Converter gas holder position prediction method and device and storage medium

Abstract

The embodiment of the invention provides a holder position prediction method and device for a converter gas holder and a storage medium, and belongs to the field of holder position prediction of converter gas holders. The holder position prediction method comprises the following steps: acquiring historical production data and historical holder position data of a converter gas holder; the correlation degree between the historical production data and the historical cabinet position data is calculated, the historical production data with the correlation degree meeting a preset threshold value is screened out, and the historical production data and the historical cabinet position data are jointly input into a cabinet position prediction model; and on the basis of the cabinet position prediction model, encoding input data into a global context vector with a fixed length, and generating a multi-step cabinet position prediction sequence according to the global context vector. The method solves the problem of low prediction accuracy caused by the influence of working condition change on the converter gas holder in traditional time sequence prediction, effectively improves the prediction precision and efficiency, and is suitable for intelligent scheduling and safety control of a converter gas system.

Description

Technical Field

[0001] This invention relates to the field of converter gas holder position prediction technology, specifically to a converter gas holder position prediction method, equipment and storage medium. Background Technology

[0002] Converter gas is an important secondary energy source in the steel production process, and its efficient recovery and precise control are of great significance for reducing energy consumption and carbon emissions. As a key buffer device, the gas holder is responsible for the storage and regulation of gas, and accurate prediction of its position is fundamental to achieving balanced gas scheduling and energy management.

[0003] Because converter gas recovery is an intermittent operation, and converter gas consumption is affected by changes in operating conditions, the amount of gas in the gas holder fluctuates significantly and is difficult to predict accurately. Currently, gas management mainly relies on manual dispatching, with dispatchers typically making predictions based on experience. However, this method suffers from latency and low prediction accuracy. Furthermore, due to differences in operator experience, prediction results are prone to inconsistencies, affecting the efficiency and accuracy of dispatching.

[0004] In the prior art, such as the invention patent with publication number CN103942422A, "a converter gas holder for metallurgical enterprises based on particle size calculation" is disclosed. This method uses the fuzzy C-means method to cluster data to obtain membership degree and cluster center, and then performs fuzzy inference and makes predictions through the center defuzzification method.

[0005] However, the fuzzy C-means clustering method used in this method requires manual setting of initial values ​​and number of clusters, which has a certain degree of subjectivity. Moreover, as an unsupervised learning method, it requires high-quality data. However, sensors in the steel industry often need to work in harsh environments with high temperature, high pressure and high corrosion. Over time, sensors inevitably experience abnormal data loss, noise and other issues, and the directly transmitted data is difficult to meet the requirements.

[0006] Secondly, the prediction process of traditional autoregressive models involves iteration, and the prediction error increases with each iteration, resulting in low prediction accuracy. Furthermore, while traditional autoregressive models perform relatively accurately under normal operating conditions, the prediction results often deviate when operating conditions change, significantly reducing accuracy. Summary of the Invention

[0007] The purpose of this invention is to provide a method, device, and storage medium for predicting the position of converter gas holders, in order to solve the technical problems existing in the background art, realize multi-step prediction of converter gas holder positions, provide scientific data support for subsequent scheduling, and reduce enterprise operating costs.

[0008] To achieve the above objectives, embodiments of the present invention provide a method for predicting the position of a converter gas holder. The method includes: acquiring historical production data and historical position data of the converter gas holder, wherein the historical production data includes production and consumption user data and pressure data of the converter gas holder; calculating the correlation between the historical production data and the historical position data, filtering out historical production data whose correlation meets a preset threshold, and inputting them together with the historical position data into a position prediction model; and based on the position prediction model, encoding the input data into a fixed-length global context vector, and generating a multi-step position prediction sequence based on the global context vector.

[0009] Optionally, before calculating the correlation between the historical production data and the historical counter location data, the counter location prediction method further includes: performing data removal, filling, and normalization processing on the historical production data and the historical counter location data.

