Glass kiln temperature intelligent prediction control method based on hybrid neural network

Abstract

The invention relates to the technical field of glass kiln temperature prediction control, in particular to an intelligent glass kiln temperature prediction control method based on a hybrid neural network, which solves the problem of single model in the prior art and improves the accuracy of glass kiln temperature prediction. Data acquisition and preprocessing: acquiring and preprocessing production historical data related to glass kiln temperature prediction control; s2, CNN feature extraction: extracting space and time features of the preprocessed data by using a CNN; s3, data serialization: converting the data into a serialized form for input of the LSTM long short-term memory network; s4, LSTM time sequence prediction: inputting the serialized data into an LSTM long short-term memory network model for prediction; and S5, controlling the glass kiln: calculating and controlling data such as natural gas, oxygen and oil quantity of the glass kiln in real time on the basis of the predicted temperature in combination with the current state of the smelting furnace.

Description

Technical Field

[0001] The present invention relates to the technical field of glass furnace temperature prediction and control, and more specifically, to an intelligent prediction and control method for glass furnace temperature based on a hybrid neural network. Background Art

[0002] With the continuous progress of industrial technology, the glass manufacturing industry has increasingly higher requirements for the accuracy of furnace temperature control. As a key device in the glass production process, the temperature stability and uniformity of the glass furnace directly affect the quality, production efficiency, and production cost of glass products. Traditional glass furnace temperature control mainly relies on manual experience to adjust fuel or power output or uses a computer for PID control. However, these methods have significant instability and limitations, making it difficult to ensure the precise control of the temperature in each temperature zone of the furnace, resulting in unstable glass product performance and increased production costs. To improve this problem, researchers have begun to explore and develop artificial intelligence-based furnace temperature prediction and control methods for automatically predicting and controlling the furnace temperature. Utilizing advanced artificial intelligence technology to process and analyze furnace temperature data is crucial for improving prediction efficiency, prediction accuracy, and furnace stability. With the rapid development of deep learning and artificial intelligence technologies, automatic, efficient, and accurate furnace temperature prediction has become possible.

[0003] Currently, there are already some artificial intelligence technologies that can be used for the prediction and control of furnace temperature. For example, some methods use deep learning technologies to analyze and predict furnace temperature data, such as using convolutional neural networks (CNNs) or long short-term memory networks (LSTMs) to analyze and predict furnace temperature data. However, these methods all use a single model and generally have certain limitations in terms of applicability and accuracy.

[0004] In view of this, we propose an intelligent prediction and control method for glass furnace temperature based on a hybrid neural network. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent prediction and control method for glass furnace temperature based on a hybrid neural network, aiming to solve the problem that current furnace temperature prediction technologies, although they can use artificial intelligence technology to automatically analyze and predict furnace temperature to achieve the control of furnace temperature, still have a significant problem, that is, they only use a single model, which is prone to insufficient feature extraction and poor model generalization ability, thus affecting the accuracy of prediction.

[0006] To solve the above technical problems, the present invention provides the following technical solution: an intelligent prediction and control method for glass furnace temperature based on a hybrid neural network, the method comprising the following steps:

[0007] S1, Data Acquisition and Preprocessing: Collect and preprocess the production historical data related to the temperature prediction and control of the glass furnace;

[0008] S2, CNN Feature Extraction: Use the CNN convolutional neural network to extract the spatial and temporal features of the preprocessed data;

[0009] S3, Data Serialization: Convert the data into a serialized form for input to the LSTM long short-term memory network;

[0010] S4, LSTM Time Series Prediction: Input the serialized data into the LSTM long short-term memory network model for prediction;

[0011] S5, Glass Furnace Control: Based on the predicted temperature, combined with the current furnace state, calculate in real time the data such as natural gas, oxygen, and oil volume for controlling the glass furnace.

[0012] Preferably, the specific steps in the above step S1 include the following steps:

[0013] S1.1: Use the sensor network to collect in real time the historical data of the crown temperature of the glass furnace, the historical data of the bottom temperature of the glass melting furnace, the natural gas flow rate, the oxygen flow rate, and the oil output volume historical data;

[0014] S1.2: Remove the outliers and fill in the missing data in the above historical data;

[0015] S1.3: Convert all the above data to the same scale to improve the efficiency and accuracy of model training.

