Steel furnace temperature prediction method and system, computer equipment and storage medium

Abstract

The invention relates to the technical field of temperature prediction, and discloses a steel furnace temperature prediction method and system, computer equipment and a storage medium, and the method comprises the steps: obtaining the historical operation data of a steel furnace, and carrying out the data processing of the historical operation data, and obtaining a model data set; the model data set is used for training a temperature prediction model, a trained temperature prediction model is obtained, the temperature prediction model comprises a composite embedding module, an encoder module and a decoder module, and the composite embedding module comprises a composite convolution layer, a category embedding layer, an attention layer and a splicing layer; and acquiring real-time operation data of the steelmaking furnace, and inputting the real-time operation data into the trained temperature prediction model to obtain a temperature prediction result of the steelmaking furnace. According to the method, the precision of a model prediction result is improved by comprehensively processing the data of the multi-dimensional sensor, so that the temperature control precision in the steelmaking process is improved, the energy consumption is reduced through a temperature feedback mechanism, and the steel quality is further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of temperature prediction, and in particular to a method, system, computer equipment and storage medium for predicting the temperature of a steelmaking furnace. Background Art

[0002] In the steelmaking process, precise control of furnace temperature has a vital impact on the quality of steel. Traditional temperature control methods mainly rely on manual experience and simple feedback control, which have problems such as slow response and low precision. With the development of artificial intelligence and big data technology, using advanced machine learning algorithms to predict the temperature of steelmaking furnaces has become a new solution. This method can establish an accurate temperature prediction model by analyzing historical data, thereby achieving precise control of the steelmaking process. However, the current temperature prediction method based on machine learning also has certain limitations when applied to the temperature prediction of steelmaking furnaces. On the one hand, the model based on data statistics for analysis and prediction or regression analysis is relatively simple, and it is difficult to grasp the complex nonlinear relationship in the steelmaking process. On the other hand, the conventional model structure does not consider the time-related features or does not quantify the time-related features, and cannot capture the impact of time series on steelmaking temperature, resulting in insufficient accuracy of model prediction. Summary of the invention

[0003] In order to solve the above technical problems, the present invention provides a steelmaking furnace temperature prediction method, system, computer equipment and storage medium, so as to solve the problem of insufficient prediction accuracy of existing models, and achieve the technical effect of improving the temperature control accuracy in the steelmaking process, reducing energy consumption and improving steel quality.

[0004] In a first aspect, the present invention provides a method for predicting temperature of a steelmaking furnace, the method comprising:

[0005] Acquire historical operation data of the steelmaking furnace, and perform data processing on the historical operation data to obtain a model data set, wherein the model data set includes the historical operation data, the shifted historical operation data, and the offset acquisition time;

[0006] Using the model data set to train a pre-established temperature prediction model to obtain a trained temperature prediction model, wherein the temperature prediction model includes a composite embedding module, an encoder module, and a decoder module, wherein the composite embedding module includes a composite convolutional layer, a category embedding layer, an attention layer, and a splicing layer;

[0007] The real-time operation data of the steelmaking furnace is obtained, and the real-time operation data is input into a trained temperature prediction model to obtain a temperature prediction result of the steelmaking furnace.

[0008] Furthermore, the step of acquiring historical operation data of the steelmaking furnace and performing data processing on the historical operation data to obtain the model data set includes:

[0009] Performing data preprocessing on historical operation data of the steelmaking furnace, wherein the historical operation data includes temperature data, gas composition data, and raw material composition data, wherein the data preprocessing includes missing value processing, data alignment based on dynamic time warping, outlier processing, and data standardization;

[0010] Shifting the preprocessed historical operation data rightwards according to the time dimension to obtain shifted historical operation data;

[0011] The time difference between the collection times of the historical operation data is obtained and normalized to obtain the offset collection time.

[0012] Further, the composite embedding module includes a first composite embedding module and a second composite embedding module, the composite convolution layer includes a first composite convolution layer and a second composite convolution layer, the category embedding layer includes a first category embedding layer and a second category embedding layer, the attention layer includes a first attention layer and a second attention layer, and the splicing layer includes a first splicing layer and a second splicing layer;

[0013] The first composite embedding module includes a plurality of first composite convolutional layers, a plurality of first category embedding layers, a first attention layer and a first concatenation layer, and the first composite convolutional layers and the first category embedding layers correspond one to one;

[0014] The second composite embedding module includes a plurality of second composite convolutional layers, a plurality of second category embedding layers, a second attention layer, and a second concatenation layer, and the second composite convolutional layers and the second category embedding layers correspond one to one;

