low-dimensional feature sequence output by a previous step and carrying a future instability trend. The embodiment equally divides the predicted state evolution trajectory on the time axis according to a preset time sampling step (for example, 10 seconds), to extract the trajectory feature vector sequence. The selection of the time sampling step is based on balancing the smoothness of deduction and the calculation efficiency, to ensure that no key state turning point is lost in the restoration process. Each vector in the trajectory feature vector sequence represents a state representation in the hidden layer space at a specific time.

[0077] In an implementation manner, the trajectory feature vector sequence is input into a preset inverse mapping function constructed by the decoder part by the embodiment. The decoder part is constructed as a deep neural network symmetrical to the structure of the encoder part, and includes three one-dimensional transpose convolution layers. The embodiment performs dimension recovery operation by using the decoder part. Specifically, the model uses the weight parameters generated by learning the distribution of a large amount of physical space data in the pre-training stage, to project the 256-dimensional hidden variables back to the original high-dimensional feature space by performing transpose convolution operation, to obtain the initial physical space coordinate set. Since the infrared image features are introduced at the input end, the coordinate set not only includes the preliminary restored smelting temperature values and furnace pressure values, but also includes feature components representing the spatial distribution.

[0078] It should be noted that the embodiment utilizes preset physical constraint conditions to perform iterative correction processing on the initial physical space coordinate set. The preset physical constraint conditions include mass conservation constraint and energy balance constraint. The mass conservation constraint specifically requires that the absolute value of the difference between the total input and the total output and the residual amount of each key element (such as carbon and oxygen) in the metallurgical reaction be within a preset residual range; the energy balance constraint specifically requires that the heat input (such as arc heat production and chemical reaction heat release) and the heat output (such as cooling water heat removal, furnace heat dissipation, and steel liquid sensible heat) meet the thermodynamic equilibrium relationship. The significance of this step is to correct the non-physical distortion that may be generated by the neural network under pure data driving, and to ensure that the simulation result meets the metallurgical mechanism.

[0079] It should be noted that the embodiment performs iterative correction processing in the following specific manner: a penalty function including multiple physical mechanism deviations is constructed. The penalty function is composed of the weighted sum of the absolute values of the deviations of the aforementioned physical constraint conditions. The embodiment uses the gradient descent method to fine-tune the initial physical space coordinate set, performs recursive iteration operation by minimizing the penalty function, and stops until the coordinate value change amount of two consecutive iterations is less than a preset convergence threshold, thereby obtaining the corrected physical space coordinates.

[0080] In an implementation manner, the convergence threshold is set to 0.15. The selection of this value is based on the static measurement accuracy (for example, 1.5 degrees Celsius) of core hardware sensors such as platinum rhodium thermocouples in the multi-source sensor cluster, and the convergence threshold is set to 0.1 times of the lowest accuracy, so as to ensure that the corrected value has substantial physical meaning and will not fall below the level of hardware measurement noise due to excessive iteration.

[0081] For example, if the initial smelting temperature value at a certain time restored by the decoder is 1605 degrees Celsius, but the energy balance equation calculated according to the current power supply power and cooling intensity shows that there is a large heat surplus, the embodiment fine-tunes this temperature value to 1598 degrees Celsius in the direction of satisfying the heat balance through iterative correction processing, so as to serve as the corrected physical space coordinate at this time.

[0082] It should be noted that the embodiment integrates the time step information and the physical component in the corrected physical space coordinates, and maps and synthesizes the multi-path state representation in the physical space. The embodiment associates each corrected physical space coordinate at a time step with its corresponding time stamp, and reorganizes them according to the time axis dimension to finally construct a whole-process trajectory including multi-dimensional physical attribute values and meeting the timing logic, which is determined as the multi-path state representation in the physical space. Through the above recursive restoration and physical correction logic, the embodiment eliminates the numerical drift problem commonly seen in traditional simulation, and realizes accurate regression of the virtual space to the physical law.

