Position prediction method of sintering end point, storage medium and computer equipment
By constructing a sintering endpoint position prediction model based on channel-independent timing convolution network, the problem of difficulty in capturing long-term dependence on multiple parameter measurement and timing models in the existing methods is solved, and efficient and balanced sintering endpoint prediction is achieved.
Patent Information
- Application Number
- CN202411692095.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-25
- Publication Date
- 2025-03-18
AI Technical Summary
Existing methods for sintering end point position prediction rely on measurements of multiple production and state parameters, and timing models are difficult to capture long-term dependencies or consume a large amount of resources.
A channel-independent timing convolutional network (TCN) sintering end point position prediction model is used to obtain and pretreat flue exhaust gas temperature data, use quadratic curve fitting to perform soft measurements, build sample sets and train to predict the sintering end point position.
Improves the efficiency and performance balance of sintering endpoint forecasting, reduces dependence on production and state parameter measurements, can effectively capture long-term dependencies and reduce resource consumption.
Smart Images
Figure CN119646396A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of sintering ore, and in particular to a sintering end point position prediction method, a storage medium and a computer device. Background Art
[0002] The position of the sintering endpoint (burning through point, BTP) is an important indicator for measuring the output and quality of sintered ore, and is also a sign of whether the sintering process is normal. The sintering endpoint prediction technology can prevent the endpoint from being advanced or delayed, and reduce the hysteresis of human control, which is of great significance for improving the output and quality of sintered ore, ensuring the stability of the sintering process, and reducing energy consumption.
[0003] The main technical means for predicting the sintering endpoint position are currently to establish mathematical models based on physical and chemical reaction mechanisms, expert systems based on metallurgical control expert knowledge and skilled operator experience, and computational models based on artificial intelligence algorithms such as support vector machines, neural networks and other machine learning methods. The sintering process is a complex physical and chemical reaction process with nonlinear, strong coupling, and large hysteresis characteristics, which is difficult to accurately simulate using mathematical models; expert models have great uncertainty and ambiguity in the basis of manual judgment due to differences in the experience of different operators; computational models such as SVM, BP network, and RBF network that rely on state control parameters such as temperature field, permeability, and trolley speed have poor versatility due to differences in selected parameters, measurement accuracy, and hysteresis; while time series models RNN and LSTM have difficulty in capturing long-distance dependencies, and Transformer has performance requirements that consume a lot of computing resources and memory. Summary of the invention
[0004] The purpose of the present invention is to provide a sintering endpoint position prediction method, storage medium and computer equipment to improve the efficiency and performance balance of sintering endpoint prediction, solve the problem of existing methods relying on the measurement of various production and state parameters in the sintering process, and the difficulty of time series models in capturing long-term dependence or consuming large resources. The specific technical solution is as follows:
[0005] A method for predicting a sintering endpoint position, the method comprising the following steps:
[0006] S100, obtaining historically collected flue gas temperature data and performing preprocessing;
[0007] S200, using a soft measurement method of curve fitting to calculate the exhaust gas temperature inflection point, the exhaust gas temperature rise point and the sintering end point for the pre-treated flue gas temperature data;
[0008] S300, performing sliding window slicing processing on the position data of the historical exhaust gas temperature inflection point, the exhaust gas temperature rise point and the sintering end point obtained by soft measurement, and obtaining soft measurement sequence slices of the positions of the historical exhaust gas temperature inflection point, the exhaust gas temperature rise point and the sintering end point as a sample set;
[0009] S400, constructing a sintering endpoint position prediction model based on a channel-independent temporal convolutional network;
[0010] S500, using the sample set to train a channel-independent temporal convolutional network sintering endpoint position prediction model to obtain a trained sintering endpoint position prediction model;
[0011] S600, inputting the soft measurement sequence slices of the current exhaust gas temperature inflection point, exhaust gas temperature rise point and sintering end point into the trained sintering end point position prediction model, and outputting the current predicted exhaust gas temperature inflection point, exhaust gas temperature rise point and sintering end point position sequence.
