Temperature prediction method, system and equipment of hot blast stove vault and medium

Abstract

The invention provides a temperature prediction method, system and device for a hot blast stove vault and a medium, and relates to the technical field of hot blast stove temperature prediction.The method comprises the steps that operation parameter data of a hot blast stove at the current time step is obtained; constructing a temperature characteristic matrix based on the operation parameter data, wherein the temperature characteristic matrix comprises the operation parameter data, and an air-fuel ratio trend characteristic, a vault temperature multi-scale characteristic and an exhaust gas temperature multi-scale characteristic which are determined based on the operation parameter data; extracting multi-interval context vectors corresponding to the temperature feature matrix on the basis of a pre-trained multi-granularity space-time attention mechanism through a pre-trained temperature prediction model, and performing gating weight linear combination on the multi-interval context vectors, and predicting an arch crown temperature prediction value and an exhaust gas temperature prediction value of the hot blast stove in a future time step. Accurate prediction of the temperature of the arch crown and the temperature of the waste gas of the hot blast stove is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of hot blast stove temperature prediction, and in particular to a method, system, equipment and medium for predicting the temperature of a hot blast stove dome. Background Art

[0002] Hot blast furnaces are a type of high-temperature furnace widely used in industries such as metallurgy. Accurately predicting dome and exhaust gas temperatures is crucial during hot blast furnace operation, directly impacting production safety. Traditional temperature control methods rely on sensors to monitor dome and exhaust gas temperatures in real time before controlling and adjusting the hot blast furnace, making them difficult to adapt to rapidly changing production needs.

[0003] In recent years, with the development of artificial intelligence (AI) technology, data-driven temperature prediction methods have begun to be researched and developed. These methods leverage historical operating data to predict future temperatures, enabling more precise temperature control. While these methods have achieved some progress, they still have limitations. For example, using only basic time series features can overlook other feature representations; the mixing of short-term and long-term trend features in the hidden layer can hinder future temperature prediction; and a single-step prediction model can lead to error accumulation. Summary of the Invention

[0004] In view of this, the object of the present invention is to provide a method, system, device and medium for predicting the temperature of a hot blast stove dome, thereby achieving accurate prediction of the hot blast stove dome temperature and the exhaust gas temperature.

[0005] In a first aspect, the present invention provides a method for predicting the temperature of a hot blast stove dome, comprising: Obtaining the operating parameter data of the hot blast furnace at the current time step, the operating parameter data including the dome temperature time series data, the exhaust gas temperature time series data, the air-fuel ratio time series data, the air flow time series data, and the gas flow time series data; Determining an air-fuel ratio trend feature based on air-fuel ratio time series data in the operating parameter data, and determining a dome temperature multi-scale feature and an exhaust gas temperature multi-scale feature based on dome temperature time series data and exhaust gas temperature time series data in the operating parameter data, respectively, and constructing a temperature feature matrix by combining the operating parameter data, the air-fuel ratio trend feature, the dome temperature multi-scale feature, and the exhaust gas temperature multi-scale feature; Through the pre-trained temperature prediction model, the multi-interval context vectors corresponding to the temperature feature matrix are extracted based on the pre-trained multi-granularity spatiotemporal attention mechanism, and the multi-interval context vectors are linearly combined with gated weights to predict the dome temperature prediction value and exhaust gas temperature prediction value of the hot blast furnace in the future time step.

[0006] In one embodiment, an air-fuel ratio trend feature is determined based on air-fuel ratio time series data in the operating parameter data, and a dome temperature multi-scale feature and an exhaust gas temperature multi-scale feature are determined based on dome temperature time series data and exhaust gas temperature time series data in the operating parameter data, respectively. A temperature feature matrix is ​​constructed by combining the operating parameter data, the air-fuel ratio trend feature, the dome temperature multi-scale feature, and the exhaust gas temperature multi-scale feature, including: For the air-fuel ratio at any moment other than the first moment in the air-fuel ratio time series data, based on a preset air-fuel ratio reference value, determine the adjacent air-fuel ratio difference feature between the air-fuel ratio at the current moment and the air-fuel ratio at the previous moment, and construct the air-fuel ratio trend feature based on the adjacent air-fuel ratio difference features at all moments; For any of the dome temperature time series data and the exhaust gas temperature time series data, determine the window ratio corresponding to the time series data at different scales, and determine the mean feature and standard deviation feature corresponding to the time series data according to the window ratio to construct the multi-scale feature corresponding to the time series data. The multi-scale feature is divided into the dome temperature multi-scale feature and the exhaust gas temperature multi-scale feature; The dome temperature time series data, exhaust gas temperature time series data, air-fuel ratio time series data, air flow time series data, gas flow time series data, dome temperature multi-scale features and exhaust gas temperature multi-scale features are spliced ​​to obtain the temperature feature matrix.

[0007] In one embodiment, the temperature prediction model includes an input layer, an improved LSTM layer, and a fully connected output layer. The pre-trained temperature prediction model uses a pre-trained multi-granularity spatiotemporal attention mechanism to extract multi-interval context vectors corresponding to the temperature feature matrix, and performs a gated weighted linear combination on the multi-interval context vectors to predict the dome temperature prediction value and / or exhaust gas temperature prediction value of the hot blast stove at a future time step, including: Through the input layer, receive the temperature feature matrix; Through the improved LSTM layer, the pre-trained multi-granularity spatiotemporal attention mechanism is used to extract the short-term context vector, medium-term context vector, and long-term context vector from the temperature feature matrix. The short-term context vector, medium-term context vector, and long-term context vector are then linearly combined with gated weights to obtain the target context vector. Through the fully connected output layer, based on the target context vector, the dome temperature prediction value and / or exhaust gas temperature prediction value of the hot blast stove at the future time step are predicted.