[0083] In step S15, according to the multi-path state representation in the physical space, the preset associated constraint is fused to perform potential risk path screening, and a risk path set is obtained, including:

[0084] The multi-path state representation in the physical space is retrieved, from which temperature gradient values, pressure fluctuation values and component concentration values are separated to obtain a coupling feature vector;

[0085] The coupling feature vector is used to perform a matrix mapping operation to determine a state evolution manifold under a normal working condition;

[0086] The geodesic distance of the multi-path state representation in the physical space relative to the state evolution manifold is calculated to obtain an abnormal evolution metric value;

[0087] If the abnormal evolution metric value exceeds a preset critical threshold boundary, path filtering processing is performed according to a pre-calculated cumulative risk weight to obtain the risk path set.

[0088] It should be noted that the preset associated constraint includes a process coupling constraint and a safety boundary constraint. The process coupling constraint is based on a metallurgical mechanism, for example, an empirical correlation between gas flow and top pressure in a blast furnace, a quantitative relationship model between oxygen blowing intensity and decarburization rate. The safety boundary constraint is a hard limit set according to equipment manuals and production regulations, for example, the hot spot temperature of the furnace lining must not exceed the maximum allowable temperature of the refractory material; the carbon monoxide concentration in the flue gas must not be higher than 50% of the lower explosive limit. When screening the risk path, the system will check whether the multi-path state representation violates these constraints. Any violation at any time point will result in the path being marked as a potential risk.

[0089] In an implementation manner, the embodiment retrieves the multi-path state representation in the physical space. The multi-path state representation in the physical space is a time series trajectory with a clear physical dimension produced by a previous step. The embodiment separates temperature gradient values, pressure fluctuation values and component concentration values from the multi-path state representation to obtain a coupling feature vector sequence.

[0090] In an implementation manner, the embodiment performs spatial feature extraction on the multi-path state representation in the physical space. For the physical state of each time step, the temperature gradient value is obtained by performing a first-order partial differential operation on adjacent sampling points. The pressure fluctuation value is determined by using a high-pass filtering algorithm to separate high-frequency components from the total pressure signal. The time-varying data of a key chemical component (such as the percentage of carbon content in molten steel) is extracted as the component concentration value. The above three types of values are spliced to construct the multi-dimensional coupling feature vector sequence.

[0091] It should be noted that the embodiment utilizes the coupling feature vector sequence to perform a local linear embedding (LLE) mapping operation to determine a state evolution manifold under normal working conditions. The state evolution manifold represents the essential topological structure of the metallurgical production in the stable operation interval. The embodiment calls a pre-trained local linear embedding mapping matrix to project the coupling feature vector sequence into a low-dimensional manifold coordinate system.

[0092] It should be noted that the mapping matrix used in the local linear embedding (LLE) mapping operation is obtained by minimizing manifold reconstruction residual fitting in the pre-training stage. In the embodiment, the engineering calibration method is used to determine the manifold neighbor number parameter k as 15. The specific determination method is to perform a grid search algorithm in the offline stage to find the number of nearest neighbor points that minimizes the reconstruction residual variance, ensuring that the evolution manifold can filter out sensor random noise and accurately retain the dynamic continuity of the metallurgical reaction.

[0093] It should be noted that the embodiment calculates the geodesic distance of the multi-path state representation in the physical space with respect to the state evolution manifold to obtain an abnormal evolution metric value. Compared with the Euclidean distance in linear space, the geodesic distance can measure the actual evolution length of the path along the nonlinear manifold surface, and more accurately reflect the degree of deviation of the metallurgical state. First, a k-neighbor graph is constructed using the low-dimensional manifold coordinate point set obtained by the local linear embedding (LLE) mapping. Each manifold coordinate point is connected to its k nearest points (k = 15) in Euclidean distance, and the edge weight is the Euclidean distance between the two points. Then, the mean point of all points on the normal working condition manifold is taken as the source point, and the projection point of the simulated state point to be evaluated on the manifold (determined by LLE inverse transformation or nearest neighbor search) is taken as the target point. The Dijkstra algorithm is used to calculate the shortest path length on the k-neighbor graph, which is the geodesic distance as the abnormal evolution metric value.

[0094] It should be noted that the embodiment determines whether the abnormal evolution metric value exceeds a preset critical threshold boundary. The preset critical threshold boundary is determined by statistical analysis. The geodesic distance distribution corresponding to the preset historical stable running data set is searched, and the 99.5th percentile of the distribution is extracted as the abnormal triggering critical value. If the current abnormal evolution metric value exceeds the critical value, it is determined that the corresponding path has potential risks.

