Coal gas / oxygen consumption prediction method and system suitable for steel plant and medium

Abstract

The invention discloses a coal gas / oxygen consumption prediction method and system suitable for a steel plant and a medium. The method comprises the steps that the coal gas / oxygen consumption of the steel plant and historical sequence data of influence indexes highly related to the coal gas / oxygen consumption are obtained; training an attention mechanism-based improved Elman neural network by using the acquired historical sequence data to obtain a coal gas / oxygen consumption prediction model; and acquiring real-time monitoring sequence data of the coal gas / oxygen consumption and the influence indexes highly related to the coal gas / oxygen consumption, and inputting the real-time monitoring sequence data into the coal gas / oxygen consumption prediction model to obtain a predicted value of the coal gas / oxygen consumption at a future moment. According to the method, the coal gas / oxygen consumption and the influence index data highly related to the coal gas / oxygen consumption are used as model input, and an attention mechanism is introduced to adopt different attention degrees for various inputs, so that the coal gas / oxygen consumption prediction precision can be further improved, and accurate and rapid real-time coal gas / oxygen consumption prediction is realized.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing, and in particular to a method, system and medium for predicting gas / oxygen consumption applicable to a steel plant. Background Art

[0002] Resource scheduling and optimal allocation are important components of the intelligent manufacturing process. Monitoring and estimating resource consumption is an important basis for achieving resource scheduling and optimal allocation. Steel companies are large energy consumers. Continuously improving the energy conservation and consumption reduction level of steel companies is of great significance to the sustainable development of enterprises. Monitoring and predicting the real-time consumption of gas and oxygen not only helps to grasp the trend of energy consumption, reduce resource waste, and ensure accurate energy supply, but also helps to maintain the stable operation of the intelligent manufacturing system.

[0003] The main methods for predicting energy consumption of domestic and foreign enterprises include causal relationship prediction method, energy consumption elasticity coefficient prediction method, neural network prediction method, etc. Since the energy consumption prediction model of steel enterprises is not only complex but also often nonlinear, and neural networks have strong ability to fit nonlinear mapping and learning ability, compared with other prediction methods, the prediction method based on neural networks has higher prediction accuracy and higher reliability of prediction results.

[0004] At present, the Elman neural network is used. The network was proposed by Jeffrey L.Elman. The algorithm is simple and easy to implement. The Elman neural network can be regarded as a recursive neural network with local memory units and local feedback connections. On the basis of the basic structure of the BP network, it adds a succession layer as a one-step delay operator to achieve the purpose of memory, so that the network has the ability to adapt to time-varying characteristics. Compared with the feedforward neural network, it has stronger computing power and short-term memory function. The Elman neural network consists of four layers, namely the input layer, hidden layer, output layer, and succession layer. The model inputs the original input data through the input layer and accesses the output of the hidden layer at the previous moment through the succession layer. The input layer data and the succession layer data are spliced ​​and input into the hidden layer for feature extraction, and the features are input to the output layer, and the output layer outputs the prediction results.

[0005] When the Elman neural network is conventionally used to predict the consumption of gas and oxygen, only the time series of gas and oxygen consumption is used as input to predict the consumption of gas and oxygen at future moments. There are few reference factors, and the prediction accuracy of gas and oxygen consumption needs to be improved.

[0006] In addition, as technology improves, the distribution of coal gas and oxygen consumption data of steel enterprises may change, and the same model will have worse performance on data with different distributions from the training data. Therefore, it is necessary to use the continuously generated new data to update the model parameters. However, for general deep learning algorithms based on backpropagation, after training on new data, the performance of the model on old data will drop significantly, which will lead to catastrophic forgetting. Summary of the invention

[0007] The object of the present invention is to provide a method, system and medium for predicting gas / oxygen consumption applicable to steel plants, so as to improve the prediction accuracy of gas and oxygen consumption.