[0008] In one embodiment, a pre-trained multi-granularity spatiotemporal attention mechanism is used to extract a short-term context vector, a medium-term context vector, and a long-term context vector from the temperature feature matrix, and a gated weighted linear combination of the short-term context vector, the medium-term context vector, and the long-term context vector is performed to obtain a target context vector, including: According to the preset multi-interval ratio, the hidden features of the temperature feature matrix are divided into short-term feature matrix, medium-term feature matrix and long-term feature matrix; Perform regional attention calculations on the short-term feature matrix, medium-term feature matrix, and long-term feature matrix respectively to obtain the short-term context vector, medium-term context vector, and long-term context vector; Determine first weight matrices for the short-term context vector, the medium-term context vector, and the long-term context vector, respectively, to perform gated weighted linear combination on the short-term context vector, the medium-term context vector, and the long-term context vector to obtain a target context vector.

[0009] In one embodiment, regional attention calculation is performed on the short-term feature matrix, the medium-term feature matrix, and the long-term feature matrix to obtain a short-term context vector, a medium-term context vector, and a long-term context vector, including: For any of the short-term feature matrix, medium-term feature matrix, and long-term feature matrix, perform the following operations: Perform nonlinear mapping on the hidden state of the feature matrix to obtain mapping features; A second weight matrix of the hidden state corresponding to the mapped feature is determined based on the mapped feature, and is used to perform weighted aggregation on the hidden state to obtain a context vector corresponding to the feature matrix.

[0010] In one embodiment, after performing regional attention calculations on the short-term feature matrix, the medium-term feature matrix, and the long-term feature matrix to obtain the short-term context vector, the medium-term context vector, and the long-term context vector, the method further includes: Detecting abnormal events based on short-term context vectors; If yes, then the authenticity of the abnormal event is verified based on the similarity between the short-term context vector and the medium-term context vector; In the case that the abnormal event is a true abnormality, the short-term context vector is corrected, and a new long-term context vector is determined based on the corrected short-term context vector and the long-term context vector.

[0011] In one embodiment, the method further comprises: If the dome temperature prediction value and / or the exhaust gas temperature prediction value are higher than the dynamic warning threshold, a temperature alarm message is generated.

[0012] In a second aspect, the present invention further provides a temperature prediction device for a hot blast stove dome, comprising: The data acquisition module is used to obtain the operating parameter data of the hot blast furnace at the current time step, and the operating parameter data includes the time series data of the dome temperature, the time series data of the exhaust gas temperature, the time series data of the air-fuel ratio, the time series data of the air flow and the time series data of the gas flow; a feature construction module for determining an air-fuel ratio trend feature based on the air-fuel ratio time series data in the operating parameter data, and determining a dome temperature multi-scale feature and an exhaust gas temperature multi-scale feature based on the dome temperature time series data and the exhaust gas temperature time series data in the operating parameter data, and constructing a temperature feature matrix by combining the operating parameter data, the air-fuel ratio trend feature, the dome temperature multi-scale feature, and the exhaust gas temperature multi-scale feature; The temperature prediction module is used to extract the multi-interval context vectors corresponding to the temperature feature matrix based on the pre-trained multi-granularity spatiotemporal attention mechanism through the pre-trained temperature prediction model, and perform gated weighted linear combination of the multi-interval context vectors to predict the dome temperature prediction value and / or exhaust gas temperature prediction value of the hot blast furnace in the future time step.

[0013] In a third aspect, the present invention further provides an electronic device comprising a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement any one of the methods provided in the first aspect.

[0014] In a fourth aspect, the present invention further provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement any one of the methods provided in the first aspect.

[0015] The present invention provides a method, system, device and medium for predicting the temperature of a hot blast stove dome. First, the operating parameter data of the hot blast stove at the current time step is obtained, and the operating parameter data includes dome temperature time series data, exhaust gas temperature time series data, air-fuel ratio time series data, air flow time series data and gas flow time series data; then, a temperature feature matrix is ​​constructed based on the operating parameter data, and the temperature feature matrix includes the operating parameter data and air-fuel ratio trend characteristics, dome temperature multi-scale characteristics, and exhaust gas temperature multi-scale characteristics determined based on the operating parameter data; finally, a pre-trained temperature prediction model is used to extract multi-interval context vectors corresponding to the temperature feature matrix based on a pre-trained multi-granularity spatiotemporal attention mechanism, and the multi-interval context vectors are gated weighted linear combination to predict the dome temperature prediction value and exhaust gas temperature prediction value of the hot blast stove at future time steps. The above method constructs a temperature feature matrix including operating parameter data and air-fuel ratio trend characteristics, dome temperature multi-scale characteristics, and exhaust gas temperature multi-scale characteristics determined based on the operating parameter data, and uses the temperature prediction model to perform multi-granularity spatiotemporal attention processing to form a multi-interval context vector. The gating mechanism is then used to perform gated weighted linear combination of the multi-interval context vectors to achieve accurate prediction of the dome temperature and exhaust gas temperature of the hot blast furnace in future time steps, significantly improving the reliability of temperature prediction, improving combustion efficiency, and reducing energy consumption.

