Method and system for water-based production

Abstract

The application discloses a method and system for water-based production, which comprises the following steps: collecting research area data, constructing a water volume multi-scene set of each period in the future of the research area based on historical data of upstream water sources and water supply systems and calculating a confidence water volume envelope; constructing a process-water consumption response surface based on historical operation data of production lines under different working conditions, generating a process optimization scheme and a process optimization coefficient; constructing a water network diagram and a production network diagram respectively, coupling the water network diagram and the production network diagram based on the confidence water volume envelope, constructing a water network-production network double-layer structure and calculating a production limit priority; calculating a conservative production capacity upper limit based on different confidence lower limits, real-time water consumption coefficient expectations and process optimization coefficients, generating a multi-level robust production scheduling scheme based on the conservative production capacity upper limit, an aggressive production capacity upper limit and the production limit priority and carrying out linkage control. The application realizes dynamic adaptation of water resource supply and production activities.

Description

Technical Field

[0001] This invention relates to the intersection of industrial production scheduling and water resource optimization, and particularly to a method and system for "production determined by water availability". Background Technology

[0002] Water resources are an indispensable key resource in industrial production. "Production based on water availability" has become the core principle for ensuring sustainable production of enterprises and alleviating the contradiction between water supply and demand. Existing technical solutions related to production based on water availability have many defects and are difficult to adapt to complex industrial production scenarios and the uncertain characteristics of water resources.

[0003] In existing solutions, some technical approaches treat the available water volume for a given period as a fixed value reported by the water supply unit and directly substitute it into the capacity limit calculation formula. This approach does not adequately consider uncertainties such as fluctuations in water inflow, changes in reclaimed water production, and water supply facility malfunctions. In actual operation, this may lead to situations where production is scheduled according to the algorithm but the actual water supply is insufficient, affecting production stability. Furthermore, existing solutions only provide a single capacity limit and do not offer scheduling options for different risk levels, failing to meet the production needs of enterprises with varying risk appetites. In existing technologies, the "process optimization coefficient" is often arbitrarily set as a fixed constant, lacking clear physical or data support, making it impossible to explain the rationality of the coefficient's value. More importantly, existing solutions do not... Establishing a correlation between process parameters and water consumption fails to provide specific solutions on "which process parameters need to be adjusted to achieve water-saving targets and by what extent," causing process optimization to remain at the conceptual level and difficult to implement. Current production-limiting ranking rules are mostly based on the single-point indicator of "water consumption / output value ratio," treating each process as an independent entity and ignoring two key coupling relationships between processes: first, water network coupling, where the wastewater from one process may be the recycled water source for another process, and shutting down the former will reduce the usable water of the latter; second, production network coupling, where shutting down the preceding process may force the subsequent high-output production line to shut down, causing output value losses far exceeding those of the process itself. This locally optimal decision-making approach often leads to damage to overall production efficiency.

[0004] Therefore, there is an urgent need to construct a multi-dimensional optimization method for production based on water availability that can accurately characterize the uncertainty of water sources, scientifically generate process optimization schemes, and systematically formulate production restriction decisions. This will address the shortcomings of existing technologies and enhance the scientific rigor, robustness, and operability of enterprises' production based on water availability.

[0005] This invention proposes a method and system for "production determined by water availability" to solve the aforementioned problems, enabling production scheduling schemes to dynamically adapt to water source fluctuations and changes in operating conditions, thereby improving the long-term stability and adaptability of the system. Summary of the Invention

[0006] Objective of the invention: To provide a method for "production determined by water availability" to solve the aforementioned problems in the prior art. Furthermore, to provide a system for "production determined by water availability".

[0007] Technical solution: The "production determined by water availability" method includes the following steps:

[0008] Step S1: Collect data of the study area, including historical data of upstream water sources and water supply systems, and construct a multi-scenario set of water volume for different time periods in the study area, and calculate the confidence water volume envelope.

[0009] Step S2: Based on the historical operating data of the production line under different operating conditions, construct a process-water consumption response surface model and generate process optimization coefficients;

[0010] Step S3: Construct water network diagram and production network diagram respectively. Based on the confidence water volume envelope band, couple water network diagram and production network diagram to construct a two-layer structure of water network-production network and calculate production restriction priority.

[0011] Step S4: Calculate the expected real-time water consumption coefficient based on the real-time operation data of the production line; determine the aggressive capacity limit based on the historical maximum capacity of the study area, market demand, and rated capacity of the equipment; calculate the conservative capacity limit based on different confidence lower limits in the confidence water volume envelope, the expected real-time water consumption coefficient, and the process optimization coefficient; and generate a multi-level robust production scheduling scheme and link it for control based on the conservative capacity limit, the aggressive capacity limit, and the production restriction priority.

[0012] According to one aspect of this application, step S1 is further comprising:

[0013] Step S11: Collect historical data of upstream water sources, historical data of water supply systems, basic data of processes, and historical operating data of production lines under different operating conditions in the study area. The historical data of upstream water sources and historical data of water supply systems include time series of water inflow and supply in the same season, statistical data of water supply facility failures, and data on fluctuations in unconventional water production.

[0014] Step S12: Based on historical data of upstream water sources and historical data of water supply systems, construct a set of multiple water volume scenarios for each future time period, and calculate the probability of each scenario.

[0015] Step S13: For each future time period, based on its corresponding water volume multi-scenario set and scenario probability, calculate at least two lower limits for water volume at different confidence levels, including the lower limit for high confidence level and the lower limit for low confidence level, to form the confidence water volume envelope for that time period.

[0016] According to one aspect of this application, step S12 further comprises:

[0017] Step S12a: Based on the pre-constructed correlation model between historical precipitation and inflow, divide the water into three levels: abundant water, normal water, and dry water, and set the natural inflow fluctuation range for each level.

[0018] Step S12b: Based on fault statistics, classify the faults into four levels: no faults, minor faults, general faults, and serious faults, and set the reduction range for each level.

[0019] Step S12c: Based on the historical production data of the reclaimed water station, the fluctuation level is divided, and the natural water inflow fluctuation scenario, the water supply facility failure scenario and the unconventional water production fluctuation scenario are combined in all dimensions to generate a basic scenario set, and the water volume multi-scenario set corresponding to each future time period is obtained respectively.

[0020] Step S12d: Based on historical data, the occurrence frequency of each basic scenario is statistically analyzed, combined with industry prior probabilities, and the historical frequencies are corrected to obtain the final scenario probability.

[0021] According to one aspect of this application, step S2 further comprises:

[0022] Step S21: Extract historical operating data of the production line under different operating conditions, construct an operating sample set, and use data-driven fitting to obtain the process-water consumption response surface model;

[0023] Step S22: Continuously collect new operating data sample sets, correct the process-water consumption response surface model, identify whether the current operating conditions deviate from the training data range, and obtain the optimal process-water consumption response surface model.

[0024] Step S23: Calculate the partial derivatives of the optimal process-water consumption response surface model with respect to each process parameter. Based on the feasible range of process parameters, production load constraints, and product quality constraints, construct a constrained optimization problem and solve it to obtain the optimal process parameters.

[0025] Step S24: Based on the optimal process parameters and the unit water consumption corresponding to the current process parameters, calculate the process optimization coefficient and generate specific control instructions, i.e., the process optimization scheme.

[0026] According to one aspect of this application, step S21 further comprises:

[0027] Step S21a: Extract historical operating data of the production line under different operating conditions and construct an operating sample set;

[0028] Step S21b: Construct process-water consumption response surface models based on different operating conditions using multinomial response surface model, random forest tree model, and BP neural network model respectively;

[0029] Step S21c: Divide the running sample set into a training set, a validation set, and a test set in a 6:2:2 ratio. Train the process-water consumption response surface model, calculate the goodness of fit, root mean square error, and mean absolute error, and verify the model accuracy. For models with accuracy less than the preset accuracy threshold, improve the model accuracy by adding parameter interaction terms, standardizing features, and weighted fusion of multinomial model and random forest to obtain the process-water consumption response surface model. The preset accuracy threshold includes the goodness of fit threshold and the root mean square error threshold.

[0030] According to one aspect of this application, step S22 further comprises:

[0031] Step S22a: Collect new operational data;

[0032] Step S22b: Set the scrolling window size and the corresponding window increment threshold. When the number of newly collected samples reaches the window increment threshold, remove the earliest n groups of samples in the window, add new samples, and obtain a new set of running data samples. n is a positive integer greater than 20 and less than the window increment threshold.

[0033] Step S22c: Update the parameters of the process-water consumption response surface model using the least squares incremental method and the incremental learning algorithm respectively, and set an accuracy warning threshold. When the model accuracy is lower than the accuracy warning threshold, trigger a full retraining of the model.

[0034] Step S22d: Collect the feature cluster centers of all historical normal operating conditions, construct the operating condition benchmark library, calculate the Euclidean distance between the current operating condition features and the cluster centers of the benchmark library, and determine the operating condition deviation when the distance is greater than the threshold. Correct the process-water consumption response surface model until no deviation occurs, and obtain the optimal process-water consumption response surface model.

[0035] According to one aspect of this application, step S3 further comprises:

[0036] Step S31: Construct a water network diagram based on the basic data of the process, with the process or device as the node and various water flows as the edge. The node attributes include water consumption, reuse amount, and water quality requirements, and the edge attributes include water quantity and water quality changes.

[0037] Step S32: Construct a production network diagram based on process basic data, with processes or production lines as nodes and material dependencies as edges. Node attributes include output value, output, and order dependency degree.

