Zinc rotary kiln digital twin model parameter updating method, device, equipment and medium

Abstract

The present invention discloses a method, device, equipment and medium for updating parameters of a zinc rotary kiln digital twin model. The method comprises: constructing a target mechanism model of the zinc rotary kiln by analyzing the multi-field coupling mechanism of the zinc rotary kiln; acquiring the observation data of the zinc rotary kiln in the form of discrete time series data, and performing time series data analysis and image processing on the observation data to extract the complementary data corresponding to the state of the zinc rotary kiln; constructing an improved LSTM twin parameter generation network according to the reaction mechanism analysis of the zinc rotary kiln; inputting the complementary data into the improved LSTM twin parameter generation network for data training, and dynamically updating the parameters of the target mechanism model according to the training results. Since the present invention performs time series data analysis and image processing on the observation data, and then inputs the complementary data corresponding to the state of the zinc rotary kiln into the improved LSTM twin parameter generation network for training, the effectiveness of the parameter update of the zinc rotary kiln digital twin model is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of metallurgical equipment modeling and application technology, and in particular to a method, device, equipment and medium for updating parameters of a digital twin model of a zinc rotary kiln. Background Art

[0002] The zinc rotary kiln is a key equipment for zinc recovery in the zinc smelting industry. Its stable operation and efficient zinc recovery are of great significance for energy conservation and carbon reduction, which requires accurate monitoring of key variables in its operation process. However, the zinc rotary kiln has a large axial length, closed internal space, complex reaction mechanism, and unclear internal working conditions, making it difficult to directly measure the key variables of the reaction operation conditions. Virtual-real fusion and virtual-real synchronization digital. It is a necessary means to achieve real-time and accurate monitoring of key variables of the zinc rotary kiln.

[0003] However, in the actual production process, model parameters will change with factors such as environmental changes, operating conditions, and equipment wear. Therefore, the model needs to be corrected in real time to meet the virtual-real synchronization requirements of digital twins. At present, the dynamic update of industrial equipment digital twins mainly includes data-driven updates, that is, using real-time sensor data or monitoring information to dynamically update the parameters or status of the digital twin model. This includes using machine learning algorithms to analyze real-time data, and then updating the parameters or status of the digital twin model based on the analysis results; real-time optimization algorithm updates, that is, using real-time optimization algorithms, such as model predictive control (MPC), to optimize the digital twin model online. These algorithms can adjust the parameters of the digital twin model according to the real-time changing process conditions and equipment status to achieve the best control effect. However, the existing digital twin model update methods also have some shortcomings. The existing methods lack sufficient model adaptability and cannot respond to environmental changes and system dynamics in real time.

[0004] Therefore, there is an urgent need for a parameter updating method for the digital twin model of a zinc rotary kiln, which can solve the technical problem that the parameter update of the existing digital twin model of a zinc rotary kiln cannot respond to environmental changes and system dynamics in real time. Summary of the invention

[0005] The main purpose of the present invention is to provide a method, device, equipment and medium for updating the parameters of a zinc rotary kiln digital twin model, aiming to solve the technical problem that the parameter update of the existing zinc rotary kiln digital twin model cannot respond to environmental changes and system dynamics in real time.

[0006] To achieve the above object, the present invention provides a method for updating parameters of a zinc rotary kiln digital twin model, the method comprising the following steps:

[0007] By analyzing the multi-field coupling mechanism of the zinc rotary kiln, a target mechanism model of the zinc rotary kiln is constructed;

[0008] Acquire observation data of the zinc rotary kiln in the form of discrete time series data, perform time series data analysis and image processing on the observation data, and extract complementary data corresponding to the state of the zinc rotary kiln;

[0009] According to the reaction mechanism analysis of the zinc rotary kiln, an improved LSTM twin parameter generation network is constructed, wherein the improved LSTM twin parameter generation network includes a rolling window technology or an instant learning adaptive mechanism;

[0010] The completed data is input into the improved LSTM twin parameter generation network for data training, and the parameters of the target mechanism model are dynamically updated according to the training results.

[0011] Optionally, the supplementary data includes DCS data features and image features, and after the steps of acquiring the observation data of the zinc rotary kiln in the form of discrete time series data, performing time series data analysis and image processing on the observation data, and extracting the supplementary data corresponding to the state of the zinc rotary kiln, the method further includes:

[0012] Performing cross-modal feature fusion on the DCS data features and the image features using cross-modal attention bilateral residual feature fusion to obtain a first data fusion feature and a second data fusion feature;

[0013] Respectively obtaining residual information of the first data fusion feature and residual information of the second data fusion feature through a frame-level residual learning function;

[0014] The residual information of the first data fusion feature and the residual information of the second data fusion feature are multimodally fused to obtain the completed data after feature fusion.

[0015] Optionally, after the step of performing cross-modal feature fusion on the DCS data features and the image features using cross-modal attention bilateral residual feature fusion to obtain a first data fusion feature and a second data fusion feature, the step further includes:

[0016] Determining an average value of the first data fusion feature and an average value of the second data fusion feature along a time dimension;

[0017] Normalizing an average value of the first data fusion feature, an average value of the second data fusion feature, and a preset bias to obtain a target weight value of multimodal fusion;

[0018] Accordingly, the step of performing multimodal fusion on the residual information of the first data fusion feature and the residual information of the second data fusion feature to obtain the completed data after feature fusion includes:

[0019] Based on the target weight value, the residual information of the first data fusion feature and the residual information of the second data fusion feature are multimodally fused to obtain the completed data after feature fusion.

[0020] Optionally, after the step of acquiring the observation data of the zinc rotary kiln in the form of discrete time series data, performing time series data analysis and image processing on the observation data, and extracting the supplementary data corresponding to the state of the zinc rotary kiln, the method further includes:

[0021] Extracting the short-term operation characteristics and the long-term operation characteristics of the zinc rotary kiln in the supplemented data according to a preset time axis;

[0022] Determining the fusion weights of the short-term running features and the long-term running features through a sequence weighted attention mechanism;

[0023] Based on the fusion weight, the short-term operation feature and the long-term operation feature are fused to obtain the completed data after feature fusion.