[0010] Optionally, the data removal and filling includes: replacing the data identified as abnormal in the historical production data and historical counter data with the average data of the previous and next time moments, and / or filling the missing data at a certain time moment by using linear interpolation. The normalization process includes: performing a linear transformation on the data after the data removal and imputation process, mapping the data to the range [0,1]. The calculation formula is as follows:

[0011] in, This represents the normalized data. This indicates the data after data removal and filling processes. This indicates the maximum value in the data after data removal and imputation. This represents the minimum value in the data after data removal and imputation.

[0012] Optionally, calculating the correlation between the historical production data and the historical counter location data includes: calculating the difference between the historical production data and the historical counter location data, and determining the extreme value of the difference; calculating the correlation coefficient based on the extreme value, and calculating the correlation between the historical production data and the historical counter location data based on the correlation coefficient.

[0013] Optionally, the formula for calculating the correlation coefficient is:

[0014] in, Represents the correlation coefficient. This represents historical counter location data. Represents historical production data. This represents the resolution coefficient.

[0015] Optionally, the correlation coefficient is used to calculate the correlation degree between the historical production data and the historical counter location data, and the calculation formula is as follows:

[0016] in, Indicates the degree of relevance. Indicates the total length of the data sequence. Indicates the first The correlation coefficient at each time point.

[0017] Optionally, the cabinet prediction model is a long short-term memory network model based on an encoder-decoder framework, including an encoder LSTM and a decoder LSTM, wherein the fully connected layers of the long short-term memory network model are embedded with dropout layers to suppress overfitting during training.

[0018] Optionally, encoding the input data into a fixed-length global context vector and generating a multi-step cabinet location prediction sequence based on the global context vector includes: processing the input data sequentially according to time steps using the encoder LSTM, encoding the temporal features in the input data and the influence relationships between data to obtain the fixed-length global context vector; the decoder LSTM using the global context vector as the initial state to progressively generate prediction data for each time step, and outputting the prediction data for all time steps through a time-distributed fully connected layer to obtain the multi-step cabinet location prediction sequence.

[0019] Optionally, after generating the multi-step cabinet location prediction sequence based on the global context vector, the method further includes: performing inverse normalization calculation on the multi-step cabinet location prediction sequence to obtain the actual prediction value.

[0020] Secondly, embodiments of the present invention also provide a converter gas holder position prediction device, the position prediction device comprising: a memory; and a processor configured to execute instructions stored in the memory to perform the above-described converter gas holder position prediction method.

[0021] Thirdly, embodiments of the present invention also provide a machine-readable storage medium storing instructions for causing a machine to execute the converter gas holder position prediction method of any of the above claims.

[0022] Through the above technical solution, the embodiments of the present invention introduce correlation analysis to filter multivariate historical production data, effectively identify and retain the key influencing factors most relevant to cabinet position fluctuations, reduce the redundancy and noise of model input, and provide high-quality data input for subsequent prediction.

[0023] Furthermore, a prediction model based on an encoder-decoder architecture is adopted. The encoder encodes the filtered multivariate time-series data into a fixed-length context vector with global information. The decoder then directly generates future multi-step cabinet position sequences based on this vector. This effectively avoids the error accumulation problem caused by step-by-step iteration in traditional rolling prediction, achieving accurate mapping from historical sequences to future multi-step sequences. Overall, this invention improves prediction accuracy and stability while enhancing the model's adaptability to operating condition fluctuations, providing reliable technical support for real-time scheduling and safety control of gas systems.