[0038] Step S33: Determine water resource constraints based on the confidence water volume envelope. For each candidate production restriction process, simulate the impact of partial or complete shutdown on the water network and production network. Calculate the changes in water consumption and output value of the entire system. Calculate the marginal water benefit index based on the ratio of the changes in water consumption to the changes in output value of the entire system. Determine the production restriction priority based on the ranking of marginal water benefit index values.

[0039] According to one aspect of this application, step S33 further comprises:

[0040] Step S33a: Determine different water resource constraint levels based on the confidence water volume envelope, including: severe water shortage, moderate water shortage and mild water shortage. Design simulation scenarios for each constraint level, wherein different reduction ratio gradients are set for each candidate production restriction process, and simulate the system response under different reduction degrees.

[0041] Step S33b: Simulate the difference between the total water consumption and total output value of the entire system before and after the production reduction, calculate the changes in water consumption and output value of the entire system, calculate the impact of core orders based on the ratio of the number of core orders affected by production restrictions to the total number of core orders, and calculate the production stability index based on the connectivity of the production network and the continuity of material flow.

[0042] Step S33c: Calculate the marginal water benefit index using a weighted summation method based on the changes in water consumption, output value, core order impact, and production stability indicators of the entire system; Step S33d: Sort the marginal water benefit indexes of all candidate processes in descending order, generate a production restriction priority list, and obtain the production restriction order, optimal production reduction ratio, and triggering conditions for each process.

[0043] According to one aspect of this application, step S4 further comprises:

[0044] Step S41: Calculate the expected real-time water consumption coefficient based on the real-time operation data of the production line; calculate the conservative upper limit of production capacity based on the different lower limits of confidence in the confidence water volume envelope, the expected real-time water consumption coefficient, and the process optimization coefficient.

[0045] Step S42: Determine the aggressive production capacity limit based on the historical maximum production capacity of the study area, the total market demand, and the rated production capacity of the equipment. Based on the conservative production capacity limit, the aggressive production capacity limit, and the production restriction priority, generate a conservative production scheduling plan, an aggressive production scheduling plan, and a combination of phased production restriction and process adjustment.

[0046] Step S43: Link the production scheduling plan with the production scheduling system and process control system to automatically switch or smoothly adjust the plan according to the real-time water supply situation, and trigger the emergency adjustment plan in extreme water shortage scenarios.

[0047] According to another aspect of this application, a "production determined by water availability" system is provided, comprising:

[0048] At least one processor; and

[0049] A memory communicatively connected to at least one of the processors; wherein,

[0050] The memory stores instructions that can be executed by the processor to implement the "production determined by water" method described in any of the above technical solutions.

[0051] Beneficial effects: By adopting the "production determined by water" approach and through the whole-process technical design of "water resource constraint perception - process water consumption optimization - network collaborative simulation - precise production restriction scheduling", dynamic adaptation and optimal matching of water resource supply and production activities are achieved. Attached Figure Description

[0052] Figure 1 This is a flowchart of the present invention.

[0053] Figure 2 This is a flowchart of step S1 of the present invention.

[0054] Figure 3 This is a flowchart of step S2 of the present invention.

[0055] Figure 4 This is a flowchart of step S3 of the present invention.

[0056] Figure 5 This is a flowchart of step S4 of the present invention. Detailed Implementation

[0057] like Figure 1 As shown, the following technical solution is proposed. According to one aspect of this application, a method for "production determined by water availability" is provided, comprising the following steps:

[0058] Step S1: Collect data of the study area. Based on historical data of upstream water sources and water supply systems, construct a multi-scenario set of water volume for different time periods in the study area and calculate the confidence water volume envelope.

[0059] Step S2: Based on the historical operating data of the production line under different operating conditions, construct a process-water consumption response surface model and generate process optimization schemes and process optimization coefficients;

[0060] Step S3: Construct water network diagram and production network diagram respectively. Based on the confidence water volume envelope band, couple water network diagram and production network diagram to construct a two-layer structure of water network-production network and calculate production restriction priority.

[0061] Step S4: Calculate the expected real-time water consumption coefficient based on the real-time operation data of the production line; determine the aggressive capacity limit based on the historical maximum capacity of the study area, market demand, and rated capacity of the equipment; calculate the conservative capacity limit based on different confidence lower limits in the confidence water volume envelope, the expected real-time water consumption coefficient, and the process optimization coefficient; and generate a multi-level robust production scheduling scheme and link it for control based on the conservative capacity limit, the aggressive capacity limit, and the production restriction priority.

[0062] like Figure 2 As shown, according to one aspect of this application, step S1 further comprises:

[0063] Step S11: Collect historical data of upstream water sources, historical data of water supply systems, basic data of processes, and historical operating data of production lines under different operating conditions in the study area. The historical data of upstream water sources and historical data of water supply systems include time series of water inflow and supply in the same season, statistical data of water supply facility failures, and data on fluctuations in unconventional water production.

[0064] Step S12: Based on historical data of upstream water sources and historical data of water supply systems, construct a set of multiple water volume scenarios for each future time period, and calculate the probability of each scenario.

[0065] Step S13: For each future time period, based on its corresponding water volume multi-scenario set and scenario probability, calculate at least two lower limits for water volume at different confidence levels, including the lower limit for high confidence level and the lower limit for low confidence level, to form the confidence water volume envelope for that time period.

[0066] First, the water volume scenarios are sorted in ascending order of water volume value. A discrete cumulative probability distribution function of water volume is constructed by accumulating the scenario probabilities in ascending order. Then, the water volume quantiles at continuous confidence levels are calculated from the discrete scenario set using the left-hand cumulative interpolation method combined with linear interpolation technology. The lower limit of water volume corresponding to each confidence level is determined (the probability that the actual water volume is ≥ the lower limit is not lower than the target confidence level). Among them, 95% confidence level is selected for high confidence level and 80% confidence level is selected for low confidence level. The selection is based on the industry technical specifications for water resource planning and water supply security. 95% confidence level is suitable for the high standard risk prevention and control requirements of extreme water shortage conditions, and 80% confidence level is suitable for the routine water supply security requirements of daily scheduling. The combination of the two forms a confidence water volume envelope covering extreme and routine conditions. A 95% high confidence level is a high-standard risk control threshold in the industry, corresponding to the water supply guarantee requirements under extreme water shortage conditions. It meets the core needs of flood control and drought relief, urban and rural water supply security, and rigid water supply for key water users, ensuring that the water volume is not lower than this lower limit with a 95% probability, and avoiding the risk of water shortage with an extremely low probability.

[0067] The 80% low confidence level is within the industry's standard water supply guarantee threshold, corresponding to the general needs of daily water resource scheduling and water use planning. It balances water supply guarantee efficiency with the rationality of water resource utilization and is suitable for routine scheduling scenarios under non-extreme operating conditions.

[0068] In one embodiment, specifically:

[0069] Arrange the water volume values ​​of each scenario in the future time period's multi-scenario water volume set in ascending order from smallest to largest to obtain an ordered water volume sequence Q1≤Q2≤…≤Q n The probabilities for each scenario are P1, P2, ..., P. n Where n is the total number of scenes, the scene probabilities are accumulated sequentially in ascending order of water volume to obtain the discrete cumulative water volume probability distribution function:

[0070] F(Q i )=P(Q≤Q i )= ;

[0071] Among them, F(Q) i This indicates that the actual water consumption is no greater than Q. i The cumulative probability, 1-F(Q) i ) represents the cumulative probability that the actual water consumption is greater than Qi.

[0072] The confidence level is the probability threshold for ensuring water supply, requiring that the probability of actual water supply being greater than or equal to the lower limit of water supply is not less than the target confidence level α, i.e., satisfying the constraint:

[0073] 1-F(Q α )≧α;

[0074] Based on the aforementioned discrete cumulative distribution function, the lower limit of water quantity Q corresponding to the confidence level α is calculated using the left-hand cumulative interpolation method. α Specifically:

[0075] If there exists scenario i such that 1-F(Q) i If ) = α, then directly take Q. i This serves as the lower limit for water volume at the confidence level α.

[0076] If there is no cumulative probability of exact matching, select the one that satisfies 1-F(Q). α The minimum water volume value Q ≥ α i Based on this, and combining the water volume and cumulative probability of adjacent scenes, the quantiles at continuous confidence levels are calculated:

[0077] ;

[0078] Three types of basic data were collected for the study area: historical data of upstream water sources, historical data of the water supply system, and historical operating data of the production line under different operating conditions.

[0079] The historical data on upstream water sources mainly includes inflow data from major reservoirs and rivers within the basin (such as average monthly inflow, average daily inflow, and interannual inflow variation data), precipitation data (such as hourly precipitation, daily precipitation, and monthly precipitation data), and evaporation data; historical data on the water supply system covers time series of inflow and supply volumes for the same season (used to analyze seasonal water supply patterns), and statistical data on water supply facility malfunctions (including malfunction types of pumps, pipe networks, and water treatment equipment, such as mechanical malfunctions, electrical malfunctions, and water quality non-compliance malfunctions, as well as the occurrence time and duration of each malfunction). Data on unconventional water production fluctuations (including daily and monthly production data of reclaimed water plants, rainwater collection data, and factors affecting production fluctuations such as influent water quality, treatment process parameters, and weather conditions); historical operating data of production lines under different operating conditions, including process parameters (such as reaction temperature, pressure, feed rate, residence time, and raw material composition), production parameters (such as output, load rate, and equipment operating status), and water consumption parameters (such as water consumption per unit product, total water consumption, and reuse rate).

[0080] According to one aspect of this application, step S12 further comprises:

[0081] Step S12a: Based on the pre-constructed correlation model between historical precipitation and inflow, divide the water into three levels: abundant water, normal water, and dry water, and set the natural inflow fluctuation range for each level.