[0012] Furthermore, in step S200, the positions of the exhaust gas temperature inflection point, the exhaust gas temperature rise point and the sintering end point are calculated by using a soft measurement method of quadratic curve fitting, specifically including: by using a soft measurement method, a number of points are selected near the exhaust gas temperature inflection point, the exhaust gas temperature rise point and the exhaust gas temperature vertex for fitting, and the exhaust gas temperature T i and bellows position x i It is approximately the following quadratic fitting relationship:
[0013]
[0014] Among them, a, b, c are fitting coefficients, and m is the total number of bellows;
[0015] Then, the positions of the exhaust gas temperature inflection point, the exhaust gas temperature rise point and the sintering end point are solved according to the fitting curve using soft measurement method.
[0016] Further, solving the position of the exhaust gas temperature inflection point specifically includes: when the difference between the exhaust gas temperature at a certain wind box position and the exhaust gas temperature at the previous wind box position is less than the set threshold value T threshold1 , and the difference between the exhaust gas temperature at the next wind box position and the exhaust gas temperature at this wind box position is greater than the set threshold T threshold2 When:
[0017] T i -T i-1 <T threshld1
[0018] T i+1 -T i >T threshold2
[0019] Select several points near this position for quadratic curve fitting, and solve for the slope of the set value K trp The position at which the exhaust gas temperature turns is the position of the exhaust gas temperature inflection point, and the solution formula is as follows:
[0020] K trp =2ax trp +b
[0021] Among them, x trp is the position of the exhaust gas temperature inflection point.
[0022] Furthermore, the location of the exhaust gas temperature rise point is specifically determined by: the exhaust gas temperature of the wind box at the rear end of the trolley reaches the set rising temperature T brp When the position is selected, select the one closest to the set temperature T brp Perform quadratic curve fitting on several points near the bellows position and solve the function value to the set value T brp The position at which the exhaust gas temperature rises is the position of the exhaust gas temperature rising point, and the solution formula is as follows:
[0023]
[0024] Among them, x brp is the location of the exhaust gas temperature rise point.
[0025] Furthermore, solving the position of the sintering end point specifically includes: the position of the highest point of the exhaust gas temperature of the wind box at the rear end of the trolley, selecting several points near the wind box position with the highest temperature to perform quadratic curve fitting, and solving the vertex position, which is the position of the current sintering end point. The solution formula is as follows:
[0026] x btp =-b / 2a
[0027] Among them, x btp is the position of the sintering end point.
[0028] Furthermore, the step S300 is specifically as follows: taking L as the window size and 1 as the step size, for the historical soft measurement data of the exhaust gas temperature inflection point, the exhaust gas temperature rising point and the sintering end point at any moment, take the soft measurement data of the moment and the previous L historical moments as a two-dimensional input in the form of 3*L, take the soft measurement data of T moments after the moment as a two-dimensional output in the form of 3*T, and such input and output constitute the sample of the moment.
[0029] Furthermore, in step S400, the channel-independent temporal convolutional network sintering endpoint position prediction model includes temporal embedding, a backbone network for extracting temporal, feature, and channel dimension information, and a task head for performing prediction. The construction process is specifically as follows: a temporal convolutional network TCN is used for modeling, and its temporal embedding adopts Patch Embedding technology to divide the input sequence into multiple subsequences of fixed size and map them to the target embedding space. This process processes each variable separately to achieve channel independence; the backbone network is formed by stacking multiple backbone modules, and timing, feature, and channel information is extracted through deep separable convolution. In the first step, each backbone module extracts variable timing information through a deep convolution DWConv module using a large convolution kernel. In the second step, two grouped convolution modules ConvFFN are used to extract feature interaction information and channel interaction information respectively, and then a backbone module output is obtained through residual connection; finally, the output of the backbone network is flattened through the prediction head and the result is mapped to the prediction sequence of each variable.
