Deep learning-based closed blast furnace fault prediction method and related device

Abstract

The invention discloses a deep learning-based closed blast furnace fault prediction method and a related device, and the method comprises the steps: carrying out the collection and processing of data corresponding to each position based on sensors corresponding to eight monitoring points on a closed blast furnace, and forming eight groups of monitoring data, wherein the monitoring sensors on each monitoring point form a group of monitoring data; a data aggregation node uploads the eight groups of monitoring data to a far-end server, and performs data matrix construction processing on the eight groups of monitoring data on the far-end server to form data matrix information; and the data matrix information is input into a convergent deep learning model for fault prediction processing of the closed blast furnace, and a fault prediction result of the closed blast furnace is obtained. In the embodiment of the invention, the fault of the closed blast furnace can be predicted more accurately, so that the fault of the closed blast furnace can be cleaned in time, and the production efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence, and particularly to a fault prediction method and related device for an ISF (Innovative Shaft Furnace) based on deep learning. Background Art

[0002] In the lead-zinc smelting industry, the ISF (Innovative Shaft Furnace) is a key smelting device, and its operating conditions are directly related to the output and quality of lead and zinc, as well as energy consumption and environmental protection indicators; currently, in the prediction of the operation process of the ISF furnace, there are many technical problems to be solved; currently, in the smelting production process of the ISF furnace, with the progress of the production process, there will be a large amount of sintered ore and coke combustion, and the generation of furnace tumors will occur during the combustion process. As the number of furnace tumors increases, the cross-section of the furnace body gradually becomes narrower, leading to abnormalities in a series of production evaluation indicators such as blast pressure, affecting the lead-zinc production process; the purpose of the present invention is to develop a prediction algorithm for the ISF furnace, by introducing advanced data processing technologies and intelligent algorithms, effectively solving the defects of traditional prediction methods in dealing with ore composition fluctuations, multi-parameter coupling, and the utilization of multi-source data, realizing accurate prediction of the operating state of the ISF furnace, and improving the stability, efficiency, and safety of lead-zinc smelting production. Summary of the Invention

[0003] The purpose of the present invention is to overcome the deficiencies of the prior art. The present invention provides a fault prediction method and related device for an ISF based on deep learning, to achieve more accurate prediction of the faults of the ISF, so as to timely clear the faults of the ISF, thereby improving production efficiency.

[0004] To solve the above technical problems, an embodiment of the present invention provides a fault prediction method for an ISF based on deep learning, which is applied to an ISF. The ISF is sequentially provided with a burden heating zone, a reoxidation zone, a reduction zone, and a slag melting zone from top to bottom, and eight monitoring points are arranged on the ISF, and a corresponding sensor is arranged at each monitoring point. The method includes:

[0005] Based on the sensors corresponding to the eight monitoring points on the ISF, the data corresponding to each position are respectively collected and processed to form eight groups of monitoring data, wherein the monitoring sensors at each monitoring point form a group of monitoring data;

[0006] The data aggregation node uploads the eight groups of monitoring data to the remote server, and performs data matrix construction processing on the eight groups of monitoring data on the remote server to form data matrix information;

[0007] Input the data matrix information into a converged deep learning model for fault prediction processing of a closed blast furnace, and obtain the fault prediction result of the closed blast furnace. The deep learning model is successively composed of a convolutional neural network, a first long short-term neural memory network, an attention mechanism network, and a second long short-term neural memory network.

[0008] Optionally, eight monitoring points are arranged on the closed blast furnace, including:

[0009] Temperature sensors and pressure sensors are respectively arranged on two monitoring points on the burden heating zone of the closed blast furnace for monitoring the top temperature data and top pressure data of the closed blast furnace;

[0010] Zinc vapor concentration sensors and nitrogen dioxide concentration sensors are respectively arranged on two monitoring points on the reoxidation zone of the closed blast furnace for monitoring the zinc vapor concentration data and nitrogen dioxide concentration data of the reoxidation zone;

[0011] First wind pressure sensors and first air volume sensors are respectively arranged on two monitoring points on the secondary air pipeline of the reduction zone of the closed blast furnace for monitoring the wind pressure data and air volume data of the secondary air pipeline;

[0012] Second wind pressure sensors and second air volume sensors are respectively arranged on two monitoring points at 16 tuyeres of the slag melting zone of the closed blast furnace for monitoring the wind pressure data and total air volume data at 16 tuyeres.

[0013] Optionally, based on the sensors corresponding to the eight monitoring points on the closed blast furnace, the data corresponding to each position is respectively collected and processed to form eight groups of monitoring data, including:

[0014] During the operation of the closed blast furnace, control the sensors corresponding to the eight monitoring points on the closed blast furnace to collect and process data at the same frequency, and obtain the collected monitoring data collected by the sensors corresponding to the eight monitoring points;

[0015] The sensors corresponding to the eight monitoring points converge the collected monitoring data to the data convergence node, and the sensors corresponding to the eight monitoring points are communicatively connected to the data convergence node;

[0016] After receiving the collected monitoring data, the data convergence node writes the collected monitoring data into the corresponding empty array to form eight groups of monitoring data, where each empty array is bound to the ID of a sensor and is used to store the collected monitoring data collected by the sensor with the bound ID, and the lengths of the eight empty arrays are the same.

[0017] Optionally, the data convergence node uploads the eight groups of monitoring data to a remote server, including:

[0018] The data aggregation node establishes a communication connection with the remote server, and after the eight groups of monitoring data are obtained at the data aggregation node, the eight groups of monitoring data are uploaded to the remote server based on the communication connection.

[0019] Optionally, on the remote server, data matrix construction processing is performed on the eight groups of monitoring data to form data matrix information, including:

[0020] On the remote server, for each group of detection data in the eight groups of monitoring data, calculation processing of standard deviation, average value, and covariance value is respectively performed to obtain the standard deviation, average value, and covariance value corresponding to each group of detection data;

[0021] Based on the joint probability density function, joint probability density calculation is performed using the average value and covariance value corresponding to each group of detection data to obtain the joint probability density corresponding to each group of detection data;

[0022] Based on the joint probability density and the standard deviation corresponding to each group of detection data, outlier screening processing is performed to obtain the outliers in each group of monitoring data;

[0023] The outliers in each group of monitoring data are removed, and interpolation processing is performed on each group of monitoring data with outliers removed using the difference algorithm to obtain each group of interpolated monitoring data;

[0024] Based on each group of interpolated monitoring data, data matrix construction processing is performed in matrix form to form data matrix information.