[0024] Optionally, after the step of inputting the completed data into the improved LSTM twin parameter generation network for data training and dynamically updating the parameters of the target mechanism model according to the training results, the method further includes:

[0025] Simulating and solving the dynamically updated target mechanism model according to the equipment control parameters of the zinc rotary kiln and the external disturbance data to obtain simulation results;

[0026] Iteratively verifying the consistency between the simulation result and the actual situation of the zinc rotary kiln to achieve dynamic tracking of the zinc rotary kiln;

[0027] The operating status of the zinc rotary kiln is monitored and predicted based on the verification results.

[0028] Optionally, the step of acquiring the observation data of the zinc rotary kiln in the form of discrete time series data, performing time series data analysis and image processing on the observation data, and extracting the complementary data corresponding to the state of the zinc rotary kiln includes:

[0029] Acquiring observation data of the zinc rotary kiln in the form of discrete time series data through a DCS system, wherein the observation data includes DCS data and flame images during the operation of the zinc rotary kiln;

[0030] Based on the experience of on-site fire monitoring and control of the operation mode of the zinc rotary kiln, the working condition analysis and guidance of the zinc rotary kiln are carried out, the DCS data and the flame image are analyzed and processed, and the processed DCS data and the processed flame image are obtained;

[0031] Through time series data analysis and image processing, feature extraction is performed on the processed DCS data and the processed flame image to obtain complementary data corresponding to the zinc rotary kiln state.

[0032] Optionally, the supplemented data includes time domain features, frequency domain features and time-frequency domain features of the DCS data and dynamic features and static features of the flame image.

[0033] In addition, to achieve the above purpose, the present invention also proposes a zinc rotary kiln digital twin model parameter updating device, the device comprising:

[0034] A model building module, used for building a target mechanism model of the zinc rotary kiln by analyzing the multi-field coupling mechanism of the zinc rotary kiln;

[0035] A data processing module, used to obtain observation data of the zinc rotary kiln in the form of discrete time series data, and perform time series data analysis and image processing on the observation data to extract complementary data corresponding to the state of the zinc rotary kiln;

[0036] A network construction module, used for constructing an improved LSTM twin parameter generation network according to the reaction mechanism analysis of the zinc rotary kiln, wherein the improved LSTM twin parameter generation network includes a rolling window technology or an instant learning adaptive mechanism;

[0037] A parameter updating module is used to input the completed data into the improved LSTM twin parameter generation network for data training, and dynamically update the parameters of the target mechanism model according to the training results.

[0038] In addition, to achieve the above-mentioned purpose, the present invention also proposes a zinc rotary kiln digital twin model parameter updating device, the device comprising: a memory, a processor, and a zinc rotary kiln digital twin model parameter updating program stored in the memory and executable on the processor, the zinc rotary kiln digital twin model parameter updating program being configured to implement the steps of the zinc rotary kiln digital twin model parameter updating method as described above.

[0039] In addition, to achieve the above-mentioned purpose, the present invention also proposes a medium, on which a zinc rotary kiln digital twin model parameter update program is stored. When the zinc rotary kiln digital twin model parameter update program is executed by a processor, the steps of the zinc rotary kiln digital twin model parameter update method as described above are implemented.

[0040] The present invention discloses a method for constructing a target mechanism model of a zinc rotary kiln by analyzing a multi-field coupling mechanism of a zinc rotary kiln; obtaining observation data of the zinc rotary kiln in the form of discrete time series data, performing time series data analysis and image processing on the observation data, and extracting complementary data corresponding to the state of the zinc rotary kiln; constructing an improved LSTM twin parameter generation network according to the reaction mechanism analysis of the zinc rotary kiln, wherein the improved LSTM twin parameter generation network includes a rolling window technology or an instant learning adaptive mechanism; inputting the complementary data into the improved LSTM twin parameter generation network for data training, and dynamically updating the parameters of the target mechanism model according to the training results. Since the present invention obtains the observation data of the zinc rotary kiln in the form of discrete time series data, performs time series data analysis and image processing on the observation data, extracts the complementary data corresponding to the state of the zinc rotary kiln, and then inputs the complementary data into an improved LSTM twin parameter generation network including a rolling window technology or an instant learning adaptive mechanism for training, and dynamically updates the parameters of the target mechanism model of the zinc rotary kiln according to the training results, compared with the prior art, the present invention ensures the virtual-real synchronization of the digital twin of the zinc rotary kiln, and solves the technical problem that the parameter update of the existing zinc rotary kiln digital twin model cannot respond to environmental changes and system dynamics in real time. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic flow chart of the first embodiment of the method for updating parameters of the digital twin model of a zinc rotary kiln according to the present invention;

[0042] Figure 2 This is a multi-source observation data analysis diagram in the parameter updating method of the zinc rotary kiln digital twin model of the present invention;

[0043] Figure 3 The improved LSTM twin parameter generation network of the present invention includes a schematic diagram of the rolling window technology;

[0044] Figure 4 The improved LSTM twin parameter generation network of the present invention includes a schematic diagram of an instant learning adaptive mechanism;

[0045] Figure 5 It is a flow chart of the second embodiment of the method for updating parameters of the digital twin model of a zinc rotary kiln according to the present invention;

[0046] Figure 6 It is a schematic diagram of a multimodal data fusion framework in the parameter updating method of the digital twin model of a zinc rotary kiln of the present invention;

[0047] Figure 7 It is a schematic flow chart of a third embodiment of a method for updating parameters of a digital twin model of a zinc rotary kiln according to the present invention;

[0048] Figure 8It is a diagram of the zinc rotary kiln state perception algorithm in the zinc rotary kiln digital twin model parameter updating method of the present invention;

[0049] Fig. 9 It is a structural block diagram of the first embodiment of the digital twin model parameter updating device for zinc rotary kiln of the present invention;

[0050] Fig.10 It is a structural schematic diagram of a zinc rotary kiln digital twin model parameter updating device in a hardware operating environment involved in an embodiment of the present invention.