[0016] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or understood by practicing the present invention. The purposes and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 A schematic flow chart of a method for predicting the temperature of a hot blast stove dome provided by an embodiment of the present invention; Figure 2 A schematic diagram of a method for predicting the temperature of a hot blast stove dome provided by an embodiment of the present invention; Figure 3A schematic diagram of an LSTM-MGSTA network provided in an embodiment of the present invention; Figure 4 A schematic structural diagram of a temperature prediction device for a hot blast stove dome provided by an embodiment of the present invention; Figure 5 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0021] To address the poor accuracy of predicting the coordinated dome and exhaust gas temperatures during hot blast furnace operation in the metallurgical industry, the present invention provides a hot blast furnace dome temperature prediction method, system, device, and medium, enabling accurate prediction of both dome and exhaust gas temperatures. This embodiment of the invention utilizes a long short-term memory (LSTM) network combined with a multi-granularity spatiotemporal attention mechanism to address the technical bottlenecks of traditional methods, such as delayed response to sudden temperature changes and insufficient capture of long-term trends. The technical solution of this embodiment of the invention is applicable to temperature prediction in blast furnace hot blast furnaces, providing core algorithmic support for equipment safety warnings.

[0022] To facilitate understanding of this embodiment, a method for predicting the temperature of a hot blast stove dome disclosed in an embodiment of the present invention is first described in detail. Figure 1 The flowchart of a method for predicting the temperature of a hot blast stove dome is shown, and the method mainly includes the following steps S102 to S106: Step S102: obtaining the operating parameter data of the hot blast stove at the current time step.

[0023] The operating parameter data includes dome temperature time series data, exhaust gas temperature time series data, air-fuel ratio time series data, air flow time series data and gas flow time series data.

[0024] Step S104, determining the air-fuel ratio trend characteristics based on the air-fuel ratio time series data in the operating parameter data, and determining the dome temperature multi-scale characteristics and the exhaust gas temperature multi-scale characteristics based on the dome temperature time series data and the exhaust gas temperature time series data in the operating parameter data, respectively, and constructing a temperature feature matrix by combining the operating parameter data, the air-fuel ratio trend characteristics, the dome temperature multi-scale characteristics and the exhaust gas temperature multi-scale characteristics.

[0025] Among them, the temperature feature matrix includes operating parameter data and air-fuel ratio trend characteristics, dome temperature multi-scale characteristics, and exhaust gas temperature multi-scale characteristics determined based on the operating parameter data. The dome temperature multi-scale characteristics include the mean characteristics and standard deviation characteristics of the dome temperature time series data at microscale, mesoscale, and macroscale. The exhaust gas temperature multi-scale characteristics include the mean characteristics and standard deviation characteristics of the exhaust gas temperature time series data at microscale, mesoscale, and macroscale.

[0026] In step S106, the pre-trained temperature prediction model is used to extract the multi-interval context vectors corresponding to the temperature feature matrix based on the pre-trained multi-granularity spatiotemporal attention mechanism, and the multi-interval context vectors are linearly combined with gated weights to predict the dome temperature prediction value and the exhaust gas temperature prediction value of the hot blast furnace in the future time step.

[0027] The multi-interval context vectors include short-term context vectors, medium-term context vectors, and long-term context vectors. In one embodiment, an LSTM-MGSTA network is constructed as a temperature prediction model. The input layer receives the temperature feature matrix. A multi-granularity spatiotemporal attention mechanism is added after the improved LSTM layer in the model. The hidden layer vectors of the M time steps output by the improved LSTM layer are divided into three intervals: short-term, medium-term, and long-term. Regional attention calculations are performed on each interval to obtain regional context vectors. The final context is generated by linearly combining the gated weights through a gating mechanism. The final generated context is passed through a fully connected layer to predict the vault temperature and exhaust gas temperature for the next N time steps.

[0028] The temperature prediction method for the dome of a hot blast stove provided in an embodiment of the present invention constructs a temperature feature matrix including operating parameter data and air-fuel ratio trend characteristics, dome temperature multi-scale characteristics, and exhaust gas temperature multi-scale characteristics determined based on the operating parameter data, and uses a temperature prediction model to perform multi-granularity spatiotemporal attention processing to form a multi-interval context vector. The gating mechanism is then combined to perform gated weighted linear combination of the multi-interval context vectors, thereby achieving accurate prediction of the dome temperature and exhaust gas temperature of the hot blast stove in future time steps, significantly improving the reliability of temperature prediction, improving combustion efficiency, and reducing energy consumption.

[0029] For ease of understanding, the embodiments of the present invention provide Figure 2 The diagram shows a method for predicting the temperature of a hot blast furnace dome, which involves two stages: training and prediction. The training stage includes processes such as raw data feature engineering, model construction, training and optimization, and the prediction stage includes processes such as real-time prediction, early warning and alarm.

[0030] The embodiment of the present invention takes the prediction stage as an example to explain the specific implementation process of the temperature prediction method of the hot blast stove dome.

[0031] (1) Obtain the operating parameter data of the hot blast furnace at the current time step, that is, collect the operating parameter data of the hot blast furnace, including gas flow, air flow, dome temperature, and exhaust gas temperature, and calculate the air-fuel ratio according to the following formula : , is the gas flow value, is the air flow value.