[0025] Optionally, the data matrix information is input into a converged deep learning model for fault prediction processing of the closed blast furnace to obtain a fault prediction result of the closed blast furnace, including:

[0026] The data matrix information is input into a converged deep learning model, and downsampling processing of the data matrix information is performed in the convolutional neural network in the converged deep learning model to obtain downsampled feature data;

[0027] The downsampled feature data is input into the first long short-term neural memory network for time series prediction processing to obtain a predicted time series;

[0028] The predicted time series is input into the attention mechanism network, with Key representing the fluctuation characteristics during the generation process of the closed blast furnace, Value representing the data generation range, and Query representing the attention optimization processing of the predicted time series by the attention mechanism network to obtain an optimized predicted time series;

[0029] Input the optimized predicted time series into the second long short-term neural memory network for fault fusion prediction processing of the closed blast furnace, and output the fault prediction result of the closed blast furnace through the fully connected layer of the converged deep learning model to obtain the fault prediction result of the closed blast furnace.

[0030] Optionally, the convolutional neural network in the converged deep learning model performs downsampling processing on the data matrix information to obtain downsampled feature matrix data, including:

[0031] The convolutional neural network in the converged deep learning model performs a convolution operation on the data matrix information using a convolutional layer, and extracts feature data information from the data matrix information based on the convolution operation;

[0032] Perform a downsampling operation on the extracted feature data information in the downsampling layer of the convolutional neural network, and output it through the fully connected layer in the convolutional neural network to form downsampled feature data.

[0033] In addition, an embodiment of the present invention further provides a closed blast furnace fault prediction device based on deep learning, which is applied to a closed blast furnace. The closed blast furnace is sequentially provided with a burden heating zone, a reoxidation zone, a reduction zone, and a slag melting zone from top to bottom, and eight monitoring points are provided on the closed blast furnace, and a corresponding sensor is provided at each monitoring point. The device includes:

[0034] Data acquisition module: used to collect and process the data corresponding to each position based on the sensors corresponding to the eight monitoring points on the closed blast furnace to form eight groups of monitoring data, where the monitoring sensors at each monitoring point form a group of monitoring data;

[0035] Matrix construction module: used for the data aggregation node to upload the eight groups of monitoring data to the remote server, and perform data matrix construction processing on the eight groups of monitoring data on the remote server to form data matrix information;

[0036] Fault prediction module: used to input the data matrix information into the converged deep learning model for fault prediction processing of the closed blast furnace to obtain the fault prediction result of the closed blast furnace. The deep learning model is sequentially composed of a convolutional neural network, a first long short-term neural memory network, an attention mechanism network, and a second long short-term neural memory network.

[0037] In addition, an embodiment of the present invention further provides an electronic device, including a processor and a memory. The processor runs a computer program or code stored in the memory to implement the closed blast furnace fault prediction method as described in any one of the above.

[0038] In addition, an embodiment of the present invention further provides a computer-readable storage medium for storing a computer program or code. When the computer program or code is executed by a processor, it implements the method for predicting the failure of an airtight blast furnace as described in any one of the above.

[0039] In an embodiment of the present invention, sensors corresponding to eight monitoring points on the airtight blast furnace are used to collect and process data corresponding to each position respectively, forming eight groups of monitoring data. Among them, the monitoring sensors on each monitoring point form a group of monitoring data; the data aggregation node uploads the eight groups of monitoring data to a remote server, and constructs a data matrix for the eight groups of monitoring data on the remote server to form data matrix information; inputs the data matrix information into a converged deep learning model for failure prediction processing of the airtight blast furnace to obtain a failure prediction result of the airtight blast furnace; realizes more accurate prediction of the failure of the airtight blast furnace, so that the failure of the airtight blast furnace can be cleared in time, thereby improving production efficiency and maintaining the health of the furnace body. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0041] Figure 1 is a schematic flow chart of the method for predicting the failure of an airtight blast furnace based on deep learning in an embodiment of the present invention;

[0042] Figure 2 is a schematic flow chart of the method for predicting the failure of an airtight blast furnace based on deep learning in another embodiment of the present invention;

[0043] Figure 3 is a schematic structural composition diagram of the device for predicting the failure of an airtight blast furnace based on deep learning in an embodiment of the present invention;

[0044] Figure 4 is a schematic structural composition diagram of an electronic device in an embodiment of the present invention;

[0045] Figure 5 is a schematic structural composition diagram of an airtight blast furnace in an embodiment of the present invention;

[0046] Figure 6 is a schematic structural composition diagram of a deep learning model in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0048] Example 1, please refer to Figure 1 , Figure 1 which is a schematic flow chart of a fault prediction method for a closed blast furnace based on deep learning in the embodiments of the present invention.

[0049] As Figure 1 shown, a fault prediction method for a closed blast furnace based on deep learning is applied to a closed blast furnace. The closed blast furnace is sequentially provided with a burden heating zone, a reoxidation zone, a reduction zone, and a slag melting zone from top to bottom. And eight monitoring points are arranged on the closed blast furnace, and a corresponding sensor is arranged at each monitoring point. The method includes:

[0050] S101: Based on the sensors corresponding to the eight monitoring points on the closed blast furnace, the data corresponding to each position are respectively collected and processed to form eight groups of monitoring data, where the monitoring sensors at each monitoring point form a group of monitoring data;

[0051] In the specific implementation process of the present invention, the eight monitoring points arranged on the closed blast furnace include: temperature sensors and pressure sensors are respectively arranged on two monitoring points on the burden heating zone of the closed blast furnace for monitoring the top temperature data and top pressure data of the closed blast furnace; zinc vapor concentration sensors and nitrogen dioxide concentration sensors are respectively arranged on two monitoring points on the reoxidation zone of the closed blast furnace for monitoring the zinc vapor concentration data and nitrogen dioxide concentration data of the reoxidation zone; a first wind pressure sensor and a first air volume sensor are respectively arranged on two monitoring points on the secondary air duct of the reduction zone of the closed blast furnace for monitoring the wind pressure data and air volume data of the secondary air duct; a second wind pressure sensor and a second air volume sensor are respectively arranged on two monitoring points at 16 tuyeres of the slag melting zone of the closed blast furnace for monitoring the wind pressure data and total air volume data at 16 tuyeres.