[0030] Furthermore, the data processing steps of the Embedding layer are as follows: for an input X containing M variables of length L and shape (M, L) in , and the dimension is increased to facilitate mapping to the embedding space, and the input X with shape (M, 1, L) is obtained. in ; Perform padding and patching operations on each variable sequence, and select a sequence of length P as a subsequence every step S. In particular, for X in The last value of is repeated (PS) times as a subsequence, then the length of the new sequence is N = [L / S], [·] represents the rounding operation, and the input X with shape (M, 1, N) is obtained. in ; Through one-dimensional convolution, each channel (variable) is kept independent, and the input is mapped to the target embedding space with dimension D, where D is the feature dimension after embedding, and the input X with shape (M, D, N) is obtained. emb ;
[0031] The backbone network is composed of multiple backbone modules stacked together, including DWConv modules, ConvFFN1 modules and ConvFFN2 modules connected in sequence. The deep separable convolution technology is used to learn the time series information of variables / features and channel interactions in two separate modules. The DWConv module implements the first step in the deep separable convolution, merging the channel dimension and the feature dimension to obtain data with a shape of (M*D). One-dimensional convolution with a large convolution kernel is used to capture long-distance dependency information in the time dimension. ConvFFN1 and ConvFFN2 are designed for information learning of different channel dimensions. The ConvFFN module consists of two 1*1 point convolution PWConv modules. The point convolution uses grouped convolution technology. The two ConvFFN modules use different grouping logics: the number of groups in ConvFFN1 is set to M, indicating that the feature interaction information of the same variable is learned, and the number of groups in ConvFFN2 is set to D, indicating that the variable interaction information of the same feature is learned. After two ConvFFN modules, the output result of a backbone module is obtained through residual connection.
[0032] After passing through the backbone network, the output of shape (M, D, N) is obtained, which is flattened in the flattening layer to obtain an output of shape (M, D*N), and then fully connected linear mapping is performed in the task head to obtain a prediction sequence output (M, T) with a target length of T;
[0033] In the input stage, each input data is Z-standardized and the mean and standard deviation are recorded; in the output stage, each output data is de-standardized.
[0034] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the sintering endpoint position prediction method as described above are implemented.
[0035] The present invention also provides a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the sintering endpoint position prediction method as described above when executing the program.
[0036] The present invention provides a method for predicting the sintering endpoint position, which has the following beneficial effects:
[0037] The present invention provides a method for predicting the position of a sintering endpoint, which obtains historically collected flue gas temperature data and performs preprocessing, calculates the flue gas temperature inflection point, the waste gas temperature rise point and the position of the sintering endpoint by using a soft measurement method of quadratic curve fitting on the preprocessed flue gas temperature data, performs sliding window slicing processing on the historical waste gas temperature inflection point, waste gas temperature rise point and the position data of the sintering endpoint obtained by soft measurement, obtains soft measurement sequence slices of the historical waste gas temperature inflection point, waste gas temperature rise point and the position of the sintering endpoint as a sample set, and constructs a sintering endpoint position prediction model based on a channel-independent temporal convolutional network. The sample set is used to train the sintering end point position prediction model based on the channel-independent time series convolutional network to obtain a trained sintering end point position prediction model, and the soft measurement sequence slices of the current exhaust gas temperature inflection point, exhaust gas temperature rising point and sintering end point are input into the trained sintering end point position prediction model, and the current predicted exhaust gas temperature inflection point, exhaust gas temperature rising point and sintering end point position sequence are output, which can improve the efficiency and performance balance of sintering end point prediction, solve the problems of existing methods' dependence on the measurement of various production and state parameters in the sintering process, and the difficulty of time series models in capturing long-term dependence or consuming large resources. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 A schematic flow chart of a sintering endpoint position prediction method provided by the present invention;
[0039] Figure 2 It is a schematic diagram of the overall process of an embodiment of the present invention;
[0040] Figure 3 Schematic diagram of a sintering endpoint position prediction model based on a channel-independent temporal convolutional network according to an embodiment of the present invention;
[0041] Figure 4 It is a structural diagram of a backbone module according to an embodiment of the present invention;
[0042] Figure 5 It is a structural schematic diagram of a grouping module according to an embodiment of the present invention;
[0043] Figure 6 4 is a structural block diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0044] The following will be combined with the accompanying drawings provided by the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. According to the following description, the advantages and features of the present invention will be more clear. It should be noted that the accompanying drawings are all in a very simplified form and are not in precise proportions, which are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention.
[0045] Example 1
[0046] This embodiment provides a method for predicting the sintering end point position. Figure 1 , 2 As shown, the method comprises the following steps:
[0047] S100, obtaining historically collected flue gas temperature data and performing preprocessing.