[0008] In a first aspect, a method for predicting gas / oxygen consumption applicable to a steel plant is provided, comprising:

[0009] Obtain historical series data on gas / oxygen consumption of steel plants and influencing indicators related to high gas / oxygen consumption;

[0010] The gas / oxygen consumption from time t-T+1 to time t and the sequence data of influencing indicators highly related to gas / oxygen consumption are used as input, and the gas / oxygen consumption at time t+1 is used as output. The improved Elman neural network based on the attention mechanism is trained using the acquired historical sequence data to obtain the gas / oxygen consumption prediction model, where T is the sequence length considered for prediction;

[0011] The real-time monitoring sequence data of gas / oxygen consumption and influencing indicators related to gas / oxygen consumption are obtained, and input into the gas / oxygen consumption prediction model to obtain the predicted value of gas / oxygen consumption at a future time.

[0012] Furthermore, it also includes:

[0013] The gas / oxygen consumption and influencing indicator data related to high gas / oxygen consumption are obtained online. Whenever the length of the newly added data obtained online reaches N, the N+1th to Bth samples in the old data are extracted and merged with the newly added data to form a new data set to train the gas / oxygen consumption prediction model so as to update the gas / oxygen consumption prediction model; wherein B and N are both preset values.

[0014] Furthermore, the influencing indicators related to high gas consumption include one or more of oxygen consumption, blast furnace blast volume, blast furnace air pressure, and blast furnace air temperature; the influencing indicators related to high oxygen consumption include one or more of gas consumption, blast furnace blast volume, blast furnace air pressure, and blast furnace air temperature.

[0015] Furthermore, the improved Elman neural network based on the attention mechanism includes:

[0016] The input layer introduces the current input data, including gas / oxygen consumption and highly related influencing index data, and outputs a feature matrix that integrates multiple data to eliminate the dimension effect;

[0017] The attention layer calculates the attention value of the feature matrix output by the input layer and outputs a new time series value;

[0018] The receiving layer receives the output of the previous hidden layer. The output of the receiving layer and the output of the attention layer are input into the current hidden layer.

[0019] Hidden layer, which performs nonlinear transformation on the output of the attention layer and the follow-up layer to extract the temporal features of the sequence and output the hidden state matrix;

[0020] The output layer performs nonlinear calculations on the hidden state matrix to output the predicted value of gas / oxygen consumption at the next moment.

[0021] Furthermore, the input layer inputs M T×1 vectors, where M represents the total number of categories including gas / oxygen consumption and influencing indicators highly correlated therewith; the M input vectors are standardized respectively to obtain M dimensionless vectors; the M dimensionless vectors are concatenated to obtain an initial feature matrix X(t) of dimension T×N; then the input layer output is defined as a0(t)=X(t).

[0022] Furthermore, the attention layer input is the input layer output, and the output is defined as Q=X(t)W Q , K = X(t)W K , V = X(t)W V , W Q , W K , W V are the coefficient matrices used to calculate the Q, K, and V matrices, respectively, and their dimensions are N×T, N×T, and N×1, respectively. X(t) represents the initial feature matrix output by the input layer, and d k is the length of the sequence considered, which is T; the attention layer output a1(t) is a T×1 column vector.

[0023] Furthermore, the receiving layer input is the output of the hidden layer at the previous moment, and the output is defined as u(t)=a2(t-1); the output u(t) is a column vector of L×1.

[0024] Furthermore, the hidden layer input is the attention layer output and the receiving layer output, and the output is defined as a2(t)=σ(z2(t)), u(t) is the output of the receiving layer at the current moment, a1(t) is the output of the attention layer, and z2(t) represents the intermediate output of the hidden layer with a dimension of L×1. b2 represents the trainable parameters with dimensions of L×T, L×L, and L×1 respectively; σ is the activation function, which adopts the ReLU function; the output a2(t) is an L×1 column vector.

[0025] Furthermore, the output layer input is the output of the hidden layer, and the output is defined as σ is the activation function, using the ReLU function; z3(t) represents the intermediate output of the output layer with a dimension of 1×1, b3 represents the trainable parameters with dimensions of 1×L and 1×1 respectively; a2(t) represents the output of the hidden layer.