[0051] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0052] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0053] The embodiment of the present invention provides a method for updating parameters of a zinc rotary kiln digital twin model, referring to Figure 1 , Figure 1 It is a schematic flow chart of the first embodiment of the method for updating parameters of the digital twin model of a zinc rotary kiln according to the present invention.

[0054] In this embodiment, the method for updating parameters of the zinc rotary kiln digital twin model comprises the following steps:

[0055] Step S10: constructing a target mechanism model of the zinc rotary kiln by analyzing the multi-field coupling mechanism of the zinc rotary kiln.

[0056] It should be noted that the execution subject of this embodiment can be a computer server device with data processing, network communication and program running functions, such as a personal computer, a laptop computer, etc., or an electronic device capable of realizing the above functions, a zinc rotary kiln digital twin model parameter updating device, etc. The following takes a system including a zinc rotary kiln digital twin model parameter updating device as an example to illustrate this embodiment and the following embodiments.

[0057] It needs to be explained that the multi-field coupling mechanism analysis involves the interaction and influence between multiple physical fields in the zinc rotary kiln, which can be temperature field, concentration field, pressure field, etc. Multi-field coupling mechanism analysis is an important means to understand and predict the performance of complex systems. By deeply studying the types, mechanisms and analysis methods of multi-field coupling problems, we can better guide engineering design and optimization and improve product performance and reliability.

[0058] It should be understood that the target mechanism model of the zinc rotary kiln is the digital twin model corresponding to the zinc rotary kiln. The principle of the digital twin model is to use the physical model for digital modeling, map the physical production line or equipment to the virtual space, and realize the bridge between the physical world and the digital world by integrating advanced technologies such as big data, cloud computing, the Internet of Things, and artificial intelligence.

[0059] Step S20: acquiring observation data of the zinc rotary kiln in the form of discrete time series data, performing time series data analysis and image processing on the observation data, and extracting complementary data corresponding to the state of the zinc rotary kiln.

[0060] It should be noted that, considering that the changes in the parameters of the zinc rotary kiln under actual working conditions have a certain time series, the observation data of the zinc rotary kiln are obtained in the form of discrete time series data.

[0061] Discrete time series data is in the form of a series of data points arranged in time order, each data point corresponds to a specific time point or time index. These data points can be any quantifiable measurement value, such as temperature, pressure, etc.

[0062] In a specific implementation, the observation data of the zinc rotary kiln can be obtained in the form of discrete time series data through the DCS system, and the observation data includes DCS data and flame images during the operation of the zinc rotary kiln; based on the experience of on-site fire observation and control of the operation mode of the zinc rotary kiln, the working condition analysis guidance of the zinc rotary kiln is carried out, and the DCS data and flame images are analyzed and processed to obtain processed DCS data and processed flame images; through time series data analysis and image processing, feature extraction is performed on the processed DCS data and the processed flame images to obtain complete data corresponding to the state of the zinc rotary kiln.

[0063] It should be understood that a DCS system, or distributed control system, is a control system widely used in the field of industrial automation.

[0064] It is understandable that the DCS data may include the kiln tail temperature (outlet flue gas temperature), kiln head temperature and outlet CO concentration (outlet carbon monoxide concentration) of the zinc rotary kiln.

[0065] It should be noted that in order to reduce the impact of noise and outliers in the data on subsequent analysis, after obtaining the observation data of the zinc rotary kiln, the collected observation data can be preprocessed, including data cleaning, denoising, normalization and other operations.

[0066] It should be explained that the supplemented data includes the time domain features, frequency domain features and time-frequency domain features of the DCS data and the dynamic features and static features of the flame image.

[0067] Step S30: According to the reaction mechanism analysis of the zinc rotary kiln, an improved LSTM twin parameter generation network is constructed, and the improved LSTM twin parameter generation network includes a rolling window technology or an instant learning adaptive mechanism.

[0068] It should be understood that the reaction mechanism of the zinc rotary kiln includes reduction reaction, oxidation reaction and combustion reaction etc. occurring in the zinc rotary kiln.

[0069] It should be noted that since the changes in the parameters of the zinc rotary kiln under actual working conditions have a certain time series, a rolling window technology or an instant learning adaptive mechanism is added to the LSTM twin parameter generation network to obtain an improved LSTM twin parameter generation network, and the parameters of the target mechanism model are dynamically updated according to the improved LSTM twin parameter generation network.

[0070] It should be understood that rolling window technology and just-in-time learning adaptive mechanism are both effective means to cope with dynamically changing environments. Rolling window technology achieves real-time response and optimization to dynamic environments by continuously rolling and updating windows. In each window, the system or algorithm will make decisions or optimizations based on current information and constraints, and then apply the results to the actual system or process. As time goes by, the window will roll forward, new data will be taken into account, and old data will be gradually eliminated. Just-in-time learning adaptive mechanism collects and analyzes data in real time, and continuously adjusts and optimizes the behavior and parameters of the system or algorithm to adapt to changes in the environment.

[0071] Step S40: input the completed data into the improved LSTM twin parameter generation network for data training, and dynamically update the parameters of the target mechanism model according to the training results.