[0095] In an implementation, the embodiment performs path filtering processing according to a pre-calculated cumulative risk weight to obtain the risk path set. The pre-calculated cumulative risk weight is determined by an engineering calibration method. Specifically, the embodiment performs sensitivity analysis by using a partial derivative matrix method, determines a risk coefficient of each dimension by calculating a partial derivative of a terminal finished product cost function with respect to a temperature gradient, a pressure fluctuation, and a concentration anomaly, and performs linear weighted summation in combination with a time step length of anomaly persistence. The embodiment performs risk score accumulation on a path where an anomaly is detected, retains a path with a total score higher than a preset safety threshold, and thereby constructs the risk path set.

[0096] For example, if a path has a temperature gradient anomaly for 30 seconds, a geodesic distance continuously remains 1.2 times or more than the critical threshold boundary, and a risk score corresponding to the path reaches a preset high risk level (for example, the score is higher than a set threshold 4.0) after accumulation calculation, the embodiment determines the path as a risk path and stores the path in the risk path set.

[0097] It should be noted that the embodiment realizes automatic high-precision filtering of massive multi-path evolution results under the premise of ensuring physical logic rigor by using the above filtering mechanism based on local linear embedding manifold learning and geodesic distance, and greatly reduces the computational overhead of subsequent virtual simulation scene construction.

[0098] In step S16, a key evolution node is extracted from the risk path set, and a pre-trained node expansion generative adversarial model is used to perform data enhancement mapping processing on the key evolution node to obtain a virtual simulation scene, including:

[0099] Identifying a state mutation position in the risk path set to determine the key evolution node;

[0100] Inputting the key evolution node into the node expansion generative adversarial model to perform feature derivation mapping processing and output an expansion node data set topologically associated with the key evolution node;

[0101] Reformatting and recombining the expansion node data set and the key evolution node to obtain virtual scene data, and determining a scene integrity measurement value according to a preset data coverage rate of the virtual scene data in a feature space;

[0102] If the scene integrity measurement value reaches a preset scene integrity threshold, the virtual simulation scene is output.

[0103] In an implementation manner, the embodiment identifies a state mutation position in the set of risk paths, and determines the key evolution node. The state mutation position represents a moment when a physical feature in a metallurgical evolution path has a nonlinear dramatic change. The embodiment performs a second-order time derivative operation on each physical quantity in the set of risk paths, and calculates a state evolution acceleration sequence.

[0104] In an implementation manner, the embodiment determines a mutation threshold value by statistical analysis. The embodiment extracts a state evolution acceleration sequence in a normal working condition in historical production data, and calculates a standard deviation of each dimension ; the embodiment sets the mutation threshold value as. If the absolute value of the evolution acceleration of a node exceeds the mutation threshold value, the embodiment determines that the position is the state mutation position, and determines the physical state data corresponding to the position as the key evolution node. By introducing a mutation determination logic based on a standard deviation multiple, the embodiment can automatically and objectively capture an evolution inflection point with high risk representation significance.

[0105] It should be noted that the embodiment inputs the key evolution node into the pre-trained node expansion generative adversarial model, performs feature derivation mapping processing, and outputs an expansion node data set topologically associated with the key evolution node. Since the spatial features of the infrared image are extracted by the mixed convolutional neural network in S11, the key evolution node not only contains time series information, but also carries spatial field features. The node expansion generative adversarial model is constructed as a conditional generative adversarial network, and the generator network of the model adopts a multi-layer fully connected residual structure. The embodiment performs feature derivation mapping processing, uses the spatio-temporal physical components of the key evolution node as a condition vector, performs local manifold sampling in the latent space of the model, and outputs a plurality of derived state points in the topological field of the key node that satisfy physical continuity through the generator network, thereby constructing the expansion node data set. The feature derivation processing complements the evolution details missing in the risk path due to the limitation of the sampling frequency; specifically, the feature vector of the key evolution node is input into the generator network as a condition, and a random noise vector is sampled from the standard normal distribution and concatenated with the condition vector in the latent space; the generator network generates a new feature vector similar to the key node in the feature space (i.e., the Euclidean distance is less than a preset radius) and consistent with the physical distribution learned from the training data based on the concatenated vector, i.e., an expansion node, by forward propagation, and a set of topologically associated (i.e., in the local neighborhood of the feature space) expansion node data sets can be generated by repeating this process and fine-tuning the noise input.