[0024] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0025] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a method for predicting the position of a converter gas holder provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the converter gas system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the process for calculating grey relational degree provided in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the algorithm principle of the LSTM network provided in this embodiment of the invention; Figure 5 This is a schematic diagram illustrating the algorithm principle of the cabinet location prediction model provided in this embodiment of the invention; Figure 6 This is a flowchart illustrating another method for predicting the position of a converter gas holder provided in an embodiment of the present invention. Detailed Implementation

[0026] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0027] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with relevant laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0028] Figure 1 This is a flowchart illustrating a method for predicting the position of a converter gas holder according to an embodiment of the present invention. Figure 1 As shown, the cabinet location prediction method includes steps S1-S3: Step S1: Obtain historical production data and historical gas holder location data of the converter gas holder. The historical production data includes the production, consumption, and user data of the converter gas holder, as well as the pressure data of the gas holder. Step S2: Calculate the correlation between the historical production data and the historical counter location data, filter out the historical production data whose correlation meets the preset threshold, and input them together with the historical counter location data into the counter location prediction model; Step S3: Based on the cabinet location prediction model, the input data is encoded into a fixed-length global context vector, and a multi-step cabinet location prediction sequence is generated based on the global context vector.

[0029] In step S1, the historical production data of the converter gas holder includes the production and consumption user data of the converter gas holder and the holder pressure data. The production and consumption user data refers to the data of the production users and the consumption users. Figure 2 This is a schematic diagram of the converter gas system provided in an embodiment of the present invention. Specifically, as shown... Figure 2 As shown, the converter gas system includes the converter gas recovery main as the production user, and the consuming users include sintering, desulfurization and denitrification, two hot blast stoves, primary dust removal for steelmaking, converter ladle baking, continuous casting ladle baking, bar and wire rod co-firing, double high-temperature wire rod, double high-temperature wire rod co-firing, single high-temperature wire rod, single high-temperature wire rod co-firing, lime kiln, pellet plant, venting tower and two generator sets.

[0030] In this embodiment, the acquired production and consumption user data, cabinet pressure data, and historical cabinet location data are historical data from the past month, with a collection interval of 1 minute.

[0031] Because sensors in the steel industry often operate in harsh environments with high temperatures, high pressures, and high corrosion, they inevitably experience data loss and damage over time, requiring replacement. Problematic data poses challenges to model training; therefore, it is necessary to identify and process outliers in the original data to ensure data reliability.

[0032] Therefore, in some embodiments, before calculating the correlation between historical production data and historical counter location data, the counter location prediction method further includes: performing data removal, filling, and normalization processing on the historical production data and historical counter location data.

[0033] Furthermore, the data removal and filling includes: replacing the data identified as abnormal in the historical production data and historical counter data with the average data of the previous and next time moments, and / or filling the missing data at a certain time moment using linear interpolation. Data deemed abnormal refers to outliers that do not conform to common sense, such as unrealistic negative values ​​or values ​​outside the normal range. For sintering data, the consumption range should be between 0 and 14000; values ​​outside this range are considered outliers.

[0034] Different characteristics affecting coal gas typically have different dimensions and orders of magnitude, and the significant differences between them can greatly impact data analysis. Therefore, a normalization method is used to transform dimensional expressions into dimensionless expressions to eliminate the influence between different dimensions.

[0035] The normalization process includes: performing a linear transformation on the data after the data removal and imputation process, mapping the data to the range [0,1]. The calculation formula is as follows:

[0036] in, This represents the normalized data. This indicates the data after data removal and filling processes. This indicates the maximum value in the data after data removal and imputation. This represents the minimum value in the data after data removal and imputation.

[0037] The MinMaxScaler function is a tool in Scikit-Learn used for data normalization. In this embodiment, this function is used for normalization and denormalization.

[0038] Many factors affect the fluctuation of the converter gas holder position, such as the gas inflow rate affected by raw material conditions, air temperature and pressure, the gas consumption rate affected by equipment operating conditions, and the holder pressure, all of which directly affect the rise and fall of the gas holder position.

[0039] Therefore, in step S2, the correlation degree between historical production data and historical counter data is calculated using the grey relational algorithm to eliminate users with low correlation degree, thereby simplifying the data input of the model, avoiding the impact of redundant input on modeling accuracy, and improving the calculation speed.