[0082] Step S12b: Based on fault statistics, classify the faults into four levels: no faults, minor faults, general faults, and serious faults, and set the reduction range for each level.

[0083] Step S12c: Based on the historical production data of the reclaimed water station, divide it into three levels: high fluctuation, medium fluctuation and low fluctuation. Combine the natural water fluctuation scenario, the water supply facility failure scenario and the unconventional water production fluctuation scenario in all dimensions to generate a basic scenario set, and obtain the water volume multi-scenario set corresponding to each future time period.

[0084] Step S12d: Based on historical data, the occurrence frequency of each basic scenario is statistically analyzed, combined with industry prior probabilities, and the historical frequencies are corrected to obtain the final scenario probability.

[0085] Based on preprocessed historical data of upstream water sources and water supply systems, a multi-dimensional combination method is used to construct a set of multiple water volume scenarios for each future time period. The scenario probabilities are then calculated through historical statistics and prior correction. Specifically:

[0086] Based on a correlation model between historical precipitation and upstream water inflow, the quantitative relationship between precipitation and inflow is clarified. According to the statistical characteristics of historical inflow data (such as mean, standard deviation, and quantiles), three levels of water levels are defined: abundant water, normal water, and dry water. For example, using the multi-year average inflow as a benchmark, an inflow ≥ 1.2 times the multi-year average is considered abundant water, 0.8-1.2 times the multi-year average is considered normal water, and ≤ 0.8 times the multi-year average is considered dry water. For each level, natural inflow fluctuation ranges are set based on historical fluctuation data, such as ±5% during abundant water season (relatively stable inflow), ±8% during normal water season, and ±12% during dry water season (greater fluctuations during dry water season).

[0087] Based on statistical data on water supply facility failures, the frequency and impact of different types of failures were analyzed, and four levels were classified: no failure, minor failure, general failure, and severe failure. No failure means the facility is operating normally with no loss of water supply; minor failure means the failure has a small impact, reducing water supply by 5%-10% (e.g., failure of a single small water pump); general failure means a reduction in water supply of 10%-20% (e.g., partial failure of multiple pumps operating in parallel, or localized pipeline leakage); severe failure means a reduction in water supply of more than 20% (e.g., failure of a main water pump, rupture of the main pipeline, shutdown of the water treatment system). For each failure level, a specific reduction in water supply was set.

[0088] Based on historical production data of reclaimed water stations and rainwater harvesting systems, the statistical patterns of production fluctuations were analyzed and classified into three levels: high fluctuation, medium fluctuation, and low fluctuation. High fluctuation is defined as a production fluctuation of ≥15% (such as sudden changes in the quality of reclaimed water influent or a sudden increase in rainwater collection caused by heavy rain), medium fluctuation is 8%-15%, and low fluctuation is ≤8% (normal operating conditions).

[0089] The above-mentioned natural water flow fluctuation scenarios, water supply facility failure scenarios, and unconventional water production fluctuation scenarios are combined in all dimensions to generate a basic scenario set, such as abundant water + no failure + low fluctuation, dry water + severe failure + high fluctuation. The water volume multi-scenario set corresponding to each future period is composed of these basic scenarios to ensure coverage of all possible water supply fluctuation situations.

[0090] First, the frequency of occurrence of each basic scenario is statistically analyzed based on historical data. For example, the number of occurrences of the "dry water + severe failure + high fluctuation" scenario from 2018 to 2023 is statistically analyzed and divided by the length of the presidential time to obtain the historical frequency. Since some extreme scenarios (such as severe failure combined with high fluctuation) occur less frequently in historical data, calculations based solely on historical frequencies will lead to large probability deviations. Therefore, it is necessary to combine industry prior probabilities (based on water supply fluctuation statistics of similar industrial parks, determined by industry experts) to correct the historical frequencies. The correction method can adopt a weighted average method, for example, with a historical frequency weight of 0.7 and an industry prior probability weight of 0.3, to finally obtain the final scenario probability for each basic scenario, ensuring the scientificity and reliability of the probability calculation.

[0091] For each future time period, based on its corresponding set of multiple water volume scenarios and the final probability of each scenario, the lower limit of water volume under different confidence levels is calculated to form a confidence water volume envelope. In this embodiment, the predicted water supply value under each basic scenario is first defined. This value is obtained by deducting the water supply loss caused by faults from the natural water supply, the water supply from conventional water supply facilities, and the water supply from unconventional water supply. All basic scenarios in this time period are sorted in ascending order of predicted water supply value, and the cumulative probability is calculated based on the scenario probability. At least two different confidence levels are set according to the control requirements, usually including a high confidence level (such as 9). 5% (to ensure water supply security in extreme scenarios) and low confidence levels (e.g., 80%, to balance water supply security and production efficiency); find the water supply value corresponding to the cumulative probability reaching the set confidence level, which is the lower limit of water volume at that confidence level. The lower limit of water volume corresponding to the high confidence level is a more conservative water supply security value, and the lower limit of water volume corresponding to the low confidence level is a more lenient water supply security value; use the lower limits of water volume corresponding to two or more confidence levels as boundaries to form the confidence water volume envelope for that period. For example, the confidence water volume envelope for April 2024 (dry season) is 580,000-650,000 m³. 3 580,000 m 3 The lower confidence limit is 95%, 650,000 m³ 3 The 80% confidence lower limit is used to clarify the range of water supply capacity under different levels of protection during this period.

[0092] like Figure 3 As shown, according to one aspect of this application, step S2 further comprises:

[0093] Step S21: Extract historical operating data of the production line under different operating conditions, construct an operating sample set, and use data-driven fitting to obtain the process-water consumption response surface model;

[0094] Step S22: Continuously collect new operating data sample sets, correct the process-water consumption response surface model, identify whether the current operating conditions deviate from the training data range, and obtain the optimal process-water consumption response surface model.

[0095] Step S23: Calculate the partial derivatives of the optimal process-water consumption response surface model with respect to each process parameter. Based on the feasible range of process parameters, production load constraints, and product quality constraints, construct a constrained optimization problem and solve it to obtain the optimal process parameters.

[0096] Step S24: Based on the optimal process parameters and the unit water consumption corresponding to the current process parameters, calculate the process optimization coefficient and generate specific control instructions, i.e., the process optimization scheme.

[0097] Among them, the control commands adopt a standardized structured format of "head-body-tail", and achieve accurate conversion of optimal process parameters into control commands through a four-step mapping logic of "parameter standardization → threshold verification → format encapsulation → command verification".

[0098] The control commands adopt a "standardized structured data format". The command header occupies 8 bytes and contains the command identifier, priority, timestamp, and core metadata. The command body is the control target value and parameter number of the optimal process parameters, which is organized in the form of "parameter number-target value" key-value pairs. The command tail occupies 2 bytes and is the command end identifier, which is fixed as "FF-FF" and is used by the distributed control system (DCS system) to confirm that the command parsing is complete.

[0099] For example, under normal priority conditions, adjust the reaction temperature to 85.50℃ and the feed flow rate to 12.30m³. 3 / h control command format (simplified hexadecimal representation): Header: 4F50432D3031-02 20260815143025123-0A Body: 02-001-00855001-002-00123002 Tail: FFFF.

[0100] In this embodiment, a four-step mapping process of "parameter standardization → threshold verification → format encapsulation → instruction verification" is adopted, and the specific process is as follows:

[0101] Standardize the parameter numerical format and convert all parameters into structured data of "numerical value + unit". The unit must be consistent with the parameter unit in the DCS system.

[0102] According to the coding rules of the production line process parameters, each optimal parameter is matched with a unique parameter number to ensure that it corresponds one-to-one with the parameter identifier of the DCS system.

[0103] Based on the safe operating thresholds of production line equipment (preset in the system, including upper and lower limits of parameters and maximum adjustment range), the standardized optimal process parameters are subjected to safety verification, including:

[0104] Verify whether the target value of the parameter is within the safe threshold range. If it exceeds the range, trigger an alarm, terminate the mapping process, and return parameter optimization suggestions.

[0105] Verify whether the difference between the optimal parameters and the current process parameters exceeds the maximum adjustment range (e.g., the maximum single temperature adjustment should not exceed 5℃, and the maximum single flow rate adjustment should not exceed 2m). 3 If the value exceeds / h, the mapping is split into multi-step progressive control commands to avoid equipment failure or process fluctuations caused by parameter mutations.

[0106] In accordance with the above-mentioned format requirements of "instruction header-instruction body-instruction tail", the optimal process parameters that have passed the verification are encapsulated into structured control instructions.

[0107] The complete control command after encapsulation is subjected to integrity verification. The check code is calculated and compared with the check bit in the command header. If they match, the command is deemed valid and proceeds to the subsequent interface transmission stage. If they do not match, the encapsulation process is re-executed until a valid command is generated.

[0108] The interface adopts a unified architecture protocol for open platform communication that is common in the industry and complies with the IEC62541 standard. Specifically, it includes interface communication parameters, data interaction process and security specifications to ensure the parseability of control commands, transmission security and execution reliability.

[0109] In this embodiment, a multidimensional response surface of "process parameter vector p → unit water consumption f(p)" is constructed using data-driven or mechanistic models. By utilizing the gradient and constraint optimization capabilities of this model, a specific solution for "how to adjust process parameters such as temperature, flow rate, and dosage to save water" is automatically provided, and the process optimization coefficient is calculated accordingly, instead of relying on constants set arbitrarily.

[0110] The original "process optimization coefficient" was often set to a fixed value such as 0.95 or 0.9, lacking a clear physical or data basis, and it did not explain from which process data it could be derived. As a result, on the one hand, it was impossible to explain "why it is 0.95 instead of 0.93", and on the other hand, it was impossible to directly deduce "which process parameters need to be adjusted and by how much to achieve this 5% water saving". As a result, process optimization remained at the conceptual level and did not form an executable control strategy.