[0015] The number of the first composite convolutional layers is consistent with the number of data categories in the historical running data, the input data of each of the first composite convolutional layers is the historical running data of the corresponding category, and the output data of the first composite convolutional layer is the input data of the corresponding first category embedding layer;

[0016] The input data of the first attention layer is the output data of each of the first category embedding layers, the input data of the first splicing layer is the output data of the first attention layer and the offset acquisition time, and the output data of the first splicing layer is the input data of the encoder module;

[0017] The number of the second composite convolutional layers is consistent with the number of data categories in the shifted historical running data, the input data of each second composite convolutional layer is the shifted historical running data of the corresponding category, and the output data of the second composite convolutional layer is the input data of the corresponding second category embedding layer;

[0018] The input data of the second attention layer is the output data of each second category embedding layer, the input data of the second splicing layer is the output data of the second attention layer and the offset prediction time, and the output data of the second splicing layer is the input data of the decoder module, wherein the offset prediction time is obtained by the previous model prediction result.

[0019] Furthermore, the composite convolutional layer includes several parallel sub-convolutional layers and a sub-splicing layer connected to each of the convolutional layers, the features of each first category embedding layer and the corresponding second category embedding layer are shared, and the attention layer is composed of a fully connected layer and a normalization layer.

[0020] Furthermore, the composite convolution layer includes three parallel sub-convolution layers, the convolution kernel sizes of each convolution layer are 5*h*32, 3*h*32, and 1*h*32 respectively, and the embedding dimension of each category embedding layer is n d / 3, the dimension of the fully connected layer is n d *3, where h represents feature data of different categories, n d Indicates the dimension of the input data.

[0021] Further, the encoder module includes a first linear addition layer, a first position coding layer and an encoder, the decoder module includes a second linear addition layer, a second position coding layer, a decoder, a fully connected layer and an output layer, and the encoder and the decoder are connected;

[0022] The input data of the first linear addition layer is the output data of the first position encoding layer and the output data of the first composite embedding module, and the output data of the first linear addition layer is the input data of the encoder;

[0023] The input data of the second linear addition layer is the output data of the second position encoding layer and the output data of the second composite embedding module, and the output data of the second linear addition layer is the input data of the decoder.

[0024] Furthermore, after the step of obtaining the temperature prediction result of the steelmaking furnace, the method further comprises:

[0025] A temperature deviation value is calculated based on the temperature prediction result and the process recommended temperature, and a heating strategy of the steelmaking furnace is optimized based on the temperature deviation value.

[0026] In a second aspect, the present invention provides a steelmaking furnace temperature prediction system, the system comprising:

[0027] A data set construction module is used to obtain historical operation data of the steelmaking furnace, and perform data processing on the historical operation data to obtain a model data set, wherein the model data set includes historical operation data, shifted historical operation data and offset collection time;

[0028] A model building module, used to train a pre-established temperature prediction model using the model data set to obtain a trained temperature prediction model, wherein the temperature prediction model includes a composite embedding module, an encoder module and a decoder module, wherein the composite embedding module includes a composite convolutional layer, a category embedding layer, an attention layer and a splicing layer;

[0029] The temperature prediction module is used to obtain the real-time operation data of the steelmaking furnace, and input the real-time operation data into the trained temperature prediction model to obtain the temperature prediction result of the steelmaking furnace.

[0030] In a third aspect, an embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0031] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0032] The present invention provides a method, system, computer device and storage medium for predicting the temperature of a steelmaking furnace. Through the method, the multi-dimensional sensor data of the steelmaking furnace is comprehensively processed, which can solve the problem that conventional models cannot effectively process continuous data, the feature embedding representation degree is not high, and the problem that the internal relationship of the embedded features of each category is disturbed due to the premature merging of features of different categories, thereby improving the accuracy of the model prediction results. The present invention can improve the temperature control accuracy in the steelmaking process, reduce energy consumption, and thus improve the quality of steel. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 1 is a schematic flow chart of a method for predicting temperature of a steelmaking furnace in an embodiment of the present invention;

[0034] Figure 2 is a schematic diagram of the structure of a temperature prediction model in an embodiment of the present invention;

[0035] Figure 3 is a schematic diagram of the structure of a composite embedded module of a temperature prediction model in an embodiment of the present invention;

[0036] Figure 4 is a structural schematic diagram of a steelmaking furnace temperature prediction system in an embodiment of the present invention;