[0052] Further, sensors corresponding to eight monitoring points on the closed blast furnace respectively collect and process data corresponding to each position, forming eight groups of monitoring data, including: during the operation of the closed blast furnace, control the sensors corresponding to the eight monitoring points on the closed blast furnace to collect and process data at the same frequency, and obtain the collected monitoring data collected by the sensors corresponding to the eight monitoring points; the sensors corresponding to the eight monitoring points converge the collected monitoring data to the data convergence node, where the sensors corresponding to the eight monitoring points are communicatively connected to the data convergence node; after receiving the collected monitoring data, the data convergence node writes the collected monitoring data into the corresponding empty array, forming eight groups of monitoring data, where each empty array is bound to the ID of a sensor and is used to store the collected monitoring data collected by the sensor with the bound ID, and the lengths of the eight empty arrays are the same.

[0053] Specifically, as Figure 5 shown, the general structure of the closed blast furnace is successively provided with a burden heating zone, a reoxidation zone, a reduction zone, and a slag melting zone from top to bottom; and eight monitoring points are successively arranged, and a corresponding sensor is arranged at each monitoring point for collecting relevant data of the corresponding monitoring point.

[0054] That is, a temperature sensor and a pressure sensor are respectively arranged at two monitoring points on the burden heating zone of the closed blast furnace for monitoring and collecting the top temperature data and top pressure data of the closed blast furnace; a zinc vapor concentration sensor and a nitrogen dioxide concentration sensor are respectively arranged at two monitoring points on the reoxidation zone of the closed blast furnace for monitoring and collecting the zinc vapor concentration data and nitrogen dioxide concentration data of the reoxidation zone; a first wind pressure sensor and a first air volume sensor are respectively arranged at two monitoring points on the secondary air duct in the reduction zone of the closed blast furnace for monitoring and collecting the wind pressure data and air volume data of the secondary air duct; a second wind pressure sensor and a second air volume sensor are respectively arranged at two monitoring points at 16 tuyeres in the slag melting zone of the closed blast furnace for monitoring and collecting the wind pressure data and total air volume data at 16 tuyeres; in this way, corresponding sensors can be arranged at eight monitoring points, and the required data information can be collected by using the sensors arranged at the eight monitoring points.

[0055] During the operation of the closed blast furnace, sensors installed at eight monitoring points will perform data acquisition operations. At this time, it is necessary to control the sensors corresponding to the eight monitoring points on the closed blast furnace to perform data acquisition and processing at the same frequency, so as to obtain the acquisition monitoring data collected by the sensors corresponding to the eight monitoring points. The specific implementation method can be achieved by synchronizing the acquisition clock frequencies of the sensors corresponding to the eight monitoring points and performing data acquisition operations at the same acquisition frequency. It can also be achieved by the upper device communicating with the data aggregation node and controlling the sensors corresponding to the eight monitoring points through the data aggregation node to achieve synchronous data acquisition.

[0056] After obtaining the acquisition monitoring data collected by the sensors corresponding to the eight monitoring points, it is necessary to converge the acquisition monitoring data collected by the sensors corresponding to the eight monitoring points to the data aggregation node. The sensors corresponding to the eight monitoring points are communicatively connected to the data aggregation node. Specifically, the communication connection can be established wirelessly or wired. The data aggregation node is also installed outside the closed blast furnace. And after the data aggregation node receives the acquisition monitoring data, at this time, it is necessary to write the acquisition monitoring data into the corresponding empty array to form eight groups of monitoring data. The empty array is set in advance. And each empty array is bound to the ID of a sensor, which is used to store the acquisition monitoring data collected by the sensor with the bound ID. The lengths of the eight empty arrays are the same.

[0057] S102: The data aggregation node uploads the eight groups of monitoring data to the remote server and performs data matrix construction processing on the eight groups of monitoring data on the remote server to form data matrix information.

[0058] In the specific implementation process of the present invention, the data aggregation node uploading the eight groups of monitoring data to the remote server includes: the data aggregation node establishing a communication connection with the remote server, and after the data aggregation node obtains the eight groups of monitoring data, uploading the eight groups of monitoring data to the remote server based on the communication connection.

[0059] Further, data matrix construction processing is performed on the eight groups of monitoring data on the remote server to form data matrix information, including: calculating the standard deviation value, average value, and covariance value for each group of detection data in the eight groups of monitoring data on the remote server to obtain the standard deviation value, average value, and covariance value corresponding to each group of detection data; performing joint probability density calculation using the average value and covariance value corresponding to each group of detection data based on the joint probability density function to obtain the joint probability density corresponding to each group of detection data; performing outlier screening processing based on the joint probability density and the standard deviation value corresponding to each group of detection data to obtain the outliers in each group of monitoring data; removing the outliers in each group of monitoring data and performing interpolation processing on each group of monitoring data with outliers removed using the difference algorithm to obtain each group of interpolated monitoring data; performing data matrix construction processing in matrix form based on each group of interpolated monitoring data to form data matrix information.

[0060] Specifically, after the eight groups of monitoring data are formed at the data aggregation node, these eight groups of monitoring data need to be uploaded to the remote server, and some subsequent processing will be executed on the remote server; at this time, a communication connection is established between the data aggregation node and the remote server through a gateway. After the communication connection is established, after the data aggregation node obtains the eight groups of monitoring data, the eight groups of monitoring data are uploaded to the remote server through the communication connection.

[0061] After the remote server receives the eight groups of monitoring data, data matrix construction is required, and before data matrix construction, corresponding processing needs to be performed on each group of data in the eight groups of monitoring data. Specifically, the outliers in each group of data are removed, and then interpolation work is performed on the removed outliers.

[0062] That is, the standard deviation value, average value, and covariance value are respectively calculated for each group of detection data in the eight groups of monitoring data on the remote server. This calculation needs to be implemented by calling the standard deviation formula, average value formula, and covariance formula, so as to obtain the standard deviation value, average value, and covariance value corresponding to each group of detection data; then the joint probability density function is called on the remote server, and the joint probability density calculation is performed using the average value and covariance value corresponding to each group of detection data through the joint probability density function to obtain the joint probability density corresponding to each data in each group of detection data; then outlier screening processing is performed through the joint probability density and the standard deviation value corresponding to each group of detection data to obtain the outliers in each group of monitoring data; after obtaining the outliers, the outliers in each group of monitoring data are removed, and interpolation processing is performed on each group of monitoring data with outliers removed using the difference algorithm to obtain each group of interpolated monitoring data; finally, data matrix construction processing is performed in matrix form to form data matrix information.

