Evaporation process technological index estimation method and device and storage medium

Abstract

The invention provides an evaporation process technological index estimation method and device, a medium and equipment, and relates to the technical field of industrial processes, and the method comprises the steps: collecting historical dynamic information data in an evaporation process, and carrying out the preprocessing and dynamic feature extraction to obtain dynamic features; embedding the dynamic features into different categories of the LSTM network to obtain an improved LSTM network; constructing a multiple dynamic information integration model based on the improved LSTM network; taking historical dynamic information data as input to train the multiple dynamic information integration model to obtain a trained multiple dynamic information integration model; and collecting real-time dynamic information data in the evaporation process, inputting the real-time dynamic information data into the trained multiple dynamic information integration model, obtaining an evaporation process technological index estimation result, confirming the evaporation process technological index, and realizing accurate estimation in a complex and changeable industrial environment by a traditional technological index estimation method.

Description

Technical Field

[0001] The present invention relates to the field of industrial process technology, and in particular to a method, device and storage medium for estimating process indicators of an evaporation process. Background Art

[0002] In the alumina production process, the evaporation process is a key link in purifying and concentrating bauxite slurry, which directly affects the final product quality and production efficiency. The concentration of the outlet sodium aluminate solution is one of the most important process indicators of the evaporation process. In order to ensure the stability and qualification of the evaporated sodium aluminate solution concentration, it is usually necessary for on-site operators to adjust industrial parameters according to the actual solution concentration test results. However, since the evaporation process involves a cascade of multiple production equipment, it takes a certain amount of time for the material to go from the inlet to the outlet. In addition, there is a large lag in sampling and testing analysis, which leads to delays in on-site operations, affects stable production, and may cause waste of resources and energy.

[0003] In fact, the variability, complexity and nonlinear characteristics of the evaporation process make it difficult for traditional detection methods to meet the actual production needs. The various factors involved, such as changes in raw materials, fluctuations in operating conditions and aging of equipment, may affect the concentration of the outlet liquid. The interaction of these factors makes it difficult for a simple single adjustment method to achieve accurate control and estimation of the concentration. Summary of the invention

[0004] In order to overcome the disadvantage that the conventional process index estimation method in the prior art is difficult to achieve accurate estimation in the face of a complex and changeable industrial environment, the main purpose of the present invention is to provide a method and device for estimating process indexes of an evaporation process.

[0005] To achieve the above object, the present invention adopts the following technical scheme, a method for estimating process indicators of an evaporation process, comprising the following steps:

[0006] Collect historical dynamic data during the evaporation process, perform preprocessing and dynamic feature extraction, obtain growth rate, exponential difference, logarithmic difference and ratio, and construct multi-dynamic information;

[0007] Embed multiple dynamic information into different categories of LSTM network to obtain an improved LSTM network;

[0008] Building a multiple dynamic information integration model based on the improved LSTM network; and training the multiple dynamic information integration model using historical dynamic information data as input to obtain a trained multiple dynamic information integration model;

[0009] The real-time dynamic information data collected during the evaporation process is input into the trained multiple dynamic information integration model to obtain the estimated results of the process indicators of the evaporation process and confirm the process indicators of the evaporation process.

[0010] The dynamic data include raw material flow rate, evaporator temperature, evaporator liquid level, evaporator pressure and outlet solution concentration;

[0011] The multi-dynamic information includes: growth rate, exponential difference, logarithmic difference and ratio, and the formula is expressed as:

[0012]

[0013] x t ′=x t -αx t-1

[0014]

[0015] Among them, g t Expressed as growth rate, Δlog(x t ) is expressed as logarithmic difference, x t Represents the data value at the current moment, x t-1 represents the data value of the previous moment, α represents the smoothing coefficient, which usually ranges from 0 to 1 and is set to 0.5 in the experiment; t ′ represents the exponential difference, r t Indicates a ratio.