[0037] Figure 5 It is a diagram of the internal structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] See also Figure 1 The first embodiment of the present invention provides a method for predicting the temperature of a steelmaking furnace, which includes steps S10 to S30:

[0040] Step S10, acquiring historical operation data of the steelmaking furnace, and performing data processing on the historical operation data to obtain a model data set, wherein the model data set includes the historical operation data, the shifted historical operation data, and the offset collection time;

[0041] Step S20, using the model data set to train a pre-established temperature prediction model to obtain a trained temperature prediction model, wherein the temperature prediction model includes a composite embedding module, an encoder module and a decoder module, and the composite embedding module includes a composite convolutional layer, a category embedding layer, an attention layer and a splicing layer;

[0042] Step S30, acquiring real-time operation data of the steelmaking furnace, and inputting the real-time operation data into a trained temperature prediction model to obtain a temperature prediction result of the steelmaking furnace.

[0043] The present invention provides a steelmaking furnace temperature prediction method based on an improved Transformer neural network. Compared with the traditional LSTM model, the Transformer model introduces a self-attention mechanism (Self-Attention), which can better capture the global correlation between time steps in a long time series, and is particularly suitable for scenes such as steelmaking furnace temperature with multi-sensor input and complex time series dependence. However, there are still some problems when the traditional Transformer model is used for steelmaking furnace temperature prediction. On the one hand, the main purpose of the traditional Transformer is to process data related to natural language processing, and its input is discrete, mainly based on one-hot encoding, while in steelmaking furnace temperature When predicting problems, continuous data is mainly processed, and its original dimension is relatively low. If it is used directly to generate feature embedding, the degree of representation will be unfavorable for subsequent prediction tasks. On the other hand, the dimensional values ​​of the multi-dimensional one-hot features input by the traditional Transformer model do not need to distinguish categories, while in the steelmaking furnace temperature prediction problem, different categories of features are distinguished, including temperature, gas composition, raw material composition, etc. If these data are merged to generate embedded features too early, it will interfere with the internal relationship of each category of features. In addition, in natural language processing tasks, the input and output are in different semantic spaces, so two independent sets of embedded representations of input and output will be generated, while in the steelmaking furnace temperature prediction problem, the representation meaning of each data is no different in the input and output.

[0044] In order to solve the above problems, the present invention provides an improved model suitable for the steelmaking furnace temperature prediction problem. Before describing the structure of the model, the steps for constructing the data set used for model training are first described. The data set is constructed using the following steps:

[0045] Performing data preprocessing on historical operation data of the steelmaking furnace, wherein the historical operation data includes temperature data, gas composition data, and raw material composition data, wherein the data preprocessing includes missing value processing, data alignment based on dynamic time warping, outlier processing, and data standardization;

[0046] Shifting the preprocessed historical operation data rightwards according to the time dimension to obtain shifted historical operation data;

[0047] The time difference between the collection times of the historical operation data is obtained and normalized to obtain the offset collection time.

[0048] In this embodiment, the historical operation data of the steelmaking furnace is first obtained, and the safe is classified according to temperature, gas composition, and raw material composition, that is, the historical operation data includes temperature data, gas composition data, and raw material composition data, wherein the temperature data refers to the temperature in the furnace, which can be collected using a temperature sensor, and the unit is Celsius; the gas composition data is collected using a gas analyzer, mainly including oxygen, carbon dioxide, carbon monoxide, and nitrogen, expressed in gas volume percentage; the raw material composition data is collected using a spectrometer, mainly including iron, carbon, manganese, silicon, phosphorus, and sulfur, expressed in mass percentage. At the same time, the collection time of each collection is recorded, and all data collected at the same collection time are a sample. Assuming that each sample involves 12 attribute data, the sample form can be expressed as: (collection time, furnace temperature, oxygen volume percentage, carbon dioxide volume percentage, carbon monoxide volume percentage, nitrogen volume percentage, iron mass percentage, carbon mass percentage, manganese mass percentage, silicon mass percentage, phosphorus mass percentage, sulfur mass percentage).

[0049] Then, the collected historical operation data is preprocessed. The preprocessing here includes missing value processing, data alignment based on dynamic time warping, outlier processing and data standardization. Specifically, it includes:

[0050] ① Missing value processing

[0051] Data loss may occur during data collection due to various reasons. Such data will have a negative impact on model training and therefore needs to be processed. Preferably, data samples with attribute missing rate of no less than 30% are eliminated.