[0026] In a second aspect, a gas / oxygen consumption prediction system applicable to a steel plant is provided, comprising:

[0027] A data monitoring module is used to obtain real-time monitoring sequence data of gas / oxygen consumption and influencing indicators related to high gas / oxygen consumption, and transmit it to a processing module;

[0028] A processing module is configured with a gas / oxygen consumption prediction model trained in any of the above-mentioned gas / oxygen consumption prediction methods applicable to steel plants; and is used to input the data transmitted by the data monitoring module into the gas / oxygen consumption prediction model and output the predicted value of gas / oxygen consumption at a future time.

[0029] In a third aspect, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that when the computer program is executed by a processor, it implements any of the above-mentioned methods for predicting gas / oxygen consumption applicable to steel plants.

[0030] The present invention proposes a method, system and medium for predicting gas / oxygen consumption in steel plants, which has the following advantages:

[0031] (1) The current prediction method, on the one hand, does not consider other influencing factors related to oxygen and gas consumption in the prediction process; on the other hand, does not adopt different attention levels for different parts of the time series, resulting in the prediction accuracy to be improved. In order to solve this problem, in the present invention, gas / oxygen consumption and influencing index data highly related to gas / oxygen consumption are simultaneously used as model inputs, and an attention mechanism is introduced to adopt different attention levels for each type of input, thereby further improving the prediction accuracy of gas / oxygen consumption and realizing accurate and fast real-time gas or oxygen consumption prediction.

[0032] (2) The Attention mechanism calculates the correlation between different elements in the time series, and thus adopts different degrees of attention to different parts of the time series according to the correlation matrix. However, in the existing Attention mechanism implementation process, the dimensions of the matrices K, Q, and V are the same, and cannot be directly embedded in the Elman neural network used in the present invention. Therefore, the present invention uses a V matrix calculation with a different dimension from the conventional attention, so that the Attention mechanism can be embedded in the Elman neural network used in the present invention, thereby adopting different degrees of attention to various types of inputs to improve the prediction accuracy of gas / oxygen consumption.

[0033] (3) In the present invention, new data is continuously collected to replace the oldest data, and the gas / oxygen consumption prediction model training and updating are performed. The idea is to retrain the current model by jointly storing the old training set subset and the new task. Thus, the model can retain the performance on the old data while adapting to the new data, and the model's adaptability to time-varying gas or oxygen consumption conditions is improved. The present invention can solve the catastrophic forgetting problem of ordinary deep learning models after training on new tasks or new data. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 This is a general framework diagram of a method for predicting gas / oxygen consumption applicable to a steel plant provided by an embodiment of the present invention;

[0036] Figure 2 It is a structural diagram of an improved Elman neural network provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0037] To make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described in detail below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0038] like Figure 1 As shown, the present invention provides a method for predicting gas / oxygen consumption applicable to a steel plant, comprising:

[0039] S1: Obtain historical series data of the steel plant's gas / oxygen consumption and influencing indicators related to high gas / oxygen consumption.

[0040] During implementation, the influencing indexes related to high gas consumption include one or more of oxygen consumption, blast furnace blast volume, blast furnace air pressure, and blast furnace air temperature; the influencing indexes related to high oxygen consumption include one or more of gas consumption, blast furnace blast volume, blast furnace air pressure, and blast furnace air temperature. The reason for selecting the three influencing factors of blast furnace blast volume, blast furnace air pressure, and blast furnace air temperature is that the furnace thermal system can directly reflect the working state of the furnace, and the air supply system affects the stable operation of the furnace condition, thereby being able to characterize the gas / oxygen consumption to a certain extent. Gas consumption and oxygen consumption are interrelated, and the two are highly correlated influencing indicators. Therefore, when predicting gas or oxygen consumption, the other can be considered as a highly correlated influencing indicator.

[0041] It should also be noted that what is obtained is sequence data, and the data needs to be processed for subsequent model training. When processing the data, starting from the first sampling point, a sliding window is used to take samples, the sample length is T+1, and the sliding step is 1. The values ​​of the first T (that is, the sequence length considered for prediction) sampling points are used as feature values, and the value of the T+1 sampling point is used as the label of the sample. For example, a time series data X of oxygen consumption contains 5000 sampling points (and sequence length), that is, X = [x1, x2,…, x 5000 ], after processing the data set, the obtained data set is:

[0042] {(x1x2 …x T ,x T+1 ); (x2x3 …x T+1 ,x T+2 );…;(x 5000-T x 5001-T …x 4999 ,x 5000 )}, this operation is applied to the historical series data of gas and oxygen consumption and highly correlated influencing indicators respectively, and then a sample data set can be obtained.