[0072] For example, refer to Figure 2 , Figure 2This is a multi-source observation data analysis diagram in the parameter updating method of the digital twin model of the zinc rotary kiln of the present invention. The limited observation data during the operation of the zinc rotary kiln include DCS data and flame images, wherein the DCS data include kiln head temperature, kiln tail temperature, outlet CO concentration, etc. The observation data is collected and stored in the form of discrete time series data through the DCS system. On-site workers adjust the rotary kiln by observing the fire at the kiln head to judge the combustion status, so the flame image contains a lot of working condition information of the reaction state in the kiln. Therefore, combined with the experience of workers on-site fire and regulating the operation mode of the rotary kiln, the rotary kiln working condition analysis guidance is carried out, and multi-source observation data such as DCS data and flame images are analyzed and processed. Through time series data analysis and image processing, the supplementary data related to the state of the zinc rotary kiln is extracted, and the supplementary data includes DCS data features (DCS data features include time domain features, frequency domain features and time-frequency domain features of DCS data, etc.) and image features (i.e. flame features, including flame brightness, material area, particle size and optical flow vector, etc.). Then, based on the analysis of the reaction mechanism of the zinc rotary kiln, an improved LSTM twin parameter generation network was constructed, and the completed data extracted from the observed data was used as the input feature of the improved LSTM twin parameter generation network. After data training, the dynamic parameters of the zinc rotary kiln digital twin model (i.e., the target mechanism model) were obtained to update the mechanism model coefficients, coupling coefficients and boundary conditions. The dynamic parameters of the zinc rotary kiln digital twin model include mass transfer model parameters, energy transfer model parameters, heat transfer model parameters, chemical reaction parameters, boundary conditions and physical property parameters.

[0073] It should be noted that in order to achieve dynamic tracking of the on-site zinc rotary kiln and ensure the virtual-real synchronization of the digital twin, after step S40, it also includes: simulating and solving the dynamically updated target mechanism model according to the equipment control parameters of the zinc rotary kiln and the external disturbance data to obtain simulation results; iteratively verifying the consistency of the simulation results with the actual situation of the zinc rotary kiln to achieve dynamic tracking of the zinc rotary kiln; and monitoring and predicting the operating status of the zinc rotary kiln based on the verification results.

[0074] For example, refer to Figure 3 and Figure 4 , Figure 3The improved LSTM twin parameter generation network of the present invention includes a schematic diagram of the rolling window technology; the time series data (input variable) is cut into the collected data samples according to the window length by the rolling window technology, and the data in each window is input into the LSTM model as a sample. For example, if the value of a future time point (for example, the temperature field) is to be predicted, the data of the first n time points can be used as input, and these data points are organized into a sample by the rolling window technology, and the model is trained using historical data, and the weights are adjusted by back propagation and gradient descent (or other optimizers) to minimize the prediction error. The loss function (such as the mean square error MSE) is used to measure the difference between the predicted value and the true value. The rolling window technology can be used for continuous multi-step prediction. That is, for each new prediction, the predicted value of the previous step training LSTM model (that is, the improved LSTM twin parameter generation network), such as the temperature field prediction (t), is used as one of the inputs of the next prediction to retrain the LSTM model, and the prediction is continuously "rolled" to obtain the temperature field prediction (t+1). The prediction results can be updated in real time through the rolling window technology, and the dynamic changes in the data can be adapted. Figure 4 The improved LSTM twin parameter generation network of the present invention includes a schematic diagram of the instant learning adaptive mechanism. Before the LSTM model processes new data, the instant learning adaptive mechanism is first used to pre-process the historical data, extract key features, and construct a similarity metric. When new data (e.g. , ) arrives, the historical data set is measured for similarity with the new data according to the similarity metric to find samples similar to the new data. Based on the similar samples found, the LSTM model can be trained online to specifically process the new data, or the parameters of the global LSTM model can be locally adjusted. The local model or parameter adjustment can retrain the LSTM model based on the prediction results or error feedback of similar samples. The prediction results of the local model or the adjusted global LSTM model are fused to obtain the final prediction result, such as the temperature field prediction (t+1). The prediction results can be updated in real time and further adaptively adjusted based on the feedback of the new data. Figure 3 and Figure 4 The improved LSTM twin parameter generation network is formed by stacking an input layer, an output layer, and multiple hidden layers, and the neurons in the layers are independent of each other, while the neurons between the layers are connected to each other.

[0075] It should be noted that Long Short-Term Memory (LSTM) is a recurrent neural network (RNN) that solves time series problems. It effectively solves the gradient explosion and gradient disappearance problems in RNN. Controlled by two gates. One is the forget gate, which determines the cell state at the last moment. The other is the input gate, which determines how much network input there is at the current moment. is saved to the unit state The current output value of LSTM The output gate and the cell state LSTM is divided into an encoder and a decoder. The encoder implements isometric learning of the input feature data, and the decoder calculates the reconstruction error. Assume that the input sample …, is defined as , and the observation window size is denoted by t. If the type of study .exist Any sequence in of The encoder hidden state vector corresponding to the column is shown in formula (1):

[0076] (1)

[0077] in, is at time n−1 The output state vector of the encoder unit. The input vector , for Step and The coefficient weight matrix of order . Function Usually the activation function is "tanh". Each column vector in is used as the input of the encoder, and the following result is obtained, as shown in formula (2).

[0078] (2)

[0079] in refers to the time at n The output of the encoding unit. Is the parameter set for the encoding part. It is usually set to the tanh activation function. After the entire sequence is input to the encoder, the output sequence set can be obtained Then, the merging step is performed as follows.

[0080] (3)

[0081] (4)

[0082] (5)

[0083] where y is After the pooling step, Entering the decoder. The input can be reconstructed as formulas (6) and (7).

[0084] (6)

[0085] (7)

[0086] in, Reconstructing data. is the decoder implicit state vector. and It is usually set to the activation function "tanh". The LSTM model is ultimately implemented by minimizing the function Obtain and realize dynamic update of digital twin parameters of zinc rotary kiln equipment.

[0087] The present embodiment discloses that a target mechanism model of a zinc rotary kiln is constructed by analyzing the multi-field coupling mechanism of the zinc rotary kiln; observation data of the zinc rotary kiln is obtained in the form of discrete time series data, and time series data analysis and image processing are performed on the observation data to extract the completed data corresponding to the state of the zinc rotary kiln; based on the reaction mechanism analysis of the zinc rotary kiln, an improved LSTM twin parameter generation network is constructed, and the improved LSTM twin parameter generation network includes a rolling window technology or an instant learning adaptive mechanism; the completed data is input into the improved LSTM twin parameter generation network for data training, and the parameters of the target mechanism model are dynamically updated according to the training results. Since this embodiment obtains the observation data of the zinc rotary kiln in the form of discrete time series data, performs time series data analysis and image processing on the observation data, extracts the complementary data corresponding to the state of the zinc rotary kiln, and then inputs the complementary data into an improved LSTM twin parameter generation network including a rolling window technology or an instant learning adaptive mechanism for training, and dynamically updates the parameters of the target mechanism model of the zinc rotary kiln according to the training results, compared with the prior art, this embodiment ensures the virtual-real synchronization of the zinc rotary kiln digital twin, and solves the technical problem that the parameter update of the existing zinc rotary kiln digital twin model cannot respond to environmental changes and system dynamics in real time.