[0052] ②Data alignment based on dynamic time warping

[0053] The steelmaking furnace environment is complex. It is usually impossible to obtain the collection results of all devices at the same time during data collection. At the same time, there may be delays in data transmission. If it is directly based on simple timestamp alignment, it will cause errors. Therefore, dynamic time warping (DTW) is used to dynamically align the data collected by different devices on the time axis. Specifically, the data of different devices are regarded as time series, and the longest sequence without missing values ​​is s l , and compare other sequences with s l Align them and use DTW to calculate the "best matching path" between them. For data points x1 and x2 collected by two devices, the absolute value distance |x1-x2| is used to calculate the distance in DTW.

[0054] By aligning the timing of temperature data with other data such as gas composition through DTW, the synchronization of data used for model training and prediction can be improved, thereby improving the accuracy of temperature prediction.

[0055] ③ Outlier processing

[0056] Abnormal data may appear during data collection, and a box plot is used to remove abnormal values. i , count the 25th percentile p of its attribute value 25 and the 75th percentile p 75 , calculate the interquartile range IQR = p 75 -p 25 , if x i <p 25 -1.5IQR or x i >p 75 +1.5IQR, then x i If the outlier attribute in the sample is less than 30%, the median is used for filling; otherwise, the sample is removed.

[0057] ④Data standardization

[0058] In the collected data, the magnitude of each data varies greatly. For example, the temperature of the converter is between 1600℃ and 1700℃, while the mass percentage of phosphorus is usually less than 0.03%. Therefore, the data needs to be normalized. Specifically, for attribute x i , calculate its mean μ in the data set i With standard deviation σ i , each attribute takes a value x ij Standardize according to the following formula:

[0059]

[0060] After preprocessing the historical operation data, the data set can be constructed. In this embodiment, the data set includes historical operation data, shifted historical operation data and offset acquisition time, wherein the historical operation data refers to the historical operation data after the above data preprocessing. It is assumed that k samples are selected from the collected data every s samples to form training samples. The k sample sequence is in the form of (X1, X2, ..., X k ), where sample X i The format is consistent with the above sample format, and (X1, X2, ..., X k-1 ) Prediction X k The temperature y is y. s is called the step length, k-1 is called the sequence length, and s=25, k=61.

[0061] The shifted historical running data refers to the data after the historical running data after data preprocessing is shifted right once according to the time dimension. The offset collection time refers to the offset of any collection time t in the training sample. i Calculate the time difference between adjacent samples, i.e. ti ′=t i -t i-1 ,The offset acquisition time here still needs to be standardized, that is, the offset time in all training samples is taken, the mean and standard deviation are calculated, and ,then standardized according to the above standardization formula before adding it to the model data set.

[0062] Based on the above model data set, the temperature prediction model in this embodiment is constructed based on the improved Transformer neural network model, see Figure 2 The temperature prediction model includes a composite embedding module, an encoder module and a decoder module. It can be seen that the encoder module and the decoder module are actually the Transformer model architecture. In this embodiment, a composite embedding module is added to the original model architecture to process the input data, and then the input data is input into the Transformer model. The reason for adding a composite embedding layer to the original model is mainly to solve the problems mentioned above that the traditional Transformer model has a poor representation degree for continuous data, which is not conducive to subsequent prediction, and cannot handle classified data well.

[0063] In a preferred embodiment, the specific structure of the composite embedded module is:

[0064] The composite embedding module includes a first composite embedding module and a second composite embedding module, the composite convolution layer includes a first composite convolution layer and a second composite convolution layer, the category embedding layer includes a first category embedding layer and a second category embedding layer, the attention layer includes a first attention layer and a second attention layer, and the splicing layer includes a first splicing layer and a second splicing layer;

[0065] The first composite embedding module includes a plurality of first composite convolutional layers, a plurality of first category embedding layers, a first attention layer and a first concatenation layer, and the first composite convolutional layers and the first category embedding layers correspond one to one;

[0066] The second composite embedding module includes a plurality of second composite convolutional layers, a plurality of second category embedding layers, a second attention layer, and a second concatenation layer, and the second composite convolutional layers and the second category embedding layers correspond one to one;

[0067] The number of the first composite convolutional layers is consistent with the number of data categories in the historical running data, the input data of each of the first composite convolutional layers is the historical running data of the corresponding category, and the output data of the first composite convolutional layer is the input data of the corresponding first category embedding layer;

[0068] The input data of the first attention layer is the output data of each of the first category embedding layers, the input data of the first splicing layer is the output data of the first attention layer and the offset acquisition time, and the output data of the first splicing layer is the input data of the encoder module;

[0069] The number of the second composite convolutional layers is consistent with the number of data categories in the shifted historical running data, the input data of each second composite convolutional layer is the shifted historical running data of the corresponding category, and the output data of the second composite convolutional layer is the input data of the corresponding second category embedding layer;

[0070] The input data of the second attention layer is the output data of each second category embedding layer, the input data of the second splicing layer is the output data of the second attention layer and the offset prediction time, and the output data of the second splicing layer is the input data of the decoder module, wherein the offset prediction time is obtained by the previous model prediction result.

