Zinc hydrometallurgy digital twinning construction method based on time sequence model

Abstract

The invention discloses a zinc hydrometallurgy digital twinning construction method based on a time sequence model, which comprises the following steps: step 1, acquiring historical ion concentration data of each process of zinc hydrometallurgy as a data sample, and obtaining data sample characteristics; 2, learning the VAE variational auto-encoder model based on the data sample features, and enhancing the data sample through the learned VAE variational auto-encoder model; 3, training the data sample enhanced in each process through an LSTM (Long Short-Term Memory) network time sequence prediction model; 4, configuring the trained LSTM long short-term memory network time sequence prediction model corresponding to each process in the virtual process scene; and 5, performing prediction by using the corresponding trained LSTM long short-term memory network time sequence prediction model, and presenting a prediction result in a virtual scene. The method can improve and optimize the production strategy of the zinc hydrometallurgy process.

Description

Technical Field

[0001] The present invention relates to the technical field of digital twins, and in particular to a method for constructing a digital twin of zinc hydrometallurgy based on a timing model. Background Art

[0002] Hydrometallurgy is the main method of modern zinc smelting, but in the zinc smelting process, the output of hazardous waste anode mud is large, the treatment cost is high, resulting in resource waste and environmental problems. In the impurity removal process, there are many types of impurity ions, and the concentration difference with the main metal zinc ions is large, making it difficult to distinguish the ion concentration of the reaction solution in the system and the corresponding reaction.

[0003] Industrial data has the characteristics of large data volume and high noise. The generative model VAE (Variational Autoencoder) can learn the distribution pattern of data and the intrinsic structure and characteristics of data, and can generate new data samples that conform to the pattern based on the learned content. Therefore, it has excellent data enhancement capabilities and can fill in the missing values ​​of industrial data samples, correct outliers, and enhance data while conforming to the original data distribution structure, so that the data meets the requirements of time series model training.

[0004] The LSTM time series prediction model has a strong feature extraction capability. It can learn the data distribution and data regularity of the ion data in the hydrometallurgical zinc smelting process enhanced by VAE through a neural network, and realize the prediction of future ion concentrations.

[0005] Digital twin is a technology that simulates and analyzes physical entities or systems in the real world in real time through digital models. It connects the real world with the digital world by collecting real-time data of physical entities, thereby simulating real scenes in virtual space. Through digital twin technology, a three-dimensional visualization digital twin scene of all elements of hydrometallurgy can be built, and combined with the LSTM time series prediction model, the data of hydrometallurgy leaching, purification and electrolysis can be visualized and predicted to optimize production strategies. Summary of the invention

[0006] The present invention provides a method for constructing a digital twin of hydrozinc smelting based on a time series model, builds a digital twin scene of hydrozinc smelting, realizes the simulation and data visualization of each process flow of hydrozinc smelting, and predicts the future ion concentration, so as to help improve and optimize the production strategy of hydrozinc smelting process.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is: A method for constructing a digital twin of hydrometallurgy of zinc based on a time series model comprises the following steps: Step 1, obtaining the historical ion concentration data of each process of the factory's hydrometallurgical zinc smelting as data samples, analyzing the data samples of each process respectively, and obtaining the data sample characteristics corresponding to each process, wherein the data sample characteristics include ion concentration distribution, ion concentration variation law, the connection between sample characteristics of different processes and abnormal samples, and then proceeding to step 2; Step 2: According to the data sample characteristics of each process obtained in step 1, the VAE variational autoencoder model is used to learn the data sample distribution law of each process for the historical ion concentration data of each process, and then the data samples corresponding to each process are filled with missing values ​​and corrected for outliers through the learned VAE variational autoencoder model, thereby obtaining enhanced data samples, and then proceeding to step 3; Step 3: For the enhanced data samples of each process, different LSTM long short-term memory network time series models are used for training and parameter tuning. The LSTM long short-term memory network time series prediction model after parameter tuning has the ability to predict the future ion concentration for the historical ion concentration data of the corresponding process, that is, the ion concentration at a certain time or time period in the future is predicted based on the ion concentration data of the corresponding process at a certain time in the operation of the factory, and then the process goes to step 4; Step 4: Build virtual scenes of each process in the process flow of the hydrometallurgical zinc smelting plant, and configure the trained and parameter-optimized LSTM long short-term memory network time series prediction model corresponding to each process in the corresponding virtual process scene to predict the future ion concentration of each process when the process flow in the virtual digital twin scene is running, so as to realize the construction of the hydrometallurgical zinc smelting digital twin scene framework, and then proceed to step 5; Step 5. Given the ion concentration data of each process, for different processes, use the corresponding trained LSTM long short-term memory network time series prediction model to predict the ion concentration data of the corresponding process within a certain time in the future based on the ion concentration data of the corresponding process, and present the predicted results in the virtual scene of step 4. In the virtual scene, process flow simulation and process operation prediction are performed based on the predicted ion concentration data.