[0063] Among them, the joint probability density function is as follows:

[0064]

[0065] Among them, x i is the first i-th data in each group of detection data; P(x i ) is the joint probability density corresponding to the first i-th data in each group of detection data; Σ is the covariance value; μ is the average value; n is the number of data in each group of detection data.

[0066] When screening for outliers, it is necessary to screen and process outliers by taking the data corresponding to the joint probability density less than the preset value and / or greater than three times the standard deviation as outliers.

[0067] S103: Input the data matrix information into the converged deep learning model for fault prediction processing of the closed blast furnace, and obtain the fault prediction result of the closed blast furnace. The deep learning model is successively composed of a convolutional neural network, a first long short-term neural memory network, an attention mechanism network, and a second long short-term neural memory network.

[0068] In the specific implementation process of the present invention, the step of inputting the data matrix information into the converged deep learning model for fault prediction processing of the closed blast furnace to obtain the fault prediction result of the closed blast furnace includes: inputting the data matrix information into the converged deep learning model, and performing downsampling processing on the data matrix information in the convolutional neural network of the converged deep learning model to obtain downsampled feature data; inputting the downsampled feature data into the first long short-term neural memory network for time series prediction processing to obtain a predicted time series; inputting the predicted time series into the attention mechanism network, using Key to represent the fluctuation characteristics of the closed blast furnace during the generation process, using Value to represent the data generation range, and using Query to perform attention optimization processing on the predicted time series by the attention mechanism network to obtain an optimized predicted time series; inputting the optimized predicted time series into the second long short-term neural memory network for fault fusion prediction processing of the closed blast furnace, and outputting the fault prediction result of the closed blast furnace through the fully connected layer of the converged deep learning model to obtain the fault prediction result of the closed blast furnace.

[0069] Further, the convolutional neural network in the converged deep learning model performs downsampling processing on the data matrix information to obtain downsampled feature matrix data, including: the convolutional neural network in the converged deep learning model uses a convolutional layer to perform a convolution operation on the data matrix information, and extracts feature data information in the data matrix information based on the convolution operation; the extracted feature data information is subjected to a downsampling operation in the downsampling layer in the convolutional neural network, and is output through the fully connected layer in the convolutional neural network to form downsampled feature data.

[0070] Specifically, as Figure 6 shown, the deep learning model is successively composed of a convolutional neural network, a first long short-term neural memory network, an attention mechanism network, and a second long short-term neural memory network; and the end of the second long short-term neural network is connected to the fully connected layer, and the final fault prediction result of the closed blast furnace is output through the fully connected layer.

[0071] Among them, the training data of the deep learning model mainly comes from the measured data of the Third Smelter of Baiyin Nonferrous Group Co., Ltd.; these data are used to construct the training data, which is also in the form of a data matrix and is input into the deep learning model for deep learning training, and finally a converged deep learning model is formed.

[0072] When performing fault prediction on the closed blast furnace, it is necessary to input the formed data matrix information into the converged deep learning model for fault prediction; at this time, in the converged deep learning model, the data matrix information first passes through the convolutional neural network. In the convolutional neural network, first, the data matrix information is subjected to data feature extraction processing to extract the corresponding feature data information. Then, in order to prevent overfitting and improve the prediction accuracy of the subsequent time series, the feature data information is subjected to corresponding downsampling operations through the downsampling module in the convolutional neural network to form downsampled feature data; then, the downsampled feature data is input into the first long short-term neural memory network through the fully connected layer in the convolutional neural network, and the first long short-term neural memory network is used to build the time series using the downsampled feature data, so as to complete the prediction on the time series and obtain the predicted time series; among them, after the downsampled feature data enters the first long short-term neural memory network, the downsampled data features will flow in the form of cell states, and are constructed into time series information through the forget gate, input gate, and output gate, so as to obtain the predicted time series.

[0073] In order to be able to focus on the corresponding partial feature data content in the prediction time, it is necessary to introduce an attention mechanism network; and input the prediction time series into the attention mechanism network, use Key to represent the fluctuation characteristics during the generation process of the closed blast furnace, use Value to represent the data generation range, and use Query to represent the attention optimization processing of the prediction time series by the attention mechanism network, so as to obtain an optimized prediction time series;

[0074] Due to the temporality and strong non-linearity of the data generated in the industry, it is necessary to input the optimized prediction time series into the second long short-term neural memory network again for the fault fusion prediction processing of the closed blast furnace. Then, after obtaining the fault prediction result of the closed blast furnace, output the fault prediction result of the closed blast furnace through the fully connected layer of the convergent deep learning model, and finally obtain the fault prediction result of the closed blast furnace; the obtained fault prediction result will be more in line with the time curve of the closed blast furnace failing during actual production.

[0075] In the embodiment of the present invention, sensors corresponding to eight monitoring points on the closed blast furnace are used to collect and process the data corresponding to each position respectively, forming eight groups of monitoring data, where the monitoring sensors on each monitoring point form a group of monitoring data; the data aggregation node uploads the eight groups of monitoring data to the remote server, and constructs a data matrix for the eight groups of monitoring data on the remote server to form data matrix information; input the data matrix information into the convergent deep learning model for the fault prediction processing of the closed blast furnace, and obtain the fault prediction result of the closed blast furnace; realize more accurate prediction of the faults of the closed blast furnace, so as to timely clear the faults of the closed blast furnace, thereby improving production efficiency and maintaining the health of the furnace body.

[0076] Embodiment 2, please refer to Figure 2 , Figure 2 which is a schematic flowchart of a method for predicting faults of a closed blast furnace based on deep learning in another embodiment of the present invention.

[0077] As Figure 2 shown, a method for predicting faults of a closed blast furnace based on deep learning is applied to a closed blast furnace. The closed blast furnace is sequentially provided with a burden heating zone, a reoxidation zone, a reduction zone, and a slag melting zone from top to bottom, and eight monitoring points are provided on the closed blast furnace, and a corresponding sensor is provided at each monitoring point. The method includes:

[0078] S201: Based on the sensors corresponding to the eight monitoring points on the closed blast furnace, collect and process the data corresponding to each position respectively, forming eight groups of monitoring data, where the monitoring sensors on each monitoring point form a group of monitoring data;

[0079] S202: The data aggregation node uploads the eight groups of monitoring data to the remote server, and constructs a data matrix for the eight groups of monitoring data on the remote server to form data matrix information;

[0080] S203: Input the data matrix information into the converged deep learning model, and perform downsampling processing on the data matrix information in the convolutional neural network in the converged deep learning model to obtain downsampled feature data;

[0081] S204: Input the downsampled feature data into the first long short-term neural memory network for time series prediction processing to obtain a predicted time series;

[0082] S205: Input the predicted time series into the attention mechanism network, use Key to represent the fluctuation characteristics during the generation process of the closed blast furnace, use Value to represent the data generation range, and use Query to perform attention optimization processing on the predicted time series by the attention mechanism network to obtain an optimized predicted time series;

[0083] S206: Input the optimized predicted time series into the second long short-term neural memory network for fault fusion prediction processing of the closed blast furnace, and output the fault prediction result of the closed blast furnace through the fully connected layer of the converged deep learning model to obtain the fault prediction result of the closed blast furnace.

