Aluminum slag solid waste rotary kiln fuel addition control method and system

Abstract

The application provides an aluminum slag solid waste rotary kiln fuel adding control method and system, relates to the rotary kiln control technical field, and the method comprises the following steps: obtaining feed flow data, fuel flow data, tail gas temperature data and preheating temperature data; through a temperature prediction model, obtaining predicted tail gas temperature data and predicted preheating temperature data at the current moment, and judging whether the temperature prediction model needs to be trained; if so, training according to the feed flow data, the fuel flow data, the preheating temperature data and the tail gas temperature data, predicting the predicted tail gas temperature data and the predicted preheating temperature data at the next moment after training, and determining the predicted fuel flow data at the next moment. According to the application, whether the fuel flow is appropriate can be automatically determined, and the fuel flow is automatically adjusted, the dependence on manual adjustment is reduced, the timeliness and accuracy of temperature control are improved, and the processing quality of aluminum slag solid waste and the quality of output are improved.

Description

Technical Field

[0001] This invention relates to the field of rotary kiln control technology, and in particular to a method and system for controlling fuel addition in a rotary kiln for aluminum slag solid waste. Background Technology

[0002] In related technologies, the temperature control of rotary kilns is one of the key factors in whether a rotary kiln can obtain high-quality products. The temperature of a rotary kiln is related to many factors, such as the fuel flow rate and the feed flow rate at the feed inlet. However, in related technologies, the fuel flow rate is usually set to a default value or manually controlled, which results in insufficient and untimely temperature control of the rotary kiln when the feed flow rate changes, which may lead to a decline in the quality of the products.

[0003] The information disclosed in the background section of this application is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0004] This invention provides a method and system for controlling fuel addition in a rotary kiln for aluminum slag solid waste, which can solve the technical problem that the temperature control of the rotary kiln is not precise and timely when the feed flow rate changes.

[0005] According to a first aspect of the present invention, a method for controlling fuel addition in a rotary kiln for aluminum slag solid waste is provided, comprising: acquiring, at each moment after the rotary kiln starts operating, feed flow rate data, fuel flow rate data, exhaust gas temperature data at the rotary kiln outlet, and preheating temperature data of preheating air before fuel combustion heating; processing the feed flow rate data and fuel flow rate data from multiple moments prior to the current moment using a temperature prediction model to obtain predicted exhaust gas temperature data and predicted preheating temperature data for the current moment; determining whether the temperature prediction model needs training based on the predicted exhaust gas temperature data, predicted preheating temperature data, exhaust gas temperature data, and preheating temperature data for the current moment; if the temperature prediction model needs training, then based on the feed flow rate data and fuel flow rate data from multiple moments prior to the current moment... The temperature prediction model is trained using data, preheating temperature data, and exhaust gas temperature data to obtain a trained temperature prediction model. If the temperature prediction model does not require training, it is used as the trained temperature prediction model. The trained temperature prediction model is then used to process the feed flow rate data and fuel flow rate data from the current moment and several previous moments to obtain the predicted exhaust gas temperature data and predicted preheating temperature data for the next moment. Based on the predicted feed flow rate data, predicted exhaust gas temperature data, and predicted preheating temperature data for the next moment, as well as the feed flow rate data, exhaust gas temperature data, and preheating temperature data from the current moment and several previous moments, the predicted fuel flow rate data for the next moment is determined. At the next moment, the fuel flow rate of the rotary kiln is set to the predicted fuel flow rate data.

[0006] According to the present invention, determining whether a temperature prediction model needs training based on the predicted exhaust gas temperature data, predicted preheating temperature data, exhaust gas temperature data, and preheating temperature data at the current moment includes: determining a temperature state vector at the current moment based on the exhaust gas temperature data and preheating temperature data at the current moment; determining a predicted temperature state vector at the current moment based on the predicted exhaust gas temperature data and predicted preheating temperature data at the current moment; and determining that the temperature prediction model needs training if the cosine similarity between the temperature state vector at the current moment and the predicted temperature state vector is less than or equal to a first similarity threshold.

[0007] According to the present invention, if the temperature prediction model needs to be trained, the temperature prediction model is trained based on feed flow rate data, fuel flow rate data, preheating temperature data, and exhaust gas temperature data from the current time and multiple previous times to obtain a trained temperature prediction model. This includes: constructing a flow state vector at time ik from the feed flow rate data and fuel flow rate data at time ik, where time ik is a time before the current time, 1 ≤ k ≤ n, n is the number of input ports of the temperature prediction model, and i, k, and n are all positive integers; and inputting the flow state vectors from time in (i) to time in (i-1) into a first fully connected layer for processing to obtain the flow state vectors from time in (i) to time in (i-1). The flow rate input vector at time i-1 is obtained; the flow rate input vector from time in to time i-1 is processed by the temperature prediction model to obtain the output feature vector at time i-1; the output feature vector at time i-1 is processed by the second fully connected layer and the first activation layer to obtain the predicted exhaust gas temperature data and predicted preheating temperature data at time i; based on the predicted exhaust gas temperature data and predicted preheating temperature data at time i, the first loss function of the temperature prediction model is determined; based on the first loss function, the temperature prediction model is trained to obtain the trained temperature prediction model.

[0008] According to the present invention, based on the predicted exhaust gas temperature data and predicted preheating temperature data at the i-th time point, and the exhaust gas temperature data and preheating temperature data at the i-th time point, the first loss function of the temperature prediction model is determined, including: according to the formula Determine the first loss function of the temperature prediction model. ,in, Let i be the predicted exhaust gas temperature data at time i. For the predicted preheating temperature data at time i, The exhaust gas temperature data is at time i. The preheating temperature data is for the i-th time step. Let N be the number of training data points at the current time. , i, N and All are positive integers.

[0009] According to the present invention, determining the predicted fuel flow rate data for the next time moment based on the predicted feed flow rate data, predicted exhaust gas temperature data, and predicted preheating temperature data for the next time moment, as well as the feed flow rate data, exhaust gas temperature data, and preheating temperature data for the current time moment and multiple time moments prior to it, includes: obtaining the fuel prediction input vector for the current time moment and multiple time moments prior to it based on the feed flow rate data, predicted exhaust gas temperature data, and predicted preheating temperature data for the next time moment, and determining the fuel prediction input vector for the next time moment based on the predicted feed flow rate data, predicted exhaust gas temperature data, and predicted preheating temperature data for the next time moment; and then... The fuel prediction input vectors at the current time and multiple time steps prior to the current time step, as well as the fuel prediction input vector for the next time step, are input into the third fully connected layer for processing to obtain the fuel prediction state vectors at the current time step and multiple time steps prior to the current time step, and the fuel prediction state vector for the next time step. These fuel prediction state vectors are then input into the trained fuel flow prediction model to obtain the fuel flow prediction feature vector for the next time step. Finally, the fuel flow prediction feature vector for the next time step is input into the fourth fully connected layer and the second activation layer for processing to obtain the predicted fuel flow data for the next time step.

