Back pressure prediction optimization method and system for oxygen-nitrogen liquefaction device

Abstract

The application provides a back pressure prediction optimization method and system of an oxygen-nitrogen liquefaction device, the method comprising: acquiring real-time operation parameters of an air separation device and the oxygen-nitrogen liquefaction device in a target period, and an environmental temperature parameter corresponding to the period; performing data preprocessing on the real-time operation parameters and the environmental temperature parameter, extracting statistical features, and generating standardized feature samples; based on SPC statistical process control theory, performing process stability determination and degradation trend identification on the standardized feature samples, and outputting a device operation state result; inputting the standardized feature samples and the device operation state result into a pre-trained BP neural network back pressure prediction model, and outputting a back pressure prediction value of the oxygen-nitrogen liquefaction device in a next period; and based on the back pressure prediction value, combining preset oxygen-nitrogen ratio process rules and air separation load scheduling logic, generating a back pressure optimization adjustment scheme and abnormal early warning information. The scheme combines SPC and a BP neural network, can identify a device operation degradation trend, and realizes back pressure prediction, optimization and early warning.

Description

Technical Field

[0001] This application relates to the field of back pressure prediction technology for air separation liquefaction units, and more specifically, to a method and system for back pressure prediction optimization of oxygen and nitrogen liquefaction units. Background Technology

[0002] In the iron and steel metallurgical industry, the combined production mode of air separation units and oxygen-nitrogen liquefaction units is widely used. The oxygen and nitrogen produced by the air separation unit are transported to the oxygen-nitrogen liquefaction unit for liquefaction. The compressor back pressure of the liquefaction unit is a key process parameter that directly affects the unit's energy consumption, product output, and equipment operating safety. Abnormal back pressure fluctuations not only cause the release of nitrogen and oxygen gas on the air separation side, increasing production losses, but also put the liquefaction unit under high-cost operating conditions, and in severe cases, lead to equipment failure.

[0003] Currently, most production sites use traditional monitoring platforms to monitor oxygen and nitrogen liquefaction units. These platforms can only display process parameter curves and provide alarms for parameters exceeding limits. Alarms are only triggered when process parameters deviate from the allowable range, failing to provide early warnings of faults or identify trends in performance degradation. Operators can only react passively after an alarm occurs, often missing the optimal time for adjusting operating conditions and intervening in faults. This can easily lead to problems such as large-scale nitrogen and oxygen gas emissions from the air separation unit and long-term high-energy-consumption operation of the liquefaction unit, making it difficult to meet the steel industry's demand for efficient and low-cost production control of air separation-liquefaction combined systems.

[0004] Furthermore, existing technologies lack a technical solution that combines process statistical analysis with intelligent prediction algorithms to conduct trend analysis and predictive optimization of back pressure in liquefaction units. This makes it difficult to predict subsequent changes in back pressure in advance and to proactively provide back pressure optimization scheduling solutions that are suitable for the current operating conditions. Relying on manual experience for load adjustment results in strong adjustment lag, which restricts the further improvement of the production efficiency and economic operation level of oxygen and nitrogen liquefaction units.

[0005] Therefore, there is an urgent need for a technical solution that can identify the degradation trend of the device and realize back pressure prediction and early warning. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the purpose of this application is to provide a back pressure prediction and optimization method and system for oxygen and nitrogen liquefaction units. By combining SPC statistical process control theory with BP neural network back pressure prediction model, the method can identify the deterioration trend of unit operation, realize back pressure prediction, optimization and anomaly early warning, reduce nitrogen and oxygen gas venting and emission, reduce the operating cost of oxygen and nitrogen liquefaction units, and improve production management efficiency.

[0007] To solve the above-mentioned technical problems, this application provides the following technical solution: According to a first aspect of the embodiments of this application, a method for predicting and optimizing the back pressure of an oxygen-nitrogen liquefaction device is provided, the method comprising: Step S1: Obtain the real-time operating parameters of the air separation unit and the nitrogen-oxygen liquefaction unit during the target time period, as well as the ambient temperature parameters for the corresponding time period; Step S2: Perform data preprocessing on the real-time operating parameters and the ambient temperature parameters, extract statistical features, and generate standardized feature samples; Step S3: Based on SPC statistical process control theory, determine the process stability and identify the degradation trend of the standardized feature samples, and output the device operation status results; Step S4: Input the standardized feature samples and the device operating status results into the pre-trained BP neural network back pressure prediction model, and output the predicted back pressure value of the oxygen and nitrogen liquefaction device for the next time period; Step S5: Based on the predicted back pressure value, combined with the preset oxygen-nitrogen ratio process rules and air separation load scheduling logic, generate a back pressure optimization adjustment scheme and abnormal early warning information.

[0008] Furthermore, the real-time operating parameters include parameters of the target air separation unit, parameters of other air separation units, and parameters of the liquefaction unit; The target air separation unit parameters include hourly oxygen production, hourly nitrogen production, oxygen venting opening degree, nitrogen venting opening degree, air compressor power, hourly air compressor power consumption, booster motor power consumption, and whether the nitrogen level has reached the target value. The other air separation unit parameters include the hourly oxygen production, the hourly nitrogen production, the oxygen venting opening degree, the nitrogen venting opening degree, the air compressor power, and the air compressor hourly power consumption; The parameters of the liquefaction unit include hourly output of liquid oxygen, hourly output of liquid nitrogen, compressor power, and historical value of compressor back pressure. The ambient temperature parameter is used to define the nitrogen and oxygen production threshold, which serves as the basis for the judgment of the air separation load scheduling logic. Wherein, the target air separation unit parameters are the operating parameters corresponding to the target air separation unit, and the other air separation unit parameters are the operating parameters corresponding to air separation units operating other than the target air separation unit.

[0009] Further, step S2 specifically includes: The real-time operating parameters and the ambient temperature parameters obtained in step S1 are subjected to outlier detection and missing value processing respectively. Abnormal operating condition parameters belonging to shutdown and debugging are removed from the real-time operating parameters, and the ambient temperature parameters corresponding to the abnormal operating condition parameters at the same time are discarded simultaneously. The statistical features are extracted from the real-time operating parameters after the outlier detection, missing value processing and abnormal operating condition parameter removal are completed, and the sample statistical duration is adjusted according to different parameter types to generate feature samples with different time granularities. The feature samples, together with the environmental temperature parameters after the outlier detection and missing value processing, are standardized to obtain the standardized feature samples.