[0016] Based on the structure of the LSTM network unit, the growth rate is embedded into the input gate, and the formula is expressed as:

[0017] i t =σ(W i ·[x(t),h(t-1)]+x gro (t)·W gro +b i )

[0018] f t =σ(W f ·[x(t),h(t-1)]+b f )

[0019] o t =σ(W o ·[x(t),h(t-1)]+b o )

[0020]

[0021] Among them, x(t) and h(t-1) are the input data at the current time t and the hidden state at the previous time; the function σ is the Sigmoid activation function, which limits the value between 0 and 1, and the function σ c is the hyperbolic tangent activation function, which limits the value to between -1 and 1; it 、f t , o t , Respectively represent the input gate, forget gate, output gate and internal state of the LSTM network unit; W i ,W f ,W o ,W c are the weight matrices of the output gate, forget gate, output gate, and internal state respectively; b i ,b f ,b o ,b c Respectively represent the bias items of the LSTM network unit input gate, forget gate, output gate and internal state; x gro (t) is the growth rate information input, W gro is its weight coefficient.

[0022] Based on the structure of the LSTM network unit, the growth rate is embedded into the input gate and the logarithmic difference is embedded into the output gate. The formula is expressed as:

[0023] i t =σ(W i ·[x(t),h(t-1)]+x g (t)·W g +b i )

[0024] f t =σ(W f ·[x(t),h(t-1)]+b f )

[0025] o t =σ(W o ·[x(t),h(t-1)]+x log_diff (t)·W log_diff +b o )

[0026]

[0027] Among them, x(t) and h(t-1) are the input data at the current time t and the hidden state at the previous time; the function σ is the Sigmoid activation function, which limits the value between 0 and 1, and the function σ c is the hyperbolic tangent activation function, which limits the value to between -1 and 1; i t 、f t , o t , Respectively represent the input gate, forget gate, output gate and internal state of the LSTM network unit; W i ,W f ,Wo ,W c are the weight matrices of the output gate, forget gate, output gate, and internal state respectively; b i ,b f ,b o ,b c Respectively represent the bias items of the LSTM network unit input gate, forget gate, output gate and internal state; x gro (t) is the growth rate information input, W gro is its weight coefficient; x log_diff (t) is the exponential difference information input, W log_diff The weight coefficient for exponential difference information input; x erp_diff (t) is the logarithmic difference information input, W erp_diff The weight factor for the log-difference information input.

[0028] Based on the structure of the LSTM network unit, the growth rate, logarithmic difference and exponential difference are integrated into the input gate, output gate and internal state of the LSTM unit respectively. The formula is expressed as follows:

[0029] i t =σ(W i ·[x(t),h(t-1)]+x gro (t)·W gro +b i )

[0030] f t =σ(W f ·[x(t),h(t-1)]+b f )

[0031] o t =σ(W o ·[x(t),h(t-1)]+x log_diff (t)·W log_diff +b o )

[0032]

[0033] Among them, x(t) and h(t-1) are the input data at the current time t and the hidden state at the previous time; the function σ is the Sigmoid activation function, which limits the value between 0 and 1, and the function σ c is the hyperbolic tangent activation function, which limits the value to between -1 and 1; i t 、f t , o t , Respectively represent the input gate, forget gate, output gate and internal state of the LSTM network unit; W i ,W f,W o ,W c are the weight matrices of the output gate, forget gate, output gate, and internal state respectively; b i ,b f ,b o ,b c Respectively represent the bias items of the LSTM network unit input gate, forget gate, output gate and internal state; x gro (t) is the growth rate information input, W gro is its weight coefficient; x log_diff (t) is the exponential difference information input, W log_diff The weight coefficient for exponential difference information input; x erp_diff (t) is the logarithmic difference information input, W erp_diff The weight factor for the log-difference information input.

[0034] The method of embedding dynamic information into different categories of LSTM network includes the following steps:

[0035] Based on the structure of LSTM network unit, the growth rate is embedded into the input gate, the exponential difference is embedded into the unit state, the logarithmic difference is embedded into the output gate, and then embedded into each category according to the set rule ratio to obtain the improved LSTM network.