[0084] For the specific implementation manner of the second embodiment, reference may be made to the above embodiment, which will not be elaborated herein.

[0085] Embodiment 3, please refer to Figure 3 , Figure 3 which is a schematic structural composition diagram of a closed blast furnace fault prediction device based on deep learning in the embodiments of the present invention.

[0086] As Figure 3 shown, a closed blast furnace fault prediction device based on deep learning is applied to a closed blast furnace. The closed blast furnace is sequentially provided with a burden heating zone, a reoxidation zone, a reduction zone, and a slag melting zone from top to bottom, and eight monitoring points are provided on the closed blast furnace. A corresponding sensor is provided at each monitoring point. The device includes:

[0087] A data acquisition module 301: configured to collect and process the data corresponding to each position based on the sensors corresponding to the eight monitoring points on the closed blast furnace to form eight groups of monitoring data, where the monitoring sensors at each monitoring point form a group of monitoring data;

[0088] In the specific implementation process of the present invention, eight monitoring points are provided on the closed blast furnace, including: temperature sensors and pressure sensors are respectively arranged on two monitoring points on the charge heating zone of the closed blast furnace for monitoring the furnace top temperature data and furnace top pressure data of the closed blast furnace; zinc vapor concentration sensors and nitrogen dioxide concentration sensors are respectively arranged on two monitoring points on the reoxidation zone of the closed blast furnace for monitoring the zinc vapor concentration data and nitrogen dioxide concentration data of the reoxidation zone; a first wind pressure sensor and a first air volume sensor are respectively arranged on two monitoring points on the secondary air pipeline in the reduction zone of the closed blast furnace for monitoring the wind pressure data and air volume data of the secondary air pipeline; a second wind pressure sensor and a second air volume sensor are respectively arranged on two monitoring points at 16 tuyeres in the slag melting zone of the closed blast furnace for monitoring the wind pressure data and total air volume data at 16 tuyeres.

[0089] Further, the sensors corresponding to the eight monitoring points on the closed blast furnace are respectively used to collect and process the data corresponding to each position, forming eight groups of monitoring data, including: during the operation of the closed blast furnace, controlling the sensors corresponding to the eight monitoring points on the closed blast furnace to collect and process data at the same frequency to obtain the collected monitoring data collected by the sensors corresponding to the eight monitoring points; the sensors corresponding to the eight monitoring points converge the collected monitoring data to the data convergence node, where the sensors corresponding to the eight monitoring points are communicatively connected to the data convergence node; after receiving the collected monitoring data, the data convergence node writes the collected monitoring data into the corresponding empty array to form eight groups of monitoring data, where each empty array is bound to the ID of a sensor and is used to store the collected monitoring data collected by the sensor with the bound ID, and the lengths of the eight empty arrays are the same.

[0090] Specifically, as Figure 5 shown, the general structure of the closed blast furnace is successively provided with a charge heating zone, a reoxidation zone, a reduction zone, and a slag melting zone from top to bottom; and eight monitoring points are successively arranged, and a corresponding sensor is arranged at each monitoring point for collecting relevant data of the corresponding monitoring point.

[0091] That is, temperature sensors and pressure sensors are respectively arranged at two monitoring points on the charge heating zone of the closed blast furnace to monitor and collect the top temperature data and top pressure data of the closed blast furnace; zinc vapor concentration sensors and nitrogen dioxide concentration sensors are respectively arranged at two monitoring points on the reoxidation zone of the closed blast furnace to monitor and collect the zinc vapor concentration data and nitrogen dioxide concentration data of the reoxidation zone; a first wind pressure sensor and a first air volume sensor are respectively arranged at two monitoring points on the secondary air pipeline of the reduction zone of the closed blast furnace to monitor and collect the wind pressure data and air volume data of the secondary air pipeline; second wind pressure sensors and second air volume sensors are respectively arranged at two monitoring points at 16 tuyeres of the slag melting zone of the closed blast furnace to monitor and collect the wind pressure data and total air volume data at 16 tuyeres; in this way, corresponding sensors can be arranged at eight monitoring points, and the sensors arranged at the eight monitoring points can be used to collect corresponding required data information.

[0092] During the operation of the closed blast furnace, the sensors arranged at the eight monitoring points will perform data collection operations; at this time, it is necessary to control the sensors corresponding to the eight monitoring points arranged on the closed blast furnace to perform data collection and processing at the same frequency, so as to obtain the collection monitoring data collected by the sensors corresponding to the eight monitoring points; the specific implementation method can be achieved by synchronizing the collection clock frequencies of the sensors corresponding to the eight monitoring points and performing data collection operations at the same collection frequency; it can also be achieved by the upper device communicating with the data aggregation node and controlling the sensors corresponding to the eight monitoring points through the data aggregation node to achieve synchronous data collection.

[0093] After obtaining the collection monitoring data collected by the sensors corresponding to the eight monitoring points, it is necessary to converge the collection monitoring data collected by the sensors corresponding to the eight monitoring points to the data aggregation node, where the sensors corresponding to the eight monitoring points are communicatively connected to the data aggregation node, and specifically, a communication connection can be established in a wireless manner or a wired manner, and the data aggregation node is also arranged outside the closed blast furnace; and after the data aggregation node receives the collection monitoring data, at this time, it is necessary to write the collection monitoring data into the corresponding empty array to form eight groups of monitoring data, where the empty array is set in advance; and each empty array is bound to the ID of a sensor to store the collection monitoring data collected by the sensor with the bound ID, and the lengths of the eight empty arrays are the same.