[0071] In this embodiment, the composite embedded module can be divided into a first composite embedded module for processing historical operation data and a second composite embedded module for processing shifted historical operation data. The two composite embedded modules have the same structure. The structure of the composite embedded module for processing historical operation data is taken as an example for explanation below.

[0072] Specifically, the composite embedding module is composed of multiple composite convolutional layers, multiple category embeddings corresponding to the composite convolutional layers one by one, an attention layer and a splicing layer, wherein the number of composite convolutional layers is consistent with the number of data types of the input data. Taking the above-mentioned data set including three types of data: temperature, gas composition and raw material composition as an example, the composite convolutional layer in each composite embedding module in this embodiment is set to three according to the data classification, which are used to perform convolution processing on different types of input data, namely, the temperature input data, gas composition input data and raw material composition input data in the figure. Similarly, for the composite embedding module that processes the shift history operation data, its three composite convolutional layers are used to process the temperature shift data, the gas composition shift data and the raw material composition shift data.

[0073] Correspondingly, the category embedding layers in the composite embedding module are also set to three, namely, the temperature embedding layer, the gas composition embedding layer and the raw material composition embedding layer. The input of each category embedding layer is the output of the corresponding composite convolutional layer. The output results of these three category embedding layers will be input into the splicing layer together. Different types of data will be spliced ​​in the splicing layer, and then spliced ​​with the corresponding offset acquisition time in the data set, that is, the offset acquisition time is part of the input data of the splicing layer of the composite embedding module, and the other part of the input data of the splicing layer is the output data of each category embedding layer. It should be noted here that for the composite embedding module that processes the shifted historical operation data, the offset time of the splicing layer input is not the offset acquisition time in the data set, but the offset prediction time determined by the time difference between the time in the prediction result output last time during the model training process and the label time.

[0074] See also Figure 3 The composite convolutional layer in this embodiment is composed of multiple parallel sub-convolutional layers and a sub-convolutional layer, i.e. Figure 3 The composite embedding layer mainly extracts the association information between data with similar positions through convolutions of different sizes. Preferably, three sub-convolution layers are set, and the convolution kernel sizes are set to 5*h*32, 3*h*32, and 1*h*32, respectively, where h is the number of different category features, and the values ​​of temperature, gas composition, and raw material composition are 1, 4, and 6, respectively. After the features of each category are extracted through convolution, they are spliced ​​according to the dimension of the number of features of each category.

[0075] The embedding dimension of each category embedding layer is n d / 3, and the feature sharing of the category embedding layer with consistent classification of historical operation data and shifted historical operation data. This is because in the steelmaking furnace temperature prediction problem, the meaning of each data is the same in input and output, so the embedding parameters can be reduced.

[0076] The attention layer in the composite embedding module consists of a fully connected layer and a normalized layer. d *3's fully connected layer fully connects the outputs of the three category embedding layers and then normalizes them. The concatenation layer multiplies the result of the attention layer with the embedded features of each category and concatenates them according to the embedding feature dimension. At the same time, the offset acquisition time (for input data) or the offset prediction time (for output data) is also concatenated.

[0077] The composite embedding module of this embodiment is used to process continuous feature data of different categories when predicting the temperature of the steelmaking furnace, thereby avoiding the situation where the original low-dimensional feature embedding has insufficient representation, and the embedded features generated by merging different categories of features too early will interfere with the internal relationships of the features of each category, thereby improving the accuracy of subsequent prediction results.

[0078] The data processed by the composite embedding module will be input into the encoder module and the decoder module for processing, and the prediction result will be output. The encoder module includes a linear addition layer, a position coding layer and an encoder. In addition to the linear addition layer and the position coding layer, the decoder module also includes a decoder corresponding to the encoder. Specifically, both position coding layers use trigonometric function position coding, and the coding formula is:

[0079]

[0080] In the formula, pos is the position number of the input data, i is the dimension number of the encoding, and n d Indicates the dimension of the encoding.