[0008] In the further step 3, each process is configured with a different LSTM long short-term memory network timing model for training.

[0009] Furthermore, it also includes step 6: simulating the process simulation content through the constructed digital twin model to optimize the process strategy, and using the trained LSTM long short-term memory network time series prediction model to predict the ion data of the corresponding process within a certain period of time according to the historical data of each process provided by the hydrometallurgical zinc smelting plant in the virtual scene to perform process simulation. When the actual process flow runs to the moment of the ion concentration predicted in step 5, the ion concentration data of the actual process at that moment is transmitted to the virtual scene, the predicted data at that moment and the model's future predicted data are corrected, and presented in the virtual scene.

[0010] Compared with the prior art, the present invention has the following advantages: The digital twin construction method of the present invention predicts the ion concentration data of different processes of hydrometallurgical zinc smelting, and presents the prediction results in a virtual scene for visualization and simulation, which can provide assistance and reference for the person in charge of the process to improve and optimize the production strategy of the hydrometallurgical zinc smelting process. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a flow chart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0012] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0013] like Figure 1 As shown, this embodiment discloses a method for constructing a digital twin of hydrometallurgy based on a time series model, comprising the following steps: Step 1: Obtain historical ion concentration data of each process of hydrometallurgical zinc smelting from a metallurgical plant as data samples, wherein the processes include leaching, purification, and electrolysis.

[0014] The data samples of each process are analyzed by statistical techniques such as descriptive statistics, time series analysis, and regression analysis to obtain the data sample characteristics of each process. The data sample characteristics include ion concentration distribution, ion concentration change rules, the relationship between sample characteristics, and abnormal samples. The abnormal samples include missing characteristic values, abnormal characteristic values, or abnormal fluctuations in the characteristic values ​​of the previous and next samples in the same time period. Then proceed to step 2.

[0015] Step 2: According to the data sample characteristics of each process obtained in step 1, the VAE variational autoencoder model is used to learn the data sample distribution law of each process for the historical ion concentration data of each process. That is, the VAE variational autoencoder model is trained with the data samples of each process so that the VAE variational autoencoder model learns the potential distribution and characteristics of the data in the data samples of each process.

[0016] Next, the data samples of each process are filled with missing values ​​and corrected for outliers in accordance with the distribution law of the initial data through the learned VAE variational autoencoder model, thereby obtaining enhanced data samples. Specifically, when filling missing values, the samples containing missing values ​​are input into the trained VAE variational autoencoder model. The VAE variational autoencoder model generates data through the decoder to supplement the corresponding missing values, identifies outliers, and replaces them with new values ​​that are more in line with the potential distribution to ensure the rationality of the data. After the missing value filling and outlier correction are completed, proceed to step 3.

[0017] Step 3: For the enhanced data samples of each process, different LSTM long short-term memory network time series prediction models are trained respectively, and the parameters of the corresponding LSTM long short-term memory network time series prediction models are tuned.

[0018] In this embodiment, taking into account the large differences in the specific operations of different process flows, resulting in large differences in the structure and size of the ion data of different processes, the LSTM long short-term memory network time series prediction model trained under a certain process data is difficult to use to predict other process data. Therefore, at least one LSTM long short-term memory network time series prediction model is set for each process data for training and parameter adjustment.

[0019] The LSTM long short-term memory network time series prediction model can capture the long-term dependencies and periodic changes in time series data. By optimizing the LSTM long short-term memory network time series prediction model structure and hyperparameters, the performance of the LSTM long short-term memory network time series model is gradually improved. After training and parameter tuning, the obtained LSTM long short-term memory network time series prediction model has the ability to predict the future ion concentration of the corresponding process ion data, that is, it has the ability to predict the ion concentration at a certain time or time period in the future based on the ion concentration data of the corresponding process at a certain time. After the model training and evaluation are completed, the model is exported as a file for deployment in actual applications, and then proceed to step 4; Step 4: Build a virtual scene of the leaching, purification, and electrolysis processes in the process flow of the hydrometallurgical zinc smelting plant, and configure the trained and parameter-optimized LSTM long short-term memory network timing model corresponding to each process in the corresponding virtual process scene. This is used to predict the future ion concentration of each process when the process flow in the virtual digital twin scene is running, realize the construction of the hydrometallurgical zinc smelting digital twin scene framework, and then proceed to step 5.