[0048] Specifically, for one or more abandoned temperature measurement points at the same wind box position, the abnormal values should be removed, the average value should be taken, and then stored in the time series database.
[0049] S200, using a soft measurement method of curve fitting to calculate the flue gas temperature inflection point (TRP), the waste gas temperature rise point (BRP) and the sintering endpoint (BTP) for the pre-treated flue gas temperature data.
[0050] In a preferred embodiment, the positions of the exhaust gas temperature inflection point, the exhaust gas temperature rise point and the sintering end point are calculated by using a soft measurement method of quadratic curve fitting.
[0051] Specifically, through the soft measurement method, several points are selected near the exhaust gas temperature inflection point, the exhaust gas temperature rising point and the exhaust gas temperature peak for fitting. i and bellows position x i It is approximately the following quadratic fitting relationship:
[0052]
[0053] Among them, a, b, c are fitting coefficients, and m is the total number of bellows.
[0054] Then, the positions of the exhaust gas temperature inflection point, exhaust gas temperature rise point and sintering end point are solved by soft measurement according to the fitting curve, specifically:
[0055] (1) Calculation of the location of the exhaust gas temperature inflection point
[0056] The exhaust gas temperature of the bellows in the middle of the trolley meets the condition that the temperature of the front bellows rises slowly, while the temperature of the rear bellows rises rapidly, and the position is fitted.
[0057] When the difference between the exhaust gas temperature at a certain wind box position and the exhaust gas temperature at the previous wind box position is less than the set threshold T threshold1 , and the difference between the exhaust gas temperature at the next wind box position and the exhaust gas temperature at this wind box position is greater than the set threshold T threshold2 When:
[0058] T i -T i-1 <T threshold1
[0059] T i+1 -T i >Tthreshold2
[0060] Select several points near this position for quadratic curve fitting, and solve for the slope of the set value K trp The position at which the exhaust gas temperature turns is the position of the exhaust gas temperature inflection point, and the solution formula is as follows:
[0061] K trp =2ax trp +b
[0062] Among them, x trp is the position of the exhaust gas temperature inflection point.
[0063] Among them, T threshold1 , T threshold2 According to actual production experience, the value range is 10~20℃; K trp It is set according to actual production experience, with a range of 5-10℃ / m.
[0064] (2) Calculation of the location of the exhaust gas temperature rise point
[0065] The exhaust gas temperature of the wind box at the rear end of the trolley reaches the set rising temperature T brp When the position is selected, select the one closest to the set temperature T brp Perform quadratic curve fitting on several points near the bellows position and solve the function value to the set value T brp The position at which the exhaust gas temperature rises is the position of the exhaust gas temperature rising point, and the solution formula is as follows:
[0066]
[0067] Among them, x brp is the location of the exhaust gas temperature rise point.
[0068] Among them, T brp Determined based on actual operating experience, the range is 160-200℃.
[0069] (3) Calculation of the position of the sintering end point
[0070] The position of the highest point of the exhaust gas temperature of the wind box at the rear end of the trolley, select several points near the wind box position with the highest temperature to perform quadratic curve fitting, and solve the vertex position, which is the position of the current sintering end point. The solution formula is as follows:
[0071] x btp =-b / 2a
[0072] Among them, x btp is the position of the sintering end point.
[0073] S300, performing sliding window slicing processing on the position data of historical exhaust gas temperature inflection points, exhaust gas temperature rise points and sintering end points obtained by soft measurement, and obtaining soft measurement sequence slices of the positions of historical exhaust gas temperature inflection points, exhaust gas temperature rise points and sintering end points as a sample set.
[0074] Specifically, with L as the window size and 1 as the step size, for the historical soft measurement data of the exhaust gas temperature inflection point, the exhaust gas temperature rising point and the sintering end point at any moment, the soft measurement data of that moment and the previous L historical moments are taken as a two-dimensional input in the form of 3*L, and the soft measurement data of T moments after that moment are taken as a two-dimensional output in the form of 3*T. Such input and output constitute the sample at that moment.