[0032] (2) Construct a temperature feature matrix based on the operating parameter data. Construct a feature tensor that contains the following time-step aligned features: (2.1) Basic time series features: The current dome temperature, exhaust gas temperature, air-fuel ratio, air flow rate, and gas flow rate form an M × 5 dimensional matrix.

[0033] (2.2) For the air-fuel ratio at any moment other than the first moment in the air-fuel ratio time series data, based on the preset air-fuel ratio reference value, determine the adjacent air-fuel ratio difference characteristics between the air-fuel ratio at the current moment and the air-fuel ratio at the previous moment, and construct the air-fuel ratio trend characteristics based on the adjacent air-fuel ratio difference characteristics at all moments. Specifically: The adjacent air-fuel ratio difference characteristics are calculated for the 2nd to Mth time steps, and the first time step is padded with zeros to form an M×1 dimensional matrix; the characteristic formula is: ; Where t=2,3,...,M is the tth moment, is the adjacent air-fuel ratio difference characteristic at time t, is the air-fuel ratio at time t, For the tth The air-fuel ratio at the moment, It is the air-fuel ratio for the best combustion efficiency according to the current working conditions of the hot blast stove.

[0034] (2.3) For any of the dome temperature time series data and the exhaust gas temperature time series data, determine the window ratio corresponding to the time series data at different scales, and determine the mean feature and standard deviation feature corresponding to the time series data according to the window ratio to construct the multi-scale feature corresponding to the time series data. The multi-scale feature is divided into the dome temperature multi-scale feature and the exhaust gas temperature multi-scale feature.

[0035] Multi-scale features, also known as multi-scale local binary pattern (MSD-LBP) features, calculate the mean and standard deviation of the micro, meso, and macro scales of the dome temperature and exhaust gas temperature series, and then broadcast them to all time steps to form an M×12 dimensional matrix. Specifically, it includes: Adopt a dynamic window ratio associated with the time series length M to avoid the limitations of fixed windows: ; Among them, the scale , express 、 、 , representing the length of time series at micro, meso and macro scales, respectively.

[0036] The LBP features at the micro, meso, and macro scales are calculated for the dome temperature and exhaust gas temperature, respectively. The formula is as follows: ; in, For scale 1 in the time window The LBP features under express 、 、 , representing the length of time series at micro, medium and macro scales, respectively, represents the time window Middle Temperature values, is a step function, For example, assuming the time window length is The temperature value in this time window is , we can get .

[0037] Extract the mean and standard deviation of the LBP features of each scale, that is, the LBP sequence of each scale Convert to , That is, the mean, That is, the standard deviation. The feature dimensions of the dome temperature and the exhaust gas temperature are M×6, respectively. The MSD-LBP features of the dome temperature and the exhaust gas temperature are concatenated to obtain a 12-dimensional feature vector. The 12-dimensional feature vector is broadcasted to M time steps to obtain an M×12-dimensional feature vector.

[0038] (2.4) The dome temperature time series data, exhaust gas temperature time series data, air-fuel ratio time series data, air flow time series data, gas flow time series data, dome temperature multi-scale features, and exhaust gas temperature multi-scale features are spliced ​​together to obtain a temperature feature matrix. That is, the above features are spliced ​​into an M×18-dimensional feature matrix.

[0039] Furthermore, each feature column can be normalized to its maximum and minimum values, and the normalization parameters can be recorded to facilitate the subsequent denormalization of the prediction results.

[0040] (3) Using the pre-trained temperature prediction model, the pre-trained multi-granularity spatiotemporal attention mechanism is used to extract the multi-interval context vectors corresponding to the temperature feature matrix, and the multi-interval context vectors are linearly combined with gated weights to predict the dome temperature prediction value and / or exhaust gas temperature prediction value of the hot blast furnace in the future time step.

[0041] First, the temperature prediction model is explained. It consists of an input layer, an improved LSTM layer, and a fully connected output layer. Specifically, the input layer receives a tensor of shape [M, 18]. The LSTM layer leverages the time series modeling capabilities of the LSTM network to extract long-term dependencies and deep temporal features from the input features. A spatiotemporal attention mechanism is introduced after the LSTM layer, and a multi-granular spatiotemporal attention mechanism is added after the LSTM layer. This multi-granular spatiotemporal attention mechanism divides the hidden layer vectors of M time steps into three intervals: short-term, medium-term, and long-term. Feature projection, regional attention weight calculation, and context vector generation are performed on each interval. After concatenating the context vectors for each interval, fusion weights are generated using a trainable parameter matrix. Finally, the final context is generated using a linear combination of gated weights. The fully connected layer outputs the features processed by the multi-granular spatiotemporal attention mechanism as the predicted temperature value for the next N minutes.

[0042] The training process of the temperature prediction model is then explained. To improve the model's generalization, a K-fold crossover strategy is used to partition the historical data into five equal-length blocks based on temporal continuity, each containing a complete combustion cycle. Within the solution space, a grid search method is used to search for the optimal hyperparameter combination, including the learning rate and number of hidden units, with the goal of minimizing the MAE on the validation set. The model is then trained based on the resulting optimal hyperparameter combination. During training, a teacher-forcing strategy is used to input true values ​​into the network with a probability p, which decays linearly from 1.0 to 0.5 with the number of training epochs and the loss value. If the MAE on the validation set does not decrease for five consecutive epochs, the current model parameters are saved and training terminates. The trained model structure and weights are saved to the file system for later use. This save includes the model structure, namely the configuration of the LSTM network and the multi-granular spatiotemporal attention mechanism; the weight parameters; and the data normalization parameters used to denormalize the predicted values.