[0081] The linear addition layer is used to add vectors or matrices position by position, wherein the input data of the linear addition layer of the encoder module is the output data of the position coding layer of the encoder module and the output data of the first composite embedding module, and the input data of the linear addition layer of the decoder module is the output data of the position coding layer of the decoder module and the output data of the second composite embedding module.

[0082] The encoder in the encoder module includes a multi-head attention layer, a linear addition layer, a normalization layer, a fully connected layer, a linear addition layer, and a normalization layer. The encoder in the encoder module includes a masked multi-head attention layer, a linear addition layer, a normalization layer, a multi-head attention layer, a linear addition layer, a normalization layer, a fully connected layer, and a linear addition layer. The connection relationship between each layer is as follows Figure 2 As shown, the encoder and decoder can be one or more. Preferably, there are three encoders and three decoders respectively. The three encoders are connected in sequence, and the three decoders are connected in sequence. The output of the encoder at the top layer is in accordance with Figure 2 The connection relationship between the encoder and decoder is input into the three decoders respectively. The output of the top decoder will be input into the fully connected layer, and finally the predicted temperature result will be output through the output layer.

[0083] Furthermore, the multi-head attention layer and masked multi-head attention layer in the encoder are mainly used to encode data so that the model can automatically learn and focus on important data. Assuming that the data input (masked) to the multi-head attention layer is x, from the above description, we know that the dimension of x is n d . Introduce 3 parameter matrices W q ,W k ,W v , whose dimensions are all n d ×n k . Q, K, and V are calculated as follows:

[0084]

[0085] The attention of the multi-head attention layer is calculated according to the following formula:

[0086]

[0087] The attention value will be n d / n k The results are concatenated to get the output of the multi-head attention layer.

[0088] The purpose of the masked multi-head attention layer in the decoder is to prevent the model from obtaining data after the prediction time during training. Let U be an n k ×n k The lower triangular matrix of , then the masked attention value is calculated as follows:

[0089]

[0090] The normalization layer in the encoder and decoder is used to normalize its input data to ensure that the distribution of the data remains stable. The calculation formula is as follows:

[0091]

[0092] Where μ, σ are the mean and variance of the layer calculated according to the hidden dimension, ∈ = 1e-7 to prevent division by zero errors, and γ, β are trainable parameters.

[0093] The fully connected layer of the temperature prediction model further enhances the representation of features in the form of FFN(x)=max(0,xW1+b1)W2+b2, with n hidden units. f , W1 and W2 are matrices, b1 and b2 are parameters, and x represents the input data of the fully connected layer.

[0094] The output layer of the temperature prediction model is also a fully connected layer, which is mainly used to output the temperature prediction results. The number of hidden layer units is 1. In the actual temperature prediction process, n d =512,n k =64,n f =1024.

[0095] After the data set and temperature prediction model are built, the model data set can be used to train the model. A preferred training method is to divide the model data set into a training set and a validation set, randomly extract 70% of the data as the training set, and the remaining 30% of the data as the validation set. The extraction process is repeated 5 times to obtain

[0096] For each pair of training set and validation set above, training is performed according to the following strategy:

[0097] The back propagation algorithm is used to optimize the above model. The loss function uses the mean square error. The output temperature of the model is set to The actual temperature is y, and the loss function is expressed as:

[0098]

[0099] Specifically, the optimizer used is the Adaptive Moment Estimation (Adam) algorithm, the initial learning rate is set to 0.001, and the training is performed for 50 rounds. At the same time, the early stopping mechanism is used, that is, the model of the most recently trained 5 rounds is used in the validation set D. val If the loss function value no longer decreases, the training is stopped, and the model with the smallest loss function value of the validation set during the training process is taken as the current optimal model.

[0100] For the trained temperature prediction model, the prediction process is to collect k-1 sample data of the steelmaking furnace and the sample offset collection time to be predicted, input the model for prediction, and then gradually predict, and finally obtain the predicted temperature corresponding to the sample. For the temperature predicted by the model, in a preferred embodiment, the present invention also provides a corresponding temperature feedback control strategy, and the specific steps include:

[0101] A temperature deviation value is calculated based on the temperature prediction result and the process recommended temperature, and a heating strategy of the steelmaking furnace is optimized based on the temperature deviation value.

[0102] In this embodiment, the predicted temperature y pred Compare with the process recommended temperature s and set the deviation value sy pred This is fed back to the operator or temperature controller to optimize the heating strategy and ensure the temperature is within the optimal range.