[0075] After the above sliding window slicing is performed on the exhaust gas temperature inflection point, the exhaust gas temperature rise point and the sintering end point, the sample set containing the input and output at each moment is obtained, and the training set, validation set and test set are randomly selected in proportion.
[0076] S400, construct a sintering endpoint position prediction model based on a channel-independent temporal convolutional network.
[0077] Specifically, the sintering endpoint position prediction model based on channel-independent temporal convolutional network includes temporal embedding, a backbone network for extracting temporal, feature, and channel dimension information, and a task head for performing prediction. The construction process is specifically as follows: a temporal convolutional network TCN is used for modeling, and its temporal embedding adopts Patch Embedding technology to divide the input sequence into multiple subsequences of fixed size and map them to the target embedding space. This process processes each variable separately to achieve channel independence; the backbone network is formed by stacking multiple backbone modules, and the timing, feature, and channel information are extracted through deep separable convolution. In the first step, each backbone module extracts variable timing information through a deep convolution DWConv module using a large convolution kernel. Specifically, the large convolution kernel is a convolution kernel greater than or equal to 31*31. In the second step, two grouped convolution modules ConvFFN are used to extract feature interaction information and channel interaction information respectively, and then a backbone module output is obtained through residual connection; finally, the output of the backbone network is flattened through the prediction head and the result is mapped to the prediction sequence of each variable.
[0078] References Figure 3 As shown, the specific structure of the sintering endpoint position prediction model based on the channel-independent temporal convolutional network is as follows:
[0079] (1) Embedding layer
[0080] a. For an input X containing M variables of length L and shape (M, L) in , and the dimension is increased to facilitate mapping to the embedding space, and the input X with shape (M, 1, L) is obtained.in ;
[0081] b. Perform padding and patching operations on each variable sequence, and select a sequence of length P as a subsequence every step S. In particular, for X in The last value of is repeated (PS) times as a subsequence, then the length of the new sequence is N = [L / S], [·] represents the rounding operation, and the input X with shape (M, 1, N) is obtained. in ;
[0082] c. Through one-dimensional convolution, each channel (variable) is kept independent, and the input is mapped to the target embedding space with dimension D, where D is the feature dimension after embedding, and the input X with shape (M, D, N) is obtained. emb .
[0083] (2) Backbone network
[0084] The backbone network is composed of multiple backbone modules stacked together. The backbone module structure is as follows: Figure 4 As shown, it includes a DWConv module, a ConvFFN1 module and a ConvFFN2 module connected in sequence, and adopts a deep separable convolution technique to learn the temporal information of variables / features and the channel interaction using two separate modules.
[0085] a. The DWConv module implements the first step in the depthwise separable convolution. It combines the channel dimension and the feature dimension to obtain data of shape (M*D). It uses a one-dimensional convolution with a large convolution kernel to capture long-distance dependency information in the time dimension. Specifically, the large convolution kernel is a convolution kernel greater than or equal to 31*31.
[0086] b. ConvFFN1 and ConvFFN2 are designed for learning information of different channel dimensions. The ConvFFN module consists of two 1*1 point convolution PWConvs, such as Figure 5 As shown;
[0087] Point convolution uses grouped convolution technology. The two ConvFFNs use different grouping logics: the number of groups in ConvFFN1 is set to M, which means learning the feature interaction information of the same variable, and the number of groups in ConvFFN2 is set to D, which means learning the variable interaction information of the same feature. After two ConvFFNs, the output result of a backbone module is obtained through residual connection;
[0088] (3) Flatten layer
[0089] After passing through the backbone network, an output of the shape (M, D, N) is obtained, which is flattened to obtain an output of the shape (M, D*N), and a prediction sequence output (M, T) with a target length of T is obtained through a fully connected linear mapping in the task head;
[0090] (4) RevIN Standardization
[0091] In the input stage, each input data is Z-standardized and the mean and standard deviation are recorded; in the output stage, each output data is de-standardized.
[0092] S500, using the sample set to train the channel-independent temporal convolutional network sintering end point position prediction model to obtain a trained sintering end point position prediction model.