[0036] The method of training the multiple dynamic information integration model by taking the historical dynamic data as input comprises the following steps:

[0037] Divide historical dynamic data into training set and validation set;

[0038] The training set is used as the input of the multiple dynamic information integration model, and the LSTM network is trained using a back-propagation algorithm. The network weights and biases are continuously adjusted until the loss function is minimized, thereby obtaining the trained multiple dynamic information integration model.

[0039] The method of using historical dynamic information data as input to train the multiple dynamic information integration model also includes: using the verification set as the trained multiple dynamic information integration model to verify the evaporation process process index estimation result, comparing the verification evaporation process process index estimation result with the true value, and confirming the evaporation process process index.

[0040] The pre-processing comprises the following steps:

[0041] Standardize the historical dynamic information data collected during the evaporation process to obtain standardized historical dynamic information data;

[0042] The standardized historical dynamic information data is differentially processed to obtain processed historical dynamic information data.

[0043] A device for estimating process indicators of an evaporation process, characterized by comprising:

[0044] The acquisition module is used to collect historical dynamic information data during the evaporation process, and perform preprocessing and dynamic feature extraction to obtain dynamic features;

[0045] A model building module is used to embed dynamic features into different categories of LSTM networks to obtain an improved LSTM network; construct a multiple dynamic information integration model based on the improved LSTM network; and train the multiple dynamic information integration model using historical dynamic information data as input to obtain a trained multiple dynamic information integration model;

[0046] The process processing module is used to collect real-time dynamic information data during the evaporation process and input it into the trained multiple dynamic information integration model to obtain the estimation results of the process indicators of the evaporation process and confirm the process indicators of the evaporation process.

[0047] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, a method for estimating a process index of an evaporation process is implemented.

[0048] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, a method for estimating process indicators of an evaporation process is implemented.

[0049] Compared with the prior art, the beneficial effects of the present invention are as follows: the multiple dynamic information integration of the present invention utilizes the unique characteristics of each dynamic indicator, making the LSTM network more robust when processing time series data, reducing the prediction error caused by data fluctuations and noise, and ensuring the stability and reliability of the prediction results. By embedding a variety of dynamic information such as growth rate, exponential difference, logarithmic difference and ratio into different categories of the LSTM network, the model's ability to capture complex dynamic changes is improved, and a high-precision estimation of the concentration of the sodium aluminate solution at the outlet of the evaporation process is achieved. In view of the various complex factors in the evaporation process, an LSTM model with multiple dynamic information integration is constructed, which can flexibly adapt to various dynamic changes in the production process, provide reliable estimation results, and meet the needs of actual production. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The drawings described herein are used to provide further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.

[0051] Figure 1 It is a flowchart of the LSTM model estimation task process of multiple dynamic information integration of the present invention;

[0052] Figure 2 It is a structural diagram of a variable dynamic information acquisition unit of the present invention;

[0053] Figure 3 It is a schematic diagram of the variable dynamic information structure of the present invention;

[0054] Figure 4 It is a construction diagram of the multi-information fusion model of the present invention.

[0055] Figure 5 It is a schematic diagram of model verification of the present invention;

[0056] Figure 6 It is a structural diagram of the LSTM unit with multiple dynamic information inputs of the present invention;

[0057] Figure 7 It is a line graph of the estimation results of the comparison model of the present invention and a bar graph of the evaluation index;

[0058] Figure 8 It is a performance evaluation diagram of the multi-dynamic information integration LSTM model of the present invention. DETAILED DESCRIPTION

[0059] The present invention will be further described below in conjunction with the accompanying drawings and implementation modes.

[0060] The present invention provides a method for estimating process indicators of an evaporation process based on multiple dynamic information integration LSTM, comprising the following steps:

[0061] Step 1: Collect and preprocess the data of the relevant variables involved in the evaporation production process before the experiment, and further extract the dynamic features of the target variables. The specific steps are as follows:

[0062] S1. Collect relevant historical data from sensors and control systems of alumina production evaporation equipment, including raw material flow, evaporator temperature, evaporator liquid level, evaporator pressure and outlet feed liquid concentration. These relevant variables will constitute the feature set of the model.