[0103] The present embodiment provides a method for predicting the temperature of a steelmaking furnace. The present invention proposes an improved model suitable for the problem of predicting the temperature of a steelmaking furnace. By comprehensively processing multi-dimensional sensor data, the improved model effectively solves the problems that conventional models cannot effectively process continuous data, the feature embedding representation degree is not high, and the premature merging of features of different categories leads to interference in the internal relationships of embedded features of each category. The accuracy of the model prediction results is improved, thereby improving the temperature control accuracy in the steelmaking process, and reducing energy consumption through the temperature feedback mechanism. The stable furnace temperature control further improves the quality of steel.

[0104] See also Figure 4 Based on the same inventive concept, a steelmaking furnace temperature prediction system proposed in the second embodiment of the present invention includes:

[0105] The data set construction module 10 is used to obtain the historical operation data of the steelmaking furnace, and perform data processing on the historical operation data to obtain a model data set, wherein the model data set includes the historical operation data, the shifted historical operation data and the offset acquisition time;

[0106] A model building module 20, used to train a pre-established temperature prediction model using the model data set to obtain a trained temperature prediction model, wherein the temperature prediction model includes a composite embedding module, an encoder module and a decoder module, wherein the composite embedding module includes a composite convolutional layer, a category embedding layer, an attention layer and a splicing layer;

[0107] The temperature prediction module 30 is used to obtain the real-time operation data of the steelmaking furnace, and input the real-time operation data into the trained temperature prediction model to obtain the temperature prediction result of the steelmaking furnace.

[0108] The technical features and technical effects of the steelmaking furnace temperature prediction system proposed in the embodiment of the present invention are the same as those of the method proposed in the embodiment of the present invention, and are not described in detail here. Each module in the above-mentioned steelmaking furnace temperature prediction system can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0109] In addition, an embodiment of the present invention further provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0110] See also Figure 5 , an internal structure diagram of a computer device in an embodiment, the computer device can specifically be a terminal or a server. The computer device includes a processor, a memory, a network interface, a display and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a steelmaking furnace temperature prediction method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0111] It can be understood by those skilled in the art that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computing device may include more or fewer components than shown in the figure, or combine certain components, or have the same component arrangement.

[0112] In addition, an embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0113] In summary, the embodiment of the present invention proposes a method, system, computer device and storage medium for predicting the temperature of a steelmaking furnace. The method obtains the historical operation data of the steelmaking furnace and processes the historical operation data to obtain a model data set, wherein the model data set includes historical operation data, shifted historical operation data and offset acquisition time; the pre-established temperature prediction model is trained using the model data set to obtain a trained temperature prediction model, wherein the temperature prediction model includes a composite embedding module, an encoder module and a decoder module, wherein the composite embedding module includes a composite convolutional layer, a category embedding layer, an attention layer and a splicing layer; the real-time operation data of the steelmaking furnace is obtained, and the real-time operation data is input into the trained temperature prediction model to obtain the temperature prediction result of the steelmaking furnace. The present invention improves the accuracy of the model prediction result by comprehensively processing the multi-dimensional sensor data, thereby improving the temperature control accuracy in the steelmaking process, and reduces energy consumption through the temperature feedback mechanism, thereby further improving the quality of steel.

[0114] Each embodiment in this specification is described in a progressive manner, and the same or similar parts of each embodiment can be directly referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0115] The above-mentioned embodiments only express several preferred implementation modes of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in the technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be regarded as the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be based on the protection scope of the claims.

Claims

1. A method for predicting temperature of a steelmaking furnace, characterized in that: include: Acquire historical operation data of the steelmaking furnace, and perform data processing on the historical operation data to obtain a model data set, wherein the model data set includes the historical operation data, the shifted historical operation data, and the offset acquisition time; Using the model data set to train a pre-established temperature prediction model to obtain a trained temperature prediction model, wherein the temperature prediction model includes a composite embedding module, an encoder module, and a decoder module, wherein the composite embedding module includes a composite convolutional layer, a category embedding layer, an attention layer, and a splicing layer; The real-time operation data of the steelmaking furnace is obtained, and the real-time operation data is input into a trained temperature prediction model to obtain a temperature prediction result of the steelmaking furnace.

2. The method for predicting temperature of a steelmaking furnace according to claim 1, characterized in that: The step of obtaining the historical operation data of the steelmaking furnace and performing data processing on the historical operation data to obtain the model data set includes: Performing data preprocessing on historical operation data of the steelmaking furnace, wherein the historical operation data includes temperature data, gas composition data, and raw material composition data, wherein the data preprocessing includes missing value processing, data alignment based on dynamic time warping, outlier processing, and data standardization; Shifting the preprocessed historical operation data rightwards according to the time dimension to obtain shifted historical operation data; The time difference between the collection times of the historical operation data is obtained and normalized to obtain the offset collection time.