[0088] refer to Figure 5 , Figure 5 It is a schematic flow chart of the second embodiment of the method for updating parameters of the digital twin model of a zinc rotary kiln according to the present invention.

[0089] Based on the first embodiment, in this embodiment, after step S20, the following steps are further included:

[0090] Step S211: perform cross-modal feature fusion on the DCS data features and the image features using cross-modal attentive bilateral residual feature fusion to obtain a first data fusion feature and a second data fusion feature.

[0091] It should be noted that the completion data includes DCS data features and image features. The DCS data features include the time domain features, frequency domain features and time-frequency domain features of the DCS data. The image features include the dynamic features and static features of the flame image. Cross-modal attentive bilateral residual feature fusion (AttentionBRFF) is a cross-modal attentive bilateral residual feature fusion (AttentionBRFF) that uses simplified residual learning in the residual network to introduce residual features as exogenous variables into their respective modal modules. The DCS data features and image features in the completion data are fused using cross-modal attentive bilateral residual feature fusion. Decomposed into two sub-operations:

[0092] ;

[0093] .

[0094] In the formula, They represent fusion features, DCS data features and visual image features respectively. captures the DCS data features as a fusion of invariants, while operates in a similar manner. This bifurcation ensures the preservation of the essential features of each feature modality while simplifying the learning of robust fusion representations. In the AttentionBRFF structure, the extracted residual features are input into the complementary modality network. The fusion learning between the two modalities (i.e., DCS data features and visual image features) is accomplished using the cross-modal attention mechanism in the Transformer architecture. This internal strategy can effectively enhance the model's ability to adaptively focus on salient features, thereby improving the efficiency of the entire fusion process.

[0095] Step S212: Obtain the residual information of the first data fusion feature and the residual information of the second data fusion feature respectively through a frame-level residual learning function.

[0096] It should be noted that the frame-level residual learning function aims to learn auxiliary information that can fine-tune the basic modality features frame by frame. However, due to different sampling rates and potential blanks at the beginning or end of the data, image features and DCS data features often show mismatched time lengths. To solve this problem, an attention-based method can be used. In the specific implementation, for the basic modality Each time step in the high-level features of , using the attention mechanism to calculate the most relevant residual information is:

[0097] = ;

[0098] Among them, the basic mode The high-level feature can be the first data fusion feature or the second data fusion feature, n is the length of the time dimension, for and The softmax normalized similarity between (i.e., attention weight). Attention weight Calculated as:

[0099] ;

[0100] It should be understood that the above formula can be used to subsequently enable the improved LSTM twin parameter generation network to automatically and accurately locate the most relevant features in each frame of the auxiliary modality in the basic modality, thereby effectively solving the time misalignment problem.

[0101] Step S213: performing multimodal fusion on the residual information of the first data fusion feature and the residual information of the second data fusion feature to obtain complementary data after feature fusion.

[0102] It should be noted that in order to improve the prediction accuracy of the dynamic update of the parameters of the target mechanism model, after step S211, it also includes: determining the average value of the first data fusion feature and the average value of the second data fusion feature along the time dimension; normalizing the average value of the first data fusion feature, the average value of the second data fusion feature and the preset bias to obtain the target weight value of multimodal fusion.

[0103] The generation of the target weight value (i.e., adaptive weight parameter) can be expressed as:

[0104] ;

[0105] Where M is a function that computes the average along the time dimension, which allows for dynamic weighting of the fused features, allowing for more adaptive and robust decisions.

[0106] Correspondingly, the step of performing multimodal fusion on the residual information of the first data fusion feature and the residual information of the second data fusion feature to obtain the completed data after feature fusion includes: based on the target weight value, performing multimodal fusion on the residual information of the first data fusion feature and the residual information of the second data fusion feature to obtain the completed data after feature fusion.

[0107] It should be noted that by determining the average value of the first data fusion feature and the average value of the second data fusion feature along the time dimension; normalizing the average value of the first data fusion feature, the average value of the second data fusion feature and the preset bias, the two output features from the AttentionBRFF mechanism are fused to calculate the target weight value of multimodal fusion, thus realizing a dynamic adaptive decision fusion process. In multimodal fusion, adaptive decision fusion means that the fusion representation obtained after AttentionBRFF processing maintains the same shape as the original representation. This compatibility allows the features to be directly input into the original unimodal model for prediction.

[0108] Based on the target weight value, the residual information of the first data fusion feature and the residual information of the second data fusion feature are multimodally fused, that is, the category-level weighting is achieved, which can be expressed as:

[0109] ;

[0110] in , , They are the completed data after feature fusion, the residual information of the first data fusion feature, and the residual information of the second data fusion feature. ⊙ represents the Hadamard product. Based on the existing fusion features, dynamic weights are further calculated at the category level, so that the model can seamlessly switch between various fine-grained tasks with enhanced robustness. In addition, it effectively alleviates the challenges brought by category imbalance.

[0111] For example, refer to Figure 6 , Figure 6 The schematic diagram of the multimodal data fusion framework in the parameter update method of the digital twin model of the zinc rotary kiln of the present invention is shown. The framework contains two feature extraction modules for flame images and DCS data. These networks independently extract high-level features from the original DCS data as well as the flame image of the zinc rotary kiln. The single constraint imposed at this stage is to encapsulate each time step feature as a one-dimensional vector. Its architecture consists of two semi-independent modules, each of which is customized for isolated learning and optimization of different modalities. In contrast to being completely disconnected, these modules are interconnected through a complex three-layer fusion mechanism (data layer fusion, feature layer fusion, and decision layer fusion). The mechanism includes a bilateral residual feature fusion strategy with cross-modal attention, adaptive weight fusion (adaptive decision fusion), and category-level weighting. This multi-faceted fusion scheme promotes effective multimodal interaction, thereby enriching the quality of the overall multimodal representation.