[0043] S2: Taking the gas / oxygen consumption from time t-T+1 to time t and the sequence data of influencing indicators highly correlated with gas / oxygen consumption as input, and the gas / oxygen consumption at time t+1 as output, the improved Elman neural network based on the attention mechanism is trained using the acquired historical sequence data to obtain the gas / oxygen consumption prediction model.

[0044] like Figure 2As shown, the improved Elman neural network based on the attention mechanism includes an input layer, an attention layer, a receiving layer, a hidden layer, and an output layer. The specific description is as follows:

[0045] The input layer introduces the input data at the current moment, including gas / oxygen consumption and highly correlated influencing index data, and outputs a feature matrix that integrates multiple data to eliminate dimensional effects.

[0046] The input layer inputs M T×1 vectors x 1 (t), x 2 (t), …, x N (t); M represents the total number of categories including gas / oxygen consumption and the influencing indicators highly related thereto. The following takes M as 4 as an example for explanation, including gas / oxygen consumption and blast furnace blast volume, blast furnace air pressure, and blast furnace air temperature; vector x i (t) means the sampling value of the i-th influencing indicator (including gas / oxygen consumption) considered T times before the current time. During implementation, the value of T can be 5, 8, 10, 20, etc. The specific value is selected according to actual needs.

[0047] The input layer normalizes the M input vectors separately. The normalization calculation of each input vector is defined as is the jth element of the normalized i-th input vector, is the jth element of the ith input vector, is the mean of all elements of the i-th input vector, σ i is the standard deviation of the i-th input vector element, 1≤i≤N,1≤j≤T; after standardization, we get M T×1 dimensionless column vectors x 1 ′(t),x 2 ′(t),…,x N ′(t); concatenate the M dimensionless vectors by column to obtain the initial feature matrix X(t) of dimension T×N as the input layer output, then the input layer output is defined as a0(t)=X(t).

[0048] The attention layer calculates the attention value of the feature matrix output by the input layer, uses a V matrix calculation with a different dimension from the conventional attention, and outputs a new time series value.

[0049] The input of the attention layer is the output of the input layer a0(t) = [x 1 ′(t),x 2 ′(t),…,x N ′(t)], from Q = X(t)W Q , K = X(t)W K, V = X(t)W V Calculate the Q, K, and V matrices respectively; W Q , W K , W V are the coefficient matrices used to calculate the Q, K, and V matrices, with dimensions of N×T, N×T, and N×1 respectively; the dimensions of the Q, K, and V matrices are T×T, T×T, and T×1 respectively. The output of the attention layer is defined as d k is the length of the sequence under consideration, which is T; where QK T Calculate the similarity relationship between T sampling points and divide by After normalization, the softmax operation converts it into a probability value, which represents the correlation between T sampling points. It is multiplied by the V matrix to obtain a new time feature after weighted summation. The attention layer outputs a1(t) as a T×1 column vector.

[0050] The receiving layer receives the output of the hidden layer at the previous moment. The output of the receiving layer and the output of the attention layer are input into the hidden layer at the current moment.

[0051] The input of the receiving layer is the L×1 column vector a2(t-1), which receives the output of the hidden layer at the previous moment. The output of the receiving layer is u(t)=a2(t-1).

[0052] The hidden layer performs nonlinear transformation on the output of the attention layer and the follow-up layer to extract the temporal features of the sequence and output the hidden state matrix.

[0053] The hidden layer input is the T×1-dimensional column vector a1(t) output by the attention layer and the L×1-dimensional column vector u(t) output by the receiving layer. The hidden layer output is defined as a2(t)=σ(z2(t)). z2(t) represents the intermediate output of the hidden layer with dimension L×1, b2 represents the trainable parameters with dimensions of L×T, L×L, and L×1 respectively; σ is the activation function, which adopts the ReLU function and is defined as g(z)=max(0,z); the hidden layer output a2(t) is an L×1 column vector.