[0094] Matrix construction module 302: It is used for the data aggregation node to upload the eight groups of monitoring data to a remote server, and perform data matrix construction processing on the eight groups of monitoring data on the remote server to form data matrix information;

[0095] In the specific implementation process of the present invention, the data aggregation node uploads the eight groups of monitoring data to the remote server, including: the data aggregation node establishes a communication connection with the remote server, and after the data aggregation node obtains the eight groups of monitoring data, uploads the eight groups of monitoring data to the remote server based on the communication connection.

[0096] Further, on the remote server, data matrix construction processing is performed on the eight groups of monitoring data to form data matrix information, including: calculating the standard deviation, average value, and covariance value for each group of detection data in the eight groups of monitoring data on the remote server to obtain the standard deviation, average value, and covariance value corresponding to each group of detection data; performing joint probability density calculation using the average value and covariance value corresponding to each group of detection data based on the joint probability density function to obtain the joint probability density corresponding to each group of detection data; performing outlier screening processing based on the joint probability density and the standard deviation corresponding to each group of detection data to obtain the outliers in each group of monitoring data; removing the outliers in each group of monitoring data, and performing interpolation processing on each group of monitoring data with outliers removed using the difference algorithm to obtain each group of interpolated monitoring data; performing data matrix construction processing on each group of interpolated monitoring data in matrix form to form data matrix information.

[0097] Specifically, after the eight groups of monitoring data are formed at the data aggregation node, the eight groups of monitoring data need to be uploaded to the remote server, and some subsequent processing will be performed on the remote server; at this time, a communication connection is established between the data aggregation node and the remote server through a gateway. After the communication connection is established, after the data aggregation node obtains the eight groups of monitoring data, the eight groups of monitoring data are uploaded to the remote server through the communication connection.

[0098] After the remote server receives the eight groups of monitoring data, it is necessary to construct a data matrix, and before constructing the data matrix, corresponding processing needs to be performed on each group of data in the eight groups of monitoring data. Specifically, the outliers in each group of data are removed, and then interpolation work is performed on the removed outliers.

[0099] That is, on the remote server, the standard deviation value, average value, and covariance value of each set of eight groups of monitoring data are calculated respectively. This calculation needs to call the standard deviation formula, average value formula, and covariance formula to obtain the standard deviation value, average value, and covariance value corresponding to each set of monitoring data. Then, on the remote server, the joint probability density function is called, and the joint probability density calculation is performed using the average value and covariance value corresponding to each set of monitoring data through the joint probability density function to obtain the joint probability density corresponding to each data in each set of monitoring data. Then, outlier screening is performed through the joint probability density and the standard deviation value corresponding to each set of monitoring data to obtain the outliers in each set of monitoring data. After obtaining the outliers, the outliers in each set of monitoring data are removed, and the interpolation algorithm is used to interpolate each set of monitoring data after removing the outliers to obtain each set of interpolated monitoring data. Finally, data matrix construction processing is performed in the form of a matrix to form data matrix information.

[0100] Among them, the joint probability density function is as follows:

[0101]

[0102] Among them, x i is the i-th data in each set of monitoring data; P(x i ) is the joint probability density corresponding to the i-th data in each set of monitoring data; Σ is the covariance value; μ is the average value; n is the number of data in each set of monitoring data.

[0103] When screening whether it is an outlier, the corresponding data with a joint probability density less than the preset value and / or greater than three times the standard deviation needs to be used as an outlier for outlier screening.

[0104] The fault prediction module 303: is used to input the data matrix information into the converged deep learning model for fault prediction processing of the closed blast furnace, and obtain the fault prediction result of the closed blast furnace. The deep learning model is successively composed of a convolutional neural network, a first long short-term memory network, an attention mechanism network, and a second long short-term memory network.

[0105] In the specific implementation process of the present invention, inputting the data matrix information into a converged deep learning model for fault prediction processing of a closed blast furnace to obtain a fault prediction result of the closed blast furnace includes: inputting the data matrix information into the converged deep learning model, performing downsampling processing on the data matrix information in the convolutional neural network in the converged deep learning model to obtain downsampled feature data; inputting the downsampled feature data into the first long short-term neural memory network for time series prediction processing to obtain a predicted time series; inputting the predicted time series into the attention mechanism network, using Key to represent the fluctuation characteristics of the closed blast furnace during the generation process, using Value to represent the data generation range, and using Query to perform attention optimization processing on the predicted time series by the attention mechanism network to obtain an optimized predicted time series; inputting the optimized predicted time series into the second long short-term neural memory network for fault fusion prediction processing of the closed blast furnace, and outputting the fault prediction result of the closed blast furnace through the fully connected layer of the converged deep learning model to obtain the fault prediction result of the closed blast furnace.

[0106] Further, performing downsampling processing on the data matrix information in the convolutional neural network in the converged deep learning model to obtain downsampled feature matrix data includes: using the convolutional layer in the convolutional neural network in the converged deep learning model to perform a convolution operation on the data matrix information, and extracting feature data information in the data matrix information based on the convolution operation; performing a downsampling operation on the extracted feature data information in the downsampling layer in the convolutional neural network, and outputting through the fully connected layer in the convolutional neural network to form downsampled feature data.

[0107] Specifically, as Figure 6 shown, the deep learning model is sequentially composed of a convolutional neural network, a first long short-term neural memory network, an attention mechanism network, and a second long short-term neural memory network; and the end of the second long short-term neural network is connected to the fully connected layer, and the final fault prediction result of the closed blast furnace is output through the fully connected layer.

[0108] Among them, the training data of the deep learning model mainly comes from the measured data of the Third Smelter of Baiyin Nonferrous Group Co., Ltd.; using these data to construct training data, which is also in the form of a data matrix, input into the deep learning model for deep learning training, and finally form a converged deep learning model.

[0109] When conducting fault prediction for a closed - blast furnace, it is necessary to input the formed data matrix information into a converged deep - learning model for fault prediction. At this time, in the converged deep - learning model, the data matrix information first passes through a convolutional neural network. In the convolutional neural network, first, the data matrix information is processed to extract data features, thereby extracting corresponding feature data information. Then, in order to prevent overfitting and improve the prediction accuracy of subsequent time series, the feature data information is subjected to corresponding down - sampling operations through the down - sampling module in the convolutional neural network to form down - sampled feature data. Then, the down - sampled feature data is input into the first long - short - term neural memory network through the fully - connected layer in the convolutional neural network. The first long - short - term neural memory network uses the down - sampled feature data to construct a time series, thereby completing the prediction on the time series and obtaining a predicted time series. Among them, after the down - sampled feature data enters the first long - short - term neural memory network, the down - sampled data features will flow in the form of cell states and are constructed into time - series information through forget gates, input gates, and output gates, thus obtaining a predicted time series.