[0112] This embodiment discloses the use of bilateral residual feature fusion with cross-modal attention to perform cross-modal feature fusion on the DCS data features and the image features to obtain a first data fusion feature and a second data fusion feature; the residual information of the first data fusion feature and the residual information of the second data fusion feature are respectively obtained through a frame-level residual learning function; the residual information of the first data fusion feature and the residual information of the second data fusion feature are multimodally fused to obtain the completed data after feature fusion. Since this embodiment uses bilateral residual feature fusion with cross-modal attention to perform cross-modal feature fusion on the DCS data features and the image features, and then obtains the residual information of the first data fusion feature and the residual information of the second data fusion feature through a frame-level residual learning function, and finally performs multimodal fusion on the residual information, compared with the prior art, the present invention not only improves the efficiency of multimodal fusion, but also ensures the accuracy of dynamic update of the target mechanism model parameters.

[0113] refer to Figure 7 , Figure 7 It is a schematic flow chart of the third embodiment of the method for updating parameters of the digital twin model of a zinc rotary kiln according to the present invention.

[0114] Based on the above embodiments, in this embodiment, after step S20, the following steps are further included:

[0115] Step S214: extracting the short-term operation characteristics and the long-term operation characteristics of the zinc rotary kiln in the supplemented data according to a preset time axis.

[0116] Step S215: Determine the fusion weights of the short-term running features and the long-term running features through a sequence weighted attention mechanism.

[0117] Step S216: Based on the fusion weight, the short-term operation feature and the long-term operation feature are fused to obtain the completed data after feature fusion.

[0118] For example, refer to Figure 8 , Figure 8For the state perception algorithm diagram of the zinc rotary kiln in the parameter update method of the digital twin model of the zinc rotary kiln of the present invention, an improved LSTM (long short-term memory) multi-data fusion algorithm is constructed. The algorithm takes the kiln head / kiln tail temperature, inlet and outlet pressure / concentration, kiln head flame state, empirical parameters and other data collected during the production process of the zinc rotary kiln equipment as input, and extracts the short-term and long-term operation characteristics of the zinc rotary kiln according to the preset time axis. The change of the zinc rotary kiln behavior can be from long-term to short-term. The model learns to have better suggestions for kilns whose parameters change over time. The two independent blocks in the proposed architecture, after combining them together, try to extract this variable behavior pattern in the kiln. The data set records are divided into short-term and long-term parts according to a specific time axis. This separates the kiln working states in different time axes and evaluates the model based on the time axis, while improving the accuracy of the system and not encountering too many errors, so as to make appropriate decisions according to the needs of the model. In terms of dynamic model update, it plays a more effective role. Finally, the dynamic tracking of the on-site zinc rotary kiln equipment is realized, the virtual and real synchronization of the digital twin is guaranteed, and the effect of dynamic update is achieved.

[0119] This embodiment discloses extracting the short-term operation characteristics and long-term operation characteristics of the zinc rotary kiln in the supplementary data according to a preset time axis; determining the fusion weights of the short-term operation characteristics and the long-term operation characteristics through a sequence weighted attention mechanism; based on the fusion weights, fusing the short-term operation characteristics and the long-term operation characteristics to obtain the supplementary data after feature fusion. Compared with the prior art, this embodiment extracts the short-term operation characteristics and the long-term operation characteristics of the zinc rotary kiln in the supplementary data according to a preset time axis, and then based on the fusion weights, fuses the short-term operation characteristics and the long-term operation characteristics to obtain the supplementary data after feature fusion, thereby improving the accuracy of dynamic updating of the target mechanism model parameters.

[0120] In addition, an embodiment of the present invention also proposes a medium, on which a zinc rotary kiln digital twin model parameter update program is stored. When the zinc rotary kiln digital twin model parameter update program is executed by a processor, the steps of the zinc rotary kiln digital twin model parameter update method described above are implemented.

[0121] Reference Fig. 9 , Fig. 9 This is a structural block diagram of the first embodiment of the zinc rotary kiln digital twin model parameter updating device of the present invention.

[0122] like Fig. 9 As shown, the zinc rotary kiln digital twin model parameter updating device proposed in the embodiment of the present invention includes: a model construction module 901, a data processing module 902, a network construction module 903 and a parameter updating module 904.

[0123] The model building module 901 is used to build a target mechanism model of the zinc rotary kiln by analyzing the multi-field coupling mechanism of the zinc rotary kiln.

[0124] The data processing module 902 is used to obtain the observation data of the zinc rotary kiln in the form of discrete time series data, and perform time series data analysis and image processing on the observation data to extract the complementary data corresponding to the state of the zinc rotary kiln.

[0125] The network construction module 903 is used to construct an improved LSTM twin parameter generation network according to the reaction mechanism analysis of the zinc rotary kiln, and the improved LSTM twin parameter generation network includes a rolling window technology or an instant learning adaptive mechanism.

[0126] The parameter updating module 904 is used to input the completed data into the improved LSTM twin parameter generation network for data training, and dynamically update the parameters of the target mechanism model according to the training results.

[0127] The parameter updating module 904 is also used to simulate and solve the dynamically updated target mechanism model according to the equipment control parameters and external disturbance data of the zinc rotary kiln to obtain simulation results; iteratively verify the consistency of the simulation results with the actual situation of the zinc rotary kiln to achieve dynamic tracking of the zinc rotary kiln; and monitor and predict the operating status of the zinc rotary kiln based on the verification results.