[0010] According to the present invention, the training steps of the fuel flow prediction model include: obtaining the fuel prediction input vector at the j-th time step based on the feed flow rate data, exhaust gas temperature data, and preheating temperature data at the j-th time step, where the j-th time step is the time step before the current time step, m is the number of input ports of the fuel flow prediction model, 0≤s≤m-1, and j, s, and m are all integers; inputting the fuel prediction input vectors from the (j-m+1)-th time step to the j-th time step into the third fully connected layer for processing to obtain the fuel prediction state vectors from the (j-m+1)-th time step to the j-th time step; inputting the fuel prediction state vectors from the (j-m+1)-th time step to the j-th time step into the fuel flow prediction model to obtain the fuel flow prediction feature vector at the j-th time step; and inputting the fuel flow prediction feature vector at the j-th time step into the fourth fully connected layer and the... The two activation layers are used to process the data to obtain the predicted fuel flow rate at time j. Based on the predicted fuel flow rate and feed flow rate at time j, the training flow rate state vector at time j is obtained. Based on the fuel flow rate and feed flow rate data at multiple times prior to time j, the flow rate state vectors at multiple times prior to time j are obtained. Based on the training flow rate state vector at time j, the flow rate state vectors at multiple times prior to time j, and the trained temperature prediction model, the training exhaust gas temperature data at time j+1 is determined. Based on the training exhaust gas temperature data at time j+1, the preset upper limit of exhaust gas temperature, and the preset lower limit of exhaust gas temperature, the second loss function is determined. Based on the second loss function, the fuel flow rate prediction model is trained to obtain the trained fuel flow rate prediction model.

[0011] According to the present invention, a second loss function is determined based on the training exhaust gas temperature data at time j+1, a preset upper limit for exhaust gas temperature, and a preset lower limit for exhaust gas temperature, including: according to the formula Determine the second loss function ,in, The training exhaust gas temperature data is for the (j+1)th time step. The preset upper limit of exhaust gas temperature, The preset lower limit of exhaust gas temperature, This represents the minimum fuel flow rate. This represents the maximum fuel flow rate. For the fuel flow rate data at time j, For the predicted fuel flow rate data at time j, Let M be the number of training data points at the current time. And j, M and All are positive integers.

[0012] According to a second aspect of the present invention, a fuel addition control system for an aluminum slag solid waste rotary kiln is provided, comprising: an acquisition module, configured to acquire, at each moment after the rotary kiln starts operating, feed flow rate data, fuel flow rate data, exhaust gas temperature data at the rotary kiln outlet, and preheating temperature data of preheating air before fuel combustion heating; a temperature prediction module, configured to process the feed flow rate data and fuel flow rate data from multiple moments prior to the current moment using a temperature prediction model to acquire predicted exhaust gas temperature data and predicted preheating temperature data for the current moment; a judgment module, configured to determine, based on the predicted exhaust gas temperature data, predicted preheating temperature data, exhaust gas temperature data, and preheating temperature data for the current moment, whether the temperature prediction model needs training; and a training module, configured to, if the temperature prediction model needs training, based on the feed flow rate data and fuel flow rate data from multiple moments prior to the current moment... The system trains a temperature prediction model using data, preheating temperature data, and exhaust gas temperature data to obtain a trained temperature prediction model. A determination module is used to use the trained temperature prediction model if no further training is needed. A prediction module processes the feed flow rate data and fuel flow rate data from the current moment and several previous moments using the trained temperature prediction model to obtain the predicted exhaust gas temperature data and predicted preheating temperature data for the next moment. A fuel flow rate module determines the predicted fuel flow rate data for the next moment based on the predicted feed flow rate data, predicted exhaust gas temperature data, and predicted preheating temperature data for the next moment, as well as the feed flow rate data, exhaust gas temperature data, and preheating temperature data from the current moment and several previous moments. A setting module sets the fuel flow rate of the rotary kiln to the predicted fuel flow rate data at the next moment.

[0013] Technical Effects: According to the present invention, the temperatures of preheated air and exhaust gas can be predicted based on a temperature prediction model, thereby automatically determining whether the fuel flow rate is appropriate and automatically adjusting the fuel flow rate. This reduces reliance on manual adjustment, improves the timeliness and accuracy of temperature control, and helps improve the treatment quality of aluminum slag solid waste and the quality of the output. When training the temperature prediction model, the model's predictions of the current exhaust gas temperature and preheating temperature data can be used to determine whether it can adapt to the current operating conditions and whether its accuracy is sufficient. This determines whether the model needs training. If training is required, it is performed based on prediction data from multiple historical moments and actual collected data. The weights of the loss function can be reasonably set according to the proximity of historical moments to the current moment to improve training relevance and the model's adaptability to the current operating conditions. When training the fuel flow prediction model, the already trained temperature prediction model can be used for auxiliary training to obtain a second loss function. The second loss function can be calculated under various conditions. When the exhaust gas temperature fails to remain within a reasonable range, the error term can be increased to enhance training intensity, improve training relevance and efficiency, and rapidly improve the accuracy of the fuel flow prediction model.

[0014] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Other features and aspects of the invention will become clearer from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart of a method for controlling fuel addition in a rotary kiln for aluminum slag solid waste, according to an embodiment of the present invention, is shown as an example.