[0020] Step 5: The hydrometallurgical zinc smelting plant provides the ion concentration data of each process of leaching, purification, and electrolysis. For different processes, the corresponding trained LSTM long short-term memory network time series prediction model is used to predict the ion concentration data of the corresponding process within a certain time in the future based on the ion concentration data of the corresponding process, and the predicted results are presented in the virtual scene of step 4 in the form of visual charts. In the virtual scene, the process flow simulation and process operation prediction are carried out based on the predicted ion concentration data. Then go to step 6.

[0021] Step 6: The factory manager uses the constructed digital twin model to simulate the process simulation content to optimize the production strategy and assist the manager in production planning. Specifically, in the virtual scene, according to the real historical data of each process provided by the hydrometallurgical plant, the trained LSTM long short-term memory network time series prediction model is used to predict the ion data of the corresponding process within a certain period of time, and the process simulation is performed to realize the operation prediction of the factory process scene in the virtual scene. The factory manager optimizes the current production strategy through the operation prediction presented by the virtual scene prediction. When the actual process flow runs to the moment of the predicted ion concentration in step 5, the real data of the actual process at that moment is transmitted to the virtual scene, the predicted data at that moment is corrected, and presented in the virtual scene for subsequent operation prediction.

[0022] The preferred embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. The embodiments described in the present invention are merely descriptions of the preferred embodiments of the present invention, and do not limit the concept and scope of the present invention. The various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction, and such combinations should also be regarded as the contents disclosed in the present disclosure as long as they do not violate the concept of the present invention. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

[0023] The present invention is not limited to the specific details of the above-mentioned embodiments. Within the technical concept of the present invention and without departing from the design concept of the present invention, various modifications and improvements made to the technical solution of the present invention by technical personnel in this field should fall within the protection scope of the present invention. The technical contents for which protection is sought in the present invention have been fully recorded in the claims.

Claims

1. A method for constructing a digital twin of hydrometallurgical zinc smelting based on a time series model, characterized in that: The following steps are involved: Step 1, obtaining historical ion concentration data of different processes of hydrometallurgy of zinc in the factory as data samples, analyzing the data samples of each process respectively, obtaining data sample features corresponding to each process data, wherein the data sample features include ion concentration distribution, ion concentration variation law, connection between sample features and abnormal samples, and then proceeding to step 2; Step 2: According to the data sample characteristics of each process obtained in step 1, different VAE variational autoencoder models are used to learn the data sample distribution law of each process for the historical ion concentration data of each process, and then the data samples corresponding to each process are filled with missing values ​​and corrected for outliers through the learned VAE variational autoencoder model, thereby obtaining enhanced data samples, and then proceeding to step 3; Step 3: For each process enhanced data sample, different LSTM long short-term memory network time series prediction models are used for training and parameter tuning. The LSTM long short-term memory network time series prediction model after training and parameter tuning has the ability to predict the future time ion concentration of the corresponding process historical ion concentration data, that is, based on the given corresponding process ion concentration data within a certain time in the factory operation, the ion concentration within a certain time or time period in the future is predicted, and then the process goes to step 4; Step 4: Build virtual scenes of each process in the process flow of the hydrometallurgical zinc smelting plant, and configure the trained and parameter-optimized LSTM long short-term memory network time series prediction model corresponding to each process in the corresponding virtual process scene to predict the future ion concentration of each process when the process flow in the virtual digital twin scene is running, so as to realize the construction of the hydrometallurgical zinc smelting digital twin scene framework, and then proceed to step 5; Step 5. Transmit the ion concentration data of each process to the framework. For different processes, use the corresponding trained LSTM long short-term memory network time series prediction model to predict the ion concentration data of the corresponding process within a certain time in the future based on the ion concentration data of the corresponding process, and present the predicted results in the virtual scene of step 4. In the virtual scene, process flow simulation and process operation prediction are performed based on the predicted ion concentration data.

2. According to a method for constructing a digital twin of hydrometallurgy based on a time series model according to claim 1, it is characterized in that: In step 3, each process is configured with a different LSTM long short-term memory network time series prediction model for training.

3. The method for constructing a digital twin of zinc hydrometallurgy based on a time series model according to claim 1, characterized in that: It also includes step 6: simulating the process simulation content through the constructed digital twin model to optimize the process strategy. In the virtual scene, according to the historical data of each process provided by the hydrometallurgical zinc smelting plant, the trained LSTM long short-term memory network time series prediction model is used to predict the ion data of the corresponding process within a certain period of time to perform process simulation. When the actual process flow runs to the moment of the ion concentration predicted in step 5, the ion concentration data of the actual process at that moment is transmitted to the virtual scene, the predicted data at that moment and the model's future predicted data are corrected, and presented in the virtual scene.

Images