[0040] Figure 3 This is a schematic diagram of the process for calculating grey relational degree provided in an embodiment of the present invention. Figure 3 As shown, the specific process of calculating the correlation between historical production data and historical counter location data using the grey relational analysis algorithm includes: Obtain preprocessed historical production data and historical cabinet location data; By calculating the difference between historical production data and historical counter location data, a difference sequence is constructed, and the extreme values ​​in the difference are determined. Based on the extreme values, the correlation coefficient is calculated, and the correlation degree between historical production data and historical counter location data is calculated based on the correlation coefficient. Based on the correlation, historical production data with a correlation greater than a preset threshold are selected as the main influencing factors of counter location changes and output to the counter location prediction model.

[0041] Specifically, set historical cabinet data As the reference sequence, i.e., the parent sequence, historical production data is represented as follows: .in, and These represent the length of their corresponding sequences, which is also the total number of data points.

[0042] The formula for calculating the correlation coefficient is:

[0043] in, Represents the correlation coefficient. This represents historical counter location data. Represents historical production data. Represents the resolution coefficient. Its function is to improve the significance of correlation coefficients, ranging from 0 to 1, and is generally taken as 0.5.

[0044] The correlation coefficient is used to calculate the correlation between historical production data and historical counter location data. The calculation formula is as follows:

[0045] in, Indicates the degree of relevance. Indicates the total length of the data sequence. Indicates the first The correlation coefficient at each time point.

[0046] In this embodiment, the top nine historical production data with the highest correlation coefficients are selected as input data through grey relational analysis. Specifically, these data include: sintering, gasification, generator set #1, generator set #2, primary dust removal in steelmaking, cabinet pressure, bar and wire rod co-firing, continuous casting ladle baking, and sintering desulfurization and denitrification. Finally, the input data also includes historical cabinet position data, because the cabinet position at the next moment is affected by the cabinet position data at the previous moment.

[0047] LSTM networks are widely used in time series forecasting and are a special type of recurrent neural network (RNN) designed to address the vanishing and exploding gradient problems that standard RNNs encounter when handling long-term dependencies. LSTM networks improve performance when processing time series data by introducing "memory units" to retain and update information. As a supervised model, the LSTM algorithm has been applied in other industrial fields with good results.

[0048] LSTM networks can remember information for a longer period of time by using memory cells, input gates, output gates, and forget gates. These gates control the flow of information into and out of the memory cells, enabling the network to selectively store and retrieve information as needed. Figure 4 This is a schematic diagram illustrating the algorithm principle of the LSTM network provided in this embodiment of the invention. Figure 4 As shown, an LSTM network mainly consists of four components (gates) that control the flow of information: The ForgetGate controls what is retained and forgotten, determining which information is discarded from the cell state. Its input is the hidden state from the previous time step. and the input at the current moment The output is a value between 0 and 1, representing the degree to which information in each cell is retained (1 means fully retained, 0 means completely forgotten).

[0049] The calculation formula is as follows:

[0050] in It's the Sigmoid activation function. The forget gate gives... .

[0051] The input gate controls the impact of the current input, determining which new information is added to the memory unit. Its input includes the hidden state from the previous time step. and the input at the current moment The output is the new candidate memory unit. It is used to update memory units.

[0052] The calculation formula is as follows:

[0053]

[0054] The components retained after passing through the input gate are . The value can be limited to between -1 and 1.

[0055] The `CellStateUpdate` function updates the state of a memory cell using the forget gate and the input gate. The state of a memory cell is updated from the previous state. Update to current state Specifically:

[0056] Among them, the forget gate determines how much of the memory state from the previous moment is retained, and the input gate determines the amount of information added at the current moment.

[0057] The output gate is used to control the final output, i.e., the hidden state. This output will be passed to the LSTM unit at the next time step and to the final output layer of the model. Its input is the memory state at the current time step. and the hidden state of the previous moment .

[0058]

[0059]

[0060] However, the traditional LSTM algorithm has limitations in predicting converter gas holders. First, the rolling prediction model is prone to error accumulation, and the accuracy decreases as the step size increases. Second, due to the large number of parameters, it is easy to overfit in noisy industrial data, and gradient explosion is prone to occur during training, resulting in insufficient stability. Finally, the algorithm has strict requirements on the length of the input sequence and is difficult to adapt to variable-length sequences caused by fluctuations in operating conditions.