[0106] It is worth noting that the node expansion generative adversarial model is fitted by using a set of samples generated by simulating historical sparse abnormal samples in combination with mechanism models in an offline training stage. The expanded node data set is formatted and reorganized with the key evolution nodes to obtain virtual scene data. The virtual scene data represents the multi-dimensional state distribution in the metallurgical furnace in the form of a structured tensor, and the dimensions are defined as wherein is the time step sequence length, and represents the spatial grid coordinates in the furnace, represents the physical characteristic channel.

[0107] It is worth noting that the scene integrity measure value is determined according to the preset data coverage of the virtual scene data in the feature space. In order to avoid the calculation obstacle (dimension disaster) caused by directly calculating the convex hull volume in the high-dimensional feature space, the feature space coverage of the expanded node data set relative to the historical full set data is used for calculation. Specifically, the proportion of nodes in the expanded node data set that fall within the hyper-sphere boundary constructed by the historical normal samples is calculated, and this proportion value is determined as the scene integrity measure value. This measurement method not only has better operability in engineering, but also can truly reflect the coverage degree of the generated scene to the known safety boundary.

[0108] Exemplarily, the scene integrity measure value is calculated as the ratio of the number of covered grid points to the total number of grid points.

[0109] It is worth noting that the scene integrity measure value is determined by the performance driven method. In the simulation experiment stage, the contribution rate of scene data with different integrity to the downstream optimization simulation accuracy is calculated, the performance inflection point balancing sensitivity and specificity is found by detecting performance analysis, and the value corresponding to the inflection point is set as the scene integrity threshold, for example, set as 0.88. If the scene integrity measure value reaches the threshold, the virtual simulation scene is output.

[0110] In another implementation, if the scene integrity measure value does not reach the preset scene integrity threshold, the sampling parameters of the node expansion generative adversarial model are adjusted to increase the search radius of the latent space, and the feature derivation mapping process is re-executed until the scene data meets the integrity requirement. The above process realizes the deep transformation from discrete abnormal prediction to complete simulation scene with spatial dimension information.

[0111] In step S17, according to the virtual simulation scene, the preset process adjustment parameters are simulated and optimized to determine the optimized parameter configuration, including:

[0112] The virtual simulation scene is indexed and sliced according to the feature channel dimension, and a furnace temperature distribution feature vector and a component concentration time-varying sequence are extracted;

[0113] The furnace temperature distribution feature vector is subjected to reverse standardization and thermodynamic temperature conversion processing based on preset statistical parameters to obtain a furnace temperature physical distribution matrix;

[0114] The furnace temperature physical distribution matrix and the component concentration time-varying sequence are mapped to a preset metallurgical kinetics evolution model to perform simulation operation, and a reaction rate fluctuation curve and a slag layer thickness distribution atlas are obtained;

[0115] If the reaction rate fluctuation curve exceeds a preset steady-state operation interval, a key process adjustment parameter set is located from the preset process adjustment parameters according to the slag layer thickness distribution atlas;

[0116] The key process adjustment parameter set is subjected to multi-objective particle swarm optimization calculation using a normalized multi-objective fitness function, and an optimized parameter value is obtained. The preset process adjustment parameters are updated using the optimized parameter value to determine the optimized parameter configuration.

[0117] In an implementation manner, the virtual simulation scene is indexed and sliced according to the feature channel dimension. The virtual simulation scene is a structured four-dimensional tensor output by a previous step, and the dimension definition is wherein represents a time step sequence length, represents a spatial grid coordinate in the furnace, represents a physical feature channel. In this embodiment, the temperature values of all time and space points are extracted by fixing the index number of the corresponding temperature field in the dimension of , and a furnace temperature distribution feature vector is formed. Similarly, the index number of the corresponding component concentration in the dimension of is fixed to extract a component concentration time-varying sequence.