[0063] S2. Standardize the collected data to eliminate the differences between different dimensions and make the data suitable for the input of the LSTM network.

[0064] S3. Perform differential processing on the standardized data to extract various dynamic information, including growth rate, exponential difference, logarithmic difference and ratio, and the calculation is as follows. The dynamic information can better reflect the changing characteristics of the data and serve as input information for the calculation of network unit categories in the LSTM model.

[0065]

[0066] x t ′=xt -αx t-1

[0067]

[0068] Among them, x t Represents the data value at the current moment, x t-1 represents the data value of the previous moment, α represents the smoothing coefficient, which usually ranges from 0 to 1 and is set to 0.5 in the experiment; t ′ represents the exponential difference, r t Indicates a ratio.

[0069] Step 2: Build a multiple dynamic information integration model based on the LSTM network and embed different dynamic information into different categories of LSTM. The specific steps are as follows:

[0070] S1. Based on the structure of LSTM network unit, a single dynamic information fusion model is designed. Taking the growth rate information being incorporated into the input gate of LSTM unit as an example, its corresponding weight is calculated by the following formula.

[0071] i t =σ(W i ·[x(t),h(t-1)]+x gro (t)·W gro +b i )

[0072] f t =σ(W f ·[x(t),h(t-1)]+b f )

[0073] o t =σ(W o ·[x(t),h(t-1)]+b o )

[0074]

[0075] Among them, x(t) and h(t-1) are the input data at the current time t and the hidden state at the previous time; the function σ is the Sigmoid activation function, which limits the value between 0 and 1, and the function σ c is the hyperbolic tangent activation function, which limits the value to between -1 and 1; i t 、f t , o t , Respectively represent the input gate, forget gate, output gate and internal state of the LSTM network unit, W i ,W f ,W o ,W care the weight matrices of the output gate, forget gate, output gate, and internal state respectively; b i ,b f ,b o ,b c Respectively represent the bias items of the LSTM network unit input gate, forget gate, output gate and internal state; x gro (t) is the growth rate information input, W gro is its weight coefficient.

[0076] S2. Taking the addition of growth rate information to the input gate of the LSTM unit and logarithmic difference information to the output gate as an example, a dual dynamic information fusion model is designed, and its corresponding weights are calculated using the following formula. log_diff (t) is the exponential difference information input, W log_diff The weight factor for the exponential difference information input. erp_diff (t) is the logarithmic difference information input, W erp_diff The weight factor for the log-difference information input.

[0077] i t =σ(W i ·[x(t),h(t-1)]+x g (t)·W g +b i )

[0078] f t =σ(W f ·[x(t),h(t-1)]+b f )

[0079] o t =σ(W o ·[x(t),h(t-1)]+x log_diff (t)·W log_diff +b o )

[0080]

[0081] S3. Integrate the growth rate information, logarithmic difference information and exponential difference information into the input gate, output gate and internal state of the LSTM unit respectively to construct an LSTM unit with multiple dynamic information inputs. The gate calculation formula is as follows. The newly constructed LSTM unit structure is as follows: Figure 2 shown.

[0082] i t =σ(W i ·[x(t),h(t-1)]+x gro (t)·W gro +b i )

[0083] f t =σ(W f ·[x(t),h(t-1)]+b f )

[0084] o t =σ(W o ·[x(t),h(t-1)]+x log_diff (t)·W log_diff +b o )

[0085]

[0086] The extracted dynamic information is embedded into different categories of the LSTM network, and the corresponding weight coefficients are trained. The growth rate information is embedded into the input gate, the exponential difference information is embedded into the unit state, the logarithmic difference information is embedded into the output gate, and the ratio information is comprehensively considered in each category to make full use of the characteristics of each dynamic information and enhance the estimation ability of the model, making the LSTM network more flexible and accurate in processing complex dynamic data.

[0087] Step 3: After the LSTM unit structure of multiple dynamic information integration is constructed, the export sodium aluminate solution estimation model is further established. The overall flow chart of the estimation task is as follows: Figure 1 As shown, the specific steps are as follows:

[0088] S1. Design a process index estimation model structure, including an input layer, two layers of LSTM unit structure layers for multiple dynamic information integration, and an output layer.