[0128] The data processing module 902 is also used to obtain the observation data of the zinc rotary kiln in the form of discrete time series data through the DCS system, and the observation data includes DCS data and flame images during the operation of the zinc rotary kiln; based on the experience of on-site fire observation and control of the operation mode of the zinc rotary kiln, the zinc rotary kiln operating condition analysis guidance is carried out, the DCS data and flame images are analyzed and processed, and the processed DCS data and flame images are obtained; through time series data analysis and image processing, the processed DCS data and the processed flame images are feature extracted to obtain the completed data corresponding to the state of the zinc rotary kiln.

[0129] The embodiment of the device discloses that a target mechanism model of a zinc rotary kiln is constructed by analyzing the multi-field coupling mechanism of the zinc rotary kiln; observation data of the zinc rotary kiln is obtained in the form of discrete time series data, and time series data analysis and image processing are performed on the observation data to extract the complementary data corresponding to the state of the zinc rotary kiln; according to the reaction mechanism analysis of the zinc rotary kiln, an improved LSTM twin parameter generation network is constructed, and the improved LSTM twin parameter generation network includes a rolling window technology or an instant learning adaptive mechanism; the complementary data is input into the improved LSTM twin parameter generation network for data training, and the parameters of the target mechanism model are dynamically updated according to the training results. Since the embodiment of the device obtains the observation data of the zinc rotary kiln in the form of discrete time series data, performs time series data analysis and image processing on the observation data, extracts the complementary data corresponding to the state of the zinc rotary kiln, and then inputs the complementary data into an improved LSTM twin parameter generation network including a rolling window technology or an instant learning adaptive mechanism for training, and dynamically updates the parameters of the target mechanism model of the zinc rotary kiln according to the training results, compared with the prior art, the embodiment of the device ensures the virtual-real synchronization of the digital twin of the zinc rotary kiln, and solves the technical problem that the parameter update of the existing zinc rotary kiln digital twin model cannot respond to environmental changes and system dynamics in real time.

[0130] Based on the first embodiment of the device for updating parameters of the digital twin model of a zinc rotary kiln of the present invention, a second embodiment of the device for updating parameters of the digital twin model of a zinc rotary kiln of the present invention is proposed.

[0131] In this embodiment, the data processing module 902 is also used to perform cross-modal feature fusion on the DCS data features and the image features using cross-modal attention bilateral residual feature fusion to obtain a first data fusion feature and a second data fusion feature; obtain the residual information of the first data fusion feature and the residual information of the second data fusion feature respectively through a frame-level residual learning function; perform multi-modal fusion on the residual information of the first data fusion feature and the residual information of the second data fusion feature to obtain the completed data after feature fusion.

[0132] The data processing module 902 is also used to determine the average value of the first data fusion feature and the average value of the second data fusion feature along the time dimension; normalize the average value of the first data fusion feature, the average value of the second data fusion feature and a preset bias to obtain a target weight value of multimodal fusion.

[0133] The data processing module 902 is further used to perform multimodal fusion on the residual information of the first data fusion feature and the residual information of the second data fusion feature based on the target weight value to obtain the completed data after feature fusion.

[0134] Other embodiments or specific implementation methods of the digital twin model parameter updating device of the zinc rotary kiln of the present invention can refer to the above-mentioned method embodiments and will not be described in detail here.

[0135] The present application provides a zinc rotary kiln digital twin model parameter updating device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the zinc rotary kiln digital twin model parameter updating method in the above-mentioned embodiment one.

[0136] Reference below Fig.10 , which shows a schematic diagram of the structure of a digital twin model parameter update device for a zinc rotary kiln suitable for implementing an embodiment of the present application. The digital twin model parameter update device for a zinc rotary kiln in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Fig.10 The zinc rotary kiln digital twin model parameter updating device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0137] like Fig.10As shown, the parameter update device of the digital twin model of the zinc rotary kiln may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM: Read Only Memory) 1002 or the program loaded from the storage device 1003 to the random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the parameter update device of the digital twin model of the zinc rotary kiln are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. The input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the zinc rotary kiln digital twin model parameter update device to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a zinc rotary kiln digital twin model parameter update device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have alternatively.

[0138] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0139] The zinc rotary kiln digital twin model parameter update device provided in the present application adopts the zinc rotary kiln digital twin model parameter update method in the above-mentioned embodiment, which can solve the technical problem that the parameter update of the existing zinc rotary kiln digital twin model cannot respond to environmental changes and system dynamics in real time. Compared with the prior art, the beneficial effects of the zinc rotary kiln digital twin model parameter update device provided in the present application are the same as the beneficial effects of the zinc rotary kiln digital twin model parameter update method provided in the above-mentioned embodiment, and the other technical features in the zinc rotary kiln digital twin model parameter update device are the same as the features disclosed in the previous embodiment method, which will not be repeated here.

[0140] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0141] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0142] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0143] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0144] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a medium (such as a read-only memory / random access memory, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0145] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for updating parameters of digital twin model of zinc rotary kiln. It is characterized in that The method includes: By analyzing the multi-field coupling mechanism of the zinc rotary kiln, a target mechanism model of the zinc rotary kiln is constructed; Acquire observation data of the zinc rotary kiln in the form of discrete time series data, perform time series data analysis and image processing on the observation data, and extract supplementary data corresponding to the state of the zinc rotary kiln, wherein the supplementary data includes DCS data features and image features; According to the reaction mechanism analysis of the zinc rotary kiln, an improved LSTM twin parameter generation network is constructed, wherein the improved LSTM twin parameter generation network includes a rolling window technology or an instant learning adaptive mechanism; Inputting the completed data into the improved LSTM twin parameter generation network for data training, and dynamically updating the parameters of the target mechanism model according to the training results; After the step of acquiring the observation data of the zinc rotary kiln in the form of discrete time series data, performing time series data analysis and image processing on the observation data, and extracting the supplementary data corresponding to the state of the zinc rotary kiln, the method further includes: Performing cross-modal feature fusion on the DCS data features and the image features using cross-modal attention bilateral residual feature fusion to obtain a first data fusion feature and a second data fusion feature; Respectively obtaining residual information of the first data fusion feature and residual information of the second data fusion feature through a frame-level residual learning function; Performing multimodal fusion on the residual information of the first data fusion feature and the residual information of the second data fusion feature to obtain feature fused completed data; The cross-modal feature fusion of the DCS data features and the image features is performed by using the bilateral residual feature fusion of cross-modal attention. It includes two sub-operations: ； ； In the formula, Represent fusion features, DCS data features and image features respectively, represents cross-modal feature fusion, represents the bilateral residual feature fusion of cross-modal attention, Residual information representing image features, Residual information representing the characteristics of DCS data, It represents the fusion that captures the DCS data features as invariant terms. The representation captures the fusion of image features as invariants.