[0017] Figure 2 An exemplary schematic diagram illustrating the determination of predicted fuel flow data for the next moment according to an embodiment of the present invention is shown;

[0018] Figure 3 A block diagram of a fuel addition control system for an aluminum slag solid waste rotary kiln according to an embodiment of the present invention is shown as an example. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0021] Figure 1 An exemplary flowchart of a method for controlling fuel addition in a rotary kiln for aluminum slag solid waste according to an embodiment of the present invention is shown. The method includes: Step S1, acquiring feed flow rate data, fuel flow rate data, tail gas temperature data at the rotary kiln outlet, and preheating temperature data of preheating air before fuel combustion heating at each time after the rotary kiln starts working; Step S2, processing the feed flow rate data and fuel flow rate data from multiple times prior to the current time using a temperature prediction model to obtain predicted tail gas temperature data and predicted preheating temperature data for the current time; Step S3, determining whether the temperature prediction model needs to be trained based on the predicted tail gas temperature data, predicted preheating temperature data, tail gas temperature data, and preheating temperature data for the current time; Step S4, if the temperature prediction model needs to be trained, then based on the feed flow rate data from the current time and multiple times prior to the current time, Step S5: If the temperature prediction model does not need training, it is used as the already trained model. Step S6: Using the trained temperature prediction model, the feed flow rate data and fuel flow rate data from the current moment and previous moments are processed to obtain the predicted exhaust gas temperature data and predicted preheating temperature data for the next moment. Step S7: Based on the predicted feed flow rate data, predicted exhaust gas temperature data, and predicted preheating temperature data for the next moment, as well as the feed flow rate data, exhaust gas temperature data, and preheating temperature data from the current moment and previous moments, the predicted fuel flow rate data for the next moment is determined. Step S8: At the next moment, the fuel flow rate of the rotary kiln is set to the predicted fuel flow rate data.

[0022] The aluminum slag solid waste rotary kiln fuel addition control method according to an embodiment of the present invention can predict the temperature of preheating air and exhaust gas based on a temperature prediction model, thereby automatically determining whether the fuel flow rate is appropriate, and then automatically adjusting the fuel flow rate, reducing the dependence on manual adjustment, improving the timeliness and accuracy of temperature control, and helping to improve the treatment quality of aluminum slag solid waste and the quality of the output.

[0023] According to one embodiment of the present invention, in step S1, the rotary kiln can receive the input of aluminum slag solid waste, use the air heated by fuel combustion to raise the temperature of the aluminum slag solid waste, and during the heating process, the rotary kiln can rotate to make the aluminum slag solid waste heated evenly, so that the product (e.g., aluminum with fewer impurities) can be obtained at the outlet. The exhaust gas at the outlet can also be discharged after being purified by filtration and other treatments. In addition, the exhaust gas also has a high temperature and can be used to preheat the air to improve the thermal energy utilization efficiency and save fuel.

[0024] According to one embodiment of the present invention, at each moment after the rotary kiln starts working, feed flow rate data, fuel flow rate data, exhaust gas temperature data, and preheating air temperature data can be acquired. The interval between adjacent moments can be 10 minutes, 20 minutes, etc., and the present invention does not limit this. Furthermore, the exhaust gas temperature data can be used to monitor the temperature conditions inside the rotary kiln. Inside the rotary kiln, the heated air transfers heat to the aluminum slag solid waste, causing the temperature of the aluminum slag solid waste to rise and the air temperature to fall. An appropriate exhaust gas temperature indicates that the heating condition inside the rotary kiln is normal; a low exhaust gas temperature indicates that the air temperature inside the rotary kiln is also low, indicating insufficient heating; a high exhaust gas temperature indicates that the air temperature inside the rotary kiln is high, which may cause fuel waste. Moreover, both high and low exhaust gas temperatures may reduce the quality of the output. For example, incomplete melting of aluminum may lead to waste of aluminum slag, or excessively high temperatures may cause other substances to melt and mix into the aluminum. Therefore, it is necessary to maintain normal heating conditions inside the rotary kiln. Furthermore, preheating temperature data, feed flow rate data, and fuel flow rate data can all affect the heating status and temperature within the rotary kiln. For example, excessively high preheating temperature or fuel flow rate data may result in excessively high air temperature within the rotary kiln, leading to excessively high tail gas temperature. Conversely, excessively high feed flow rate may result in insufficient heat carried by the air, meaning it may not be enough to heat excessive aluminum slag solid waste to a suitable temperature, thus causing excessively low air temperature within the rotary kiln and excessively low tail gas temperature. Therefore, obtaining feed flow rate data, fuel flow rate data, tail gas temperature data, and preheating temperature data can monitor the heating status within the rotary kiln. When one or more of these data change, other data can be adjusted promptly to maintain a stable and appropriate heating status within the rotary kiln.

[0025] According to an embodiment of the present invention, in step S2, the temperature prediction model can be a recurrent neural network model, an LSTM model, etc. The present invention does not limit the specific type of temperature prediction model. It can first determine whether the prediction accuracy of the temperature prediction model at the current moment is sufficient, that is, determine whether the temperature prediction model needs to be trained. If training is required, the model is trained using data from the current moment and multiple previous moments to make the temperature prediction model more adaptable to the current operating conditions, thereby improving the accuracy of the temperature prediction model and providing an accurate data foundation for subsequent prediction of appropriate fuel flow data.

[0026] According to one embodiment of the present invention, the heat of the air is predicted to come from the exhaust gas at the rotary kiln outlet, and the preheating temperature data of the preheated air has a certain influence on the heating status inside the rotary kiln. Therefore, in addition to monitoring and predicting the exhaust gas temperature data, the preheating temperature data can also be monitored and predicted.

[0027] According to one embodiment of the present invention, the feed flow rate data and fuel flow rate data from multiple time points prior to the current time can be combined to form a flow state vector for each time point. After dimensionality increase processing by a fully connected layer, a flow input vector is obtained, which is then input into the temperature prediction model in chronological order. For example, the flow input vector at one time point and the hidden state vector output at the previous time point (if the one time point is the first of multiple time points prior to the current time point, the hidden state vector is a zero vector) are input into the temperature prediction model, and the hidden state vector corresponding to the one time point is output. This process is repeated until the hidden state vector of the time point prior to the current time point is output, which serves as the output feature vector of the time point prior to the current time point. After processing the output feature vector through a fully connected layer and an activation layer, the predicted exhaust gas temperature data and predicted preheating temperature data for the current time point are obtained. These data can be compared with the actual detected exhaust gas temperature data and preheating temperature data at the current time point to determine whether the accuracy of the temperature prediction model is sufficient, i.e., whether the temperature prediction model needs to be trained.