[0054] The output layer performs nonlinear calculations on the hidden state matrix to output the predicted value of gas / oxygen consumption at the next moment.

[0055] The output layer input is the output L×1-dimensional column vector a2(t) of the hidden layer. The output layer performs a linear transformation on the input to obtain w3 is a 1×L weight vector, and b3 is a 1×1 bias. Perform a nonlinear transformation on z3(t) to obtain the hidden layer output σ is the activation function, which uses the ReLU function; z3(t) represents the intermediate output of the output layer with a dimension of 1×1.

[0056] When performing offline training on the improved Elman neural network based on the attention mechanism, the training data needs to be read in chronological order. The current moment is defined as the current iterative training round number, and the previous moment is defined as the previous iterative training round number relative to the current iterative training round number.

[0057] The hyperparameters used during training are:

[0058] EPOCH is the training round, which indicates the number of iterations when training the model. In this embodiment, EPOCH is 1000. Mini-batch size indicates the number of samples contained in a training batch. To ensure the computer calculation speed and achieve better results, in this embodiment, the value of mini-batch size is 64. α is the learning rate, which is used to control the step size of the model to update the weight in each iteration. In the initial stage of training, the value of α should be large and decrease as the number of iterations increases. In this embodiment, the initial value of α is 0.01, and α is adjusted using the exponential learning rate decay method. The adjustment formula is: In this embodiment, γ is 0.98, epoch-number is the current iteration number, and α0 is the initial learning rate.

[0059] is the loss function, which represents the predicted value of a single sample The error between the true value y and the mean square error is used as the loss function in this embodiment, and the calculation formula is: J is the cost function, which represents the arithmetic average of the loss functions of all samples in a training batch. The calculation formula is: In the formula, y i They represent the predicted value and true value of the i-th sample respectively, and P represents the value of the mini-batch size.

[0060] Gradient descent uses the Adam optimization algorithm to accelerate model convergence and has good robustness to sparse gradients and noise.

[0061] After determining the above hyperparameters and functions, the sample data set obtained above is divided into 70% training set, 20% validation set and 10% test set; the training set is used to train the model to learn the sample distribution, the validation set is used to evaluate the model during the iteration process, and the test set is used to perform an unbiased evaluation of the model after the iteration is completed, and finally a gas / oxygen consumption prediction model is obtained.

[0062] S3: Obtain real-time monitoring sequence data of gas / oxygen consumption and influencing indicators related to gas / oxygen consumption, and input them into the gas / oxygen consumption prediction model to obtain the predicted value of gas / oxygen consumption at a future time.

[0063] After the gas / oxygen consumption prediction model is trained, it can be used to perform online real-time prediction of gas / oxygen consumption. By obtaining the real-time monitoring sequence data of gas / oxygen consumption from time t0-T+1 to time t0 and the influencing indicators related to gas / oxygen consumption, and then inputting them into the gas / oxygen consumption prediction model, the predicted value of gas / oxygen consumption at time t0+1 can be obtained, where time t0 represents the current time.

[0064] Considering the catastrophic forgetting problem of common deep learning models after training on new tasks or new data, some preferred embodiments of the present invention further include:

[0065] The gas / oxygen consumption and the influencing index data related to the high gas / oxygen consumption are obtained online. Whenever the length of the newly added data obtained online reaches N, the N+1th Bth sample in the old data is extracted and merged with the newly added data to form a new data set to train the gas / oxygen consumption prediction model to update the gas / oxygen consumption prediction model; wherein B and N are both preset values. In this implementation, B is 5000 and N is 1000 as an example for explanation. At this time, the update training iteration process is basically the same as the process of obtaining the gas / oxygen consumption prediction model by offline training, the difference lies in the selection of hyperparameters. In this embodiment, EPOCHE is 400; the initial value of the learning rate α is 1e-5. Of course, in other embodiments, the value of N and the values ​​of EPOCHE and learning rate α in the update training iteration process can be adjusted according to actual needs. It is preferred that the values ​​of EPOCHE and learning rate α in the update training iteration process are less than the values ​​in the process of obtaining the gas / oxygen consumption prediction model by offline training.