[0016] Preferably, the specific steps in the above step S2 include the following steps:

[0017] S2.1, Construct a CNN convolutional neural network model;

[0018] S2.2, The CNN layer performs feature learning in the time and space dimensions through convolutional kernels to capture the internal temperature distribution and change trend of the furnace;

[0019] S2.3, Integrate the spatio-temporal features extracted by the CNN with the historical data to form a more comprehensive input feature set.

[0020] Preferably, the specific steps in the above step S3 include the following steps:

[0021] S3.1, Sort the fused feature data according to the time series to form serialized data suitable for LSTM processing;

[0022] S3.2, Increase the amount of training data through one of generating simulated data and sliding window sampling to improve the generalization ability of the model.

[0023] Preferably, the specific steps in step S4 include the following steps:

[0024] S4.1, design the LSTM network structure;

[0025] S4.2, use the cross-validation method to evaluate the performance of the model on the training set and the validation set, and adjust the model parameters to achieve the best prediction effect;

[0026] S4.3, after preprocessing and feature extraction of the real-time collected data, input it into the trained LSTM model for temperature prediction.

[0027] Preferably, the specific steps in step S5 include the following steps:

[0028] S5.1, dynamically adjust the supply amounts of natural gas, oxygen, and oil according to the deviation between the predicted temperature and the set temperature to achieve closed-loop control;

[0029] S5.2, combine historical data and the current furnace state to predict the possible adjustment amounts in the future for a period of time and intervene in advance;

[0030] S5.3, set temperature thresholds and resource usage limits to ensure rapid response in case of anomalies and protect the safety of the kiln and equipment.

[0031] Preferably, the above step S2.2 also includes setting a pooling layer, and the specific steps are as follows:

[0032] S2.2.1, expand or modify the traditional two-dimensional convolution kernel into one-dimensional to adapt to the one-dimensional structure of temperature data;

[0033] S2.2.2, set the input layer to receive data;

[0034] S2.2.3, set the convolutional layer, and use the one-dimensional convolutional kernel to perform convolution operations on the input data to extract local features;

[0035] S2.2.4, set the activation layer, and use the ReLU activation function to increase non-linearity and improve the expression ability of the model;

[0036] S2.2.5, set the pooling layer to perform pooling operations to reduce the feature dimension, reduce the amount of calculation, and retain important information at the same time;

[0037] S2.2.6, stack multiple convolutional layers and pooling layers to extract higher-level features layer by layer;

[0038] S2.2.7, set the output layer to output the data structure for sequence prediction.

[0039] Preferably, the LSTM network structure in the above step S4.1 further includes:

[0040] Input layer: responsible for receiving time series data;

[0041] LSTM layer: The LSTM model can contain one or more LSTM layers. Multiple LSTM layers can help the model learn more complex temporal dependencies;

[0042] Dropout layer: used to prevent overfitting;

[0043] Fully connected layer: used to convert the output of the LSTM layer into the final prediction result;

[0044] Output layer: outputs the final prediction result.

[0045] Preferably, it also includes regularly collecting new production data, retraining and updating the model to adapt to changes in production conditions.

[0046] Preferably, it also includes real-time monitoring of the accuracy of model prediction, control effect, and resource consumption.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] 1. By introducing the CNN convolutional neural network, the present invention can effectively extract the spatial features in the kiln temperature data, such as the temperature distribution law in different regions, thereby improving the prediction accuracy. Traditional temperature prediction models often show poor generalization ability when facing different working conditions and process parameter changes. By constructing a hybrid neural network model and combining the advantages of CNN and LSTM, the present invention improves the adaptability and generalization ability of the model, can achieve stable and efficient temperature prediction in different production environments, solves the problem of single model in the prior art, improves the accuracy of glass kiln temperature prediction, and further realizes more precise control of other data of the glass kiln.