[0111] In this embodiment, firstly, during the offline phase, historical operating data of a process or production line under different operating conditions are collected to obtain sample pairs {p^{(n)}, f^{(n)}}, where p^{(n)} represents the combination of key process parameters such as temperature, flow rate, residence time, and dosage, and f^{(n)} represents the corresponding unit water consumption. Simultaneously, quality index q^{(n)} and safety constraint index s^{(n)} are recorded. Based on this, a polynomial response surface methodology, tree model, or neural network method is used to fit the function f. *(p), and in the modeling process, constraints q(p)≥q_{min} and s(p)≤s_{max} are added to obtain the "process-water consumption response surface" that meets the quality and safety requirements;

[0112] Secondly, during the online phase, new {p, f} data are continuously collected to calibrate the model, and f is calculated under the current operating condition p_{current}. * (p) with respect to each parameter, and construct a constrained small-scale optimization problem: minimize f̂(p) under the conditions that |p-p_{current}|≤Δp_{max}, q(p)≥q_{min}, and s(p)≤s_{max}, and obtain the local optimal process parameter p_{opt}.

[0113] Then, the process optimization coefficient is defined as β=f * (p_{opt}) / f * (p_{current}), β directly reflects the water-saving ratio that can be achieved in this process without sacrificing quality and safety;

[0114] Finally, p_{opt} is used to generate specific control instructions or operation suggestions, and β is used as input to the water-based production algorithm (such as the robust capacity limit formula in the improved method 1) to achieve closed-loop coupling between process optimization and water-constrained production scheduling.

[0115] In terms of system deployment architecture, the process-water consumption response surface model and scheduling optimization algorithm adopt a "cloud-edge collaboration" architecture: model training and full updates are completed on cloud servers, leveraging the computing power advantages of the cloud to process large-scale historical data; the trained model parameters are sent to edge computing nodes (industrial PCs or edge gateways deployed on the production site), and the edge nodes are responsible for real-time inference and incremental updates, with response latency controlled within 100ms; the edge nodes are connected to the process control system (DCS) via industrial Ethernet to realize the real-time issuance of control commands; at the same time, the edge nodes periodically (e.g., hourly) upload real-time operating data to the cloud for continuous model optimization and iteration. This deployment architecture takes into account both the computing power requirements for model training and the latency requirements for real-time control.

[0116] According to one aspect of this application, step S21 further comprises:

[0117] Step S21a: Extract historical operating data of the production line under different operating conditions and construct an operating sample set;

[0118] Step S21b: Construct process-water consumption response surface models based on different operating conditions using multinomial response surface model, random forest tree model, and BP neural network model respectively;

[0119] Step S21c: Divide the running sample set into a training set, a validation set, and a test set in a 6:2:2 ratio. Train the process-water consumption response surface model, calculate the goodness of fit, root mean square error, and mean absolute error, and verify the model accuracy. For models with accuracy less than the preset accuracy threshold, improve the model accuracy by adding parameter interaction terms, standardizing features, and weighted fusion of multinomial model and random forest to obtain the process-water consumption response surface model. The preset accuracy threshold includes the goodness of fit threshold and the root mean square error threshold.

[0120] In other alternative implementations, the following model fusion or alternative solutions may also be used:

[0121] Firstly, the Stacking ensemble method is adopted, which uses the multinomial response surface model, random forest model and BP neural network model as base learners, and trains a meta-learner (such as linear regression or logistic regression) to integrate the outputs of the base learners. This method can further improve the generalization ability of the model.

[0122] Secondly, the Bagging ensemble method is used to perform Bootstrap sampling and aggregation on the random forest model, thereby improving the model's stability;

[0123] Third, in scenarios where the relationship between process parameters and water consumption is relatively simple, a polynomial response surface model (including interaction terms) can be used to reduce model complexity and computational resource consumption.

[0124] Fourth, in scenarios with a large amount of historical data, more complex deep learning models can be used to capture temporal features, which is suitable for production lines where operating conditions change dynamically over time.

[0125] The selection of the above alternatives should be based on a comprehensive consideration of the specific operational complexity of the production line, the amount of data, and computing resources.

[0126] Establish a quantitative correlation model between process parameters and water consumption, specifically including:

[0127] Historical operating data of the production line under different operating conditions are extracted. After data preprocessing (outlier removal, missing value imputation, and data standardization), an operating sample set is constructed. The input features of the sample set are key process parameters (such as reaction temperature, pressure, feed rate, raw material composition, and residence time; parameters with significant impact on water consumption are selected through correlation analysis and importance ranking). The output label is water consumption per unit product (or total water consumption). To ensure the representativeness of the sample set, it needs to cover the main operating conditions of the production line, including normal operating conditions, load adjustment operating conditions, and raw material fluctuation operating conditions. The number of samples needs to meet the modeling requirements, usually no less than 500 sets, and it is recommended to have no less than 800 sets for complex production lines.

[0128] Based on the characteristics of different operating scenarios, three typical data-driven models are used to construct process-water consumption response surface models to take into account the correlation characteristics of different types of processes and water consumption:

[0129] The polynomial response surface model is suitable for scenarios where process parameters and water consumption are linearly or weakly nonlinearly related. It has the advantages of simple model and strong interpretability. It is constructed using a quadratic polynomial form, and its expression is:

[0130] W=a0+Σa i x i +Σa ij x i x j ;

[0131] Where W is the unit water consumption, x i x j For process parameters, a0, a i a ij These are model parameters;

[0132] The random forest tree model is suitable for scenarios where process parameters and water consumption have complex nonlinear relationships and parameter interactions. It has the advantages of strong anti-interference ability and good generalization performance. The model results are obtained by constructing multiple decision trees and integrating voting. Parameters such as the number of decision trees, maximum depth, and minimum number of sample splits are set.

[0133] The BP neural network model is suitable for scenarios with multiple process parameters and complex relationships. It has a strong nonlinear fitting ability and adopts a three-layer network structure of "input layer-hidden layer-output layer". The number of nodes in the input layer is the number of process parameters, the number of nodes in the output layer is 1 (unit water consumption), and the number of nodes in the hidden layer is determined by trial calculation.

[0134] The training parameters for the BP neural network model are set as follows: Mean squared error (MSE) is used as the loss function, expressed as:

[0135] L=(1 / n)∑(W_pred-W_real) 2 ;

[0136] Where W_pred is the model's predicted unit water consumption, W_real is the actual unit water consumption, and n is the number of samples; the Adam optimizer is used for parameter updates, with an initial learning rate of 0.001 and a learning rate decay coefficient of 0.9, decaying once every 100 epochs; the maximum number of training epochs is set to 1000, and training is stopped early when the validation set loss no longer decreases for 50 consecutive epochs to prevent overfitting; the ReLU function is used as the activation function in the hidden layer, and no activation function is set in the output layer to ensure the continuity of the output value.

[0137] The sample set was randomly divided into a training set (for model parameter learning), a validation set (for hyperparameter tuning), and a test set (for final accuracy verification) in a 6:2:2 ratio. The three models were trained using the training set, and hyperparameters (such as the order of the multinomial model, the number of decision trees in the random forest, and the number of hidden nodes and learning rate in the BP neural network) were tuned using the validation set. Model accuracy was verified using three core metrics:

[0138] R² goodness-of-fit reflects the model's ability to interpret data; the closer R² is to 1, the better the model fits. R² should be ≥ 0.85, and ≥ 0.9 is required for complex scenarios. Root mean square error (RMSE) reflects the average deviation between the model's predicted values ​​and the actual values; a smaller RMSE indicates higher accuracy. Mean absolute error (MAE) reflects the robustness of the model's predictions, avoiding interference from extreme values ​​in accuracy assessment. Models with accuracy below the threshold (e.g., R² < 0.85, RMSE > 0.05) are optimized. Optimization measures include adding parameter interaction terms (for multinomial models), standardizing features (reducing the influence of dimensions), and using a weighted fusion of multinomial models and random forests (combining the advantages of both to improve model accuracy and stability).

[0139] The selection criteria for the above precision thresholds are as follows:

[0140] Based on engineering practice in industrial process modeling, when the goodness-of-fit R² of the response surface model reaches 0.85 or higher, the prediction error of the model for changes in process parameters can usually be controlled within 5%, which can meet the accuracy requirements of process optimization. For complex scenarios with complex interactions of process parameters and high nonlinearity, raising the threshold to 0.9 is to ensure that the model still has reliable prediction capabilities under extreme conditions. The selection of the root mean square error threshold (RMSE) > 0.05 is based on the typical order of magnitude of unit water consumption. When the unit water consumption is 2-3 m³ / h... 3 When the range is / t, an RMSE of 0.05 corresponds to a relative error of about 2%, which is within an acceptable range in engineering practice. Ultimately, the model with the highest accuracy and best generalization performance was selected as the process-water consumption response surface model.

[0141] According to one aspect of this application, step S22 further comprises:

[0142] Step S22a: Collect new operational data;

[0143] Step S22b: Set the scrolling window size and the corresponding window increment threshold. When the number of newly collected samples reaches the window increment threshold, remove the earliest n groups of samples in the window, add new samples, and obtain a new set of running data samples. n is a positive integer greater than 20 and less than the window increment threshold.

[0144] Step S22c: Update the parameters of the process-water consumption response surface model using the least squares incremental method and the incremental learning algorithm respectively, and set an accuracy warning threshold. When the model accuracy is lower than the accuracy warning threshold, trigger a full retraining of the model.