[0061] Therefore, this invention provides a cabinet location prediction model, which is a Long Short-Term Memory (LSTM) network model based on an encoder-decoder framework. The encoder-decoder structure serves as the core framework of this model, incorporating an LSTM network to achieve a direct mapping from historical sequences to future sequences. Specifically, two independent LSTM networks are used to handle feature encoding and sequence generation tasks respectively.

[0062] Figure 5 This is a schematic diagram illustrating the algorithm principle of the cabinet location prediction model provided in this embodiment of the invention. Figure 5 As shown, the historical production data and historical counter location data, after preprocessing and correlation filtering, are input into the counter location prediction model, and the input data are encoded by the encoder LSTM part.

[0063] Specifically, the input data is in the form of a multivariate sequence, and the LSTM encoder sequentially processes the historical multivariate sequence (e.g., [t1, t2, ..., t...).m [The aforementioned 10 key variables at time step 1]. In this process, the encoder LSTM, through its internal gating mechanism, encodes the temporal features of the input data and the influence relationships between data points to deeply capture temporal correlations and the influence patterns of key factors, selectively retaining important information, and finally, at the last time step, its internal state (hidden state h) is revealed. m and cell state c m The output is a fixed-length global context vector, which includes the correlation patterns and global features between all key variables and cabinet positions throughout the entire historical window, providing support for subsequent predictions. This vector is then used as the encoder's state vector and input into the decoder LSTM.

[0064] The decoder LSTM, also comprising an LSTM network, has the core task of generating multi-step prediction sequences based on the global context vector provided by the encoder LSTM. This part no longer directly receives the original historical sequence, but instead uses the context vector output by the encoder LSTM as its initial hidden states and cell states. In this way, the encoder obtains vectors containing complete global features of the historical data before decoding.

[0065] Furthermore, traditional forecasting methods typically employ an autoregressive rolling forecasting mechanism. The specific process is as follows: first, the unit value for the first future time step is predicted based on historical sequences; then, this predicted value is fed back into the same model as part of the input to predict the unit value for the next time step; this process is iterated repeatedly to gradually generate forecast sequences for all future time steps. This iterative information processing method inevitably causes prediction errors from previous steps to propagate and accumulate in subsequent steps, resulting in a significant decrease in prediction accuracy as the prediction step size increases.

[0066] The prediction model based on the encoder-decoder framework used in this invention has a fundamentally different information processing logic. The encoder LSTM part processes the entire historical multivariate sequence at once, and encapsulates its temporal features into a fixed-length context vector. This context vector, as a global summary of historical information, is directly provided to the decoder LSTM part. Based on this complete context information, the decoder directly generates the prediction sequence for all future steps through the autoregressive evolution of its internal states. This process achieves an end-to-end one-time mapping from the historical sequence to the future sequence, rather than step-by-step iteration, thus avoiding the cumulative effect caused by the gradual propagation of errors in principle.

[0067] Specifically, during the training phase, the decoder LSTM receives two pieces of information at each time step: the actual cabinet position value of the previous step and the state obtained from the encoder LSTM and evolving over time. Based on this information, the cabinet position of the current step is predicted.

[0068] The loss function calculates the entire predicted sequence (e.g., t). m+1 to t m+n The overall error between the encoder and the true sequence is calculated, and all parameters of both the encoder and decoder are optimized simultaneously through backpropagation. This allows the encoder to learn to generate context vectors that are most helpful in predicting the future.

[0069] During the prediction phase, the decoder uses its own prediction output from the previous step as input for the current step, and combines it with its internal state to autoregressively generate a complete multi-step sequence. Since the decoder's initial state is an information-rich context vector, rather than a zero state, its generation process is always guided by historical patterns, effectively avoiding error drift from zero.

[0070] In addition, the cabinet prediction model provided by this invention adds Dropout regularization to the fully connected layer of the LSTM network to enhance the model's generalization ability on industrial data containing noise and outliers, and to prevent overfitting.