[0010] Furthermore, the step of extracting statistical features from the real-time operating parameters after completing the outlier detection, missing value processing, and abnormal operating condition parameter removal specifically includes: For the real-time operating parameters that have exceeded the limits after the completion of the outlier detection, missing value processing, and removal of abnormal operating condition parameters, the mean value of the time period is extracted. For the real-time operating parameters that have fluctuated after the completion of the outlier detection, missing value processing, and removal of abnormal operating condition parameters, the standard deviation of the time period is extracted.

[0011] Furthermore, the process stability determination in step S3 adopts the eight SPC discrepancy criteria, specifically including: Criterion 1: A single sample point falls outside control limit A. Criterion 2: Nine consecutive sample points fall on the same side of the center line; Criterion 3: Six consecutive sample points increasing or decreasing; Criterion 4: Adjacent points in a series of 14 consecutive sample points fluctuate alternately up and down; Rule 5: Two out of three consecutive sample points fall outside zone B on the same side of the center line; Criterion 6: Four out of five consecutive sample points fall outside zone C on the same side of the center line; Rule 7: 15 consecutive sample points fall within area C on both sides of the center line; Criterion 8: Eight consecutive sample points fall on either side of the center line and none of them are located within region C.

[0012] Furthermore, in step S3, based on the process stability determination using the eight SPC discrepancy criteria, the upper and lower limit constraints of the process are verified, and the statistical granularity of a single sample is adjusted according to the rate of change of different process parameters.

[0013] Furthermore, step S5 specifically includes: The nitrogen and oxygen production thresholds are defined based on the ambient temperature parameters. When the ambient temperature parameters are greater than a preset temperature, the first oxygen production threshold is used as the dividing point. When the ambient temperature parameters are less than or equal to the preset temperature, the second oxygen production threshold is used as the dividing point. Based on the oxygen venting status, nitrogen venting status, and the oxygen-nitrogen ratio characteristics of the target air separation unit and the other air separation units, calculate the load increment of the corresponding air separation unit. Based on the calculated load increment, a power output adjustment strategy for the target air separation unit and / or the other air separation units is determined to match the back pressure of the oxygen and nitrogen liquefaction unit to the optimal operating range, and the back pressure optimization adjustment scheme is generated accordingly. When the predicted back pressure value deviates from the preset safety boundary, the corresponding level of abnormal warning information is generated and output. Wherein, the first oxygen production is less than the second oxygen production; the oxygen-nitrogen ratio characteristic is determined based on the oxygen-nitrogen ratio process rules.

[0014] Furthermore, the pre-trained BP neural network backpressure prediction model in step S4 is a three-layer feedforward neural network structure, including an input layer, a hidden layer, and an output layer. Adjacent layer neurons are fully connected, while neurons in the same layer are not connected, and a supervised learning mode is adopted. The hyperparameters of the pre-trained BP neural network backpressure prediction model are configured as learning rate and additional momentum factor. The training process uses the backpropagation algorithm to iteratively optimize the connection weights and biases of each layer.

[0015] Furthermore, the pre-trained BP neural network backpressure prediction model is obtained through the following steps: The historical operating data and corresponding back pressure values ​​of the oxygen-nitrogen liquefaction unit under normal operating conditions are obtained to form the original sample set. Data cleaning, statistical feature extraction, and standardization are performed on the original sample set to obtain a standardized training feature set; The standardized training feature set and the corresponding true back pressure values ​​are divided into a training set and a validation set. The training set is input into the initial BP neural network model, the predicted output value is calculated through forward propagation, and then the connection weights and biases of each layer are adjusted through error backpropagation to iteratively reduce the prediction error. Training stops when the training error meets the preset accuracy requirement or reaches the upper limit of the number of iterations. The model prediction accuracy is verified using the validation set. After the verification is passed, the pre-trained BP neural network backpressure prediction model is obtained. After the model is launched and running, online incremental updates are performed based on the real-time running data to adapt to device performance drift.

[0016] According to a second aspect of the embodiments of this application, a back pressure prediction and optimization system for an oxygen-nitrogen liquefaction device is provided, implemented using any of the back pressure prediction and optimization methods for oxygen-nitrogen liquefaction devices described above, the system comprising: The parameter acquisition module is used to acquire the real-time operating parameters of the air separation unit and the nitrogen-oxygen liquefaction unit within the target time period, as well as the ambient temperature parameters for the corresponding time period; The data processing module is used to preprocess the real-time operating parameters and the ambient temperature parameters, extract statistical features, and generate standardized feature samples. The data determination module is used to determine the process stability and identify the degradation trend of the standardized feature samples based on SPC statistical process control theory, and output the device operation status results. The back pressure prediction module is used to input the standardized feature samples and the device operating status results into a pre-trained BP neural network back pressure prediction model, and output the back pressure prediction value of the oxygen and nitrogen liquefaction device in the next time period. The optimization and adjustment module is used to generate a back pressure optimization and adjustment scheme and anomaly warning information based on the back pressure prediction value, combined with the preset oxygen-nitrogen ratio process rules and air separation load scheduling logic.

[0017] Compared with the prior art, this application has the following beneficial effects: (1) The technical solution of combining SPC statistical process control theory and BP neural network back pressure prediction model is adopted. First, the stability of the device process is determined and the deterioration trend is identified. Then, back pressure prediction is carried out based on the state results. Finally, the optimization scheme and early warning information are output. This breaks through the limitation of traditional monitoring platforms that can only realize post-event alarms when parameters exceed limits. It can identify the deterioration of the working condition and predict the back pressure change in advance, supporting operators to intervene and adjust in advance.

[0018] (2) Based on the eight criteria for SPC, a new process upper and lower limit constraint verification is added to avoid the misjudgment of "the statistical process is stable but the parameters have exceeded the process qualified range". At the same time, it supports adjusting the collection and statistical granularity of a single sample according to the rate of change of different process parameters, adapting to the dynamic characteristics of different parameters, effectively reducing the false alarm rate, and improving the accuracy of process stability and degradation trend identification.

[0019] (3) Perform outlier detection and missing value processing on real-time operating parameters and ambient temperature parameters respectively, and simultaneously remove the entire set of samples corresponding to abnormal operating conditions such as machine tripping and debugging; extract two types of statistical features, time period mean and time period standard deviation, for operating parameters, support adjusting the sample statistical duration to generate multi-granularity features, and finally unify and standardize the output to provide reliable and high-quality input data for subsequent SPC analysis and neural network prediction.