[0049] 2. Although LSTM in the present invention is good at dealing with long-term dependencies in time series data, it is relatively weak in capturing local features of data. Through its convolutional layer and pooling layer structure, CNN can effectively extract local features in the input data. In the hybrid model, CNN first processes the time series data, extracts important local features, and then passes these features to LSTM for further processing and modeling. In this way, the hybrid model can capture both local details in the data and model long-term dependencies, thereby improving the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is the method flow chart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0051] Such as Figure 1As shown in the figure, the present invention relates to an intelligent prediction control method for the temperature of a glass furnace based on a hybrid neural network, and the method includes the following steps:

[0052] S1, Data acquisition and preprocessing: Collect and preprocess the production historical data related to the temperature prediction control of the glass furnace;

[0053] S2, CNN feature extraction: Use a CNN convolutional neural network to extract the spatial and temporal features of the preprocessed data;

[0054] S3, Data serialization: Convert the data into a serialized form for input to the LSTM long short-term memory network;

[0055] S4, LSTM time series prediction: Input the serialized data into the LSTM long short-term memory network model for prediction;

[0056] S5, Glass furnace control: Based on the predicted temperature, combined with the current furnace state, calculate in real time the data such as natural gas, oxygen, and oil volume for controlling the glass furnace.

[0057] Further, the specific steps in the above step S1 include the following steps:

[0058] S1.1: Use a sensor network to collect the historical data of the crown temperature of the glass furnace, the historical data of the bottom temperature of the glass melting furnace, the natural gas flow rate, the oxygen flow rate, and the oil output historical data in real time;

[0059] S1.2: Remove the outliers and fill in the missing data from the above historical data;

[0060] S1.3: Convert all the above data to the same scale to improve the efficiency and accuracy of model training;

[0061] It is necessary to collect the historical data of the crown temperature of the glass furnace, the historical data of the bottom temperature of the glass melting furnace, the natural gas flow rate, the oxygen flow rate, and the oil output historical data, and clean and preprocess the collected data, including removing noise, outlier processing, and data normalization, to ensure the data quality.

[0062] Further, the specific steps in the above step S2 include the following steps:

[0063] S2.1, Construct a CNN convolutional neural network model;

[0064] S2.2, The CNN layer performs feature learning in the time and space dimensions through convolutional kernels to capture the internal temperature distribution and change trend of the furnace;

[0065] S2.3, Integrate the spatial and temporal features extracted by the CNN with the historical data to form a more comprehensive input feature set;

[0066] Step S2.2 above further includes setting a pooling layer, and the specific steps are as follows:

[0067] S2.2.1, Expand or modify the traditional two-dimensional convolution kernel into one-dimensional to adapt to the one-dimensional structure of temperature data;

[0068] S2.2.2, Set the input layer to receive data;

[0069] S2.2.3, Set the convolutional layer, and use the one-dimensional convolution kernel to perform convolution operations on the input data to extract local features;

[0070] S2.2.4, Set the activation layer, and use the ReLU activation function to increase non-linearity and improve the expression ability of the model;

[0071] S2.2.5, Set the pooling layer to perform pooling operations to reduce the feature dimension, reduce the amount of calculation, and retain important information at the same time;

[0072] S2.2.6, Stack multiple convolutional layers and pooling layers to extract higher-level features layer by layer;

[0073] S2.2.7, Set the output layer to output the data structure for sequence prediction;

[0074] Construct a Convolutional Neural Network (CNN) model to extract spatial and temporal features from the preprocessed data. The CNN layer performs feature learning in the temporal and spatial dimensions through the convolution kernel, captures the temperature distribution and change trend inside the kiln, selects an appropriate convolution kernel size and stride, and sets a reasonable pooling layer to reduce the amount of calculation and extract key features. The basic steps are as follows: Expand or modify the traditional two-dimensional convolution kernel into one-dimensional to adapt to the one-dimensional structure of temperature data. Input layer: Receive temperature data. Convolutional layer: Use the one-dimensional convolution kernel to perform convolution operations on the input data to extract local features. The size and number of convolution kernels determine the granularity and complexity of feature extraction. Activation layer: Usually use activation functions such as ReLU to increase non-linearity and improve the expression ability of the model. Pooling layer: Reduce the feature dimension through pooling operations, reduce the amount of calculation, and retain important information at the same time. Repeated stacking: Multiple convolutional layers and pooling layers can be stacked to extract higher-level features layer by layer. Output layer: The output layer is a fully connected layer that outputs the data structure for sequence prediction.

[0075] Furthermore, step S3 above specifically includes the following steps:

[0076] S3.1, Sort the fused feature data according to the time series to form serialized data suitable for LSTM processing;

[0077] S3.2, increase the amount of training data by generating simulated data or using sliding window sampling to improve the generalization ability of the model;

[0078] Before inputting the data into the LSTM, we need to convert it into a serialized form, which can be achieved by splitting the data into sequences of fixed length or using a sliding window. The serialized data can better utilize the memory ability of the LSTM.