[0089] S2. Based on the production site detection data, after the data preprocessing in step 1, determine the feature data set as the model input, the target data set as the model output and the corresponding dynamic information data set, and divide the training set and verification set suitable for the model input.

[0090] S3. Use the training set data to train the LSTM model, adjust the model parameters through cross-validation, and use the early stopping strategy to prevent overfitting.

[0091] Step 4: After the model is trained, determine the optimal parameters, input the verification data set into the improved model, compare the estimated value and the actual value, calculate the relevant evaluation indicators of the model, and further verify the feasibility of the multiple dynamic information integration LSTM model. The specific steps are as follows:

[0092] S1. Integrate multiple dynamic information in each gate and internal state of the LSTM unit, and verify the effect of each combination through experiments, so as to enhance the model's estimation ability for time series data.

[0093] S2. Calculate the evaluation index of the estimation model for each multi-information combination to evaluate the model performance. By comparing the estimated value and the true value of the estimation model output of each multi-information combination, judge the trend learning effect of the model on the export sodium aluminate solution concentration data, and thus derive the estimation model of the optimal combination.

[0094] S3. In order to verify the effectiveness of the multiple dynamic information integration LSTM, this case was compared with five other models, including deep learning models and machine learning models. The comparison models specifically include: industrial variable estimation model based on classic LSTM, industrial variable estimation model based on RNN, industrial variable estimation model based on GRU, industrial variable estimation model based on kernel method, and industrial variable estimation model based on decision tree.

[0095] Example

[0096] According to the production site detection data, 800 sets of experimental data were selected, including raw material flow, evaporator temperature, evaporator liquid level, evaporator pressure, etc. as feature data sets, and the outlet sodium aluminate solution concentration as the target data set. Four corresponding sets of dynamic information data sets were generated according to the target data sets.

[0097] The preprocessed data is divided into a training set (80%) and a validation set (20%), and the dynamic information data set is also divided accordingly. After the model training is completed, 160 sets of data with time series are selected for verification. Table 1 lists the performance of different evaluation indicators under each combination of LSTM unit categories and multiple information variables. Figure 4 The results of model performance evaluation when growth rate, log difference, and ratio information are used in the input gate, output gate, and internal state of the LSTM unit are illustrated.

[0098] Table 1 Performance of different evaluation indicators for each combination of LSTM unit categories and multiple information variables

[0099]

[0100] Experimental results show that adding growth rate (GR), logarithmic difference (LD) and ratio (R) information to the input gate (IG), output gate (OG) and internal state (CS) of the LSTM unit can achieve the best model performance. The multiple dynamic information input model outperforms the single information input and dual information input models, and has a significant improvement in prediction performance. This is especially reflected in the reduction of errors and the improvement of the coefficient of determination, thus confirming the effectiveness of the multi-information fusion strategy in time series prediction.

[0101] The study shows that when predicting complex time series, the LSTM model with multiple dynamic information inputs outperforms the models with single and dual inputs, highlighting its superior performance in such tasks. In order to verify the effectiveness of the multiple dynamic information integrated LSTM, the experimental results of other comparative methods using the same dataset are as follows: Figure 3 The evaluation indicators are shown in Table 2.

[0102] Table 2 Evaluation indicators

[0103] Model MSE RMSE MAE R-Squared LSTM 1.057 1.028 0.809 0.114 RNN 1.004 1.002 0.8021 0.148 GRU 1.078 1.038 0.82 0.089 KM 1.021 1.016 0.807 0.131 DT 1.154 1.074 0.856 0.017

[0104] In summary, it is concluded that although the estimation methods involving deep learning and machine learning can capture some temporal trends in the data, their estimation performance is lacking, and the accuracy and stability of the estimation do not meet the expected standards. The results highlight the challenges of estimating complex time series data using only standard LSTM, RNN, GRU, kernel methods, and decision tree models, and emphasize the need for further improvement and optimization of model development.