[0093] Specifically, according to the sample set size, the number of variables and the sequence length, the step size S, window P, the number of target embedding space features D, the number of backbone module stacks and the size of the convolution kernel in DWConv are set during the embedding operation. The hyperparameters such as the learning rate and the number of iterations epoch adopted during training are selected according to the sample set size and the model complexity. The error is evaluated using loss functions such as MAE or MSE. The training set in the sample set is used as the network input, and the loss function result of the validation set is used as the optimization target for network training. The test set is used as the trained network input, and the output result of the loss function is evaluated. The network structure parameters such as length S, window P, the number of target embedding space features D, the number of backbone module stacks, the size of the convolution kernel in DWConv, as well as hyperparameters such as the learning rate and the number of iterations epoch are adjusted. The network is retrained until the test set error reaches the expected standard. The final network structure parameters and hyperparameters are saved to obtain a trained sintering end point position prediction model.
[0094] S600, inputting the soft measurement sequence slices of the current exhaust gas temperature inflection point, exhaust gas temperature rise point and sintering end point into the trained sintering end point position prediction model, and outputting the current predicted exhaust gas temperature inflection point, exhaust gas temperature rise point and sintering end point position sequence.
[0095] Specifically, in this embodiment, the currently collected flue exhaust gas temperature data is first obtained, and then the soft measurement sequence slices of the current exhaust gas temperature inflection point, exhaust gas temperature rise point and sintering end point are obtained by the same method as steps S100, S200, and S300. Then, a data set consisting of a sequence of soft measurement results of the current and historical L exhaust gas temperature inflection points, exhaust gas temperature rise points and sintering end points is used as the input of the trained sintering end point position prediction model to obtain the position prediction sequence of the exhaust gas temperature inflection point, exhaust gas temperature rise point and sintering end point at the next T moments.
[0096] The method for predicting the sintering endpoint position provided by the present invention obtains historically collected flue gas temperature data and performs preprocessing, calculates the flue gas temperature inflection point, the waste gas temperature rise point and the position of the sintering endpoint by using a soft measurement method of quadratic curve fitting on the preprocessed flue gas temperature data, performs sliding window slicing processing on the historical waste gas temperature inflection point, waste gas temperature rise point and sintering endpoint position data obtained by soft measurement, obtains soft measurement sequence slices of the historical waste gas temperature inflection point, waste gas temperature rise point and sintering endpoint position as a sample set, and constructs a sintering endpoint position prediction model based on a channel-independent temporal convolutional network. The sample set is used to train the sintering end point position prediction model based on the channel-independent time series convolutional network to obtain a trained sintering end point position prediction model, and the soft measurement sequence slices of the current exhaust gas temperature inflection point, exhaust gas temperature rise point and sintering end point are input into the trained sintering end point position prediction model, and the current predicted exhaust gas temperature inflection point, exhaust gas temperature rise point and sintering end point position sequence are output. This can improve the efficiency and performance balance of sintering end point prediction, and solve the problems of existing methods' dependence on the measurement of various production and state parameters in the sintering process, and the difficulty of time series models in capturing long-term dependence or consuming large resources.
[0097] Example 2
[0098] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the sintering endpoint position prediction method described above are implemented.
[0099] The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk (HDD) or a solid-state drive (SSD), etc.; the storage medium may also include a combination of the above-mentioned types of memory.
[0100] Example 3
[0101] This embodiment provides a computer device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the sintering endpoint position prediction method described above when executing the program.
[0102] like Figure 6As shown, the computer device may include: at least one processor 71, such as a CPU (Central Processing Unit), at least one communication interface 73, a memory 74, and at least one communication bus 72. The communication bus 72 is used to realize the connection and communication between these components. The communication interface 73 may include a display screen (Display) and a keyboard (Keyboard), and the optional communication interface 73 may also include a standard wired interface and a wireless interface. The memory 74 may be a high-speed RAM memory (Random Access Memory, volatile random access memory) or a non-volatile memory (non-volatile memory), such as at least one disk storage. The memory 74 may optionally be at least one storage device located away from the aforementioned processor 71. The memory 74 stores application programs, and the processor 71 calls the program code stored in the memory 74 to execute any of the above method steps.
[0103] The communication bus 72 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The communication bus 72 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.