[0118] In an implementation manner, the furnace temperature distribution feature vector is subjected to reverse standardization and thermodynamic temperature conversion processing based on preset statistical parameters. Since the original data is subjected to standardization processing in the input stage, the obtained feature vector is a dimensionless statistical value. In this embodiment, the arithmetic mean value and the standard deviation of a preset historical observation period (for example, the last 24 hours) stored in S11 are called to perform reverse standardization calculation: ​

[0119]

[0120] wherein, is the normalized element value in the furnace temperature profile vector. Subsequently, the reduced Celsius value is added by 273.15 to convert into Kelvin temperature conforming to the law of thermodynamics, to obtain the thermodynamic temperature , so as to construct the furnace temperature physical distribution matrix.

[0121] It should be noted that the present embodiment utilizes a preset metallurgical kinetics evolution model to perform simulation operation. The model integrates the Arrhenius equation:

[0122]

[0123] wherein, is the reaction rate constant, is a preset pre-exponential factor, is a preset reaction activation energy, the value of which is determined according to the kinetics constant of the chemical reaction (such as carbon oxidation reaction or dephosphorization reaction) dominated in the current metallurgical stage, is the ideal gas constant, is the thermodynamic temperature calculated above. The present embodiment outputs the reaction rate fluctuation curve and the slag layer thickness distribution map by solving the equation in combination with the component mass conservation equation.

[0124] It is worth noting that the preset metallurgical kinetics evolution model is a coupled equation system solver; first, the furnace temperature physical distribution matrix is input into the Arrhenius equation to calculate the reaction rate constant at different times at each spatial position, then is input into a group of mass conservation differential equations (describing the oxidation reactions of carbon, silicon, manganese and the like) together with the time-varying sequence of component concentration, numerical integration is performed by using the Runge-Kutta method, the change rate of the concentration of each component with time and space is solved, and then the new concentration distribution and the reaction rate fluctuation curve are integrated; at the same time, based on the oxide amount generated by the reaction, the slag basicity model and the simplified model of the molten pool flow, the slag layer thickness distribution map is calculated and output. The model has used a large amount of historical production data to calibrate the key parameters (such as pre-exponential factor and activation energy) in the offline stage.

[0125] It is worth noting that the embodiment determines whether the reaction rate fluctuation curve exceeds the preset steady-state operation interval. If it exceeds, the embodiment locates the process variable with the largest contribution to the current reaction fluctuation according to the thickness gradient of each spatial grid point in the slag layer thickness distribution map through a sensitivity analysis matrix. The sensitivity analysis matrix constructed by the embodiment is based on the perturbation method. By sequentially injecting a small perturbation (for example, a set value of 1%) into each process adjustment parameter in the simulation operation, the feedback rate of the reaction rate fluctuation curve is observed, so as to quantify the contribution weight of each parameter (such as oxygen supply intensity, bottom blowing flow, etc.), and determine the key process adjustment parameter set.

[0126] Exemplarily, the embodiment performs multi-objective particle swarm optimization calculation on the key process adjustment parameter set. The embodiment sets the particle swarm size to 50 and the maximum iteration number to 100. The parameter setting is based on balancing the calculation overhead and the convergence accuracy. Through experimental calibration, the algorithm can converge to the optimal solution within 2 seconds under the scale of 50 particles, meeting the real-time interaction and high-frequency tuning requirements of the digital twin system. The embodiment constructs a normalized multi-objective fitness function

[0127]

[0128] wherein, is the reaction rate target value, is the slag layer thickness distribution variance, is the historical reference variance benchmark. By introducing the relative deviation form for normalization processing, the embodiment eliminates the dimensional and numerical order gap between different physical quantities, ensuring the balance of the optimization direction. The embodiment determines the optimal parameter value by minimizing the global optimal solution of the function , and completes the configuration update of the process parameters accordingly.

[0129] It is worth noting that the embodiment solves the adaptation problem between high-dimensional statistical characteristics and physical mechanism models through the above reverse normalization conversion, sensitivity analysis based on the perturbation method, and normalization optimization logic, and realizes the precise decision-making closed loop from virtual risk deduction to entity process parameter adjustment.