[0105] The present invention also proposes a device for estimating process indicators of an evaporation process based on multiple dynamic information integration LSTM, comprising a display, a processor, and a computer program stored in a memory and executable on the processor, wherein the processor implements the steps of any of the above methods when executing the computer program.

[0106] In addition, the present invention also proposes a storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of any of the above methods are implemented.

[0107] A person of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk, etc., various media that can store program codes.

[0108] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device, which can be a personal computer, a server, or a network device, etc., to execute the methods described in each embodiment or some parts of the embodiments.

[0109] It should be noted that, in the present invention, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device 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 device.

[0110] The above embodiments are merely examples of the present invention and do not limit the protection scope of the present invention. All designs that are the same or similar to the present invention fall within the protection scope of the present invention.

Claims

1. A method for estimating process indicators of an evaporation process, characterized in that: The following steps are involved: Collect historical dynamic data during the evaporation process, perform preprocessing and dynamic feature extraction, obtain growth rate, exponential difference, logarithmic difference and ratio, and construct multi-dynamic information; Embed multiple dynamic information into different categories of LSTM network to obtain an improved LSTM network; Building a multiple dynamic information integration model based on the improved LSTM network; and training the multiple dynamic information integration model using historical dynamic data as input to obtain a trained multiple dynamic information integration model; The real-time dynamic information data collected during the evaporation process is input into the trained multiple dynamic information integration model to obtain the estimated results of the process indicators of the evaporation process and confirm the process indicators of the evaporation process.

2. The method for estimating process indicators of an evaporation process according to claim 1, characterized in that: The dynamic data include raw material flow rate, evaporator temperature, evaporator liquid level, evaporator pressure and outlet solution concentration; The multi-dynamic information includes: growth rate, exponential difference, logarithmic difference and ratio, and the formula is expressed as: x t ′=x t -αx t-1 Among them, g t Expressed as growth rate, Δlog(x t ) is expressed as logarithmic difference, x t Represents the data value at the current moment, x t-1 represents the data value of the previous moment, α represents the smoothing coefficient, usually ranging from 0 to 1; x t ′ represents the exponential difference, r t Indicates a ratio.

3. The method for estimating process indicators of an evaporation process according to claim 2, characterized in that: The structure based on the LSTM network unit embeds the growth rate into the input gate, and the formula is expressed as: i t =σ(W i ·[x(t),h(t-1)]+x gro (t)·W gro +b i ) f t =σ(W f ·[x(t),h(t-1)]+b f ) o t =σ(W o ·[x(t),h(t-1)]+b o ) Among them, x(t) and h(t-1) are the input data at the current time t and the hidden state at the previous time; the function σ is the Sigmoid activation function, which limits the value between 0 and 1, and the function σ c is the hyperbolic tangent activation function, which limits the value to between -1 and 1; i t 、f t , o t , Respectively represent the input gate, forget gate, output gate and internal state of the LSTM network unit; W i ,W f ,W o ,W c are the weight matrices of the output gate, forget gate, output gate, and internal state respectively; b i ,b f ,b o ,b c Respectively represent the bias items of the LSTM network unit input gate, forget gate, output gate and internal state; x gro (t) is the growth rate information input, W gro is its weight coefficient.

4. The method for estimating process indicators of an evaporation process according to claim 2, characterized in that: The structure based on the LSTM network unit embeds the growth rate into the input gate and the logarithmic difference into the output gate, and the formula is expressed as: i t =σ(W i ·[x(t),h(t-1)]+x g (t)·W g +b i ) f t =σ(W f ·[x(t),h(t-1)]+b f ) o t =σ(W o ·[x(t),h(t-1)]+x log_diff (t)·W log_diff +b o ) Among them, x(t) and h(t-1) are the input data at the current time t and the hidden state at the previous time; the function σ is the Sigmoid activation function, which limits the value between 0 and 1, and the function σ c is the hyperbolic tangent activation function, which limits the value to between -1 and 1; i t 、f t , o t , Respectively represent the input gate, forget gate, output gate and internal state of the LSTM network unit; W i ,W f ,W o ,W c are the weight matrices of the output gate, forget gate, output gate, and internal state respectively; b i ,b f ,b o ,b c Respectively represent the bias items of the LSTM network unit input gate, forget gate, output gate and internal state; x gro (t) is the growth rate information input, W gro is its weight coefficient; x log_diff (t) is the exponential difference information input, W log_diff The weight coefficient for exponential difference information input; x erp_diff (t) is the logarithmic difference information input, W erp_diff The weight factor for the log-difference information input.