[0110] In order to be able to focus on the corresponding partial feature data content in the predicted time, it is necessary to introduce an attention - mechanism network. And the predicted time series is input into the attention - mechanism network. Using Key to represent the fluctuation characteristics of the closed - blast furnace during the generation process, Value to represent the data generation range, and Query to represent the attention optimization processing of the predicted time series by the attention - mechanism network, thereby obtaining an optimized predicted time series.

[0111] Due to the temporal and strong non - linearity of the data generated in industry, it is necessary to input the optimized predicted time series into the second long - short - term neural memory network again for fault fusion prediction processing of the closed - blast furnace. Then, after obtaining the fault prediction result of the closed - blast furnace, the fault prediction result of the closed - blast furnace is output through the fully - connected layer of the converged deep - learning model, and finally, the fault prediction result of the closed - blast furnace is obtained. The fault prediction result obtained in this way will be more in line with the time curve of the closed - blast furnace failure during actual production.

[0112] In an embodiment of the present invention, sensors corresponding to eight monitoring points on the closed blast furnace are used to collect and process data corresponding to each position, forming eight groups of monitoring data, where the monitoring sensors at each monitoring point form a group of monitoring data; the data aggregation node uploads the eight groups of monitoring data to a remote server, and performs data matrix construction processing on the eight groups of monitoring data on the remote server to form data matrix information; the data matrix information is input into a converged deep learning model for fault prediction processing of the closed blast furnace, and a fault prediction result of the closed blast furnace is obtained; more accurate prediction of the faults of the closed blast furnace is realized, so that the faults of the closed blast furnace can be cleared in time, thereby improving production efficiency and maintaining the health of the furnace body.

[0113] A computer-readable storage medium provided by an embodiment of the present invention, on which a computer program is stored, and when the program is executed by a processor, it implements the closed blast furnace fault prediction method in any one of the above embodiments. Among them, the computer-readable storage medium includes, but is not limited to, any type of disk (including floppy disk, hard disk, optical disk, CD-ROM, and magneto-optical disk), ROM (Read-Only Memory), RAM (Random Access Memory), EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory, magnetic card or optical card. That is, the storage device includes any medium that can store or transmit information in a readable form by a device (such as a computer, mobile phone), and can be a read-only memory, a magnetic disk or an optical disk, etc.

[0114] An embodiment of the present invention also provides a computer application program, which runs on a computer and is used to execute the closed blast furnace fault prediction method in any one of the above.

[0115] In addition, Figure 4 is a schematic diagram of the structural composition of an electronic device in an embodiment of the present invention.

[0116] An embodiment of the present invention also provides an electronic device, as Figure 4 shown. The electronic device includes devices such as a processor 402, a memory 403, an input unit 404, and a display unit 405. Those skilled in the art can understand, Figure 4The structural components of the illustrated electronic device do not constitute a limitation on all devices, and may include more or fewer components than shown, or combine certain components. The memory 403 can be used to store the application program 401 and each functional module. The processor 402 runs the application program 401 stored in the memory 403, thereby executing various functional applications and data processing of the device. The memory can be an internal memory or an external memory, or include both an internal memory and an external memory. The internal memory can include a read-only memory (ROM), a programmable ROM (PROM), an electrically programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a flash memory, or a random access memory. The external memory can include a hard disk, a floppy disk, a ZIP disk, a USB flash drive, a magnetic tape, etc. The memory disclosed in the present invention includes, but is not limited to, these types of memories. The memory disclosed in the present invention is only an example and not a limitation.

[0117] The input unit 404 is used to receive the input of signals and receive the keywords input by the user. The input unit 404 can include a touch panel and other input devices. The touch panel can collect the touch operations of the user on or near it (such as the operations of the user using a finger, a stylus, or any suitable object or accessory on or near the touch panel), and drive the corresponding connection device according to a pre-set program; the other input devices can include, but are not limited to, one or more of a physical keyboard, function keys (such as play control keys, switch keys, etc.), a trackball, a mouse, a joystick, etc. The display unit 405 can be used to display the information input by the user or the information provided to the user and various menus of the terminal device. The display unit 405 can be in the form of a liquid crystal display, an organic light-emitting diode, etc. The processor 402 is the control center of the terminal device, connecting various parts of the entire device using various interfaces and lines, and executing various functions and processing data by running or executing the software programs and / or modules stored in the memory 403, and calling the data stored in the memory.

[0118] As an embodiment, the electronic device includes: one or more processors 402, a memory 403, one or more application programs 401, wherein the one or more application programs 401 are stored in the memory 403 and are configured to be executed by the one or more processors 402, and the one or more application programs 401 are configured to execute the corresponding closed blast furnace fault prediction method in any one of the above embodiments.

[0119] In an embodiment of the present invention, sensors corresponding to eight monitoring points on the closed blast furnace are used to collect and process data corresponding to each position respectively, forming eight groups of monitoring data, where the monitoring sensors at each monitoring point form a group of monitoring data; the data aggregation node uploads the eight groups of monitoring data to a remote server, and performs data matrix construction processing on the eight groups of monitoring data on the remote server to form data matrix information; the data matrix information is input into a converged deep learning model for fault prediction processing of the closed blast furnace to obtain a fault prediction result of the closed blast furnace; realizing more accurate prediction of the faults of the closed blast furnace, so that the faults of the closed blast furnace can be cleared in time, thereby improving production efficiency and maintaining the health of the furnace body.

[0120] In addition, the above has introduced in detail a method and related device for fault prediction of a closed blast furnace based on deep learning provided by the embodiments of the present invention. Specific examples should have been used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A fault prediction method for a closed blast furnace based on deep learning, characterized in that, Applied to a closed blast furnace, the closed blast furnace is successively provided with a burden heating zone, a reoxidation zone, a reduction zone, and a slag melting zone from top to bottom, and eight monitoring points are arranged on the closed blast furnace, and a corresponding sensor is arranged at each monitoring point. The method includes: Based on the sensors corresponding to the eight monitoring points on the closed blast furnace, the data corresponding to each position are respectively collected and processed to form eight groups of monitoring data, wherein the monitoring sensors at each monitoring point form a group of monitoring data; The data aggregation node uploads the eight groups of monitoring data to the remote server, and performs data matrix construction processing on the eight groups of monitoring data on the remote server to form data matrix information; Input the data matrix information into a converged deep learning model for fault prediction processing of the closed blast furnace to obtain a fault prediction result of the closed blast furnace. The deep learning model is successively composed of a convolutional neural network, a first long short-term memory network, an attention mechanism network, and a second long short-term memory network.