[0028] According to an embodiment of the present invention, in step S3, determining whether the temperature prediction model needs to be trained based on the predicted exhaust gas temperature data, predicted preheating temperature data, exhaust gas temperature data, and preheating temperature data at the current moment includes: determining the temperature state vector at the current moment based on the exhaust gas temperature data and preheating temperature data at the current moment; determining the predicted temperature state vector at the current moment based on the predicted exhaust gas temperature data and predicted preheating temperature data at the current moment; if the cosine similarity between the temperature state vector at the current moment and the predicted temperature state vector is less than or equal to a first similarity threshold, then it is determined that the temperature prediction model needs to be trained.

[0029] According to one embodiment of the present invention, the temperature state vector is a two-dimensional vector, and the data in the vector are the current exhaust gas temperature data and the preheating temperature data. Similarly, the predicted temperature state vector is also a two-dimensional vector, and the data in the vector are the current predicted exhaust gas temperature data and the predicted preheating temperature data. The cosine similarity between the temperature state vector and the predicted temperature state vector can be determined. If the cosine similarity is less than or equal to a first similarity threshold (e.g., 0.8), it indicates that the error between the temperature data predicted by the model and the actual collected temperature data is large, the temperature prediction model is not accurate enough, and training is required. Conversely, if the cosine similarity is greater than the first similarity threshold, it indicates that the temperature prediction model is accurate and training is not required.

[0030] According to an embodiment of the present invention, in step S3, if the temperature prediction model needs to be trained, the temperature prediction model is trained based on feed flow rate data, fuel flow rate data, preheating temperature data, and exhaust gas temperature data from the current time and multiple previous times to obtain a trained temperature prediction model. This includes: forming a flow state vector at the ikth time by combining the feed flow rate data and fuel flow rate data at the ikth time, where the ikth time is the time before the current time, 1≤k≤n, n is the number of input ports of the temperature prediction model, and i, k, and n are all positive integers; and inputting the flow state vectors from the inth time to the (i-1th time)th time into the first fully connected layer for processing to obtain the ikth time... The flow input vector from time in to time i-1 is obtained; the flow input vector from time in to time i-1 is processed by the temperature prediction model to obtain the output feature vector at time i-1; the output feature vector at time i-1 is processed by the second fully connected layer and the first activation layer to obtain the predicted exhaust gas temperature data and predicted preheating temperature data at time i; based on the predicted exhaust gas temperature data and predicted preheating temperature data at time i, the first loss function of the temperature prediction model is determined; based on the first loss function, the temperature prediction model is trained to obtain the trained temperature prediction model.

[0031] According to one embodiment of the present invention, the flow state vectors at time i-1, ik, and in are obtained in a similar manner to the flow state vector at the current time, and will not be repeated here. The vectors can be input into the first fully connected layer for dimensionality increase to obtain the flow input vectors from time in to time i-1, and then sequentially input into the temperature prediction model according to the time sequence to obtain the output feature vector at time i-1. The output feature vector at time i-1 is then processed by the second fully connected layer and the first activation layer (e.g., a layer processed using the ReLU activation function) to obtain the predicted exhaust gas temperature data and predicted preheating temperature data at time i. Similarly, the predicted exhaust gas temperature data and predicted preheating temperature data for the current time and multiple previous times can be obtained. Furthermore, the actual collected exhaust gas temperature data and preheating temperature data for the current time and multiple previous times can be compared with the predicted exhaust gas temperature data and predicted preheating temperature data obtained by the above model to obtain the first loss function of the temperature prediction model.

[0032] According to an embodiment of the present invention, the first loss function of the temperature prediction model is determined based on the predicted exhaust gas temperature data and the predicted preheating temperature data at the i-th time point, and the exhaust gas temperature data and the preheating temperature data at the i-th time point, including: determining the first loss function of the temperature prediction model according to formula (1). ,

[0033] (1)

[0034] in, Let i be the predicted exhaust gas temperature data at time i. For the predicted preheating temperature data at time i, The exhaust gas temperature data is at time i. The preheating temperature data is for the i-th time step. Let N be the number of training data points at the current time. , i, N and All are positive integers.

[0035] According to an embodiment of the present invention, in formula (1), This represents the error between the exhaust gas temperature data at time i and the predicted exhaust gas temperature data. This represents the error between the preheating temperature data at time i and the predicted preheating temperature data. Therefore... This represents the total prediction error of the temperature prediction model at time i.

[0036] According to one embodiment of the present invention, the closer the i-th time point is to the current time point, the more important the prediction accuracy of the temperature prediction model becomes. Firstly, the closer the i-th time point is to the current time point, the closer the operating conditions at the i-th time point are to the operating conditions at the current time point. Therefore, high prediction accuracy of the temperature prediction model indicates that the model can adapt to making predictions under the current operating conditions. Secondly, the closer the i-th time point is to the current time point, the more likely it is to continue using the prediction data from the i-th time point in subsequent predictions. Therefore, the prediction accuracy of the temperature prediction model is more important. Thus, the closer the i-th time point is to the current time point, the greater the weight can be assigned to the total prediction error at the i-th time point; that is, the weight can be increased accordingly. The higher the total prediction error at time i, the closer time i is to the current time. A larger weight results in a more targeted approach to training the temperature prediction model, thus improving the prediction accuracy for times closer to the current time. The total prediction errors at multiple times are weighted and summed according to these weights to obtain the first loss function. The temperature prediction model can then be trained using this first loss function. For example, based on the first loss function, the parameters of the temperature prediction model can be adjusted using gradient descent, and a trained temperature prediction model can be obtained after multiple training iterations. Here, e is the natural constant, approximately equal to 2.71828.

[0037] In this way, the temperature prediction model can be used to determine whether it can adapt to the current operating conditions and whether its accuracy is sufficient, based on the prediction of the exhaust gas temperature and preheating temperature data at the current moment. This determines whether the temperature prediction model needs to be trained. If training is required, it can be trained based on prediction data from multiple historical moments and actual collected data. The weights of the loss function can be reasonably set according to the closeness between historical moments and the current moment to improve the training relevance and the adaptability of the temperature prediction model to the current operating conditions.

[0038] According to an embodiment of the present invention, in step S5, if the cosine similarity between the current temperature state vector and the predicted temperature state vector is greater than the first similarity threshold, it indicates that the temperature prediction model has high accuracy and does not need to be trained. It can be directly used as a trained temperature prediction model for subsequent predictions.