3. The method for predicting temperature of a steelmaking furnace according to claim 2, characterized in that: The composite embedding module includes a first composite embedding module and a second composite embedding module, the composite convolution layer includes a first composite convolution layer and a second composite convolution layer, the category embedding layer includes a first category embedding layer and a second category embedding layer, the attention layer includes a first attention layer and a second attention layer, and the splicing layer includes a first splicing layer and a second splicing layer; The first composite embedding module includes a plurality of first composite convolutional layers, a plurality of first category embedding layers, a first attention layer and a first concatenation layer, and the first composite convolutional layers and the first category embedding layers correspond one to one; The second composite embedding module includes a plurality of second composite convolutional layers, a plurality of second category embedding layers, a second attention layer, and a second concatenation layer, and the second composite convolutional layers and the second category embedding layers correspond one to one; The number of the first composite convolutional layers is consistent with the number of data categories in the historical running data, the input data of each of the first composite convolutional layers is the historical running data of the corresponding category, and the output data of the first composite convolutional layer is the input data of the corresponding first category embedding layer; The input data of the first attention layer is the output data of each of the first category embedding layers, the input data of the first splicing layer is the output data of the first attention layer and the offset acquisition time, and the output data of the first splicing layer is the input data of the encoder module; The number of the second composite convolutional layers is consistent with the number of data categories in the shifted historical running data, the input data of each second composite convolutional layer is the shifted historical running data of the corresponding category, and the output data of the second composite convolutional layer is the input data of the corresponding second category embedding layer; The input data of the second attention layer is the output data of each second category embedding layer, the input data of the second splicing layer is the output data of the second attention layer and the offset prediction time, and the output data of the second splicing layer is the input data of the decoder module, wherein the offset prediction time is obtained by the previous model prediction result.

4. The method for predicting temperature of a steelmaking furnace according to claim 3, characterized in that: The composite convolutional layer includes several parallel sub-convolutional layers and a sub-convolutional layer connected to each of the convolutional layers. The features of each first category embedding layer and the corresponding second category embedding layer are shared, and the attention layer is composed of a fully connected layer and a normalization layer.

5. The method for predicting temperature of a steelmaking furnace according to claim 4, characterized in that: The composite convolution layer includes three parallel sub-convolution layers, the convolution kernel sizes of each convolution layer are 5*h*32, 3*h*32, and 1*h*32 respectively, and the embedding dimension of each category embedding layer is n d / 3, the dimension of the fully connected layer is n d *3, where h represents feature data of different categories, n d Indicates the dimension of the input data.

6. The method for predicting temperature of a steelmaking furnace according to claim 3, characterized in that: The encoder module includes a first linear addition layer, a first position encoding layer and an encoder, the decoder module includes a second linear addition layer, a second position encoding layer, a decoder, a fully connected layer and an output layer, and the encoder and the decoder are connected; The input data of the first linear addition layer is the output data of the first position encoding layer and the output data of the first composite embedding module, and the output data of the first linear addition layer is the input data of the encoder; The input data of the second linear addition layer is the output data of the second position encoding layer and the output data of the second composite embedding module, and the output data of the second linear addition layer is the input data of the decoder.

7. The method for predicting temperature of a steelmaking furnace according to claim 1, characterized in that: After the step of obtaining the temperature prediction result of the steelmaking furnace, the method further includes: A temperature deviation value is calculated based on the temperature prediction result and the process recommended temperature, and a heating strategy of the steelmaking furnace is optimized based on the temperature deviation value.

8. A steelmaking furnace temperature prediction system, characterized in that: include: A data set construction module is used to obtain historical operation data of the steelmaking furnace, and perform data processing on the historical operation data to obtain a model data set, wherein the model data set includes historical operation data, shifted historical operation data and offset collection time; A model building module, used to train a pre-established temperature prediction model using the model data set to obtain a trained temperature prediction model, wherein the temperature prediction model includes a composite embedding module, an encoder module and a decoder module, wherein the composite embedding module includes a composite convolutional layer, a category embedding layer, an attention layer and a splicing layer; The temperature prediction module is used to obtain the real-time operation data of the steelmaking furnace, and input the real-time operation data into the trained temperature prediction model to obtain the temperature prediction result of the steelmaking furnace.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

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 steps of the method according to any one of claims 1 to 7 are implemented.

Images