[0043] Based on the above trained temperature prediction model, real-time temperature prediction is achieved, such as Figure 3 The schematic diagram of an LSTM-MGSTA network shown in FIG. includes the following steps: (3.1) Through the input layer, receive the temperature feature matrix.

[0044] (3.2) Through the improved LSTM layer, the pre-trained multi-granularity spatiotemporal attention mechanism is used to extract the short-term context vector, medium-term context vector and long-term context vector from the temperature feature matrix, and the short-term context vector, medium-term context vector and long-term context vector are linearly combined with gated weights to obtain the target context vector.

[0045] In one example, the process of multi-granularity spatiotemporal attention is as follows: (a) Divide the hidden features of the temperature feature matrix into short-term feature matrix, medium-term feature matrix, and long-term feature matrix according to the preset multi-interval ratio. Divide the M time steps into three time intervals according to the ratio: long-term interval: the first 30% of time steps, that is, Tlong = [1, 0.3M]; medium-term interval: the middle 50% of time steps, that is, Tmid = (0.3M, 0.8M]); short-term interval: the last 20% of time steps, that is, Tshort = (0.8M, M].

[0046] (b) Perform regional attention calculations on the short-term feature matrix, the medium-term feature matrix, and the long-term feature matrix to obtain the short-term context vector, the medium-term context vector, and the long-term context vector. For any feature matrix among the short-term feature matrix, the medium-term feature matrix, and the long-term feature matrix, perform the following operations: (b1) Perform nonlinear mapping on the hidden state of the feature matrix to obtain the mapping feature. Perform nonlinear mapping on the hidden state through an independent trainable linear transformation matrix. In one example, calculate ,in, is the mapping feature, For interval 2 corresponds to the projection matrix, is hidden state, 2 represents the time interval Category, 2∈{long,mid,short}, Indicates time interval At any moment within represents the hyperbolic tangent function.

[0047] (b2) Determine the second weight matrix of the hidden state corresponding to the mapped feature, and perform weighted aggregation on the hidden state to obtain the context vector corresponding to the feature matrix. In one example, first, the mapped features are summed and normalized along the channel dimension, and then the second weight matrix of the hidden state is determined according to the following formula: : , then weighted aggregate hidden states to generate context vectors : .in, represents the second weight matrix, indicating the moment Relative to time interval The importance of The first elements, is the total number of elements contained in the mapping feature, is the context vector.

[0048] (c) Cross-scale optimization of short-term context vectors, medium-term context vectors, and long-term context vectors. Specifically: (c1) Detect whether there is an abnormal event based on the short-term context vector.

[0049] Specifically, based on the hidden state of the short-term feature matrix and the second weight matrix, the abnormality index is defined: ;in, is the abnormality index, is the second weight matrix of the short-term feature matrix, is the historical mean of the second weight matrix of the short-term feature matrix, is the hidden state of the short-term feature matrix, is the historical mean of the hidden state of the short-term feature matrix. , is the threshold; if yes, it is determined that an abnormal event exists, otherwise no abnormal event exists.

[0050] (c2) If yes, the authenticity of the abnormal event is verified based on the similarity between the short-term context vector and the medium-term context vector.

[0051] Specifically, the cosine similarity between the short-term context vector and the mid-term context vector is calculated. If the cosine similarity is greater than a preset threshold, it is determined to be a true abnormal event, otherwise it is determined to be a false abnormal event.

[0052] (c3) When the abnormal event is a true abnormality, the short-term context vector is corrected, and a new long-term context vector is determined based on the corrected short-term context vector and the long-term context vector.

[0053] Specifically, if the abnormal event is false, the second weight matrix of the short-term feature matrix is ​​weakened. If the abnormal event is real, the second weight matrix of the short-term feature matrix is ​​strengthened to correct the short-term context vector. The corrected short-term context sequence is fused with the original long-term context vector to obtain a new long-term context vector.

[0054] (d) Determine a first weight matrix for each of the short-term context vector, the medium-term context vector, and the long-term context vector, so as to perform a gated weighted linear combination of the short-term context vector, the medium-term context vector, and the long-term context vector to obtain a target context vector.

[0055] Specifically, the process of the gating mechanism is as follows: concatenate the interval context vectors into , and then generate fusion weights through the trainable parameter matrix ,in is the weight matrix, is the bias term. Finally, the target context vector is generated by linear combination of gated weights , 、 、 are long-term context vectors , mid-term context vector , short-term context vector The weight matrix of .

[0056] (3.3) Through the fully connected output layer, based on the target context vector, the dome temperature prediction value and / or exhaust gas temperature prediction value of the hot blast stove at the future time step are predicted.

[0057] In one embodiment, a temperature alarm is generated if the predicted vault temperature and / or predicted exhaust gas temperature exceed a dynamic warning threshold. In practice, predictions are made based on real-time data, and dynamic warning thresholds are set. Alarms are automatically triggered when the predicted values ​​exceed a safe range.