[0039] According to one embodiment of the present invention, in step S6, the feed flow rate data and fuel flow rate data of the current time and multiple previous time points can be processed by a trained temperature prediction model to obtain the predicted exhaust gas temperature data and predicted preheating temperature data for the next time point. In the example, the feed flow rate data and fuel flow rate data of the current time and multiple previous time points can be respectively composed into flow state vectors. After obtaining the flow input vector for each time point through the first fully connected layer, they are sequentially input into the trained temperature prediction model in time order to obtain the output feature vector for the current time point. After processing by the second fully connected layer and the first activation layer, the predicted exhaust gas temperature data and predicted preheating temperature data for the next time point are obtained.

[0040] According to one embodiment of the present invention, in step S7, the predicted fuel flow data for the next time moment can be determined by a fuel flow prediction model.

[0041] Figure 2 An exemplary schematic diagram illustrating the determination of predicted fuel flow data for the next moment according to an embodiment of the present invention is shown.

[0042] According to one embodiment of the present invention, determining the predicted fuel flow rate data for the next time step based on the predicted feed flow rate data, predicted exhaust gas temperature data, and predicted preheating temperature data for the next time step, as well as the feed flow rate data, exhaust gas temperature data, and preheating temperature data for the current time step and multiple time steps prior to the current time step, includes: obtaining a fuel prediction input vector for the current time step and multiple time steps prior to the current time step based on the feed flow rate data, exhaust gas temperature data, and preheating temperature data for the current time step and multiple time steps prior to the current time step, and determining the fuel prediction input vector for the next time step based on the predicted feed flow rate data, predicted exhaust gas temperature data, and predicted preheating temperature data for the next time step; The fuel prediction input vectors for the current time step and multiple time steps prior to the current time step, as well as the fuel prediction input vector for the next time step, are input into the third fully connected layer for processing to obtain the fuel prediction state vectors for the current time step and multiple time steps prior to the current time step, and the fuel prediction state vector for the next time step. These fuel prediction state vectors are then input into the trained fuel flow prediction model to obtain the fuel flow prediction feature vector for the next time step. Finally, the fuel flow prediction feature vector for the next time step is input into the fourth fully connected layer and the second activation layer for processing to obtain the predicted fuel flow data for the next time step.

[0043] According to one embodiment of the present invention, the feed flow rate data, exhaust gas temperature data, and preheating temperature data from the current moment and several previous moments can be used to form the fuel prediction input vector for each moment. That is, the fuel prediction input vector is a three-dimensional vector, with each dimension containing feed flow rate data, exhaust gas temperature data, and preheating temperature data, respectively. The predicted feed flow rate data for the next moment can be determined according to the production plan; that is, this data is manually set. The predicted exhaust gas temperature data and predicted preheating temperature data for the next moment are the output data from the trained temperature prediction model. These three data can be used to form the fuel prediction input vector for the next moment. The fuel flow prediction model can determine the predicted fuel flow rate data that maintains the exhaust gas temperature at an appropriate level based on the input feed flow rate data, exhaust gas temperature data, and preheating temperature data. The fuel flow prediction model can be a recurrent neural network model, an LSTM model, etc., and the present invention does not limit it to these.

[0044] According to one embodiment of the present invention, the fuel prediction input vector at each time step can be input into a third fully connected layer for processing to obtain the fuel prediction state vector at each time step. For example, the third fully connected layer can be used to increase the dimensionality and obtain a higher-dimensional fuel prediction state vector. The fuel prediction state vector at each time step can be sequentially input into a trained fuel flow prediction model in the order of time steps to obtain the fuel flow prediction feature vector at the next time step. Specifically, when predicting the fuel flow prediction feature vector at the next time step, multiple fuel prediction state vectors are used. During input, the fuel prediction state vector at the first time step and the zero vector (as the initial hidden state vector) can be input first to obtain the hidden state vector at the first time step. Then, the fuel prediction state vector at the second time step and the hidden state vector at the first time step are input into the trained fuel flow prediction model to obtain the hidden state vector at the second time step, and so on, until the fuel prediction state vector at the next time step and the hidden state vector at the current time step are input to obtain the hidden state vector at the next time step, which serves as the fuel flow prediction feature vector at the next time step. Based on this processing method, the fuel flow prediction feature vector for the next time step can integrate data such as feed flow rate, exhaust gas temperature, and preheating temperature from multiple previous time steps. It can also refer to data from multiple time steps for calculation, thereby obtaining predicted fuel flow rate data that maintains the exhaust gas temperature at an appropriate level. In other words, the predicted feed flow rate, predicted preheating temperature, and predicted fuel flow rate for the next time step all affect the exhaust gas temperature. The trained fuel flow prediction model can determine the pattern of this influence during training. Given the predicted feed flow rate, predicted preheating temperature, and predicted exhaust gas temperature data, it can calculate the predicted fuel flow rate data that maintains the exhaust gas temperature at an appropriate level. That is, if the predicted preheating temperature is at an appropriate level, the predicted fuel flow rate data ensures that subsequent predicted preheating temperature data remains at an appropriate level; if the predicted preheating temperature is not at an appropriate level, the predicted fuel flow rate data adjusts subsequent predicted preheating temperature data to an appropriate level.

[0045] According to one embodiment of the present invention, after determining the fuel flow prediction feature vector for the next time step, it can be input into a fourth fully connected layer for dimensionality reduction, and the dimensionality reduction result can be input into a second activation layer for processing through activation functions such as ReLU to obtain the predicted fuel flow data for the next time step.

[0046] According to an embodiment of the present invention, the training steps of the fuel flow prediction model include: obtaining the fuel prediction input vector at the js-th time step based on the feed flow rate data, exhaust gas temperature data, and preheating temperature data at the js-th time step, where the j-th time step is the time step before the current time step, m is the number of input ports of the fuel flow prediction model, 0≤s≤m-1, and j, s, and m are all integers; inputting the fuel prediction input vectors from the j-m+1-th time step to the j-th time step into a third fully connected layer for processing to obtain the fuel prediction state vectors from the j-m+1-th time step to the j-th time step; inputting the fuel prediction state vectors from the j-m+1-th time step to the j-th time step into the fuel flow prediction model to obtain the fuel flow prediction feature vector at the j-th time step; and inputting the fuel flow prediction feature vector at the j-th time step into a fourth fully connected layer. The system processes the data in the first and second activation layers to obtain the predicted fuel flow rate data at time j. Based on the predicted fuel flow rate data and feed flow rate data at time j, the system obtains the training flow rate state vector at time j. Based on the fuel flow rate data and feed flow rate data at multiple times prior to time j, the system obtains the flow rate state vectors at multiple times prior to time j. Based on the training flow rate state vector at time j, the flow rate state vectors at multiple times prior to time j, and the trained temperature prediction model, the system determines the training exhaust gas temperature data at time j+1. Based on the training exhaust gas temperature data at time j+1, a preset upper limit for exhaust gas temperature, and a preset lower limit for exhaust gas temperature, the system determines the second loss function. Based on the second loss function, the system trains the fuel flow rate prediction model to obtain the trained fuel flow rate prediction model.