[0020] (4) Divide the air separation operating parameters into target air separation unit parameters and other air separation unit parameters, match the configuration differences and economic performance differences of different air separation units, provide hierarchical and accurate data basis for subsequent air separation load scheduling, and adapt to the production scenario of multiple air separation units supplying materials together.

[0021] (5) A back pressure prediction model is constructed using a three-layer feedforward BP neural network. The standardized feature samples superimposed with the device operation status results are used as input, making the prediction dimension more comprehensive. The model adopts an offline training + online incremental update mode, which can adapt to the performance drift caused by long-term operation of the device and maintain stable prediction accuracy.

[0022] (6) Based on the ambient temperature threshold, the nitrogen and oxygen production boundary point is divided. Combined with the oxygen and nitrogen venting status and the oxygen and nitrogen ratio characteristics of different air separation, the load increment is calculated. The back pressure is actively matched to the optimal operating range and an adjustment plan is generated. At the same time, when the back pressure deviates from the safety boundary, the corresponding level of early warning is output. This can effectively reduce the venting and emission of nitrogen and oxygen gas, reduce the energy consumption and production cost of the unit, and improve the efficiency of production control.

[0023] (7) The whole method can be implemented as a modular back pressure prediction and optimization system. Each module has clear functional boundaries and closed-loop process connection. It can be directly connected to the on-site production control system, which is convenient for deployment and application in the air separation-liquefaction joint production scenario. Attached Figure Description

[0024] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 A schematic flowchart illustrating a back pressure prediction and optimization method for an oxygen-nitrogen liquefaction device provided in this application embodiment; Figure 2 A connection diagram of the device corresponding to the back pressure prediction and optimization method of an oxygen-nitrogen liquefaction device provided in the embodiments of this application; Figure 3 A flowchart of the model algorithm for a back pressure prediction and optimization method for an oxygen-nitrogen liquefaction device provided in this application embodiment; Figure 4 A design diagram of SPC statistical process control theory provided for an embodiment of this application; Figure 5 A schematic diagram of a BP neural network backpressure prediction model provided in an embodiment of this application; Figure 6 A flowchart illustrating the training process of a BP neural network backpressure prediction model provided in this application embodiment; Figure 7 This application provides an embodiment of an air separation and liquefaction unit with SPC back pressure optimization under unsteady-state degradation (target oxygen production 2430 Nm³). 3 Data statistics chart for / h); Figure 8 This application provides an embodiment of a back pressure optimization under unsteady-state degradation of an air separation and liquefaction unit's SPC (target oxygen production 2685 Nm³). 3Data statistics chart for / h); Figure 9 This application provides an example of SPC back pressure optimization for an air separation and liquefaction unit under steady-state health (target oxygen production 2430 Nm³). 3 Data statistics chart for / h); Figure 10 This application provides an example of SPC steady-state health back pressure optimization for an air separation and liquefaction unit (target oxygen production 2685 Nm³). 3 Data statistics chart for / h); Figure 11 This is a structural block diagram of a back pressure prediction and optimization system for an oxygen-nitrogen liquefaction device provided in an embodiment of this application. Detailed Implementation

[0025] The present application will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.

[0026] Example 1:

[0027] Please see Figures 1-3 This application provides a back pressure prediction and optimization method for an oxygen-nitrogen liquefaction unit, targeting a combined production process scenario of an upstream air separation unit and a downstream liquefaction unit in the iron and steel metallurgical industry. Through a closed-loop management system encompassing "data acquisition - preprocessing - status determination - back pressure prediction - optimized scheduling," it achieves proactive prediction and optimization of the liquefaction unit's back pressure. The method specifically includes the following steps: Step S1: Obtain the real-time operating parameters of the air separation unit and the nitrogen-oxygen liquefaction unit during the target time period, as well as the ambient temperature parameters for the corresponding time period; Step S2: Perform data preprocessing on real-time operating parameters and ambient temperature parameters, extract statistical features, and generate standardized feature samples; Step S3: Based on SPC statistical process control theory, determine the process stability and identify the degradation trend of standardized feature samples, and output the device operation status results; Step S4: Input the standardized feature samples and the device operating status results into the pre-trained BP neural network back pressure prediction model, and output the back pressure prediction value of the oxygen and nitrogen liquefaction device for the next time period. Step S5: Based on the predicted back pressure value, combined with the preset oxygen-nitrogen ratio process rules and air separation load scheduling logic, generate a back pressure optimization adjustment scheme and abnormal early warning information.

[0028] In an optional embodiment, the aforementioned real-time operating parameters include parameters of the target air separation unit, parameters of other air separation units, and parameters of the liquefaction unit; The target air separation unit parameters include hourly oxygen production, hourly nitrogen production, oxygen venting opening degree, nitrogen venting opening degree, air compressor power, hourly air compressor power consumption, booster compressor power, and whether the nitrogen production has reached the target value. Other air separation unit parameters include hourly oxygen production, hourly nitrogen production, oxygen venting opening degree, nitrogen venting opening degree, air compressor power, and air compressor hourly power consumption. The parameters of the liquefaction unit include the hourly output of liquid oxygen, the hourly output of liquid nitrogen, the compressor power, and the historical value of the compressor back pressure; Ambient temperature parameters are used to define the threshold for nitrogen and oxygen production, serving as the basis for determining the air separation load scheduling logic; Among them, the target air separation unit parameters are the operating parameters corresponding to the target air separation unit, and the other air separation unit parameters are the operating parameters corresponding to the air separation units operating other than the target air separation unit.

[0029] Specifically, the air separation unit group includes one target air separation unit (Unit 5) and three other air separation units (Units 1, 7, and 8). Unit 5 has higher unit energy consumption and is equipped with automatic load changing, operating as a peak-shaving unit. Units 1, 7, and 8 serve as baseload units, with Unit 5's capacity supplementing the gas demand when other air separation units cannot meet the demand. The nitrogen-oxygen liquefaction unit uses an LMPL-300 nitrogen expansion refrigeration cycle liquefaction unit, designed for an oxygen-nitrogen liquefaction capacity of 3000~6000 Nm³. 3 / h, total liquefaction capacity per hour is approximately 9000 Nm 3 The oxygen and nitrogen production ratio can be adjusted as needed.

[0030] All operating parameters are acquired in real time through the on-site DCS control system and PLC electrical control unit. The physical address and acquisition attributes of each process point are pre-configured, and the raw data acquisition frequency is uniformly set to 200ms.