[0130] In summary, the embodiment discloses an intelligent metallurgical process virtual simulation method based on digital twinning. The deep integration of variational autoencoder and generative adversarial model realizes the accurate restoration of complex metallurgical processes, and the coupling of metallurgical dynamics model and optimization algorithm realizes the closed-loop decision-making from virtual risk prediction to entity process parameters. Through the above technical solutions, the present application solves the technical problem that it is difficult to construct a comprehensive virtual simulation scene that truly restores the physical production law in the prior art. ​​

[0131] Referring to Figure 2 The second embodiment of the present application provides a digital-twin-based intelligent metallurgical process virtual simulation system, comprising:

[0132] a data dimension reduction module, configured to acquire multi-modal high-dimensional data of a metallurgical process, perform dimension reduction processing on the high-dimensional data by using an encoder part in a pre-trained variational autoencoder, and obtain a feature vector set in a hidden layer space;

[0133] a multi-path simulation module, configured to perform random disturbance injection simulation by using a pre-trained time series generative adversarial model according to the feature vector set in the hidden layer space, and obtain a multi-path state sequence with time dependence;

[0134] a state prediction module, configured to, if there is an abnormal path deviating from a preset path deviation threshold in the multi-path state sequence, perform time series evolution trend prediction on the abnormal path by using a pre-trained recurrent neural network, and obtain a predicted state evolution trajectory;

[0135] an inverse mapping module, configured to map the predicted state evolution trajectory to a physical space by using a preset inverse mapping function constructed by a decoder part in the variational autoencoder, and obtain a multi-path state representation in the physical space;

[0136] a risk screening module, configured to perform potential risk path screening according to the multi-path state representation in the physical space by fusing a preset correlation constraint, and obtain a risk path set;

[0137] a scenario generation module, configured to extract a key evolution node from the risk path set, and perform data augmentation mapping processing on the key evolution node by using a pre-trained node expansion generative adversarial model, and obtain a virtual simulation scenario;

[0138] a parameter optimization module, configured to perform optimization simulation on a preset process adjustment parameter according to the virtual simulation scenario, and determine an optimized parameter configuration.

[0139] It should be noted that the digital-twin-based intelligent metallurgical process virtual simulation system provided by the embodiment of the present application is used to execute all process steps of the digital-twin-based intelligent metallurgical process virtual simulation method of the above-mentioned embodiment, and the working principles and beneficial effects of the two are one-to-one correspondence, thus no longer being described.

[0140] The embodiment of the present application further provides an electronic device. The electronic device comprises a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a digital-twin-based intelligent metallurgical process virtual simulation program. The processor executes the computer program to implement the steps in each of the above-mentioned digital-twin-based intelligent metallurgical process virtual simulation method embodiments, for exampleFigure 1 The processor executes the computer program to implement the functions of the modules / units in the above-mentioned system embodiments, such as the data dimension reduction module.

[0141] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the electronic device.

[0142] The electronic device can be a desktop computer, a notebook computer, a palm computer, a smart tablet, and the like. The electronic device can include, but is not limited to, a processor, a memory. Those skilled in the art can understand that the above components are only examples of the electronic device and do not constitute a limitation on the electronic device, and the electronic device can include more or less components than the above, or combine certain components, or different components, for example, the electronic device can also include an input / output device, a network access device, a bus, and the like.

[0143] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, and the like. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, and the like. The processor is the control center of the electronic device, and connects all parts of the electronic device through various interfaces and lines.

[0144] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the electronic device by running or executing the computer program and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function (such as a sound playing function, an image playing function, etc.), and the like; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), and the like. In addition, the memory can include a high-speed random access memory, and can also include a nonvolatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0145] The modules / units integrated in the electronic device, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can realize the steps of the above-mentioned various method embodiments when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the contents included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0146] It should be noted that the system embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e. can be located in one place or distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. In addition, the connection relationship between the modules in the system embodiment provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0147] The above specific embodiments further illustrate the purpose, technical scheme and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A virtual simulation method for intelligent metallurgical processes based on digital twins, characterized in that, include: Multimodal high-dimensional data of the metallurgical process is acquired, and the high-dimensional data is reduced in dimensionality using the encoder part of a pre-trained variational autoencoder to obtain a set of feature vectors in the hidden layer space. Based on the feature vector set of the hidden layer space, a random perturbation injection simulation is performed using a pre-trained temporal generative adversarial model to obtain a time-dependent multipath state sequence. If there is an abnormal path in the multi-path state sequence that deviates from the preset path deviation threshold, the temporal evolution trend of the abnormal path is predicted by a pre-trained recurrent neural network to obtain the predicted state evolution trajectory. Using the decoder part of the variational autoencoder, a preset inverse mapping function is constructed to map the predicted state evolution trajectory to the physical space, thereby obtaining a multipath state representation in the physical space. Based on the multi-path state representation in the physical space, potential risk paths are screened by integrating preset association constraints to obtain a set of risk paths; Key evolutionary nodes are extracted from the risk path set, and data augmentation mapping is performed on the key evolutionary nodes using a pre-trained node extension generative adversarial model to obtain a virtual simulation scenario. Based on the virtual simulation scenario, the preset process adjustment parameters are optimized and simulated to determine the optimized parameter configuration.