[0047] According to one embodiment of the present invention, the fuel prediction state vector from time j (m+1) to time j is obtained in a similar manner to the fuel prediction state vectors for the current time and the previous multiple time periods and the next time period, and will not be described again here. The fuel prediction state vectors from time j (m+1) to time j can be input into the fuel flow prediction model in time sequence to obtain the fuel flow prediction feature vector for time j, and then input into the fourth fully connected layer and the second activation layer to obtain the predicted fuel flow data for time j.

[0048] According to one embodiment of the present invention, the predicted fuel flow rate data at time j and the actual fuel flow rate data at time j may be inconsistent, making it difficult to obtain the operating conditions using the predicted fuel flow rate data. In this case, the trained temperature prediction model described above can be used for auxiliary training. That is, the predicted fuel flow rate data at time j and the actual feed flow rate data at time j are combined to form the training flow rate state vector at time j. Furthermore, the fuel flow rate data and feed flow rate data from multiple times prior to time j can be combined to form the flow rate state vectors for each time moment. The trained temperature prediction model is then used to process the flow rate state vectors for each time moment and the training flow rate state vector at time j to obtain the training exhaust gas temperature data at time j+1. The processing method is similar to the method described above for determining the predicted exhaust gas temperature data for the next time moment, and will not be repeated here. Further, training exhaust gas temperature data for multiple time moments can also be determined in a similar manner.

[0049] According to one embodiment of the present invention, determining a second loss function based on the training exhaust gas temperature data at time j+1, a preset upper limit for exhaust gas temperature, and a preset lower limit for exhaust gas temperature includes: determining the second loss function according to formula (2). ,

[0050] (2)

[0051] in, The training exhaust gas temperature data is for the (j+1)th time step. The preset upper limit of exhaust gas temperature, The preset lower limit of exhaust gas temperature, This represents the minimum fuel flow rate. This represents the maximum fuel flow rate. For the fuel flow rate data at time j, For the predicted fuel flow rate data at time j, Let M be the number of training data points at the current time. And j, M and All are positive integers.

[0052] According to an embodiment of the present invention, in formula (2), To train exhaust gas temperature data, the difference between the training exhaust gas temperature data and the average of the upper and lower limits of exhaust gas temperature is used. When the training exhaust gas temperature data is between the upper and lower limits of exhaust gas temperature, it indicates that the exhaust gas temperature can be kept within a reasonable range. In this case, the difference can be used as the exhaust gas temperature error at time j+1, thereby reducing the error during training and enabling the fuel flow prediction model to output predicted fuel flow data that is closer to the midpoint between the upper and lower limits of exhaust gas temperature.

[0053] According to one embodiment of the present invention, when the training exhaust gas temperature data is lower than the lower limit of the exhaust gas temperature, the exhaust gas temperature fails to be maintained within a reasonable range. In this case, the exhaust gas temperature can be... As the exhaust gas temperature error at time j+1, where, The error between the lower limit of exhaust gas temperature and the training exhaust gas temperature data can be reduced through training, increasing the probability that the training exhaust gas temperature data will reach the upper and lower limits of the exhaust gas temperature. Amplify this error, Let be the difference between the actual fuel flow rate at time j and the lower limit of the exhaust gas temperature. Let be the difference between the predicted fuel flow rate data at time j and the lower limit of the exhaust gas temperature. This is assuming the actual exhaust gas temperature data at time j+1 is between the upper and lower limits of the exhaust gas temperature, and the training exhaust gas temperature data is below the lower limit of the exhaust gas temperature. ,therefore, ,Right now, This coefficient can be used to amplify the error between the lower limit of exhaust gas temperature and the training exhaust gas temperature data, thereby obtaining the exhaust gas temperature error at time j+1 and improving the training intensity when the exhaust gas temperature fails to be kept within a reasonable range.

[0054] According to one embodiment of the present invention, when the training exhaust gas temperature data exceeds the upper limit of the exhaust gas temperature, the exhaust gas temperature fails to be maintained within a reasonable range. In this case, the exhaust gas temperature can be... As the exhaust gas temperature error at time j+1, where, The error between the upper limit of exhaust gas temperature and the training exhaust gas temperature data can be reduced through training, increasing the probability that the training exhaust gas temperature data will reach the upper and lower limits of the exhaust gas temperature. Amplify this error, Let be the difference between the actual fuel flow rate at time j and the upper limit of the exhaust gas temperature. Let be the difference between the predicted fuel flow rate at time j and the upper limit of the exhaust gas temperature. This is assuming the actual exhaust gas temperature at time j+1 is between the upper and lower limits of the exhaust gas temperature, and the training exhaust gas temperature is higher than the lower limit of the exhaust gas temperature. ,therefore, ,Right now, This coefficient can be used to amplify the error between the upper limit of exhaust gas temperature and the training exhaust gas temperature data, and obtain the exhaust gas temperature error at the (j+1)th time, so as to improve the training intensity when the exhaust gas temperature fails to be kept within a reasonable range.

[0055] According to one embodiment of the present invention, The weights are based on time; the closer the j-th time is to the current time, the larger the weight. The meaning of this weight is the same as in formula (1). Similarly, I will not go into details here.

[0056] According to one embodiment of the present invention, a second loss function can be obtained by weighting and summing the exhaust gas temperature error at time j+1 using the above weights, and the fuel flow prediction model can be trained using the second loss function. For example, based on the second loss function, the parameters of the fuel flow prediction model can be adjusted by gradient descent, and the trained fuel flow prediction model can be obtained after multiple training sessions.

[0057] In this way, when training the fuel flow prediction model, the already trained temperature prediction model can be used for auxiliary training to obtain a second loss function. The second loss function can be calculated in various cases. When the exhaust gas temperature fails to be kept within a reasonable range, the error term can be increased in a targeted manner to enhance the training intensity, improve the training relevance and efficiency, and quickly improve the accuracy of the fuel flow prediction model.