5. The method for estimating process indicators of an evaporation process according to claim 2, characterized in that: The structure based on the LSTM network unit integrates the growth rate, logarithmic difference and exponential difference into the input gate, output gate and internal state of the LSTM unit respectively, and the formula is expressed as follows: i t =σ(W i ·[x(t),h(t-1)]+x gro (t)·W gro +b i ) f t =σ(W f ·[x(t),h(t-1)]+b f ) o t =σ(W o ·[x(t),h(t-1)]+x log_diff (t)·W log_diff +b o ) Among them, x(t) and h(t-1) are the input data at the current time t and the hidden state at the previous time; the function σ is the Sigmoid activation function, which limits the value between 0 and 1, and the function σ c is the hyperbolic tangent activation function, which limits the value to between -1 and 1; i t 、f t , o t , Respectively represent the input gate, forget gate, output gate and internal state of the LSTM network unit; W i ,W f ,W o ,W c are the weight matrices of the output gate, forget gate, output gate, and internal state respectively; b i ,b f ,b o ,b c Respectively represent the bias items of the LSTM network unit input gate, forget gate, output gate and internal state; x gro (t) is the growth rate information input, W gro is its weight coefficient; x log_diff (t) is the exponential difference information input, W log_diff The weight coefficient for exponential difference information input; x erp_diff (t) is the logarithmic difference information input, W erp_diff The weight factor for the log-difference information input.

6. The method for estimating process indicators of an evaporation process according to claim 2, characterized in that: The different categories of embedding dynamic information into LSTM networks include: Based on the structure of LSTM network unit, the growth rate is embedded into the input gate, the exponential difference is embedded into the unit state, the logarithmic difference is embedded into the output gate, and then embedded into each category according to the set rule ratio to obtain the improved LSTM network.

7. The method for estimating process indicators of an evaporation process according to claim 1, characterized in that: The method of training the multiple dynamic information integration model by taking the historical dynamic data as input comprises the following steps: Divide historical dynamic data into training set and validation set; The training set is used as the input of the multiple dynamic information integration model, and the LSTM network is trained using a back-propagation algorithm. The network weights and biases are continuously adjusted until the loss function is minimized, thereby obtaining the trained multiple dynamic information integration model.

8. The method for estimating process indicators of an evaporation process according to claim 7, characterized in that: The method of using historical dynamic information data as input to train the multiple dynamic information integration model also includes: using the verification set as the trained multiple dynamic information integration model to verify the evaporation process process index estimation result, comparing the verification evaporation process process index estimation result with the true value, and confirming the evaporation process process index.

9. The method for estimating process indicators of an evaporation process according to claim 1, characterized in that: The pre-processing comprises the following steps: Standardize the historical information data collected during the evaporation process to obtain standardized historical dynamic information data; The standardized historical dynamic information data is differentially processed to obtain processed historical dynamic information data.

10. An evaporation process process index estimation device, characterized in that: include: The acquisition module is used to collect historical dynamic information data during the evaporation process, and perform preprocessing and dynamic feature extraction to obtain dynamic features; A model building module is used to embed dynamic features into different categories of LSTM networks to obtain an improved LSTM network; construct a multiple dynamic information integration model based on the improved LSTM network; and train the multiple dynamic information integration model using historical dynamic information data as input to obtain a trained multiple dynamic information integration model; The process processing module is used to collect real-time dynamic information data during the evaporation process and input it into the trained multiple dynamic information integration model to obtain the estimation results of the process indicators of the evaporation process and confirm the process indicators of the evaporation process.

Images