[0058] In summary, the temperature prediction of the hot blast stove dome proposed in the embodiment of the present invention aims to effectively prevent the occurrence of hot blast stove burn-through accidents by accurately predicting key temperature indicators. The embodiment of the present invention not only solves the problem of insufficient single-scale modeling through multi-scale joint characterization features, but also solves the problems of insufficient model perception granularity and insufficient prediction accuracy through the calculation and gated fusion of attention in each interval, providing important technical support for the operational safety and energy utilization of hot blast stoves in industrial scenarios. Specifically, the embodiment of the present invention achieves the following significant beneficial effects by integrating multi-scale features with the spatiotemporal attention mechanism and combining the time series modeling capabilities of the LSTM network: (1) Significantly improved prediction accuracy: By introducing multi-scale features and spatiotemporal attention mechanisms, the dynamic change trends of the dome temperature and exhaust gas temperature are deeply modeled, significantly improving the prediction accuracy.

[0059] (2) Efficient prediction method: A one-time prediction method is used to directly output the temperature series of the target time period, avoiding the error accumulation problem of rolling prediction, while reducing computational overhead and improving prediction efficiency.

[0060] (3) Robustness and generalization optimization: Through K-fold cross-validation and teacher forced training strategies, the stability and generalization ability of the model under complex working conditions are optimized, enabling it to adapt to interference factors such as fluctuations in the calorific value of different fuels and changes in gas pressure.

[0061] Based on the above embodiment, the present invention provides a device for predicting the temperature of a hot blast stove dome. Figure 4 The structure diagram of a temperature prediction device for a hot blast stove dome is shown, and the device mainly includes the following parts: The data acquisition module 402 is used to obtain the operating parameter data of the hot blast stove at the current time step, and the operating parameter data includes the time series data of the dome temperature, the time series data of the exhaust gas temperature, the time series data of the air-fuel ratio, the time series data of the air flow, and the time series data of the gas flow; a feature construction module 404 for determining an air-fuel ratio trend feature based on the air-fuel ratio time series data in the operating parameter data, and determining a dome temperature multi-scale feature and an exhaust gas temperature multi-scale feature based on the dome temperature time series data and the exhaust gas temperature time series data in the operating parameter data, and constructing a temperature feature matrix by combining the operating parameter data, the air-fuel ratio trend feature, the dome temperature multi-scale feature, and the exhaust gas temperature multi-scale feature; The temperature prediction module 406 is used to extract the multi-interval context vectors corresponding to the temperature feature matrix based on the pre-trained multi-granularity spatiotemporal attention mechanism through the pre-trained temperature prediction model, and perform gated weighted linear combination on the multi-interval context vectors to predict the dome temperature prediction value and / or exhaust gas temperature prediction value of the hot blast furnace in the future time step.

[0062] The temperature prediction device for the dome of a hot blast stove provided in an embodiment of the present invention constructs a temperature feature matrix including operating parameter data and air-fuel ratio trend characteristics, dome temperature multi-scale characteristics, and exhaust gas temperature multi-scale characteristics determined based on the operating parameter data, and uses a temperature prediction model to perform multi-granularity spatiotemporal attention processing to form a multi-interval context vector. Then, combined with a gating mechanism, the multi-interval context vector is subjected to gated weighted linear combination, thereby achieving accurate prediction of the dome temperature and exhaust gas temperature of the hot blast stove in future time steps, significantly improving the reliability of temperature prediction, improving combustion efficiency, and reducing energy consumption.

[0063] In one embodiment, the operating parameter data includes dome temperature time series data, exhaust gas temperature time series data, air-fuel ratio time series data, air flow time series data, and gas flow time series data; the feature construction module 404 is specifically used to: For the air-fuel ratio at any moment other than the first moment in the air-fuel ratio time series data, based on a preset air-fuel ratio reference value, determine the adjacent air-fuel ratio difference feature between the air-fuel ratio at the current moment and the air-fuel ratio at the previous moment, and construct the air-fuel ratio trend feature based on the adjacent air-fuel ratio difference features at all moments; For any of the dome temperature time series data and the exhaust gas temperature time series data, determine the window ratio corresponding to the time series data at different scales, and determine the mean feature and standard deviation feature corresponding to the time series data according to the window ratio to construct the multi-scale feature corresponding to the time series data. The multi-scale feature is divided into the dome temperature multi-scale feature and the exhaust gas temperature multi-scale feature; The dome temperature time series data, exhaust gas temperature time series data, air-fuel ratio time series data, air flow time series data, gas flow time series data, dome temperature multi-scale features and exhaust gas temperature multi-scale features are spliced ​​to obtain the temperature feature matrix.

[0064] In one embodiment, the temperature prediction model includes an input layer, an improved LSTM layer, and a fully connected output layer; the temperature prediction module 406 is specifically used to: Through the input layer, receive the temperature feature matrix; Through the LSTM layer, the pre-trained multi-granularity spatiotemporal attention mechanism is used to extract the short-term context vector, medium-term context vector, and long-term context vector from the temperature feature matrix. The short-term context vector, medium-term context vector, and long-term context vector are then linearly combined with gated weights to obtain the target context vector. Through the fully connected output layer, based on the target context vector, the dome temperature prediction value and / or exhaust gas temperature prediction value of the hot blast stove at the future time step are predicted.