[0058] According to an embodiment of the present invention, in step S8, after obtaining the predicted fuel flow data for the next time step through the above-trained fuel flow prediction model, the fuel flow rate of the rotary kiln is set to the predicted fuel flow data when the next time step is reached.

[0059] The fuel addition control method for rotary kilns handling aluminum slag solid waste according to embodiments of the present invention can predict the temperatures of preheating air and exhaust gas based on a temperature prediction model, thereby automatically determining whether the fuel flow rate is appropriate and automatically adjusting the fuel flow rate. This reduces reliance on manual adjustment, improves the timeliness and accuracy of temperature control, and helps improve the treatment quality of aluminum slag solid waste and the quality of the output. When training the temperature prediction model, the model's predictions of the current exhaust gas temperature and preheating temperature data can be used to determine whether the model can adapt to the current operating conditions and whether its accuracy is sufficient. This determines whether the model needs training. If training is required, it is performed based on prediction data from multiple historical times and actual collected data. The weights of the loss function can be reasonably set according to the closeness between historical and current times to improve the training's relevance and the model's adaptability to the current operating conditions. When training the fuel flow prediction model, the already trained temperature prediction model can be used for auxiliary training to obtain a second loss function. The second loss function can be calculated in various cases. When the exhaust gas temperature fails to be kept within a reasonable range, the error term can be increased in a targeted manner to enhance the training intensity, improve the training relevance and efficiency, and quickly improve the accuracy of the fuel flow prediction model.

[0060] Figure 3An exemplary block diagram of a rotary kiln fuel addition control system for aluminum slag solid waste according to an embodiment of the present invention is shown. The system includes: an acquisition module, configured to acquire, at each moment after the rotary kiln starts working, the feed flow rate data, fuel flow rate data, exhaust gas temperature data at the rotary kiln outlet, and preheating temperature data of the preheating air before fuel combustion heating of aluminum slag solid waste; a temperature prediction module, configured to process the feed flow rate data and fuel flow rate data from multiple moments prior to the current moment using a temperature prediction model to acquire predicted exhaust gas temperature data and predicted preheating temperature data for the current moment; a judgment module, configured to determine whether the temperature prediction model needs training based on the predicted exhaust gas temperature data, predicted preheating temperature data, exhaust gas temperature data, and preheating temperature data for the current moment; and a training module, configured to, if the temperature prediction model needs training, determine whether training is required based on the feed flow rate data from the current moment and multiple moments prior to the current moment. The temperature prediction model is trained using fuel flow rate data, preheating temperature data, and exhaust gas temperature data to obtain a trained temperature prediction model. A determination module is used to use the trained temperature prediction model if no further training is needed. A prediction module processes the feed flow rate data and fuel flow rate data from the current moment and several previous moments using the trained temperature prediction model to obtain the predicted exhaust gas temperature data and predicted preheating temperature data for the next moment. A fuel flow rate module determines the predicted fuel flow rate data for the next moment based on the predicted feed flow rate data, predicted exhaust gas temperature data, and predicted preheating temperature data for the next moment, as well as the feed flow rate data, exhaust gas temperature data, and preheating temperature data from the current moment and several previous moments. A setting module sets the fuel flow rate of the rotary kiln to the predicted fuel flow rate data at the next moment.

[0061] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0062] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and any variations or modifications may be made to the implementation of the present invention without departing from the stated principles.

Claims

1. A method for controlling fuel addition in a rotary kiln for aluminum slag solid waste, characterized in that, include: At each moment after the rotary kiln starts operating, data on the feed flow rate of aluminum slag solid waste, fuel flow rate, exhaust gas temperature at the rotary kiln outlet, and preheating temperature of the preheating air before fuel combustion are acquired. Using a temperature prediction model, the feed flow rate and fuel flow rate data from multiple moments prior to the current moment are processed to obtain the predicted exhaust gas temperature and predicted preheating temperature data for the current moment. Based on the predicted exhaust gas temperature, predicted preheating temperature, exhaust gas temperature, and preheating temperature data for the current moment, it is determined whether the temperature prediction model needs training. If the temperature prediction model needs training, then the feed flow rate and fuel flow rate data from the current moment and multiple moments prior to the current moment are used for training. The temperature prediction model is trained using preheating temperature data and exhaust gas temperature data to obtain a trained temperature prediction model. If the temperature prediction model does not require training, it is used as the trained temperature prediction model. The trained temperature prediction model is then used to process the feed flow rate data and fuel flow rate data from the current time and several previous time points to obtain the predicted exhaust gas temperature data and predicted preheating temperature data for the next time point. Based on the predicted feed flow rate data, predicted exhaust gas temperature data, and predicted preheating temperature data for the next time point, as well as the feed flow rate data, exhaust gas temperature data, and preheating temperature data from the current time and several previous time points, the predicted fuel flow rate data for the next time point is determined. At the next moment, the fuel flow rate of the rotary kiln is set to the predicted fuel flow rate data.

2. The method for controlling fuel addition in a rotary kiln for aluminum slag solid waste according to claim 1, characterized in that, Based on the current predicted exhaust gas temperature data, predicted preheating temperature data, exhaust gas temperature data, and preheating temperature data, determine whether the temperature prediction model needs training. This includes: determining the temperature state vector at the current moment based on the current exhaust gas temperature data and preheating temperature data; determining the predicted temperature state vector at the current moment based on the current predicted exhaust gas temperature data and predicted preheating temperature data; and determining that the temperature prediction model needs training if the cosine similarity between the current temperature state vector and the predicted temperature state vector is less than or equal to a first similarity threshold.

3. The method for controlling fuel addition in a rotary kiln for aluminum slag solid waste according to claim 1, characterized in that, If the temperature prediction model needs to be trained, it is trained using feed flow rate data, fuel flow rate data, preheating temperature data, and exhaust gas temperature data from the current time and multiple previous time points. This training process includes: constructing a flow state vector at time ik from the feed flow rate data and fuel flow rate data at time ik, where time ik is the time before the current time, 1 ≤ k ≤ n, n is the number of input ports of the temperature prediction model, and i, k, and n are all positive integers; and inputting the flow state vectors from time in (i) to time i-1 into the first fully connected layer for processing to obtain the flow state vectors from time in (i) to time i-1. The flow rate input vector at time 1 is used. The flow rate input vector from time 'in' to time 'i-1' is processed using a temperature prediction model to obtain the output feature vector at time 'i-1'. The output feature vector at time 'i-1' is then processed using a second fully connected layer and a first activation layer to obtain the predicted exhaust gas temperature data and predicted preheating temperature data at time 'i'. Based on the predicted exhaust gas temperature data and predicted preheating temperature data at time 'i', a first loss function for the temperature prediction model is determined. The temperature prediction model is then trained using the first loss function to obtain a trained temperature prediction model.