[0145] Step S22d: Collect the feature cluster centers of all historical normal operating conditions, construct the operating condition benchmark library, calculate the Euclidean distance between the current operating condition features and the cluster centers of the benchmark library, and determine the operating condition deviation when the distance is greater than 3 times the standard deviation. Correct the process-water consumption response surface model until no deviation occurs, and obtain the optimal process-water consumption response surface model.

[0146] Because production line conditions can change during industrial production due to variations in raw material supply, equipment aging, and market demand, fixed models are prone to accuracy decline or even failure. Therefore, it is necessary to dynamically correct the optimal process-water consumption response surface model and identify deviations from the operating conditions in real time. In this embodiment, the specific steps are as follows:

[0147] Through the production line's real-time monitoring system, new operational data is continuously collected, including real-time process parameters, production parameters, and water consumption data, ensuring the real-time nature and accuracy of data collection, and matching the collection frequency with the precision of production control, such as collecting one set of data per hour.

[0148] Set the scrolling window size (e.g., 400 or 600 groups, determined based on the frequency of production line condition changes) and the window increment threshold (e.g., 40 or 50 groups, i.e., triggering a sample set update when the number of newly collected samples reaches this threshold); when the number of newly collected samples reaches the window increment threshold, the "first-in, first-out" principle is adopted to remove the earliest n groups of samples in the scrolling window (n is the number of newly collected samples, and n is a positive integer greater than 20 and less than the window increment threshold to ensure the timeliness and stability of the sample set), and add the newly collected samples to obtain a new running data sample set;

[0149] The parameters of the process-water consumption response surface model were updated using both the least squares incremental method (for polynomial response surface models) and incremental learning algorithms (such as incremental random forest and incremental BP neural network algorithms for nonlinear models). After the update, the accuracy index (R0) of the model was calculated. 2 RMSE, MAE), set accuracy warning thresholds (e.g., R... 2 (<0.88, RMSE>0.04) When the model accuracy falls below the warning threshold, a full retraining is triggered to ensure that the model always maintains high accuracy;

[0150] Collect feature data from all historical normal operating conditions, and use the K-means clustering algorithm to cluster the feature data to obtain feature cluster centers for different normal operating conditions, thus constructing an operating condition benchmark library. Calculate the Euclidean distance between the current operating condition feature (i.e., the current combination of process parameters) and each cluster center in the benchmark library in real time, and calculate the standard deviation of historical feature data. When the Euclidean distance between the current operating condition feature and all cluster centers is greater than 3 times the standard deviation, it is determined to be an operating condition deviation (i.e., the current operating condition exceeds the range of historical normal operating conditions, such as abnormal raw material composition or equipment failure precursors). At this time, process optimization based on the existing model needs to be suspended, an abnormal operating condition alarm needs to be triggered, and more samples need to be collected to retrain the model to ensure that the model adapts to the new operating conditions.

[0151] When a deviation from the operating condition is detected, the system enters an exception handling mode. The specific handling process is as follows:

[0152] First, pause the process optimization calculation based on the current response surface model, and keep the current process parameters unchanged or switch to the preset conservative process parameter combination;

[0153] Secondly, an abnormal operating condition alarm is triggered to notify operators to investigate the cause of the deviation (such as abnormal raw materials, early signs of equipment failure, etc.). Simultaneously, the system automatically collects operational data under abnormal conditions at a higher frequency (e.g., one set every 15 minutes). When the accumulated sample size reaches 50 sets or more, an incremental learning algorithm is used to update the response surface model. The updated model needs to pass accuracy verification (R²). 2 (≥0.85) After verification, the system will automatically exit the abnormal handling mode and resume normal process optimization functions; if the accuracy requirements are still not met after 3 consecutive model updates, a full retraining will be triggered, and it is recommended that technical personnel intervene to analyze the root cause of the deviation in operating conditions.

[0154] Calculate the partial derivatives of the optimal process-water consumption response surface model with respect to each process parameter to determine the sensitivity of each parameter to water consumption (the larger the absolute value of the partial derivative, the more significant the parameter's impact on water consumption). Based on the actual operational requirements of the process, construct a constrained optimization problem, where the objective function is to minimize unit water consumption (i.e., minW=f(x1, x2, ..., x...). n f is the optimal response surface model, x1, x2, ..., x nThe constraints include the feasible range of process parameters (e.g., reaction temperature ≤850℃, ≥780℃, pressure ≤3.0MPa, ≥2.0MPa, feed rate ≤60t / h, ≥40t / h, etc., determined by equipment rated parameters and product quality requirements), production load constraints (e.g., capacity load rate ≥70% to ensure production economy), and product quality constraints (e.g., product purity ≥99.5%, determined by market demand). An efficient optimization algorithm is used to solve this constrained optimization problem, obtaining the optimal combination of process parameters that satisfies all constraints and minimizes unit water consumption.

[0155] Based on the unit water consumption W1 corresponding to the optimal process parameters and the unit water consumption W0 corresponding to the current process parameters, the process optimization coefficient η is calculated using the formula η=W1 / W0 (η<1, the smaller the value, the better the water-saving effect); for example, the unit water consumption W0 corresponding to the current process parameters is 2.5m. 3 / t, the unit water consumption W1 = 2.1m³ corresponding to the optimal process parameters. 3 If the process optimization coefficient η = 0.84, it indicates that the unit water consumption can be reduced by 16% through process optimization. Based on the optimal process parameters obtained from the solution, specific process adjustment control instructions are generated to clarify the adjustment direction and adjustment range of each process parameter, such as "adjusting the reaction temperature from the current 810℃ to 830℃", "adjusting the pressure from 2.3MPa to 2.8MPa", and "adjusting the feed rate from 52t / h to 45t / h". The control instructions are transmitted to the process control system of the production line to guide the on-site operators or automatic control system to complete the process parameter adjustment.

[0156] In one embodiment, specifically:

[0157] Historical operating data from 2020 to 2023 were extracted from the core ethylene cracking production line among the 12 production lines in the park, including parameters such as reaction temperature (T), pressure (P), feed rate (V), and unit water consumption (W), to construct an operating sample set of 800 groups;

[0158] The process-water consumption response surface model was constructed using a multinomial response surface model (quadratic multinomial), a random forest tree model (100 decision trees), and a BP neural network model (3 hidden layers), respectively.

[0159] The dataset was divided into three sets: a training set of 480 groups, a validation set of 160 groups, and a test set of 160 groups, in a 6:2:2 ratio. After training, the accuracy metric, R0, was calculated. 2 =0.82, RMSE=0.05, MAE=0.03; Random Forest Model R 2 =0.91, RMSE=0.02, MAE=0.015; BP neural network model R 2=0.88, RMSE=0.025, MAE=0.018; due to insufficient accuracy of the polynomial model (R 2 <0.85), using a "multinomial + random forest weighted fusion" (weights 0.3:0.7), the fused model R 2 =0.93, RMSE=0.018, MAE=0.012, which meets the accuracy threshold (R 2 (≥0.9), to obtain the optimal process-water consumption response surface model.

[0160] We continuously collect operational data from ethylene cracking production lines from January to June 2024, adding 40 new samples each month.

[0161] Set the scrolling window size to 400 groups and the incremental threshold to 40 groups. When the number of new samples reaches 40 groups, remove the earliest 40 groups of samples in the window and update the sample set.

[0162] The fusion model parameters are updated using the least squares incremental method, and an accuracy warning threshold (R0) is set. 2 <0.88);

[0163] Collect historical normal operating condition feature cluster centers (based on K-means clustering, K=5) to construct an operating condition benchmark library. Calculate the Euclidean distance between the current operating condition features (T=820℃, P=2.5MPa, V=50t / h) and the cluster centers in the benchmark library for a certain period in May 2024. The result is 3.2σ (σ is the standard deviation of historical features). If the result is determined to be an operating condition deviation, the model will be retrained.

[0164] Calculate the partial derivatives of the optimal response surface model with respect to T, P, and V, and construct a constrained optimization problem (constraints: T∈[780,850]℃, P∈[2.0,3.0]MPa, V∈[40,60]t / h, objective function: minW). Solve the optimal process parameters using the particle swarm optimization algorithm: T=830℃, P=2.8MPa, V=45t / h.

[0165] The current process parameters correspond to a unit water consumption of W0 = 2.5m³. 3 / t, the unit water consumption W1 = 2.1m³ corresponding to the optimal process parameters. 3 / t, process optimization coefficient η=W1 / W0=0.84; generate specific control instructions: adjust the reaction temperature from the current 810℃ to 830℃, the pressure from 2.3MPa to 2.8MPa, and the feed rate from 52t / h to 45t / h.

[0166] After implementing the aforementioned process optimization instructions on the ethylene cracking production line, monitoring data from one month of continuous operation showed that the unit water consumption decreased from 2.5 m³ / h. 3 / t stabilized and dropped to 2.15m 3 / t, the actual optimization coefficient is 0.86, which deviates from the calculated value of 0.84 by 2.4%, meeting the accuracy requirements; the comparative production line that did not implement optimization during the same period had a unit water consumption fluctuation range of 2.4-2.6m. 3 The original production line fluctuated by 8% per ton, while the optimized line only fluctuated by 3%. Furthermore, through deviation identification, an abnormal condition in May 2024 (excessive feed impurities) was successfully detected, allowing for timely model retraining and preventing water consumption control failure. Compared to the traditional fixed-quota method, this production line saves an average of 1200 cubic meters of water per month. 3 The water-saving benefits are significant.