[0071] The multi-step gas holder prediction sequence generated by the gas holder prediction model is inversely normalized to obtain the actual prediction value, thus completing the prediction of the converter gas holder position.

[0072] Based on the above, this invention fully considers all relevant factors affecting converter gas holders, removes abnormal data through data preprocessing, interpolates missing data, and then uses the grey relational analysis method to rank all factors by correlation degree, selecting the key factors most relevant to the holder location. The provided holder location prediction model effectively solves the overfitting problem by adding Dropout regularization to the LSTM network and integrates the LSTM network into the encoder-decoder framework to achieve end-to-end direct mapping from historical segments to future trend segments. This flexibly adapts to changes in sequence length caused by operating condition fluctuations, eliminating the need for frequent model adjustments and demonstrating strong practicality. By predicting holder location trends over a future period, it can issue early warnings before the holder approaches the safety threshold, giving operators sufficient time to intervene.

[0073] Figure 6 This is a flowchart illustrating another method for predicting the position of a converter gas holder provided in an embodiment of the present invention. Figure 6 As shown, the method includes: Step S101: Obtain data, which is the historical production data and historical position data of the converter gas holder. The historical production data includes the production and consumption user data and the holder pressure data of the converter gas holder. The production and consumption user data is the data of the production user and the consumption user. In this embodiment, taking the actual production data of a converter gas system in a large steel enterprise as an example, we will analyze an 80,000 kWh converter gas system. The gas holder is used to predict the gas pressure for the next half hour. The converter gas system includes the converter gas recovery main as the production user, and the consuming users include sintering, desulfurization and denitrification, two hot blast stoves, primary dust removal in steelmaking, converter ladle baking, continuous casting ladle baking, bar and wire rod co-firing, double high-pressure wire rod, double high-pressure wire rod co-firing, single high-pressure wire rod, single high-pressure wire rod co-firing, lime kiln, pellet plant, venting tower and two generator sets. At the same time, the position and pressure data of the gas holder need to be measured and recorded. The data collection period is one month, and the data collection interval is 1 minute.

[0074] Step S102: Data preprocessing, including data removal, imputation, and normalization; Specifically, negative values ​​and inaccurate data in the acquired data are replaced with the average of the data before and after them. Missing data is then filled in using interpolation to ensure data validity. Finally, the data after removal and filling is normalized to facilitate subsequent calculations.

[0075] Step S103: For the preprocessed data, calculate the correlation between historical production data and historical counter data using grey relational analysis, and filter out historical production data whose correlation meets the preset threshold. Specifically, the process of calculating the correlation between historical production data and historical counter location data using the grey relational analysis algorithm includes: Obtain preprocessed historical production data and historical cabinet location data; By calculating the difference between historical production data and historical counter location data, a difference sequence is constructed, and the extreme values ​​in the difference are determined. Based on the extreme values, the correlation coefficient is calculated, and the correlation degree between historical production data and historical counter location data is calculated based on the correlation coefficient. Based on the correlation, historical production data with a correlation greater than a preset threshold are selected as the main influencing factors of counter location changes and output to the counter location prediction model.

[0076] Step S104: Divide the selected data into training data and test data. Train the LSTM model based on encoder-decoder using the training data. For the trained model, perform data prediction using the test data and output the predicted data. Specifically, the encoder-decoder-based LSTM model uses an encoder-decoder structure as its core framework, integrating an LSTM network to achieve a direct mapping from historical sequences to future sequences. This is achieved by using two independent LSTM networks to handle feature encoding and sequence generation, respectively. Furthermore, Dropout regularization is added to the fully connected layers of the LSTM network to enhance the model's generalization ability on noisy and outlier industrial data, preventing overfitting.

[0077] Step S105: Perform inverse normalization on the predicted data to obtain the actual predicted value of the predicted data.

[0078] Secondly, embodiments of the present invention also provide a converter gas holder position prediction device, the position prediction device comprising: a memory; and a processor configured to execute instructions stored in the memory to perform the above-described converter gas holder position prediction method.