4. The method for controlling fuel addition in a rotary kiln for aluminum slag solid waste according to claim 3, characterized in that, Based on the predicted exhaust gas temperature data and predicted preheating temperature data at time i, and the exhaust gas temperature data and preheating temperature data at time i, the first loss function of the temperature prediction model is determined, including: according to the formula Determine the first loss function of the temperature prediction model. ,in, Let i be the predicted exhaust gas temperature data at time i. For the predicted preheating temperature data at time i, The exhaust gas temperature data is at time i. The preheating temperature data is for the i-th time step. Let N be the number of training data points at the current time. , i, N and All are positive integers.

5. The method for controlling fuel addition in a rotary kiln for aluminum slag solid waste according to claim 1, characterized in that, Based on the predicted feed flow rate, predicted exhaust gas temperature, and predicted preheating temperature data for the next time step, as well as the feed flow rate, exhaust gas temperature, and preheating temperature data for the current time step and several previous time steps, the predicted fuel flow rate data for the next time step is determined. This includes: obtaining fuel prediction input vectors for the current time step and several previous time steps based on the feed flow rate, exhaust gas temperature, and preheating temperature data for the current time step and several previous time steps; and determining the fuel prediction input vector for the next time step based on the predicted feed flow rate, predicted exhaust gas temperature, and predicted preheating temperature data for the next time step; inputting the fuel prediction input vectors for the current time step and several previous time steps, as well as the fuel prediction input vector for the next time step, into the third fully connected layer for processing to obtain the fuel prediction state vectors for the current time step and several previous time steps, and the fuel prediction state vector for the next time step; inputting the fuel prediction state vectors for the current time step and several previous time steps, as well as the fuel prediction state vector for the next time step, into the trained fuel flow rate prediction model to obtain the fuel flow rate prediction feature vector for the next time step; and inputting the fuel flow rate prediction feature vector for the next time step into the fourth fully connected layer and the second activation layer for processing to obtain the predicted fuel flow rate data for the next time step.

6. The method for controlling fuel addition in a rotary kiln for aluminum slag solid waste according to claim 5, characterized in that, The training steps of the fuel flow prediction model include: obtaining the fuel prediction input vector at time js based on the feed flow rate data, exhaust gas temperature data, and preheating temperature data at time js, where time j is the time before the current time, m is the number of input ports of the fuel flow prediction model, 0≤s≤m-1, and j, s, and m are all integers; inputting the fuel prediction input vectors from time j-m+1 to time j into the third fully connected layer for processing to obtain the fuel prediction state vectors from time j-m+1 to time j; inputting the fuel prediction state vectors from time j-m+1 to time j into the fuel flow prediction model to obtain the fuel flow prediction feature vector at time j; and inputting the fuel flow prediction feature vector at time j into the fourth fully connected layer and the second activation layer. The process involves processing data at each time step to obtain the predicted fuel flow rate at time j. Based on the predicted fuel flow rate and feed flow rate at time j, a training flow rate state vector at time j is obtained. Furthermore, based on the fuel flow rate and feed flow rate data from multiple time steps prior to time j, flow rate state vectors from multiple time steps prior to time j are obtained. Based on the training flow rate state vector at time j, the flow rate state vectors from multiple time steps prior to time j, and the trained temperature prediction model, the training exhaust gas temperature data at time j+1 is determined. Based on the training exhaust gas temperature data at time j+1, a preset upper limit for exhaust gas temperature, and a preset lower limit for exhaust gas temperature, a second loss function is determined. Finally, the fuel flow rate prediction model is trained using the second loss function to obtain the trained fuel flow rate prediction model.

7. The method for controlling fuel addition in a rotary kiln for aluminum slag solid waste according to claim 6, characterized in that, Based on the exhaust gas temperature data at time j+1, the preset upper limit and lower limit of exhaust gas temperature, the second loss function is determined, including: according to the formula Determine the second loss function ,in, The training exhaust gas temperature data is for the (j+1)th time step. The preset upper limit of exhaust gas temperature, The preset lower limit of exhaust gas temperature, This represents the minimum fuel flow rate. This represents the maximum fuel flow rate. For the fuel flow rate data at time j, For the predicted fuel flow rate data at time j, Let M be the number of training data points at the current time. And j, M and All are positive integers.

8. A fuel addition control system for an aluminum slag solid waste rotary kiln, used to execute the method as described in any one of claims 1-7, characterized in that, include: The acquisition module is used to acquire data on the feed flow rate of aluminum slag solid waste, fuel flow rate, exhaust gas temperature at the rotary kiln outlet, and preheating temperature of the preheating air before fuel combustion heating at each moment after the rotary kiln starts working. The temperature prediction module is used to process the feed flow rate and fuel flow rate data from multiple moments before the current moment through a temperature prediction model to acquire the predicted exhaust gas temperature data and predicted preheating temperature data for the current moment. The judgment module is used to determine whether the temperature prediction model needs to be trained based on the predicted exhaust gas temperature data, predicted preheating temperature data, exhaust gas temperature data, and preheating temperature data for the current moment. The training module trains the temperature prediction model using feed flow rate data, fuel flow rate data, preheating temperature data, and exhaust gas temperature data from the current moment and multiple previous moments if training is required, thus obtaining a trained temperature prediction model. The determination module uses the trained temperature prediction model as the model if training is not needed. The prediction module processes the feed flow rate data and fuel flow rate data from the current moment and multiple previous moments using the trained temperature prediction model to obtain the predicted exhaust gas temperature data and predicted preheating temperature data for the next moment. The fuel flow rate module determines the predicted fuel flow rate data for the next moment based on the predicted feed flow rate data, predicted exhaust gas temperature data, and predicted preheating temperature data for the next moment, as well as the feed flow rate data, exhaust gas temperature data, and preheating temperature data from the current moment and multiple previous moments. The setting module is used to set the fuel flow rate of the rotary kiln to the predicted fuel flow rate data at the next time step.

Images