[0104] Among them, the memory 74 may include a volatile memory (English: volatile memory), such as a random access memory (English: random-access memory, abbreviated: RAM); the memory may also include a non-volatile memory (English: non-volatile memory), such as a flash memory (English: flash memory), a hard disk (English: hard disk drive, abbreviated: HDD) or a solid-state drive (English: solid-state drive, abbreviated: SSD); the memory 74 may also include a combination of the above types of memory.
[0105] The processor 71 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and a NP.
[0106] The processor 71 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.
[0107] Optionally, the memory 74 is also used to store program instructions. The processor 71 can call the program instructions to implement the sintering endpoint position prediction method of the present invention.
[0108] Those skilled in the art should understand that the present invention can be implemented in many other specific forms without departing from the spirit and scope of the present invention. Based on the embodiments of the present invention, any changes and modifications made by ordinary technicians in the field of the present invention according to the above disclosure are within the scope of protection of the claims.
Claims
1. A method for predicting the sintering endpoint position, characterized in that: The method comprises the following steps: S100, obtaining historically collected flue gas temperature data and performing preprocessing; S200, using a soft measurement method of curve fitting to calculate the exhaust gas temperature inflection point, the exhaust gas temperature rise point and the sintering end point for the pre-treated flue gas temperature data; S300, performing sliding window slicing processing on the position data of the historical exhaust gas temperature inflection point, the exhaust gas temperature rise point and the sintering end point obtained by soft measurement, and obtaining soft measurement sequence slices of the positions of the historical exhaust gas temperature inflection point, the exhaust gas temperature rise point and the sintering end point as a sample set; S400, constructing a sintering endpoint position prediction model based on a channel-independent temporal convolutional network; S500, using the sample set to train a channel-independent temporal convolutional network sintering endpoint position prediction model to obtain a trained sintering endpoint position prediction model; S600, inputting the soft measurement sequence slices of the current exhaust gas temperature inflection point, exhaust gas temperature rise point and sintering end point into the trained sintering end point position prediction model, and outputting the current predicted exhaust gas temperature inflection point, exhaust gas temperature rise point and sintering end point position sequence.
2. The method for predicting the sintering endpoint position according to claim 1, characterized in that: In step S200, the positions of the exhaust gas temperature inflection point, the exhaust gas temperature rise point and the sintering end point are calculated by using a soft measurement method of quadratic curve fitting, specifically including: by using a soft measurement method, a number of points are selected near the exhaust gas temperature inflection point, the exhaust gas temperature rise point and the exhaust gas temperature vertex for fitting, and the exhaust gas temperature T i and bellows position x i It is approximately the following quadratic fitting relationship: Among them, a, b, c are fitting coefficients, and m is the total number of bellows; Then, the positions of the exhaust gas temperature inflection point, the exhaust gas temperature rise point and the sintering end point are solved according to the fitting curve using soft measurement method.
3. The method for predicting the sintering end point position according to claim 2, characterized in that: Solving the position of the exhaust gas temperature inflection point specifically includes: when the difference between the exhaust gas temperature at a certain wind box position and the exhaust gas temperature at the previous wind box position is less than the set threshold T threshold1 , and the difference between the exhaust gas temperature at the next wind box position and the exhaust gas temperature at this wind box position is greater than the set threshold T threshold2 When: T i -T i-1 <T threshold1 T i+1 -T i >T threshold2 Select several points near this position for quadratic curve fitting, and solve for the slope of the set value K trp The position at which the exhaust gas temperature turns is the position of the exhaust gas temperature inflection point, and the solution formula is as follows: K trp =2ax trp +b Among them, x trp is the position of the exhaust gas temperature inflection point.
4. The method for predicting the sintering end point position according to claim 2, characterized in that: Solving the location of the exhaust gas temperature rise point specifically includes: the exhaust gas temperature of the wind box at the rear end of the trolley reaches the set rising temperature T brp When the position is selected, select the one closest to the set temperature T brp Perform quadratic curve fitting on several points near the bellows position and solve the function value to the set value T brp The position at which the exhaust gas temperature rises is the position of the exhaust gas temperature rising point, and the solution formula is as follows: Among them, x brp is the location of the exhaust gas temperature rise point.