[0065] In one embodiment, the temperature prediction module 406 is specifically configured to: According to the preset multi-interval ratio, the hidden features of the temperature feature matrix are divided into short-term feature matrix, medium-term feature matrix and long-term feature matrix; Perform regional attention calculations on the short-term feature matrix, medium-term feature matrix, and long-term feature matrix respectively to obtain the short-term context vector, medium-term context vector, and long-term context vector; Determine first weight matrices for the short-term context vector, the medium-term context vector, and the long-term context vector, respectively, to perform gated weighted linear combination on the short-term context vector, the medium-term context vector, and the long-term context vector to obtain a target context vector.

[0066] In one embodiment, the temperature prediction module 406 is specifically configured to: For any of the short-term feature matrix, medium-term feature matrix, and long-term feature matrix, perform the following operations: Perform nonlinear mapping on the hidden state of the feature matrix to obtain mapping features; A second weight matrix of the hidden state corresponding to the mapped feature is determined based on the mapped feature, and is used to perform weighted aggregation on the hidden state to obtain a context vector corresponding to the feature matrix.

[0067] In one embodiment, the temperature prediction module 406 is specifically configured to: Detecting abnormal events based on short-term context vectors; If yes, then the authenticity of the abnormal event is verified based on the similarity between the short-term context vector and the medium-term context vector; In the case that the abnormal event is a true abnormality, the short-term context vector is corrected, and a new long-term context vector is determined based on the corrected short-term context vector and the long-term context vector.

[0068] In one embodiment, an alarm module is further included for: If the dome temperature prediction value and / or the exhaust gas temperature prediction value are higher than the dynamic warning threshold, a temperature alarm message is generated.

[0069] The device provided in the embodiment of the present invention has the same implementation principle and technical effects as those in the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference can be made to the corresponding content in the aforementioned method embodiment.

[0070] An embodiment of the present invention provides an electronic device. Specifically, the electronic device includes a processor and a storage device. The storage device stores a computer program, and when the computer program is executed by the processor, it executes the method described in any one of the above-mentioned embodiments.

[0071] Figure 5 This is a structural diagram of an electronic device provided in an embodiment of the present invention. The electronic device 100 includes: a processor 50, a memory 51, a bus 52 and a communication interface 53. The processor 50, the communication interface 53 and the memory 51 are connected via the bus 52; the processor 50 is used to execute an executable module stored in the memory 51, such as a computer program.

[0072] The memory 51 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage. The system network element communicates with at least one other network element via at least one communication interface 53 (which may be wired or wireless), and may utilize the Internet, a wide area network, a local area network, a metropolitan area network, or the like.

[0073] The bus 52 may be an ISA bus, a PCI bus, or an EISA bus. The bus may be divided into an address bus, a data bus, a control bus, and the like. For ease of representation, Figure 5 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0074] Among them, the memory 51 is used to store programs, and the processor 50 executes the program after receiving the execution instruction. The method executed by the device for flow process definition disclosed in any embodiment of the above-mentioned embodiment of the present invention can be applied to the processor 50 or implemented by the processor 50.

[0075] The processor 50 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method may be completed by hardware integrated logic circuits or software instructions in the processor 50. The processor 50 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It may implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in conjunction with the embodiments of the present invention may be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or the like. The storage medium is located in the memory 51 , and the processor 50 reads the information in the memory 51 and completes the steps of the above method in combination with its hardware.

[0076] The computer program product of the readable storage medium provided in the embodiment of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the method described in the previous method embodiment. The specific implementation can be referred to the previous method embodiment and will not be repeated here.

[0077] If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0078] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for predicting the temperature of a hot blast stove dome, characterized in that: include: Obtaining operating parameter data of the hot blast stove at the current time step, wherein the operating parameter data includes dome temperature time series data, exhaust gas temperature time series data, air-fuel ratio time series data, air flow time series data, and gas flow time series data; Determining an air-fuel ratio trend feature based on the air-fuel ratio time series data in the operating parameter data, and determining a dome temperature multi-scale feature and an exhaust gas temperature multi-scale feature based on the dome temperature time series data and the exhaust gas temperature time series data in the operating parameter data, respectively, and constructing a temperature feature matrix by combining the operating parameter data, the air-fuel ratio trend feature, the dome temperature multi-scale feature, and the exhaust gas temperature multi-scale feature; The pre-trained temperature prediction model is used to extract the multi-interval context vectors corresponding to the temperature feature matrix based on the pre-trained multi-granularity spatiotemporal attention mechanism, and the multi-interval context vectors are linearly combined with gated weights to predict the dome temperature prediction value and exhaust gas temperature prediction value of the hot blast stove in the future time step.

2. The method for predicting the temperature of a hot blast stove dome according to claim 1, characterized in that: An air-fuel ratio trend feature is determined based on the air-fuel ratio time series data in the operating parameter data, and a dome temperature multi-scale feature and an exhaust gas temperature multi-scale feature are determined based on the dome temperature time series data and the exhaust gas temperature time series data in the operating parameter data, respectively. A temperature feature matrix is ​​constructed by combining the operating parameter data, the air-fuel ratio trend feature, the dome temperature multi-scale feature, and the exhaust gas temperature multi-scale feature, including: For the air-fuel ratio at any moment other than the first moment in the air-fuel ratio time series data, determining adjacent air-fuel ratio difference features between the air-fuel ratio at the current moment and the air-fuel ratio at the previous moment based on a preset air-fuel ratio reference value, and constructing an air-fuel ratio trend feature based on the adjacent air-fuel ratio difference features at all moments; For any of the dome temperature time series data and the exhaust gas temperature time series data, determine the window ratio corresponding to the time series data at different scales, and determine the mean feature and standard deviation feature corresponding to the time series data according to the window ratio to construct a multi-scale feature corresponding to the time series data, wherein the multi-scale feature is divided into a dome temperature multi-scale feature and an exhaust gas temperature multi-scale feature; The dome temperature time series data, the exhaust gas temperature time series data, the air-fuel ratio time series data, the air flow time series data, the gas flow time series data, the dome temperature multi-scale features and the exhaust gas temperature multi-scale features are spliced ​​to obtain a temperature feature matrix.