2. The intelligent metallurgical process virtual simulation method based on digital twins according to claim 1, characterized in that, The process of acquiring multimodal high-dimensional data of the metallurgical process involves using the encoder portion of a pre-trained variational autoencoder to perform dimensionality reduction on the high-dimensional data, resulting in a set of feature vectors in the hidden layer space, including: Acquire multimodal high-dimensional data of the metallurgical process, and construct an initial high-dimensional matrix using the multimodal high-dimensional data; The initial high-dimensional matrix is ​​input into the encoder part, and the encoder part performs multimodal hybrid convolution mapping processing to project the initial high-dimensional matrix into a low-dimensional manifold space. The latent variable distribution parameters of the low-dimensional manifold space are extracted and determined as the feature vector set of the latent layer space.

3. The intelligent metallurgical process virtual simulation method based on digital twins according to claim 1, characterized in that, The step of using a pre-trained temporal generative adversarial model to simulate random perturbation injection based on the feature vector set of the hidden layer space to obtain a time-dependent multipath state sequence includes: A preset random disturbance factor and a preset random noise vector are invoked, and the random disturbance factor is superimposed on the random noise vector to synthesize a disturbance input vector set; The set of feature vectors in the hidden layer space and the set of perturbation input vectors are synchronously input into the temporal generative adversarial model; The temporal dependency mapping is performed using the built-in time step attention mechanism of the temporal generative adversarial model to obtain a set of interrelated simulated instantaneous states; According to the generated time order, the data points in the simulated instantaneous state set are subjected to time sequence connection processing to construct the multipath state sequence.

4. The intelligent metallurgical process virtual simulation method based on digital twins according to claim 1, characterized in that, If there are abnormal paths in the multi-path state sequence that deviate from a preset path deviation threshold, a pre-trained recurrent neural network is used to predict the temporal evolution trend of the abnormal paths to obtain the predicted state evolution trajectory, including: Extract the state vector from the multipath state sequence and calculate the Euclidean distance between the state vector and the preset standard reference vector to obtain the path deviation numerical set. If the value in the set of path deviation values ​​exceeds the preset path deviation threshold, the corresponding continuous abnormal node segments in the multi-path state sequence are extracted and determined as abnormal path data to be processed. The abnormal path data to be processed is input into the pre-trained recurrent neural network to perform feature hidden layer extraction processing to obtain the hidden layer state vector; The hidden layer state vector is used to perform multiple rounds of recursive iterative calculations to obtain the continuous evolution prediction value; Sequence recombination processing is performed based on the continuous evolution prediction values ​​to obtain the predicted state evolution trajectory.

5. The intelligent metallurgical process virtual simulation method based on digital twins according to claim 1, characterized in that, The preset inverse mapping function constructed using the decoder part of the variational autoencoder maps the predicted state evolution trajectory to the physical space, obtaining a multipath state representation in the physical space, including: The predicted state evolution trajectory is decomposed into component dimensions to extract the trajectory feature vector sequence; The trajectory feature vector sequence is input into the preset inverse mapping function constructed by the decoder part, and the dimension recovery operation is performed using the hidden layer spatial manifold distribution features learned in advance by the decoder part to obtain the initial physical space coordinate set; The initial physical space coordinate set is iteratively corrected using preset physical constraints to obtain corrected physical space coordinates; By integrating the time step information and physical components in the corrected physical space coordinates, a multipath state representation in the physical space is mapped and synthesized.