2. The method according to claim 1, It is characterized in that After the step of performing cross-modal feature fusion on the DCS data features and the image features by using cross-modal attention bilateral residual feature fusion to obtain the first data fusion feature and the second data fusion feature, the method further includes: Determining an average value of the first data fusion feature and an average value of the second data fusion feature along a time dimension; Normalizing an average value of the first data fusion feature, an average value of the second data fusion feature, and a preset bias to obtain a target weight value of multimodal fusion; Accordingly, the step of performing multimodal fusion on the residual information of the first data fusion feature and the residual information of the second data fusion feature to obtain the completed data after feature fusion includes: Based on the target weight value, the residual information of the first data fusion feature and the residual information of the second data fusion feature are multimodally fused to obtain the completed data after feature fusion.

3. The method according to claim 1, It is characterized in that After the step of acquiring the observation data of the zinc rotary kiln in the form of discrete time series data, performing time series data analysis and image processing on the observation data, and extracting the supplementary data corresponding to the state of the zinc rotary kiln, the method further includes: Extracting the short-term operation characteristics and the long-term operation characteristics of the zinc rotary kiln in the supplemented data according to a preset time axis; Determining the fusion weights of the short-term running features and the long-term running features through a sequence weighted attention mechanism; Based on the fusion weight, the short-term operation feature and the long-term operation feature are fused to obtain the completed data after feature fusion.

4. The method according to claim 1, It is characterized in that After the step of inputting the completed data into the improved LSTM twin parameter generation network for data training and dynamically updating the parameters of the target mechanism model according to the training results, the method further includes: Simulating and solving the dynamically updated target mechanism model according to the equipment control parameters of the zinc rotary kiln and the external disturbance data to obtain simulation results; Iteratively verifying the consistency between the simulation result and the actual situation of the zinc rotary kiln to achieve dynamic tracking of the zinc rotary kiln; The operating status of the zinc rotary kiln is monitored and predicted based on the verification results.

5. The method according to claim 1, It is characterized in that The step of acquiring the observation data of the zinc rotary kiln in the form of discrete time series data, performing time series data analysis and image processing on the observation data, and extracting the complementary data corresponding to the state of the zinc rotary kiln includes: Acquiring observation data of the zinc rotary kiln in the form of discrete time series data through a DCS system, wherein the observation data includes DCS data and flame images during the operation of the zinc rotary kiln; Based on the experience of on-site fire monitoring and control of the operation mode of the zinc rotary kiln, the working condition analysis and guidance of the zinc rotary kiln are carried out, the DCS data and the flame image are analyzed and processed, and the processed DCS data and the processed flame image are obtained; Through time series data analysis and image processing, feature extraction is performed on the processed DCS data and the processed flame image to obtain complementary data corresponding to the zinc rotary kiln state.

6. The method according to claim 5, It is characterized in that The supplemented data includes the time domain features, frequency domain features and time-frequency domain features of the DCS data and the dynamic features and static features of the flame image.

7. A device for updating parameters of digital twin model of zinc rotary kiln, It is characterized in that The device comprises: A model building module, used for building a target mechanism model of the zinc rotary kiln by analyzing the multi-field coupling mechanism of the zinc rotary kiln; A data processing module, used for acquiring observation data of the zinc rotary kiln in the form of discrete time series data, performing time series data analysis and image processing on the observation data, and extracting supplementary data corresponding to the state of the zinc rotary kiln, wherein the supplementary data includes DCS data features and image features; A network construction module, used for constructing an improved LSTM twin parameter generation network according to the reaction mechanism analysis of the zinc rotary kiln, wherein the improved LSTM twin parameter generation network includes a rolling window technology or an instant learning adaptive mechanism; A parameter updating module, used for inputting the completed data into the improved LSTM twin parameter generation network for data training, and dynamically updating the parameters of the target mechanism model according to the training results; The data processing module is further used to perform cross-modal feature fusion on the DCS data features and the image features using cross-modal attention bilateral residual feature fusion to obtain a first data fusion feature and a second data fusion feature; obtain residual information of the first data fusion feature and residual information of the second data fusion feature respectively through a frame-level residual learning function; perform multi-modal fusion on the residual information of the first data fusion feature and the residual information of the second data fusion feature to obtain complementary data after feature fusion; The cross-modal feature fusion of the DCS data features and the image features is performed by using the bilateral residual feature fusion of cross-modal attention. It includes two sub-operations: ； ； In the formula, Represent fusion features, DCS data features and image features respectively, represents cross-modal feature fusion, represents the bilateral residual feature fusion of cross-modal attention, Residual information representing image features, Residual information representing the characteristics of DCS data, It represents the fusion that captures the DCS data features as invariant terms. The representation captures the fusion of image features as invariants.

8. A zinc rotary kiln digital twin model parameter update device, It is characterized in that The device includes: a memory, a processor, and a zinc rotary kiln digital twin model parameter update program stored in the memory and executable on the processor, wherein the zinc rotary kiln digital twin model parameter update program is configured to implement the steps of the zinc rotary kiln digital twin model parameter update method as described in any one of claims 1 to 6.

9. A medium, It is characterized in that The medium stores a zinc rotary kiln digital twin model parameter update program, and when the zinc rotary kiln digital twin model parameter update program is executed by the processor, the steps of the zinc rotary kiln digital twin model parameter update method as described in any one of claims 1 to 6 are implemented.

Images