2. The method for predicting the failure of the closed blast furnace according to claim 1, wherein The setting of the eight monitoring points on the closed blast furnace includes: A temperature sensor and a pressure sensor are respectively arranged on two monitoring points on the burden heating zone of the closed blast furnace for monitoring the top temperature data and top pressure data of the closed blast furnace; A zinc vapor concentration sensor and a nitrogen dioxide concentration sensor are respectively arranged on two monitoring points on the reoxidation zone of the closed blast furnace for monitoring the zinc vapor concentration data and nitrogen dioxide concentration data of the reoxidation zone; A first air pressure sensor and a first air volume sensor are respectively arranged on two monitoring points on the secondary air duct of the reduction zone of the closed blast furnace for monitoring the air pressure data and air volume data of the secondary air duct; A second air pressure sensor and a second air volume sensor are respectively arranged on two monitoring points at 16 tuyeres of the slag melting zone of the closed blast furnace for monitoring the air pressure data and total air volume data at 16 tuyeres.

3. The method for predicting the failure of the closed blast furnace according to claim 1, wherein The collecting and processing of the data corresponding to each position respectively by the sensors corresponding to the eight monitoring points on the closed blast furnace to form eight groups of monitoring data includes: During the operation of the closed blast furnace, control the sensors corresponding to the eight monitoring points on the closed blast furnace to collect and process data at the same frequency to obtain the collected monitoring data collected by the sensors corresponding to the eight monitoring points; The sensors corresponding to the eight monitoring points converge the collected monitoring data to the data aggregation node, wherein the sensors corresponding to the eight monitoring points are communicatively connected to the data aggregation node; After receiving the collected monitoring data, the data aggregation node writes the collected monitoring data into the corresponding empty array to form eight groups of monitoring data, wherein each empty array is bound to the ID of a sensor and is used to store the collected monitoring data collected by the sensor with the bound ID, and the lengths of the eight empty arrays are the same.

4. The method for predicting the failure of the closed blast furnace according to claim 1, characterized in that, The data aggregation node uploading the eight groups of monitoring data to the remote server includes: The data aggregation node establishes a communication connection with the remote server, and after the eight groups of monitoring data are obtained at the data aggregation node, the eight groups of monitoring data are uploaded to the remote server based on the communication connection.

5. The method for predicting the failure of a closed blast furnace according to claim 1, characterized in that Perform data matrix construction processing on the eight groups of monitoring data on the remote server to form data matrix information, including: On the remote server, calculate the standard deviation, average value, and covariance value for each group of detection data in the eight groups of monitoring data to obtain the standard deviation, average value, and covariance value corresponding to each group of detection data; Perform joint probability density calculation using the average value and covariance value corresponding to each group of detection data based on the joint probability density function to obtain the joint probability density corresponding to each group of detection data; Perform outlier screening processing based on the joint probability density and the standard deviation corresponding to each group of detection data to obtain the outliers in each group of monitoring data; Remove the outliers in each group of monitoring data, and perform interpolation processing on each group of monitoring data with outliers removed using the difference algorithm to obtain each group of interpolated monitoring data; Perform data matrix construction processing on each group of interpolated monitoring data in matrix form to form data matrix information.

6. The method for predicting the failure of the closed blast furnace according to claim 1, characterized in that, Input the data matrix information into a converged deep learning model for fault prediction processing of the closed blast furnace to obtain the fault prediction result of the closed blast furnace, including: Input the data matrix information into a converged deep learning model, and perform downsampling processing on the data matrix information in the convolutional neural network in the converged deep learning model to obtain downsampled feature data; Input the downsampled feature data into the first long short-term neural memory network for time series prediction processing to obtain a predicted time series; Input the predicted time series into the attention mechanism network, use Key to represent the fluctuation characteristics of the closed blast furnace during the generation process, use Value to represent the data generation range, and use Query to perform attention optimization processing on the predicted time series by the attention mechanism network to obtain an optimized predicted time series; Input the optimized predicted time series into the second long short-term neural memory network for fault fusion prediction processing of the closed blast furnace, and output the fault prediction result of the closed blast furnace through the fully connected layer of the converged deep learning model to obtain the fault prediction result of the closed blast furnace.

7. The method for predicting the failure of the closed blast furnace according to claim 6, wherein Perform downsampling processing on the data matrix information in the convolutional neural network in the converged deep learning model to obtain downsampled feature matrix data, including: In the convolutional neural network in the converged deep learning model, perform a convolution operation on the data matrix information using the convolutional layer, and extract the feature data information in the data matrix information based on the convolution operation; Perform a downsampling operation on the extracted feature data information in the downsampling layer of the convolutional neural network, and output through the fully connected layer in the convolutional neural network to form downsampled feature data.

8. A closed blast furnace fault prediction device based on deep learning, characterized in that, Applied to an airtight blast furnace, the airtight blast furnace is sequentially provided with a burden heating zone, a reoxidation zone, a reduction zone and a slag melting zone from top to bottom, and eight monitoring points are arranged on the airtight blast furnace, and a corresponding sensor is arranged at each monitoring point. The device includes: A data acquisition module: used to respectively collect and process the data corresponding to each position based on the sensors corresponding to the eight monitoring points on the airtight blast furnace, and form eight groups of monitoring data, wherein the monitoring sensors at each monitoring point form a group of monitoring data; A matrix construction module: used for the data convergence node to upload the eight groups of monitoring data to a remote server, and perform data matrix construction processing on the eight groups of monitoring data on the remote server to form data matrix information; A fault prediction module: used to input the data matrix information into a converged deep learning model for fault prediction processing of the airtight blast furnace, and obtain a fault prediction result of the airtight blast furnace. The deep learning model is sequentially composed of a convolutional neural network, a first long short-term neural memory network, an attention mechanism network and a second long short-term neural memory network.

9. An electronic device, comprising a processor and a memory, characterized in that, The processor runs the computer program or code stored in the memory to implement the airtight blast furnace fault prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium for storing a computer program or code, characterized in that, When the computer program or code is executed by the processor, the airtight blast furnace fault prediction method according to any one of claims 1 to 7 is implemented.

Images