[0167] like Figure 4 As shown, according to one aspect of this application, step S3 further comprises:

[0168] Step S31: Construct a water network diagram based on the basic data of the process, with the process or device as the node and various water flows as the edge. The node attributes include water consumption, reuse amount, and water quality requirements, and the edge attributes include water quantity and water quality changes.

[0169] Water network diagrams are used to accurately depict the flow, distribution, and transformation patterns of water resources within industrial systems, providing a foundation for analyzing the impact of production restrictions on water consumption. During construction, each process or independent production unit of an industrial enterprise is used as a node, covering all major water-using units; the flow paths of various water flows are used as edges, with the edge direction consistent with the water flow direction; core attributes are labeled for each node, including water consumption under normal operating conditions, reuse volume (reclaimed water reuse volume per unit time), total wastewater discharge, water quality requirements, and water treatment costs; core attributes are labeled for each edge, including water flow rate (water flow rate per unit time), water quality changes (such as changes in conductivity, chemical oxygen demand (COD), turbidity, and other water quality indicators after water flows through a process), pipeline resistance loss, and water transmission costs; a visual water network diagram is constructed using graph structure modeling tools.

[0170] Step S32: Construct a production network diagram based on process basic data, with processes or production lines as nodes and material dependencies as edges. Node attributes include output value, output, and order dependency degree.

[0171] Production network diagrams are used to depict the material flow and production relationships within an industrial system. During construction, processes or production lines are used as nodes, each covering all core production units. Material dependencies between processes are used as edges, with the edge direction aligned with the material flow direction. For example, if the ethylene cracking process supplies material to the polyethylene process, an edge is constructed pointing from the ethylene cracking node to the polyethylene node. Each node is labeled with core attributes, including output value under normal operating conditions (economic output per unit time, such as 10,000 RMB / h or 10,000 RMB / month), output (product output per unit time, such as t / h or t / month), order dependency (the proportion of core orders to the output of this process; a higher proportion indicates greater importance of this process for fulfilling core orders), production load rate, equipment depreciation cost, and labor cost. Each edge is labeled with core attributes, including material transfer rate (material transfer volume per unit time), material purity requirements, and transfer loss rate. Similarly, graph structure modeling tools are used to construct a visual production network diagram.

[0172] In this embodiment, a collaborative identification mechanism of "multi-source data fusion + rule matching + topology verification" is adopted to achieve automated and accurate identification of water flow relationships and material dependencies, as detailed below:

[0173] Automatic water flow relationship identification data acquisition foundation: Integrate three major categories of data: real-time monitoring data from DCS system, as-built drawings of water supply and drainage pipe network, and production ledgers, to construct a water flow data fusion pool;

[0174] If there is a physical pipe directly connecting the outlet of process A and the inlet of process B, and the pipe flow sensor detects continuous water flow (flow rate ≥ 0.1 m³ / h), then... 3 If the flow rate is / h, then it is determined that there is a direct water flow relationship between A and B;

[0175] If the effluent water quality parameters (such as COD, ammonia nitrogen, pH value) of process A match the influent water quality requirements of process B by ≥90%, and there is an indirect pipeline link between the two (connected through an intermediate water storage facility / treatment unit), then an indirect water flow relationship is determined to exist.

[0176] If the wastewater from a process is treated and then returned to the same process or another process, and the return pipe label is consistent with the "recycled water" record in the production ledger, then it is determined that there is a return water / reuse flow relationship.

[0177] Based on the initial topology of the water network diagram, verify whether the identified water flow relationships form a reasonable topological link;

[0178] Collect production process cards (including material input / output lists for each process), order data from the Enterprise Resource Planning (ERP) system (including material requirements planning), and production execution data from the Manufacturing Execution System (MES) system (including material transfer records between processes and material batch traceability information) to build a material data association database;

[0179] If the material input list of process B contains the core materials (accounting for ≥80%) in the material output list of process A, and the MES system has material transfer records from A to B, then A is determined to be the upstream material supply process of B, and the two have a direct material dependency relationship.

[0180] If the production tasks of two processes are both associated with the same order number, and the production completion time of process A is earlier than the production start time of process B, it is determined from the process logic that there is an indirect material dependency relationship between the two.

[0181] Based on the preset production process flow chart, verify whether the identified material dependencies conform to the process sequence. For example, the raw material pretreatment process should take precedence over the reaction process to avoid logically reversed dependencies.

[0182] The system collects records of material transfer anomalies in real time (such as material shortages and transfer delays), dynamically corrects the identified material dependencies, and marks a dependency as "weak dependency" if the material transfer corresponding to a dependency is abnormal three times in a row and triggers manual review.

[0183] Simultaneously, based on the dynamic differences in node attributes, differentiated collection frequencies are set, and a three-level update mechanism of "real-time collection - scheduled update - anomaly trigger" is established:

[0184] The water network diagram node attributes are collected as dynamic attributes (water consumption, reuse volume) and static attributes (water quality requirements).

[0185] The dynamic attributes are acquired in real time at a high frequency of 1 minute, and the data source is the real-time data of the water flow sensor in the DCS system.

[0186] Static attributes are collected at a low-frequency, periodic rate of once a month. If the production process is adjusted, immediate collection is triggered. The data sources are process technology documents and water quality test reports.

[0187] Production network diagram node attributes:

[0188] Dynamic attributes (output value, output): Medium-frequency timed data collection is adopted, with a collection frequency of 1 hour / time. The data sources are production output statistics from the MES system and output value accounting data from the ERP system.

[0189] Static attributes (order dependency): The data is collected synchronously throughout the order cycle, that is, it is collected in real time whenever a new order is received or the order status changes. The data source is the order data of the ERP system. The order dependency is quantified according to the proportion of the capacity utilization of the process by the order (e.g., if the capacity utilization is 100%, the dependency is 1, and if the capacity utilization is 50%, the dependency is 0.5).

[0190] For frequently collected dynamic attributes (such as water consumption and output), a real-time pipeline mode of "collection-cleaning-update" is adopted. After data collection, outliers are removed (such as data exceeding the sensor range), data is smoothed (to eliminate instantaneous fluctuations), and the data is immediately updated to the network graph node attribute library with an update delay of ≤3 seconds.

[0191] For attributes collected at medium and low frequencies (such as output value and water quality requirements), batch updates are performed daily from 00:00 to 02:00, and data consistency is verified synchronously (such as the matching of output value and output, and the consistency of water quality requirements with process documents). Update logs are generated and kept for future reference.

[0192] A dual storage mode of "standardized file storage + structured database storage" is adopted. The file storage uses the graphical markup language format as the standard storage file format for the network graph; the database uses a relational database to store node tables, edge tables, and attribute association tables.

[0193] Water network diagram data structure:

[0194] Node table: contains node ID (primary key), node name (process / equipment name), node type (water-using process / recycled water process / water treatment unit), water consumption, reuse volume, water quality requirements (multiple fields such as COD, ammonia nitrogen, pH value, etc.), geographical coordinates, and data update timestamp;

[0195] Edge table: contains edge ID (primary key), starting node ID (foreign key related node table), ending node ID (foreign key related node table), water flow type (fresh water / reclaimed water / wastewater), water volume, water quality change value (difference between influent and effluent water quality), pipe number, and edge status (normal / abnormal).

[0196] Attribute association table: Records the association between node / edge attributes and data acquisition sources, including attribute ID, data acquisition device ID, data source system, and acquisition frequency.

[0197] Production network diagram data structure:

[0198] Node table: contains node ID (primary key), node name (process / production line name), node type (raw material pretreatment / reaction / finished product packaging, etc.), output value, output, order dependency, order number, production status (running / stopping / maintenance), and data update timestamp;

[0199] Edge table: contains edge ID (primary key), starting node ID (foreign key related node table), ending node ID (foreign key related node table), material type, material transfer volume, dependency relationship type (strong dependency / weak dependency), and edge status (normal / interrupted);

[0200] Attribute association table: includes attribute ID, data collection source ID, data source system (ERP / MES), collection frequency, and update mechanism type.

[0201] Step S33: Determine water resource constraints based on the confidence water volume envelope. For each candidate production restriction process, simulate the impact of partial or complete shutdown on the water network and production network. Calculate the changes in water consumption and output value of the entire system. Calculate the marginal water benefit index based on the ratio of the changes in water consumption to the changes in output value of the entire system. Determine the production restriction priority based on the ranking of marginal water benefit index values.

[0202] When making production restriction decisions, each process is no longer treated as an independent entity, and the "water consumption / output value ratio" is no longer used for ranking. Instead, a two-layer structure of "water network diagram + production network diagram" is constructed. The water reuse relationship between processes and the material dependence relationship are comprehensively considered. The "marginal shutdown benefit" of each candidate shutdown process is calculated, and a more reasonable production restriction priority is determined accordingly.

[0203] The original rule of "prioritizing production restriction for high water consumption and low output value" essentially ranks processes according to a single-point indicator (such as water consumption per unit of output value). This approach ignores two types of key couplings: first, water network coupling, that is, the drainage of one process may be an important source of recycled water for another process, and stopping the former will significantly reduce the usable water of the latter; second, production network coupling, that is, once some preceding processes are stopped, the entire subsequent high-output production line will be forced to shut down, resulting in an output value loss that is far greater than the output value contribution of the process itself. Simple single-point ratio ranking often leads to a production restriction plan that "looks profitable locally but is disadvantageous overall".

[0204] In this embodiment, firstly, at the water network level, processes or devices are used as nodes, and water flows such as circulating water, reclaimed water, and cooling water are used as edges. Node attributes include water consumption, reuse volume, and water quality requirements, while edge attributes include water volume and water quality changes, thereby constructing a water network diagram within the enterprise.