[0031] in: The parameters collected by the No. 5 target air separation unit include: hourly oxygen production, hourly nitrogen production, oxygen vent valve opening, nitrogen vent valve opening, air compressor power, hourly air compressor power consumption, booster motor power, and nitrogen production compliance status flag. The parameters collected for other air separation units #1, #7, and #8 include: hourly oxygen production, hourly nitrogen production, oxygen vent valve opening, nitrogen vent valve opening, air compressor power, and hourly air compressor power consumption. The parameters collected by the liquefaction unit include: hourly output of liquid oxygen, hourly output of liquid nitrogen, instantaneous power of the compressor, and historical value of compressor back pressure; The parameters collected for the supporting compressor units include: instantaneous power, hourly pressure delivery, cumulative power consumption, and cumulative pressure delivery for each unit, which are used for full-process energy consumption calculation and auxiliary verification of operating conditions. The ambient temperature parameter is real-time monitoring data from the plant's meteorological station, serving as the boundary input condition for air separation capacity demarcation and load scheduling.

[0032] In an optional embodiment, step S2 specifically includes: The real-time operating parameters and ambient temperature parameters obtained in step S1 are subjected to outlier detection and missing value processing respectively. Abnormal operating condition parameters belonging to shutdown and debugging are removed from the real-time operating parameters, and the ambient temperature parameters corresponding to the abnormal operating condition parameters at the same time are discarded simultaneously. Statistical features are extracted from real-time operating parameters after outlier detection, missing value processing, and removal of abnormal operating condition parameters. The sample statistical duration is adjusted according to different parameter types to generate feature samples with different time granularities. The feature samples, along with the ambient temperature parameters after outlier detection and missing value handling, are standardized together to obtain standardized feature samples.

[0033] In an optional embodiment, the above-mentioned extraction of statistical features from real-time operating parameters after outlier detection, missing value handling, and removal of abnormal operating condition parameters specifically includes: For real-time operating parameters that have exceeded limits after outlier detection, missing value processing, and removal of abnormal operating condition parameters, the mean value of the time period is extracted. For real-time operating parameters that have fluctuation anomalies after outlier detection, missing value processing, and removal of abnormal operating condition parameters, the standard deviation of the time period is extracted.

[0034] Specifically, the collected real-time operating parameters and ambient temperature parameters undergo full-process data preprocessing to extract statistical features that can be used for statistical analysis and model input, ultimately generating standardized feature samples. This step includes four processing levels: data quality verification, statistical feature extraction, adaptive sample granularity construction, and feature standardization.

[0035] At the data quality verification level, a triple cleaning mechanism of "process threshold constraint + statistical criterion discrimination + operating condition legality verification" is adopted: First, the upper and lower limit thresholds of each parameter are set based on the process design range, and abnormal jump points are identified by combining the 3σ statistical criterion, and outlier marking and missing value interpolation are performed; Second, the unit operation status flag bit of the PLC control unit is read to accurately remove parameter data corresponding to abnormal operating conditions such as tripping, start-up and shutdown, debugging, and fault reset; Finally, spatiotemporal alignment processing is performed to simultaneously discard the ambient temperature parameters corresponding to the abnormal operating conditions at the same time, ensuring that the sample set is all valid data under the steady-state operation of the device.

[0036] In this embodiment, the applicable operating condition safety boundary for the model is: liquefaction unit operating load ≥ 60% of design load (≥ 1800 Nm). 3 / h), to avoid data noise introduced by instability in distillation conditions under low load; operating load ≤105% of design load (≤6300Nm 3 / h), excluding unsafe operating conditions of overload operation; when the power collection value deviates from the normal operating range by more than 50%, it is judged as equipment tripping or collection failure, and all samples in the corresponding time period are directly removed.

[0037] At the statistical feature extraction level, for the cleaned real-time operating parameters, time-domain statistical features are extracted according to anomaly patterns: for out-of-limit anomaly patterns, the mean feature of the time period is extracted to characterize the degree of parameter centering shift; for fluctuation anomaly patterns, the standard deviation feature of the time period is extracted to characterize the degree of parameter discrete fluctuation. These two types of statistical features together constitute the feature basis for SPC process judgment and neural network prediction.

[0038] At the adaptive sample granularity construction level, the statistical duration of samples is differentiated according to the dynamic response rate and change time constant of different process parameters, generating feature samples with different time granularities. For rapidly changing parameters such as pressure and flow rate, a shorter statistical window length is used to ensure early warning sensitivity; for slowly changing parameters such as power and temperature, a longer statistical window length is used to suppress random fluctuations and reduce false alarm rate. For example, the power parameter of the high-pressure nitrogen compressor uses a 30-minute moving average as a single statistical sample. When six consecutive sample points show a monotonically increasing trend, it can be determined that the power of the high-pressure nitrogen compressor has a continuous deterioration trend within 3 hours. The optimal balance between early warning sensitivity and false alarm rate is achieved through granularity adaptation.

[0039] At the feature standardization level, the constructed feature samples and the cleaned ambient temperature parameters are subjected to Z-score standardization to eliminate the differences in dimensions and numerical magnitudes between different parameters, thereby obtaining standardized feature samples, which provide a unified data benchmark for subsequent SPC control limit calculations and BP neural network inputs.

[0040] Please see Figure 4 In an optional embodiment, step S3 above is based on SPC statistical process control theory to determine the stability of the standardized feature sample process and identify the deterioration trend, and output the device operating status results to provide prior information on operating conditions for subsequent back pressure prediction.

[0041] First, based on historical data under steady-state health conditions of the device, the center line (CL), upper control limit (UCL), and lower control limit (LCL) of each parameter control chart are calculated, and the control regions are divided into three zones: C, B, and A, according to ±σ, ±2σ, and ±3σ. In this embodiment, process stability determination employs the eight SPC anomaly detection criteria to achieve multi-mode anomaly identification, specifically including: Criterion 1: A single sample point falling outside the control limit A area corresponds to a sudden over-limit fault; Criterion 2: If nine consecutive sample points fall on the same side of the center line, the corresponding process mean is systematically shifted. Criterion 3: A continuous increase or decrease in six consecutive sample points corresponds to a trend of continuous performance degradation; Criterion 4: The alternating up and down fluctuations of adjacent points in 14 consecutive sample points correspond to periodic disturbances; Criterion 5: If two out of three consecutive sample points fall outside Zone B on the same side of the center line, it corresponds to a moderate offset trend. Criterion 6: If 4 out of 5 consecutive sample points fall outside Zone C on the same side of the center line, it indicates a slight offset trend. Criterion 7: If 15 consecutive sample points fall within area C on both sides of the center line, the corresponding process variance abnormality decreases. Criterion 8: If eight consecutive sample points fall on either side of the center line and none of them are located within region C, the corresponding process variance increases.