5. The method for predicting the sintering end point position according to claim 2, characterized in that: Solving the position of the sintering end point specifically includes: the position of the highest point of the exhaust gas temperature of the wind box at the rear end of the trolley, selecting several points near the wind box position with the highest temperature to perform quadratic curve fitting, and solving the vertex position, which is the position of the current sintering end point. The solution formula is as follows: x btp =-b / 2a Among them, x btp is the position of the sintering end point.
6. The method for predicting the sintering end point position according to claim 1, characterized in that: The step S300 is specifically as follows: taking L as the window size and 1 as the step size, for the historical soft measurement data of the exhaust gas temperature inflection point, the exhaust gas temperature rising point and the sintering end point at any moment, take the soft measurement data of this moment and the previous L historical moments as a two-dimensional input in the form of 3*L, take the soft measurement data of T moments after this moment as a two-dimensional output in the form of 3*T, and such input and output constitute the sample of this moment.
7. The method for predicting the sintering end point position according to claim 1, characterized in that: In the step S400, the channel-independent temporal convolutional network sintering endpoint position prediction model includes temporal embedding, a backbone network for extracting temporal, feature, and channel dimension information, and a task head for performing prediction. The construction process is specifically as follows: a temporal convolutional network TCN is used for modeling, and its temporal embedding adopts Patch Embedding technology to divide the input sequence into multiple subsequences of fixed size and map them to the target embedding space. This process processes each variable separately to achieve channel independence; the backbone network is formed by stacking multiple backbone modules, and the timing, feature, and channel information are extracted through deep separable convolution. In the first step, each backbone module extracts variable timing information through a deep convolution DWConv module using a large convolution kernel. In the second step, two grouped convolution modules ConvFFN are used to extract feature interaction information and channel interaction information respectively, and then a backbone module output is obtained through residual connection; finally, the output of the backbone network is flattened and mapped to the prediction sequence of each variable through the prediction head.
8. The method for predicting the sintering end point position according to claim 7, characterized in that: The data processing steps of the Embedding layer are as follows: for an input X containing M variables of length L and shape (M, L) in , and the dimension is increased to facilitate mapping to the embedding space, and the input X with shape (M, 1, L) is obtained. in ; Perform padding and patching operations on each variable sequence, and select a sequence of length P as a subsequence every step S. In particular, for X in The last value of is repeated (PS) times as a subsequence, then the length of the new sequence is N = [L / S], [·] represents the rounding operation, and the input X with shape (M, 1, N) is obtained. in ; Through one-dimensional convolution, each channel (variable) is kept independent, and the input is mapped to the target embedding space with dimension D, where D is the feature dimension after embedding, and the input X with shape (M, D, N) is obtained. emb ; The backbone network is composed of multiple backbone modules stacked together, including DWConv modules, ConvFFN1 modules and ConvFFN2 modules connected in sequence. The deep separable convolution technology is used to learn the time series information of variables / features and channel interactions in two separate modules. The DWConv module implements the first step in the deep separable convolution, merging the channel dimension and the feature dimension to obtain data with a shape of (M*D). One-dimensional convolution with a large convolution kernel is used to capture long-distance dependency information in the time dimension. ConvFFN1 and ConvFFN2 are designed for information learning of different channel dimensions. The ConvFFN module consists of two 1*1 point convolution PWConv modules. The point convolution uses grouped convolution technology. The two ConvFFN modules use different grouping logics: the number of groups in ConvFFN1 is set to M, indicating that the feature interaction information of the same variable is learned, and the number of groups in ConvFFN2 is set to D, indicating that the variable interaction information of the same feature is learned. After two ConvFFN modules, the output result of a backbone module is obtained through residual connection. After passing through the backbone network, the output of shape (M, D, N) is obtained, which is flattened in the flattening layer to obtain an output of shape (M, D*N), and then fully connected linear mapping is performed in the task head to obtain a prediction sequence output (M, T) with a target length of T; In the input stage, each input data is Z-standardized and the mean and standard deviation are recorded; In the output stage, each output data is denormalized.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the sintering endpoint position prediction method according to any one of claims 1 to 8 are implemented.
10. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the sintering endpoint position prediction method according to any one of claims 1 to 8 are implemented.