3. The method for predicting the temperature of a hot blast stove dome according to claim 1, characterized in that: The temperature prediction model includes an input layer, an improved LSTM layer, and a fully connected output layer. The pre-trained temperature prediction model uses a pre-trained multi-granularity spatiotemporal attention mechanism to extract multi-interval context vectors corresponding to the temperature feature matrix, and performs gated weighted linear combination on the multi-interval context vectors to predict the dome temperature prediction value and / or exhaust gas temperature prediction value of the hot blast stove at a future time step, including: Receiving the temperature feature matrix through the input layer; By using the improved LSTM layer and a pre-trained multi-granularity spatiotemporal attention mechanism, a short-term context vector, a medium-term context vector, and a long-term context vector are extracted from the temperature feature matrix, and a gated weighted linear combination of the short-term context vector, the medium-term context vector, and the long-term context vector is performed to obtain a target context vector; The fully connected output layer is used to predict the dome temperature prediction value and / or the exhaust gas temperature prediction value of the hot blast stove at a future time step based on the target context vector.

4. The method for predicting the temperature of a hot blast stove dome according to claim 3, characterized in that: Utilizing a pre-trained multi-granularity spatiotemporal attention mechanism to extract a short-term context vector, a medium-term context vector, and a long-term context vector from the temperature feature matrix, and performing a gated weighted linear combination of the short-term context vector, the medium-term context vector, and the long-term context vector to obtain a target context vector, including: According to a preset multi-interval ratio, the hidden features of the temperature feature matrix are divided into a short-term feature matrix, a medium-term feature matrix and a long-term feature matrix; Performing regional attention calculation on the short-term feature matrix, the medium-term feature matrix, and the long-term feature matrix respectively to obtain a short-term context vector, a medium-term context vector, and a long-term context vector; Determine a first weight matrix for each of the short-term context vector, the medium-term context vector, and the long-term context vector, so as to perform a gated weighted linear combination of the short-term context vector, the medium-term context vector, and the long-term context vector to obtain a target context vector.

5. The method for predicting the temperature of a hot blast stove dome according to claim 4, characterized in that: Performing regional attention calculation on the short-term feature matrix, the medium-term feature matrix, and the long-term feature matrix respectively to obtain a short-term context vector, a medium-term context vector, and a long-term context vector, including: The following operation is performed on any one of the short-term feature matrix, the medium-term feature matrix, and the long-term feature matrix: Perform nonlinear mapping on the hidden state of the feature matrix to obtain mapping features; A second weight matrix of the hidden state corresponding to the mapping feature is determined based on the mapping feature, so as to perform weighted aggregation on the hidden state to obtain a context vector corresponding to the feature matrix.

6. The method for predicting the temperature of a hot blast stove dome according to claim 4, characterized in that: After performing regional attention calculation on the short-term feature matrix, the medium-term feature matrix, and the long-term feature matrix to obtain a short-term context vector, a medium-term context vector, and a long-term context vector, the method further includes: detecting whether an abnormal event exists based on the short-term context vector; If yes, verifying the authenticity of the abnormal event based on the similarity between the short-term context vector and the medium-term context vector; In the case that the abnormal event is a real abnormality, the short-term context vector is corrected, and a new long-term context vector is determined based on the corrected short-term context vector and the long-term context vector.

7. The method for predicting the temperature of a hot blast stove dome according to claim 1, characterized in that: The method further comprises: If the dome temperature prediction value and / or the exhaust gas temperature prediction value is higher than a dynamic warning threshold, a temperature warning message is generated.

8. A temperature prediction device for a hot blast stove dome, characterized in that: include: A data acquisition module is used to obtain the operating parameter data of the hot blast stove at the current time step, wherein the operating parameter data includes the time series data of the dome temperature, the time series data of the exhaust gas temperature, the time series data of the air-fuel ratio, the time series data of the air flow, and the time series data of the gas flow; a feature construction module, configured to determine an air-fuel ratio trend feature based on the air-fuel ratio time series data in the operating parameter data, and to determine a dome temperature multi-scale feature and an exhaust gas temperature multi-scale feature based on the dome temperature time series data and the exhaust gas temperature time series data in the operating parameter data, respectively, and to construct a temperature feature matrix by combining the operating parameter data, the air-fuel ratio trend feature, the dome temperature multi-scale feature, and the exhaust gas temperature multi-scale feature; The temperature prediction module is used to extract the multi-interval context vectors corresponding to the temperature feature matrix based on the pre-trained multi-granularity spatiotemporal attention mechanism through the pre-trained temperature prediction model, and perform gated weighted linear combination of the multi-interval context vectors to predict the dome temperature prediction value and / or exhaust gas temperature prediction value of the hot blast stove in the future time step.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the method according to any one of claims 1 to 7.

Images