[0079] Furthermore, the specific steps in the above step S4 include the following steps:

[0080] S4.1, design the LSTM network structure;

[0081] S4.2, use the cross-validation method to evaluate the performance of the model on the training set and the validation set, and adjust the model parameters to achieve the best prediction effect;

[0082] S4.3, after preprocessing and feature extraction of the real-time collected data, input it into the trained LSTM model for temperature prediction;

[0083] The LSTM network structure in the above step S4.1 also includes:

[0084] Input layer: responsible for receiving time series data;

[0085] LSTM layer: The LSTM model can contain one or more LSTM layers. Multiple LSTM layers can help the model learn more complex time dependencies;

[0086] Dropout layer: used to prevent overfitting;

[0087] Fully connected layer: used to convert the output of the LSTM layer into the final prediction result;

[0088] Output layer: output the final prediction result;

[0089] The data obtained by extracting and serializing CNN features is input into a Long Short-Term Memory (LSTM) network model. Leveraging the LSTM's powerful processing ability for time series data, the changing trend of the furnace temperature over a period of time is predicted. The LSTM model architecture includes: Input layer: Responsible for receiving time series data. For LSTM, the dimensions of the input data usually include the number of samples, time steps, and the number of features. LSTM layer: The LSTM model can contain one or more LSTM layers. Multiple LSTM layers can help the model learn more complex temporal dependencies. The number of units (also known as the number of hidden units or neurons) in each LSTM layer is an important hyperparameter, which determines the amount of information and complexity that the LSTM layer can capture. The selection of the number of units is usually based on experiments and experience. Inside the LSTM layer, specific activation functions such as Sigmoid and Tanh are used to control the flow and update of information. These activation functions are part of the LSTM architecture. The key of LSTM lies in its three internal gate structures - forget gate, input gate, and output gate. These gates control the flow of information and allow LSTM to maintain and update memories over a longer time range. Dropout layer: A regularization technique used to prevent overfitting. Adding a Dropout layer in the LSTM model can reduce the overfitting phenomenon on the training set and improve the generalization ability of the model. Fully connected layer: Used to convert the output of the LSTM layer into the final prediction result. In time series prediction tasks, the fully connected layer usually takes the output of the last time step (or a certain aggregation of the outputs of all time steps) of the LSTM layer as input and outputs the predicted value. Output layer: Outputs the final prediction result.

[0090] Furthermore, the specific steps in step S5 are as follows:

[0091] S5.1, Dynamically adjust the supply of natural gas, oxygen, and oil volume according to the deviation between the predicted temperature and the set temperature to achieve closed-loop control;

[0092] S5.2, Combine historical data and the current furnace state to predict the adjustment amount that may be required in the future and intervene in advance;

[0093] S5.3, Set temperature thresholds and resource usage limits to ensure a rapid response in case of anomalies and protect the safety of the furnace and equipment.

[0094] Furthermore, it also includes regularly collecting new production data to retrain and update the model to adapt to changes in production conditions.

[0095] Furthermore, it also includes real-time monitoring of the accuracy of model predictions, control effects, and resource consumption.

[0096] The embodiments disclosed in the present invention are preferred embodiments, but not limited thereto. Those of ordinary skill in the art can easily understand the spirit of the present invention based on the above embodiments and make different extensions and changes. However, as long as they do not depart from the spirit of the present invention, they are within the protection scope of the present invention.

Claims

1. An intelligent prediction control method for the temperature of a glass furnace based on a hybrid neural network, characterized in that The method includes the following steps: S1, Data acquisition and preprocessing: Collect and preprocess the production historical data related to the temperature prediction and control of the glass furnace; S2, CNN feature extraction: Use the CNN convolutional neural network to extract the spatial and temporal features of the preprocessed data; S3, Data serialization: Convert the data into a serialized form for input to the LSTM long short-term memory network; S4, LSTM time series prediction: Input the serialized data into the LSTM long short-term memory network model for prediction; S5, Glass furnace control: Based on the predicted temperature, combined with the current furnace state, calculate in real time the data such as natural gas, oxygen, and oil volume for controlling the glass furnace.