6. The intelligent metallurgical process virtual simulation method based on digital twins according to claim 1, characterized in that, The step involves filtering potential risk paths based on the multi-path state representation in the physical space and integrating preset association constraints to obtain a set of risk paths, including: Retrieve the multipath state representation in the physical space, and separate the temperature gradient value, pressure fluctuation value and component concentration value from it to obtain the coupled feature vector; Matrix mapping operations are performed using the coupled feature vectors to determine the state evolution manifold under normal operating conditions; Calculate the geodesic distance of the multipath state representation in the physical space relative to the state evolution manifold to obtain the anomalous evolution metric. If the abnormal evolution metric exceeds the preset critical threshold boundary, then path filtering is performed according to the pre-calculated cumulative risk weight to obtain the risk path set.

7. The intelligent metallurgical process virtual simulation method based on digital twins according to claim 1, characterized in that, The process of extracting key evolutionary nodes from the risk path set and using a pre-trained node-extended generative adversarial model to perform data augmentation mapping on the key evolutionary nodes to obtain a virtual simulation scenario includes: Identify the locations of state mutations in the set of risk paths and determine the key evolutionary nodes; The key evolution node is input into the node extension generative adversarial model, feature derivation mapping processing is performed, and an extended node data set topologically associated with the key evolution node is output. The extended node data set and the key evolution nodes are formatted and recombined to obtain virtual scene data, and the scene integrity metric is determined based on the preset data coverage rate of the virtual scene data in the feature space. If the scene integrity metric reaches the preset scene integrity threshold, the virtual simulation scene is output.

8. The intelligent metallurgical process virtual simulation method based on digital twins according to claim 1, characterized in that, The step of optimizing and simulating preset process adjustment parameters based on the virtual simulation scenario to determine the optimized parameter configuration includes: The virtual simulation scene is indexed and sliced ​​according to the feature channel dimension to extract the furnace temperature distribution feature vector and the time-varying sequence of component concentration; The furnace temperature distribution feature vector is subjected to inverse normalization and thermodynamic temperature conversion based on preset statistical parameters to obtain the furnace temperature physical distribution matrix; The physical distribution matrix of furnace temperature and the time-varying sequence of component concentration are mapped to a preset metallurgical kinetic evolution model to perform simulation calculations, thereby obtaining the reaction rate fluctuation curve and slag layer thickness distribution map. If the reaction rate fluctuation curve exceeds the preset steady-state operating range, then the set of key process adjustment parameters is located from the preset process adjustment parameters based on the slag layer thickness distribution map. The set of key process adjustment parameters is subjected to multi-objective particle swarm optimization calculation using a normalized multi-objective fitness function to obtain optimized parameter values. The optimized parameter values ​​are then used to update the preset process adjustment parameters to determine the optimized parameter configuration.

9. A digital twin-based intelligent metallurgical process virtual simulation system, characterized in that, include: The data dimensionality reduction module is used to acquire multimodal high-dimensional data of the metallurgical process, and to perform dimensionality reduction processing on the high-dimensional data using the encoder part of the pre-trained variational autoencoder to obtain a set of feature vectors in the hidden layer space. The multi-path simulation module is used to simulate random perturbation injection using a pre-trained temporal generative adversarial model based on the feature vector set of the hidden layer space, to obtain a time-dependent multi-path state sequence. The state prediction module is used to predict the temporal evolution trend of the abnormal path if there is an abnormal path in the multi-path state sequence that deviates from a preset path deviation threshold, by using a pre-trained recurrent neural network to obtain the predicted state evolution trajectory. The inverse mapping module is used to map the predicted state evolution trajectory to the physical space using a preset inverse mapping function constructed by the decoder part of the variational autoencoder, so as to obtain a multipath state representation in the physical space. The risk screening module is used to screen potential risk paths based on the multi-path state representation in the physical space and by integrating preset association constraints, thereby obtaining a set of risk paths. The scene generation module is used to extract key evolution nodes from the risk path set and use a pre-trained node extension generative adversarial model to perform data augmentation mapping on the key evolution nodes to obtain a virtual simulation scene. The parameter optimization module is used to optimize and simulate preset process adjustment parameters based on the virtual simulation scenario, and determine the optimized parameter configuration.

Images