[0205] Secondly, at the production network level, a production network diagram is constructed with processes or production lines as nodes, material sequence and process dependence as edges, and node attributes including output value, output, and key order dependence.

[0206] Then, under the given water resource constraints (which can come from the robust water quantity constraints of the improved method one), for each candidate production restriction process i, the impact of "partial or complete shutdown" on the entire water network and production network is simulated, the change in water consumption ΔW_i and the change in output value ΔV_i of the entire system are calculated, and then the "marginal water benefit" index is defined.

[0207] Finally, priority is given to implementing production restrictions on processes that result in the greatest reduction in system water consumption per unit of output loss. Based on this, and combined with the adjustable space and water constraints provided by the process optimization method, a phased combination of production restriction and process adjustment schemes is dynamically generated to improve the overall efficiency and systematic nature of production restriction decisions.

[0208] According to one aspect of this application, step S33 further comprises:

[0209] Step S33a: Determine different water resource constraint levels based on the confidence water volume envelope, including: severe water shortage, moderate water shortage and mild water shortage. Design simulation scenarios for each constraint level, wherein different reduction ratio gradients are set for each candidate production restriction process, and simulate the system response under different reduction degrees.

[0210] Severe water shortage: Water supply ≤ lower limit of high confidence water supply (e.g., ≤ 580,000 m³) 3 ( / month), the contradiction between water supply and demand is prominent, and production needs to be significantly restricted;

[0211] Moderate water shortage: Water supply falls between the high confidence lower limit and the low confidence lower limit (e.g., 580,000-650,000 m³). 3 ( / month), appropriate production restrictions are necessary;

[0212] Mild water shortage: Water supply ≥ lower confidence level (e.g., ≥ 650,000 m³) 3 ( / month), water resources basically meet the demand, and there is no need for production restrictions or only minor production restrictions;

[0213] For each constraint level, design a corresponding simulation scenario, select all possible candidate production restriction processes (usually processes with large water consumption or significant differences in output value contribution), set different production reduction ratio gradients for each candidate production restriction process (such as 10%, 20%, 30%, 50%, 100%, i.e. partial production reduction or complete shutdown), and clarify the specific conditions of each simulation scenario (such as constraint level, production restriction process, and production reduction ratio).

[0214] Step S33b: Simulate the difference between the total water consumption and total output value of the entire system before and after the production reduction, calculate the changes in water consumption and output value of the entire system, calculate the impact of core orders based on the ratio of the number of core orders affected by production restrictions to the total number of core orders, and calculate the production stability index based on the connectivity of the production network and the continuity of material flow.

[0215] Import the constructed water network diagram and production network diagram into the simulation model, and set the initial parameters of each node and edge (based on historical normal operation data); for each simulation scenario, simulate the system response of the candidate production restriction process under the set production reduction ratio, that is, calculate the changes of key indicators of the entire system before and after the production reduction:

[0216] Change in total system water consumption ΔW: The difference in total system water consumption before and after production reduction (ΔW=W0-W1, where W0 is the total water consumption before production reduction and W1 is the total water consumption after production reduction). The larger ΔW is, the better the water-saving effect.

[0217] Change in total system output value ΔV: The difference between the total output value of the entire system before and after the reduction in output value (ΔV=V0-V1, where V0 is the total output value before the reduction in output value and V1 is the total output value after the reduction in output value). The smaller ΔV is, the smaller the loss in output value.

[0218] Core Order Impact I: The proportion of core orders affected by production restrictions to the total number of core orders. The smaller the I value, the smaller the impact on the delivery of core orders.

[0219] Production stability index S: It is a comprehensive quantification using indicators such as the connectivity of the production network and the continuity of material flow. The closer S is to 1, the more stable the production network is.

[0220] The production stability index S ranges from 0 to 1, with values ​​closer to 1 indicating a more stable production network. Its comprehensive quantification revolves around network connectivity, material flow continuity, and auxiliary stability indicators, specifically:

[0221] S is decomposed into three primary core indicators: network connectivity C (weight 40%), material flow continuity, and auxiliary stability indicators. Each primary indicator is further decomposed into actionable secondary indicators and normalized to the range of 0 to 1. Network connectivity C is obtained by weighted summation of node connectivity rate C1 (weight 25%) and link connectivity reliability C2 (weight 15%). Node connectivity rate C1 is the ratio of the number of normally connected key nodes to the total number of key nodes within the statistical period. Link connectivity reliability C2 is the percentage of the average connection time of all core connection links within the statistical period. The calculation formula is C = 0.25 × C1 + 0.15 × C2.

[0222] Combining the quantitative results of material flow continuity and auxiliary stability indicators (all calculated by weighting the corresponding decomposed indicators to ensure normalization), the three primary core indicators are integrated through reasonable weight allocation to finally obtain the comprehensive quantitative indicator S. The data of all decomposed indicators come from the equipment operation, material flow, network monitoring and other records that can be collected on the production site. The weights and some calculation thresholds can be flexibly fine-tuned according to different production scenarios.

[0223] Step S33c: Based on the changes in water consumption, output value, core order impact, and production stability indicators of the entire system, a weighted summation method is used to calculate the marginal water benefit index. To comprehensively evaluate the water-saving effect, economic loss, order impact, and production stability of each simulation scenario, a marginal water benefit index M is constructed. This index is dimensionless; the larger the value, the better the overall benefit of the scenario (i.e., smaller output value loss per unit of water saving, smaller order impact, and higher production stability). First, ΔW / ΔV is normalized, and then the index calculation formula uses a weighted summation method.

[0224] M = (ΔW / ΔV) × ω 1+ (1-I)×ω2+S×ω3;

[0225] Wherein ω1, ω2, and ω3 are the weight coefficients for output-to-water consumption ratio, core order guarantee, and production stability, respectively. They are determined by industry experts through the analytic hierarchy process, for example, ω1=0.4 (prioritizing economic benefits), ω2=0.3 (emphasizing core order delivery), and ω3=0.3 (taking production stability into account). The marginal water benefit index M value of each simulation scenario (i.e., each candidate process + corresponding production reduction ratio) is calculated according to the above formula.

[0226] In one embodiment, specifically:

[0227] A two-layer hierarchical model is constructed with "marginal water benefit index weight allocation" as the target layer (A) and "output value water consumption ratio (B1)", "core order guarantee (B2)" and "production stability (B3)" as the criterion layer (B).

[0228] Five senior experts in industrial production scheduling and water resource optimization were invited to conduct pairwise comparisons and scores on the relative importance of the three factors in the criterion layer using the analytic hierarchy process (AHP) 1-9 scale (1 indicates that the two factors are equally important, 3 indicates that the former is slightly more important than the latter, 5 indicates that the former is significantly more important than the latter, 7 indicates that the former is strongly more important than the latter, 9 indicates that the former is extremely more important than the latter, 2, 4, 6, and 8 are the median values ​​of the above adjacent judgments, and the reciprocal indicates that the two factors have opposite importance). Finally, the judgment matrix was obtained by summarizing the expert opinions (using the arithmetic mean method).

[0229] Target layer A: Marginal water benefit index weight allocation; Criterion layer B: B1 (output value water consumption ratio), B2 (core order guarantee), B3 (production stability);

[0230] Among them, comparing B1 and B2: experts believe that the output value water consumption ratio is slightly more important than the core order guarantee, and take a scale of 1.33 (corresponding to the midpoint between 1 and 3 in the 1-9 scale method, reflecting the need to "prioritize economic benefits").

[0231] Comparison of B1 and B3: Same as B1 and B2, scale is 1.33;

[0232] Comparing B2 and B3: Experts believe that securing core orders is as important as production stability, with a scale of 1;

[0233] The anti-diagonal element is the reciprocal of the corresponding element (e.g., the scale of B2 relative to B1 is 1 / 1.33≈0.75).

[0234] The weights of each factor in the criterion layer are calculated using the sum-product method commonly used in the analytic hierarchy process. The sum of the elements in each column of the judgment matrix is ​​calculated, and then each element in that column is divided by the column sum to obtain a normalized matrix. The elements in each row of the normalized matrix are added together to obtain a row sum vector. Each element in the row sum vector is divided by the total row sum to obtain ω1, ω2, and ω3.

[0235] Calculate the product of the judgment matrix and the weight vector, then divide each element of the product vector by the corresponding weight to obtain the eigenvalue vector. The average value of the eigenvalue vector is the largest eigenvalue.

[0236] The consistency index and consistency ratio were calculated separately, and the judgment matrix constructed by combining the expert scores was used to finally determine the weight coefficients ω1=0.4, ω2=0.3, and ω3=0.3.

[0237] Step S33d: Sort the marginal water benefit indicators of all candidate processes in descending order, generate a production restriction priority list, and obtain the production restriction order, optimal production reduction ratio and triggering conditions for each process.

[0238] The marginal water benefit index M values ​​of all candidate processes under different production reduction ratio scenarios are sorted in descending order to select the optimal production reduction ratio (corresponding to the production reduction ratio with the largest M value) for each candidate process. Based on the M value corresponding to the optimal production reduction ratio, all candidate processes are sorted in descending order to generate a production restriction priority list. The list clearly specifies the production restriction order of each process (priority 1 is the first to be restricted, and the higher the priority, the later the restriction), the optimal production reduction ratio, and the triggering conditions (such as triggering production restriction of priority processes 1-3 when there is severe water shortage, and triggering production restriction of priority processes 1-2 when there is moderate water shortage).