[0066] After training is completed, new network parameters are obtained, and the new model is used for online prediction, which completes an update of the network.

[0067] Through the above-mentioned coal gas / oxygen consumption prediction method suitable for steel mills, it is possible to realize online and accurate prediction of blast furnace coal gas and oxygen consumption, so that it can be fed back to the coal gas and oxygen supply system for accurate supply and reduce resource waste.

[0068] The above embodiment provides a method for predicting gas / oxygen consumption applicable to a steel plant, which has the following advantages:

[0069] (1) The current prediction method, on the one hand, does not consider other influencing factors related to oxygen and gas consumption in the prediction process; on the other hand, does not adopt different attention levels for different parts of the time series, resulting in the prediction accuracy to be improved. In order to solve this problem, in the present invention, gas / oxygen consumption and influencing index data highly related to gas / oxygen consumption are simultaneously used as model inputs, and an attention mechanism is introduced to adopt different attention levels for each type of input, thereby further improving the prediction accuracy of gas / oxygen consumption and realizing accurate and fast real-time gas or oxygen consumption prediction.

[0070] (2) The Attention mechanism calculates the correlation between different elements in the time series, and thus adopts different degrees of attention to different parts of the time series according to the correlation matrix. However, in the existing Attention mechanism implementation process, the dimensions of the matrices K, Q, and V are the same, and cannot be directly embedded in the Elman neural network used in the present invention. Therefore, the present invention uses a V matrix calculation with a different dimension from the conventional attention, so that the Attention mechanism can be embedded in the Elman neural network used in the present invention, thereby adopting different degrees of attention to various types of inputs to improve the prediction accuracy of gas / oxygen consumption.

[0071] (3) In the present invention, new data is continuously collected to replace the oldest data, and the gas / oxygen consumption prediction model training and updating are performed. The idea is to retrain the current model by jointly storing the old training set subset and the new task. Thus, the model can retain the performance on the old data while adapting to the new data, and the model's adaptability to time-varying gas or oxygen consumption conditions is improved. The present invention can solve the catastrophic forgetting problem of ordinary deep learning models after training on new tasks or new data.

[0072] The embodiment of the present invention further provides a gas / oxygen consumption prediction system applicable to a steel plant, comprising:

[0073] A data monitoring module is used to obtain real-time monitoring sequence data of gas / oxygen consumption and influencing indicators related to high gas / oxygen consumption, and transmit it to a processing module;

[0074] A processing module is configured with a gas / oxygen consumption prediction model trained in the gas / oxygen consumption prediction method for steel plants described in any of the above embodiments; it is used to input the data transmitted by the data monitoring module into the gas / oxygen consumption prediction model and output the predicted value of gas / oxygen consumption at a future time.

[0075] The gas / oxygen consumption prediction model initially configured in the processing module is the gas / oxygen consumption prediction model obtained by the aforementioned offline training; in subsequent use, it is preferred to continuously collect new data to replace the oldest data and perform gas / oxygen consumption prediction model training and update.

[0076] The prediction system may also include a display module for displaying the predicted value of the gas / oxygen consumption at a future time output by the processing module.

[0077] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for predicting gas / oxygen consumption applicable to a steel plant as described in any of the above embodiments is implemented.

[0078] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0079] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0080] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0081] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0082] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0083] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for predicting gas / oxygen consumption in a steel plant, characterized in that: include: Obtain historical series data on gas / oxygen consumption of steel plants and influencing indicators related to high gas / oxygen consumption; The gas / oxygen consumption from time t-T+1 to time t and the sequence data of influencing indicators highly related to gas / oxygen consumption are used as input, and the gas / oxygen consumption at time t+1 is used as output. The improved Elman neural network based on the attention mechanism is trained using the acquired historical sequence data to obtain the gas / oxygen consumption prediction model, where T is the sequence length considered for prediction; The real-time monitoring sequence data of gas / oxygen consumption and influencing indicators related to gas / oxygen consumption are obtained, and input into the gas / oxygen consumption prediction model to obtain the predicted value of gas / oxygen consumption at a future time.