[0042] In an optional embodiment, step S3 above, in addition to using the eight SPC discrepancy criteria to determine process stability, also performs upper and lower limit constraint verification of the process and adjusts the collection statistical granularity of a single sample according to the rate of change of different process parameters.

[0043] Specifically, based on the traditional SPC out-of-place logic, this embodiment makes two targeted improvements: First, a new process upper and lower limit constraint verification is added. On the basis of statistical control judgment, the verification of the upper limit of process specification (USL) and the lower limit of specification (LSL) is superimposed to solve the defect of traditional SPC that can only judge the statistical stability of the process and cannot identify "statistically controlled but the value exceeds the process qualification range". For example, the upper limit of the compressor power of the liquefaction unit is limited to 6500kW / h, corresponding to an economic back pressure of about 440kPa, to ensure that the judgment result meets the dual requirements of statistical controllability and process qualification at the same time. Secondly, it supports adaptive adjustment of statistical granularity for single sample collection. It can flexibly configure the number of statistical points and sampling interval for a single sample according to the rate of change of different process parameters, adapting to the different rate of change characteristics of various parameters such as flow rate, pressure, power, and temperature, and significantly improving the accuracy and timeliness of degradation trend identification.

[0044] Based on the above determination, the output device operating status results include the status classification of "steady-state health / non-steady-state degradation", as well as degradation trend types such as increasing degradation, decreasing degradation, disordered fluctuation, and periodic fluctuation, and are accompanied by degradation confidence index.

[0045] Please see Figure 5In an optional embodiment, the pre-trained BP neural network backpressure prediction model in step S4 above is a three-layer feedforward neural network structure, including an input layer, a hidden layer and an output layer. Adjacent layer neurons are fully connected and neurons in the same layer are not connected, and a supervised learning mode is adopted. The hyperparameters of the pre-trained BP neural network backpressure prediction model are configured as learning rate and additional momentum factor. The training process uses the backpropagation algorithm to iteratively optimize the connection weights and biases of each layer.

[0046] Specifically, the core hyperparameters of the model are configured as follows: learning rate 0.06, additional momentum factor 0.95, hidden layer activation function using a sigmoid function, and output layer using a linear activation function to adapt to the regression prediction of continuous backpressure values.

[0047] The model's learning process consists of two stages: forward propagation and backpropagation. In the forward propagation stage, the input feature vector is passed through the input layer and, after nonlinear mapping through the hidden layer, the back pressure prediction value is calculated by the output layer. If the error between the predicted value and the true back pressure value does not meet the accuracy requirements, the backpropagation stage is entered. The prediction error is propagated back along the network layer by layer, and the gradient descent algorithm with driving terms is used to iteratively optimize the connection weights and bias terms of each layer, so that the global prediction error continues to converge.

[0048] Please see Figure 6 In an optional embodiment, the pre-trained BP neural network backpressure prediction model described above is trained through the following steps: Historical operating data and corresponding back pressure values ​​of the oxygen and nitrogen liquefaction unit under normal operating conditions were obtained to form the original sample set. Data cleaning, statistical feature extraction, and standardization are performed on the original sample set to obtain a standardized training feature set. The standardized training feature set and the corresponding true back pressure values ​​are divided into a training set and a validation set. The training set is input into the initial BP neural network model, the predicted output value is calculated through forward propagation, and then the connection weights and biases of each layer are adjusted through error backpropagation to iteratively reduce the prediction error. Training stops when the training error meets the preset accuracy requirement or reaches the upper limit of the number of iterations. The model prediction accuracy is verified using the validation set. After the verification is passed, the pre-trained BP neural network backpressure prediction model is obtained. After the model is launched and running, online incremental updates are performed based on real-time running data to adapt to device performance drift.

[0049] Specifically, this model employs a two-stage training mechanism of "offline pre-training + online incremental update": Offline pre-training phase: Historical operating data and corresponding back pressure values ​​under normal operating conditions of the oxygen and nitrogen liquefaction unit are collected to form the original sample set; after data cleaning, statistical feature extraction and standardization, a standardized training feature set is obtained; the samples are divided into training set and validation set according to the time series to avoid data leakage caused by random partitioning; the training set is input into the initial BP neural network for iterative training, and training is stopped when the training error meets the preset accuracy requirements, or reaches the maximum number of iterations, or the validation set error no longer decreases after several consecutive rounds; after the validation set accuracy is verified, the initial pre-trained model is obtained.

[0050] Online incremental update phase: After the model is launched and running, a sliding time window mechanism is adopted to periodically incorporate the latest real-time running data, fine-tune and update the network weights, and adapt to the slow time-varying drift caused by performance degradation and component aging during long-term operation of the adaptive device, so as to ensure the long-term stability of prediction accuracy.

[0051] By introducing the device's operating status results as auxiliary input into the model, the prediction network can distinguish the different data distribution characteristics of steady-state and adverse operating conditions, effectively improving the back pressure prediction accuracy under unsteady operating conditions and avoiding prediction bias of the pure data-driven model when operating conditions change abruptly.

[0052] In an optional embodiment, step S5 specifically includes: Nitrogen and oxygen production thresholds are defined based on ambient temperature parameters. When the ambient temperature parameter is greater than the preset temperature, the first oxygen production threshold is used as the threshold. When the ambient temperature parameter is less than or equal to the preset temperature, the second oxygen production threshold is used as the threshold. Calculate the load increment of the corresponding air separation unit by combining the oxygen venting state, the nitrogen venting state, and the oxygen-nitrogen ratio characteristics of the target air separation unit and other air separation units. Based on the calculated load increment, determine the output adjustment strategy for the target air separation unit and / or other air separation units to match the back pressure of the oxygen and nitrogen liquefaction unit to the optimal operating range, and generate a back pressure optimization adjustment scheme accordingly. When the predicted back pressure value deviates from the preset safety boundary, an abnormal warning message of the corresponding level is generated and output. Among them, the first oxygen production is less than the second oxygen production; the oxygen-nitrogen ratio characteristics are determined based on the oxygen-nitrogen ratio process rules.