[0079] Thirdly, embodiments of the present invention also provide a machine-readable storage medium storing instructions for causing a machine to execute the converter gas holder position prediction method of any of the above claims.

[0080] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0081] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0082] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0083] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0084] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0085] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0086] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0087] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0088] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for predicting the position of a converter gas holder, characterized in that, The cabinet location prediction method includes: Obtain historical production data and historical gas holder location data for converter gas holders. The historical production data includes the production, consumption, and user data of converter gas holders, as well as the gas holder pressure data. Calculate the correlation between the historical production data and the historical counter location data, filter out historical production data whose correlation meets a preset threshold, and input them together with the historical counter location data into the counter location prediction model; and Based on the aforementioned counter location prediction model, the input data is encoded into a fixed-length global context vector, and a multi-step counter location prediction sequence is generated based on this global context vector.

2. The cabinet location prediction method according to claim 1, characterized in that, Before calculating the correlation between the historical production data and the historical counter location data, the counter location prediction method further includes: The historical production data and historical counter data are processed by data removal, filling, and normalization.

3. The cabinet location prediction method according to claim 2, characterized in that, The data removal and filling includes: replacing the data identified as abnormal in the historical production data and historical counter data with the average data of the previous and next time moments, and / or filling the missing data at a certain time moment using linear interpolation. The normalization process includes: performing a linear transformation on the data after the data removal and imputation process, mapping the data to the range [0,1]. The calculation formula is as follows: in, This represents the normalized data. This indicates the data after data removal and filling processes. This indicates the maximum value in the data after data removal and imputation. This represents the minimum value in the data after data removal and imputation.

4. The cabinet location prediction method according to claim 1, characterized in that, The calculation of the correlation between the historical production data and the historical counter location data includes: Calculate the difference between the historical production data and the historical cabinet location data, and determine the extreme values ​​of the difference; Based on the extreme values, a correlation coefficient is calculated, and the correlation degree between the historical production data and the historical counter data is calculated based on the correlation coefficient.

5. The cabinet location prediction method according to claim 4, characterized in that, The formula for calculating the correlation coefficient is: in, Represents the correlation coefficient. This represents historical counter location data. Represents historical production data. This represents the resolution coefficient.

6. The cabinet location prediction method according to claim 4, characterized in that, The correlation coefficient is used to calculate the correlation degree between the historical production data and the historical counter location data. The calculation formula is as follows: in, Indicates the degree of relevance. Indicates the total length of the data sequence. Indicates the first The correlation coefficient at each time point.

7. The cabinet location prediction method according to claim 1, characterized in that, The cabinet prediction model is a long short-term memory network model based on an encoder-decoder framework, including an encoder LSTM and a decoder LSTM. The fully connected layers of the long short-term memory network model are embedded with dropout layers to suppress overfitting during training.

8. The cabinet location prediction method according to claim 7, characterized in that, The step of encoding the input data into a fixed-length global context vector and generating a multi-step cabinet location prediction sequence based on the global context vector includes: The encoder LSTM processes the input data sequentially according to time steps, encoding the temporal features and the influence relationships between data in the input data to obtain the fixed-length global context vector. The decoder LSTM uses the global context vector as the initial state to generate prediction data for each time step step by step. The prediction data for all time steps are output through a time-distributed fully connected layer to obtain the multi-step cabinet prediction sequence.

9. The cabinet location prediction method according to claim 1, characterized in that, After generating a multi-step cabinet location prediction sequence based on the global context vector, the method further includes: performing inverse normalization calculation on the multi-step cabinet location prediction sequence to obtain the actual prediction value.

10. A converter gas holder position prediction device, characterized in that, The cabinet location prediction device includes: Memory; and The processor is configured to execute instructions stored in memory to perform the converter gas holder position prediction method according to any one of claims 1-9.

11. A machine-readable storage medium storing instructions for causing a machine to perform the converter gas holder position prediction method of any of the preceding claims.

Images