2. The method for predicting gas / oxygen consumption in a steel plant according to claim 1, characterized in that: Also includes: The gas / oxygen consumption and influencing indicator data related to high gas / oxygen consumption are obtained online. Whenever the length of the newly added data obtained online reaches N, the N+1th Bth sample in the old data is extracted and merged with the newly added data to form a new data set to train the gas / oxygen consumption prediction model so as to update the gas / oxygen consumption prediction model; wherein B and N are both preset values.

3. The method for predicting gas / oxygen consumption in a steel plant according to claim 1, characterized in that: The influencing indicators related to high gas consumption include one or more of oxygen consumption, blast furnace blast volume, blast furnace air pressure, and blast furnace air temperature; the influencing indicators related to high oxygen consumption include one or more of gas consumption, blast furnace blast volume, blast furnace air pressure, and blast furnace air temperature.

4. The method for predicting gas / oxygen consumption in a steel plant according to claim 1, characterized in that: The improved Elman neural network based on the attention mechanism includes: The input layer introduces the current input data, including gas / oxygen consumption and highly related influencing index data, and outputs a feature matrix that integrates multiple data to eliminate the dimension effect; The attention layer calculates the attention value of the feature matrix output by the input layer and outputs a new time series value; The receiving layer receives the output of the previous hidden layer. The output of the receiving layer and the output of the attention layer are input into the current hidden layer. Hidden layer, which performs nonlinear transformation on the output of the attention layer and the follow-up layer to extract the temporal features of the sequence and output the hidden state matrix; The output layer performs nonlinear calculations on the hidden state matrix to output the predicted value of gas / oxygen consumption at the next moment.

5. The method for predicting gas / oxygen consumption in a steel plant according to claim 4, characterized in that: The input layer inputs M T×1 vectors, where M represents the total number of categories including gas / oxygen consumption and highly correlated influencing indicators; the M input vectors are standardized respectively to obtain M dimensionless vectors; the M dimensionless vectors are concatenated to obtain an initial feature matrix X(t) with a dimension of T×N; then the input layer output is defined as a0(t)=X(t).

6. The method for predicting gas / oxygen consumption in a steel plant according to claim 4, characterized in that: The attention layer input is the input layer output, and the output is defined as Q=X(t)W Q , K = X(t)W K , V = X(t)W V , W Q , W K , W V are the coefficient matrices used to calculate the Q, K, and V matrices, respectively, and their dimensions are N×T, N×T, and N×1, respectively. X(t) represents the initial feature matrix output by the input layer, and d k is the length of the sequence considered, which is T; the attention layer output a1(t) is a T×1 column vector.

7. The method for predicting gas / oxygen consumption in a steel plant according to claim 4, characterized in that: The hidden layer input is the attention layer output and the receiving layer output, and the output is defined as a2(t)=σ(z2(t)), u(t) is the output of the receiving layer at the current moment, a1(t) is the output of the attention layer, and z2(t) represents the intermediate output of the hidden layer with a dimension of L×1. b2 represents the trainable parameters with dimensions of L×T, L×L, and L×1 respectively; σ is the activation function, which adopts the ReLU function; the output a2(t) is an L×1 column vector.

8. The method for predicting gas / oxygen consumption in a steel plant according to claim 4, characterized in that: The output layer input is the output of the hidden layer, and the output is defined as σ is the activation function, using the ReLU function; z3(t) represents the intermediate output of the output layer with a dimension of 1×1, They represent trainable parameters with dimensions of 1×L and 1×1 respectively; a2(t) represents the output of the hidden layer.

9. A gas / oxygen consumption prediction system suitable for a steel plant, characterized in that: include: A data monitoring module is used to obtain real-time monitoring sequence data of gas / oxygen consumption and influencing indicators related to high gas / oxygen consumption, and transmit it to a processing module; A processing module, which is equipped with a gas / oxygen consumption prediction model trained in the gas / oxygen consumption prediction method suitable for steel plants as described in any one of claims 1 to 8; and is used to input the data transmitted by the data monitoring module into the gas / oxygen consumption prediction model and output the predicted value of gas / oxygen consumption at a future time.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting gas / oxygen consumption applicable to a steel plant as claimed in any one of claims 1 to 8 is implemented.

Images