2. The intelligent prediction control method for the temperature of a glass furnace based on a hybrid neural network according to claim 1, wherein, The specific steps in the above step S1 include the following steps: S1.1: Use the sensor network to collect in real time the historical data of the crown temperature of the glass furnace, the historical data of the bottom temperature of the glass melting furnace, the natural gas flow rate, the oxygen flow rate, and the oil output volume historical data; S1.2: Remove the outliers from the above historical data and fill in the missing data; S1.3: Convert all the above data to the same scale to improve the efficiency and accuracy of model training.

3. The intelligent prediction control method for the temperature of a glass furnace based on a hybrid neural network according to claim 1, wherein The specific steps in the above step S2 include the following steps: S2.1, Build a CNN convolutional neural network model; S2.2, The CNN layer performs feature learning in the time and space dimensions through convolutional kernels to capture the internal temperature distribution and change trend of the furnace; S2.3, Integrate the spatial and temporal features extracted by the CNN with the historical data to form a more comprehensive input feature set.

4. A method for intelligent prediction and control of the temperature of a glass furnace based on a hybrid neural network according to claim 1, characterized in that The specific steps in the above step S3 include the following steps: S3.1, Sort the fused feature data according to the time series to form serialized data suitable for LSTM processing; S3.2, Increase the amount of training data through one of generating simulated data and sliding window sampling to improve the generalization ability of the model.

5. The intelligent prediction control method for the temperature of a glass furnace based on a hybrid neural network according to claim 1, characterized in that, The specific steps in the above step S4 include the following steps: S4.1, Design the LSTM network structure; S4.2, Use the cross-validation method to evaluate the performance of the model on the training set and the validation set, and adjust the model parameters to achieve the best prediction effect; S4.3, Input the real-time collected data after preprocessing and feature extraction into the trained LSTM model for temperature prediction.

6. The intelligent prediction control method for the temperature of a glass furnace based on a hybrid neural network according to claim 1, characterized in that The specific steps in the above step S5 include the following steps: S5.1, Dynamically adjust the supply amounts of natural gas, oxygen, and oil according to the deviation between the predicted temperature and the set temperature to achieve closed-loop control; S5.2, Combine the historical data and the current furnace state to predict the possible adjustment amounts in the future period of time and intervene in advance; S5.3, Set temperature thresholds and resource usage limits to ensure rapid response in case of anomalies and protect the safety of the furnace and equipment.

7. The intelligent prediction control method for the temperature of a glass furnace based on a hybrid neural network according to claim 3, characterized in that, The above step S2.2 also includes setting a pooling layer, and the specific steps are: S2.2.1, Expand or modify the traditional two-dimensional convolutional kernel into one-dimensional to adapt to the one-dimensional structure of the temperature data; S2.2.2, Set the input layer to receive data; S2.2.3, Set the convolutional layer, and use the one-dimensional convolutional kernel to perform convolutional operations on the input data to extract local features; S2.2.4, Set the activation layer, use the ReLU activation function to increase non-linearity and improve the model's expressive power; S2.2.5, Set the pooling layer to perform pooling operations to reduce the feature dimension, reduce the computational load, and retain important information at the same time; S2.2.6, Stack multiple convolutional layers and pooling layers to extract higher-level features layer by layer; S2.2.7, Set the output layer for output, which is a data structure for sequence prediction.

8. The intelligent prediction control method for the temperature of a glass furnace based on a hybrid neural network according to claim 5, wherein, In the above step S4.1, the LSTM network structure also includes: Input layer: Responsible for receiving time series data; LSTM layer: The LSTM model can contain one or more LSTM layers. Multiple LSTM layers can help the model learn more complex time-dependent relationships; Dropout layer: Used to prevent overfitting; Fully connected layer: Used to convert the output of the LSTM layer into the final prediction result; Output layer: Output the final prediction result.

9. The intelligent prediction control method for the temperature of a glass furnace based on a hybrid neural network according to claim 1, wherein It also includes regularly collecting new production data to retrain and update the model to adapt to changes in production conditions.

10. The intelligent prediction control method for the temperature of a glass furnace based on a hybrid neural network according to claim 1, characterized in that It also includes real-time monitoring of the accuracy of model prediction, control effect, and resource consumption.

Images