[0053] Specifically, this step follows the scheduling principle of "production determined by nitrogen and priority given to high-quality machines," and the specific execution logic is as follows: Nitrogen and oxygen production thresholds were defined based on ambient temperature parameters, and the differences in distillation efficiency of air separation units at different temperatures were matched: when the ambient temperature was greater than 10℃, with an oxygen production of 53000 Nm³... 3 / h is the dividing point, corresponding to a minimum oxygen production of 50,000 Nm³. 3 / h; when the ambient temperature is less than or equal to 10℃, with an oxygen production of 55000 Nm³. 3 / h is the dividing point, corresponding to a minimum oxygen production of 52000 Nm³. 3 / h. Boundary conditions equal to 10℃ are assigned to the low-temperature branch to avoid frequent switching of scheduling logic caused by small temperature fluctuations.

[0054] The oxygen-nitrogen ratio is set according to the design characteristics, energy consumption level, and unit positioning of different air separation units: the oxygen-nitrogen ratio for the target air separation unit #5 is 1:2, that is, for every 1000 Nm³ produced... 3 / h oxygen production corresponds to 2000 Nm 3 / h nitrogen; the oxygen-nitrogen ratio for other air separation units (No. 1, No. 7, and No. 8) is 1:3, meaning that for every 1000 Nm³ produced... 3 / h oxygen production corresponds to 3000 Nm 3 / h nitrogen. The target air separation unit #5 is equipped with an automatic load change function and operates as a peak-shaving unit; air separation units #1, #7, and #8 are prioritized for stable load operation as base-load units. When the overall capacity of the base-load units cannot meet the on-site gas demand, the target air separation unit #5 will be put into operation to supplement peak-shaving output.

[0055] Based on different combinations of oxygen and nitrogen venting states, the load increment of the corresponding air separation unit is calculated for each scenario: Logic 1: If both oxygen and nitrogen are vented, the air separation load increment is 0, maintaining the current output; if neither oxygen nor nitrogen is vented, and the oxygen production of air separation unit #1 reaches 26000 Nm³. 3 When the oxygen production of the No. 5 air separation unit reaches the minimum production at the corresponding temperature, the air separation load increment is 0, indicating that the current unit output has reached the economic limit.

[0056] Logic 2: If oxygen is vented but nitrogen is not, and the temperature is above 10℃, the nitrogen replenishment amount is 6000 Nm³. 3 / h demand calculation load increment: ① When the oxygen production of No. 5 air separation unit is greater than 53000 Nm 3 At a rate of / h, the output increment coefficient for air separation unit #5 is set to 3, corresponding to an additional 6000 Nm. 3 / h nitrogen; ② When the oxygen production of air separation unit #5 is 52000 Nm³ 3 At a rate of / h, the output increment coefficient of the No. 5 air separation unit is set to 2, corresponding to an additional 4000 Nm. 3 / h nitrogen, with an output increment coefficient of 1 for air separation units #1, #7, and #8, corresponding to a supplement of approximately 3000 Nm³. 3 / h nitrogen; ③ When the oxygen production of air separation unit #5 is 51000 Nm 3 At a rate of / h, the output increment coefficient of the No. 5 air separation unit is set to 1, corresponding to an additional 2000 Nm. 3 / h nitrogen, the output increment coefficient of air separation units #1, #7, and #8 is taken as 1.5, corresponding to a supplement of 4500 Nm 3 / h nitrogen; ④ When the oxygen production of air separation unit #5 is 50000 Nm 3 At a rate of / h, the output increment coefficient for air separation unit #5 is set to 0, while the output increment coefficients for air separation units #1, #7, and #8 are set to 2, corresponding to an additional 6000 Nm. 3 / h nitrogen, and the oxygen production of No. 1 air separation unit is required to be greater than 28000 Nm³. 3 / h.

[0057] Logic 3: If oxygen is vented but nitrogen is not, and the temperature is above 10℃, the nitrogen replenishment amount is 3000 Nm³. 3 / h demand calculation load increment: ① When the oxygen production of No. 5 air separation unit is greater than 53000 Nm 3 When the output of air separation unit #5 is 52000 Nm³ / h, the output increment coefficient is taken as 1.5; ② When the oxygen production of air separation unit #5 is 52000 Nm³ / h, the output increment coefficient is taken as 1.5; 3 At a speed of / h, the output increment coefficient of air separation unit #5 is taken as 1.5, which can meet the requirement of 3000 Nm. 3 / h nitrogen demand, the output increment coefficient of air separation units 1#, 7#, and 8# is taken as 0; ③ When the oxygen output of air separation unit 5# is 51000 Nm 3 When the output of air separation unit #5 is 1, the output increment coefficient for air separation units #1, #7, and #8 is 0.5; ④ When the oxygen production of air separation unit #5 is 50000 Nm³ / h... 3 At a rate of / h, the output increment coefficient for air separation unit #5 is set to 0, while the output increment coefficients for air separation units #1, #7, and #8 are set to 1. Furthermore, the oxygen production of air separation unit #1 is required to be greater than 27000 Nm³. 3 / h.

[0058] Based on the calculated load increments of each air separation unit, an output adjustment strategy is formulated for the target air separation unit and / or other air separation units. The back pressure of the oxygen and nitrogen liquefaction unit is matched to the economically optimal operating range (corresponding to compressor power < 6500 kW / h, back pressure approximately 440 kPa), and a back pressure optimization adjustment scheme that can be directly issued and executed is generated. At the same time, multiple safety boundaries are set. When the predicted back pressure value deviates from the preset safety boundary, an abnormal warning message of the corresponding level is generated and output according to the degree of deviation, prompting the operator to intervene in advance.

[0059] Please see Figures 7-10 To verify the practical application effect of this method, a field verification was conducted in the air separation-liquefaction combined production system of a steel complex. The verification object was the LMPL-300 nitrogen-oxygen liquefaction unit, equipped with four air separation units (No. 5, No. 1, No. 7, and No. 8). Two typical operating conditions, namely unsteady-state deterioration and steady-state health, were selected, and a temperature of 2430 Nm was set. 3 / h、2685Nm 3A comparative test was conducted on two target oxygen production levels per hour, and the verification results are as follows: Verification under unsteady and adverse chemical conditions Under this operating condition, the device exhibits progressive performance degradation, and the SPC anomaly detection rule has triggered a trend anomaly alarm.

[0060] Target oxygen production: 2430 Nm 3 / h setting: The nitrogen-to-oxygen ratio adjustment range is 0.8~0.9, and the predicted oxygen target adjustment range is 2440~2420 Nm. 3 / h; In actual operation, oxygen production fluctuates between 2447 and 2413 Nm³. 3 / h, the output fluctuation range during the transition period is +7~-7Nm 3 / h. Verification results show that under this operating condition, the back pressure exhibits a continuously low trend. This method can identify the deterioration trend and issue an early warning 1.5 to 2 hours in advance, guiding operators to adjust the air separation unit output in advance and avoid increased energy consumption caused by the back pressure continuously deviating from the optimal range. The corresponding data curves are as follows: Figure 7 As shown.

[0061] Target oxygen production: 2685 Nm³ 3 / h setting: The nitrogen-oxygen ratio adjustment range is 0.7~0.8, and the predicted oxygen target adjustment range is 2680~2690 Nm. 3 / h; In actual operation, oxygen production fluctuates between 2675 and 2678 Nm³. 3 / h, the output fluctuation range during the transition period is -5 to -12 Nm 3 / h. Verification results show that under this operating condition, the downstream back pressure is significantly lower than the warning threshold of 10 kPa. This method can accurately pinpoint the degree of degradation and trigger a secondary warning, assisting maintenance personnel in quickly locating the fault, shortening fault handling time, and preventing further deterioration of the operating condition. The corresponding data curves are as follows: Figure 8 As shown.

[0062] Steady-state health condition verification Under this operating condition, the unit operates stably, and the SPC determination result is that the statistics are under control and the process is qualified.

[0063] Target oxygen production: 2430 Nm 3 / h setting: The nitrogen-oxygen ratio adjustment range is 0.8~1.0, and the predicted oxygen target adjustment range is 2425~2430 Nm. 3 / h; In actual operation, oxygen production fluctuates between 2421 and 2431 Nm³. 3 / h, the output fluctuation range during the transition period is -4 to +1Nm 3 / h. Verification results show that under this operating condition, the system flow fluctuation is small, and the subsequent back pressure stability is excellent. This method can continuously fine-tune the air distribution ratio based on the predicted value, stabilizing the back pressure within the economic operating range and achieving optimal energy consumption control. The corresponding data curves are shown below. Figure 9 As shown.

[0064] Target oxygen production: 2685 Nm³ 3 / h setting: The nitrogen-oxygen ratio adjustment range is 0.8~1.0~0.8, and the predicted oxygen target adjustment range is 2680~2685~2690 Nm. 3 / h; In actual operation, oxygen production fluctuates between 2686~2682~2691 Nm³. 3 / h, the output fluctuation range during the transition period is +6~-3~+1Nm 3 / h. Verification results show that flow fluctuations are extremely low under this operating condition, resulting in high back pressure control accuracy. This method can achieve stable and precise back pressure control, avoiding energy loss and equipment impact caused by large fluctuations in operating conditions. Corresponding data curves are shown below. Figure 10 As shown.

[0065] Based on the comprehensive field verification results, compared with the traditional over-limit alarm mode, this method can increase the fault warning advance to 1-3 hours, effectively reduce the venting and emission of nitrogen and oxygen gas in the air separation unit, reduce the unit product energy consumption of the liquefaction unit, and improve the equipment operation stability and production control efficiency, thus having significant economic benefits and engineering application value.

[0066] Example 2

[0067] Please see Figure 11 This application also provides a back pressure prediction and optimization system for an oxygen-nitrogen liquefaction device, implemented using the back pressure prediction and optimization method for an oxygen-nitrogen liquefaction device as described in any of Embodiment 1. The system includes: The parameter acquisition module is used to acquire the real-time operating parameters of the air separation unit and the nitrogen-oxygen liquefaction unit within the target time period, as well as the ambient temperature parameters for the corresponding time period; The data processing module is used to preprocess real-time operating parameters and ambient temperature parameters, extract statistical features, and generate standardized feature samples. The data judgment module is used to determine the process stability and identify the deterioration trend of standardized feature samples based on SPC statistical process control theory, and output the device operation status results. The back pressure prediction module is used to input standardized feature samples and device operating status results into a pre-trained BP neural network back pressure prediction model and output the back pressure prediction value of the oxygen and nitrogen liquefaction device for the next period. The optimization and adjustment module is used to generate back pressure optimization and adjustment schemes and abnormal early warning information based on the back pressure prediction value, combined with the preset oxygen-nitrogen ratio process rules and air separation load scheduling logic.

[0068] It should be noted that the system provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0069] The specific embodiments of this application have been described above. It should be understood that this application is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the substantive content of this application. Unless otherwise specified, the embodiments and features described in the embodiments of this application can be arbitrarily combined with each other.

Claims

1. A method for predicting and optimizing back pressure in an oxygen-nitrogen liquefaction unit, characterized in that, The method includes: Step S1: Obtain the real-time operating parameters of the air separation unit and the nitrogen-oxygen liquefaction unit during the target time period, as well as the ambient temperature parameters for the corresponding time period; Step S2: Perform data preprocessing on the real-time operating parameters and the ambient temperature parameters, extract statistical features, and generate standardized feature samples; Step S3: Based on SPC statistical process control theory, determine the process stability and identify the degradation trend of the standardized feature samples, and output the device operation status results; Step S4: Input the standardized feature samples and the device operating status results into the pre-trained BP neural network back pressure prediction model, and output the predicted back pressure value of the oxygen and nitrogen liquefaction device for the next time period; Step S5: Based on the predicted back pressure value, combined with the preset oxygen-nitrogen ratio process rules and air separation load scheduling logic, generate a back pressure optimization adjustment scheme and abnormal early warning information.

2. The back pressure prediction and optimization method for the oxygen-nitrogen liquefaction device according to claim 1, characterized in that, The real-time operating parameters include the parameters of the target air separation unit, other air separation units, and liquefaction unit; The target air separation unit parameters include hourly oxygen production, hourly nitrogen production, oxygen venting opening degree, nitrogen venting opening degree, air compressor power, hourly air compressor power consumption, booster motor power consumption, and whether the nitrogen level has reached the target value. The other air separation unit parameters include the hourly oxygen production, the hourly nitrogen production, the oxygen venting opening degree, the nitrogen venting opening degree, the air compressor power, and the air compressor hourly power consumption; The parameters of the liquefaction unit include hourly output of liquid oxygen, hourly output of liquid nitrogen, compressor power, and historical value of compressor back pressure. The ambient temperature parameter is used to define the nitrogen and oxygen production threshold, which serves as the basis for the judgment of the air separation load scheduling logic. Wherein, the target air separation unit parameters are the operating parameters corresponding to the target air separation unit, and the other air separation unit parameters are the operating parameters corresponding to air separation units operating other than the target air separation unit.

3. The back pressure prediction and optimization method for the oxygen-nitrogen liquefaction device according to claim 1, characterized in that, Step S2 specifically includes: The real-time operating parameters and the ambient temperature parameters obtained in step S1 are subjected to outlier detection and missing value processing respectively. Abnormal operating condition parameters belonging to shutdown and debugging are removed from the real-time operating parameters, and the ambient temperature parameters corresponding to the abnormal operating condition parameters at the same time are discarded simultaneously. The statistical features are extracted from the real-time operating parameters after the outlier detection, missing value processing and abnormal operating condition parameter removal are completed, and the sample statistical duration is adjusted according to different parameter types to generate feature samples with different time granularities. The feature samples, together with the environmental temperature parameters after the outlier detection and missing value processing, are standardized to obtain the standardized feature samples.

4. The back pressure prediction and optimization method for the oxygen-nitrogen liquefaction device according to claim 3, characterized in that, The step of extracting statistical features from the real-time operating parameters after completing the outlier detection, missing value processing, and abnormal operating condition parameter removal specifically includes: For the real-time operating parameters that have exceeded the limits after the completion of the outlier detection, missing value processing, and removal of abnormal operating condition parameters, the mean value of the time period is extracted. For the real-time operating parameters that have fluctuated after the completion of the outlier detection, missing value processing, and removal of abnormal operating condition parameters, the standard deviation of the time period is extracted.

5. The back pressure prediction and optimization method for the oxygen-nitrogen liquefaction device according to claim 1, characterized in that, The process stability determination in step S3 adopts the eight SPC discrepancy criteria, specifically including: Criterion 1: A single sample point falls outside control limit A. Criterion 2: Nine consecutive sample points fall on the same side of the center line; Criterion 3: Six consecutive sample points increasing or decreasing; Criterion 4: Adjacent points in a series of 14 consecutive sample points fluctuate alternately up and down; Rule 5: Two out of three consecutive sample points fall outside zone B on the same side of the center line; Criterion 6: Four out of five consecutive sample points fall outside zone C on the same side of the center line; Rule 7: 15 consecutive sample points fall within area C on both sides of the center line; Criterion 8: Eight consecutive sample points fall on either side of the center line and none of them are located within region C.

6. The back pressure prediction and optimization method for the oxygen-nitrogen liquefaction device according to claim 5, characterized in that, Step S3, based on the process stability determination using the eight SPC discrepancy criteria, also performs upper and lower limit constraint verification of the process and adjusts the statistical granularity of a single sample collection according to the rate of change of different process parameters.

7. The back pressure prediction and optimization method for the oxygen-nitrogen liquefaction device according to claim 2, characterized in that, Step S5 specifically includes: The nitrogen and oxygen production thresholds are defined based on the ambient temperature parameters. When the ambient temperature parameters are greater than a preset temperature, the first oxygen production threshold is used as the dividing point. When the ambient temperature parameters are less than or equal to the preset temperature, the second oxygen production threshold is used as the dividing point. Based on the oxygen venting status, nitrogen venting status, and the oxygen-nitrogen ratio characteristics of the target air separation unit and the other air separation units, calculate the load increment of the corresponding air separation unit. Based on the calculated load increment, a power output adjustment strategy for the target air separation unit and / or the other air separation units is determined to match the back pressure of the oxygen and nitrogen liquefaction unit to the optimal operating range, and the back pressure optimization adjustment scheme is generated accordingly. When the predicted back pressure value deviates from the preset safety boundary, the corresponding level of abnormal warning information is generated and output. Wherein, the first oxygen production is less than the second oxygen production; the oxygen-nitrogen ratio characteristic is determined based on the oxygen-nitrogen ratio process rules.

8. The back pressure prediction and optimization method for an oxygen-nitrogen liquefaction device according to any one of claims 1-7, characterized in that, The pre-trained BP neural network backpressure prediction model in step S4 is a three-layer feedforward neural network structure, including an input layer, a hidden layer, and an output layer. Adjacent layer neurons are fully connected, while neurons in the same layer are not connected, and a supervised learning mode is adopted. The hyperparameters of the pre-trained BP neural network backpressure prediction model are configured as learning rate and additional momentum factor. The training process uses the backpropagation algorithm to iteratively optimize the connection weights and biases of each layer.

9. The back pressure prediction and optimization method for the oxygen-nitrogen liquefaction device according to claim 8, characterized in that, The pre-trained BP neural network backpressure prediction model is obtained through the following steps: The historical operating data and corresponding back pressure values ​​of the oxygen-nitrogen liquefaction unit under normal operating conditions are obtained to form the original sample set. Data cleaning, statistical feature extraction, and standardization are performed on the original sample set to obtain a standardized training feature set; The standardized training feature set and the corresponding true back pressure values ​​are divided into a training set and a validation set. The training set is input into the initial BP neural network model, the predicted output value is calculated through forward propagation, and then the connection weights and biases of each layer are adjusted through error backpropagation to iteratively reduce the prediction error. Training stops when the training error meets the preset accuracy requirement or reaches the upper limit of the number of iterations. The model prediction accuracy is verified using the validation set. After the verification is passed, the pre-trained BP neural network backpressure prediction model is obtained. After the model is launched and running, online incremental updates are performed based on the real-time running data to adapt to device performance drift.

10. A back pressure prediction and optimization system for an oxygen-nitrogen liquefaction unit, implemented using the back pressure prediction and optimization method for an oxygen-nitrogen liquefaction unit as described in any one of claims 1 to 9, characterized in that, The system includes: The parameter acquisition module is used to acquire the real-time operating parameters of the air separation unit and the nitrogen-oxygen liquefaction unit within the target time period, as well as the ambient temperature parameters for the corresponding time period; The data processing module is used to preprocess the real-time operating parameters and the ambient temperature parameters, extract statistical features, and generate standardized feature samples. The data determination module is used to determine the process stability and identify the degradation trend of the standardized feature samples based on SPC statistical process control theory, and output the device operation status results. The back pressure prediction module is used to input the standardized feature samples and the device operating status results into a pre-trained BP neural network back pressure prediction model, and output the back pressure prediction value of the oxygen and nitrogen liquefaction device in the next time period. The optimization and adjustment module is used to generate a back pressure optimization and adjustment scheme and anomaly warning information based on the back pressure prediction value, combined with the preset oxygen-nitrogen ratio process rules and air separation load scheduling logic.

Images