[0239] In one embodiment, specifically:

[0240] A water network diagram is constructed, with nodes representing 28 key processes and edges representing fresh water, recycled water, and wastewater flows. Node attributes are labeled with the water consumption of each process (e.g., the ethylene cracking process uses an average of 32,000 m³ / month). 3 ), reuse volume (15,000 m³) 3 Water quality requirements (conductivity ≤ 50 μS / cm), and water flow velocity (28,000 m³ / s) should be specified in the side attribute. 3 / month), water quality changes (such as wastewater conductivity increasing to 800μS / cm);

[0241] Construct a production network diagram with nodes representing 12 production lines and 28 processes, edges representing material dependencies (e.g., the ethylene cracking process supplies materials to the polyethylene process), and node attributes labeling output value (e.g., the average monthly output value of the ethylene cracking process is 80 million yuan), output (5000t / month), and order dependency (core orders account for 65%).

[0242] Calculate the confidence water volume envelope (580,000-650,000 m³) for April 2024. 3 The water resource constraint level is determined as: severe water shortage (water supply ≤ 580,000 m³). 3 Moderate water shortage (580,000-620,000 m³) 3 Mild water shortage (620,000-650,000 m³) 3 );

[0243] For each constraint level, a simulation scenario was designed, and 10 candidate production-limiting processes were selected. For each process, three percentage gradients of production reduction were set: 10%, 20%, and 30%.

[0244] System simulation software was used to simulate the system response under different scenarios: for example, a 20% reduction in ethylene cracking production would result in a reduction of 18,000 m³ of total water consumption for the entire system. 3 / month (water consumption change), total output value decreased by RMB 16 million / month (output value change), core order impact rate 13% (percentage of affected core orders), production stability index 0.85 (out of 1.0, the lower the value, the worse the stability).

[0245] The marginal water benefit index M is constructed as follows: M = (change in output value / change in water consumption) × 0.4 + impact of core orders × 0.3 + production stability index × 0.3. The calculated value of M for a 20% reduction in ethylene cracking production is M = (1600 / 1.8) × 0.4 + 13% × 0.3 + 0.85 × 0.3 ≈ 356.9.

[0246] The 10 candidate processes are sorted in descending order of their M values ​​to generate a production reduction priority list: Priority 1 (M=420): Propanol synthesis process (20% reduction); Priority 2 (M=385): Styrene refining process (15% reduction); Priority 3 (M=356.9): Ethylene cracking process (20% reduction). The production reduction order, optimal reduction ratio, and triggering conditions for each process are determined (e.g., production reduction of priority processes 1-3 is triggered when there is a severe water shortage).

[0247] In this embodiment, a mechanism modeling method is adopted, based on the law of conservation of mass and hydraulic dynamics equations. The core description is the transmission, distribution and water quality change process of water flow between various processes / pipelines. The input parameters include water consumption, reuse amount, pipeline resistance coefficient and water quality parameters of each process. The output parameters are the total water consumption of the whole system, water distribution of each node and water quality compliance rate, and a water network simulation sub-model is constructed.

[0248] A hybrid modeling approach combining "mechanism and data-driven" is adopted. The mechanism part is built based on production process constraints, while the data-driven part trains a Long Short-Term Memory (LSTM) network model based on historical production data to correct the bias of the mechanism model and construct a production network simulation sub-model to describe the material flow and output / output value generation process between processes. The input parameters include process capacity, material dependencies, and order requirements, while the output parameters are the total output value of the entire system, the completion rate of core orders, and the production load distribution.

[0249] The coupled coordination model is constructed based on constrained optimization theory, integrating the key constraints of the water network and the production network to achieve collaborative simulation and data interaction between the water network simulation sub-model and the production network simulation sub-model. Specifically:

[0250] Using "process nodes" as the core coupling nodes, the water network and the production network achieve bidirectional data transmission through a pre-defined data interaction channel:

[0251] Data transmitted from the production network to the water network includes: real-time capacity (or reduction ratio) and production load of each process, which are used to dynamically adjust the water demand of the corresponding processes in the water network.

[0252] The data transmitted from the water network to the production network—the actual water supply and water quality compliance status of each process—serves as constraints for the production network to adjust capacity and optimize material allocation.

[0253] Import the topology and attribute data of the water network diagram and the production network diagram, and set the initial simulation conditions such as the water resource constraint level (severe / moderate / mild water shortage) and the gradient of the production reduction ratio of candidate processes;

[0254] Based on the initial conditions, the production network simulation sub-model calculates the initial capacity and water demand of each process and transmits them to the water network sub-model through the coupling channel; the water network sub-model, combined with water resource constraints, calculates the actual water supply and water quality status of each process and feeds them back to the production network sub-model.

[0255] The production network sub-model adjusts the production capacity of each process (implementing a production reduction plan) based on the water supply constraints fed back by the water network, and recalculates indicators such as output value and order completion rate; the water network sub-model responds synchronously to changes in water demand after the production capacity adjustment, and updates total water consumption and water distribution indicators.

[0256] Repeat the forward simulation and dynamic adjustment process until the fluctuation range of the output parameters (total water consumption, total output value, and order completion rate) of the two sub-models is ≤1%. Then, the simulation is considered to have converged, and the final simulation results are output.

[0257] In this embodiment, a differentiated time step and a clear accuracy standard are set based on the actual production scheduling cycle and simulation accuracy requirements. One hour is adopted as the minimum simulation time step. In key dynamic scenarios such as the triggering of production restriction schemes and the switching of water resource constraint levels, the time step is automatically shortened to 10 minutes to accurately capture the sudden changes in water consumption and output value in a short period of time. When the fluctuation range of simulation parameters is ≤1%, the time step is restored to the basic time step to improve simulation efficiency. For each simulation scenario of water resource constraint level, the total simulation time is one complete production scheduling cycle (default 24 hours).

[0258] like Figure 5 As shown, according to one aspect of this application, step S4 further comprises:

[0259] Step S41: Calculate the expected real-time water consumption coefficient based on the real-time operation data of the production line; calculate the conservative upper limit of production capacity based on the different lower limits of confidence in the confidence water volume envelope, the expected real-time water consumption coefficient, and the process optimization coefficient.

[0260] In this embodiment, the expected real-time water consumption coefficient μ refers to the expected value of water consumption per unit of product calculated based on recent real-time operation data of the production line, used to reflect the actual water consumption level of current production. The specific calculation method is as follows: collect real-time water consumption data and corresponding production data for the past 10-30 days, calculate the average actual unit water consumption W_avg, and compare it with the historical benchmark unit water consumption W_base to obtain the expected real-time water consumption coefficient μ = W_avg / W_base. The conservative production capacity ceiling refers to the maximum feasible production capacity that ensures water supply security under specific water resource constraints. Its calculation requires comprehensive consideration of water resource constraints, actual water consumption levels, and process optimization effects. The specific calculation process is as follows:

[0261] Determine the different lower confidence limits within the confidence envelope of water volume, including the lower confidence limit Q corresponding to high confidence levels (e.g., the 95% confidence limit of 580,000 m³). 3 ( / month) and the lower confidence limit Q corresponding to the low confidence level, such as the 80% confidence limit of 650,000 m 3 / moon;

[0262] Calculate the expected real-time water consumption coefficient μ, which reflects the actual water consumption level of current production. It is calculated based on the real-time water consumption data and corresponding output data of the past 10-30 days. The formula is μ = average actual unit water consumption / historical benchmark unit water consumption. The closer μ is to 1, the closer the current water consumption level is to the historical benchmark. Extract the process optimization coefficient η (reflecting the water-saving effect of process optimization).

[0263] The conservative upper limit of production capacity is calculated using a quantitative formula: conservative upper limit of production capacity Q=(Q×η×μ) / W, where Q is the selected confidence water volume lower limit (substituted into Q and Q respectively), and W is the benchmark water consumption per unit product (determined based on the average unit water consumption under historical normal operating conditions).

[0264] The high-confidence conservative production capacity upper limit Q (suitable for severe water shortage scenarios) and the low-confidence conservative production capacity upper limit Q (suitable for moderate water shortage scenarios) are calculated separately. For example, in the high-confidence scenario, Q=(58×0.84×0.98) / 2.5≈195,000 t / month, and in the low-confidence scenario, Q=(65×0.84×0.98) / 2.5≈218,000 t / month.

[0265] Step S42: Determine the aggressive production capacity limit based on the historical maximum production capacity of the study area, the total market demand, and the rated production capacity of the equipment. Based on the conservative production capacity limit, the aggressive production capacity limit, and the production restriction priority, generate a conservative production scheduling plan, an aggressive production scheduling plan, and a combination of phased production restriction and process adjustment.

[0266] The aggressive capacity ceiling is the maximum theoretical capacity threshold that can be achieved within a certain production period in the study area, taking into account the core constraints of historical peak capacity, market demand, raw material supply, and equipment rated capacity, under optimal resource supply conditions and without extreme risks. It represents the upper limit of capacity release, specifically:

[0267] C J max =min{C LS ×k1,C MQ C YL ×k2,C SB ×k3,C XS}×k4;

[0268] Among them, C J max The aggressive production capacity ceiling is the maximum aggressive production capacity ultimately approved for the target period in the study area, C. LS The historical maximum capacity represents the peak actual production capacity of the study area over the past 3-5 years. k1 is the historical capacity improvement coefficient, equal to the proportion of capacity improvement brought about by process upgrades or equipment modifications. It is 1.0 for no modifications and takes the value based on the actual gain from modifications. C MQ C represents the total market demand, i.e., the maximum effective demand from downstream market orders / industry demand within the target period. YL The maximum raw material supply capacity is represented by the maximum total supply of core raw materials within the target time period. k2 is the raw material utilization rate coefficient, an efficiency coefficient for refined raw material utilization / loss reduction, determined based on the actual loss rate of the production process; for lossless optimization, it is set to 1.0. C SB The rated total capacity of the equipment is the sum of the rated capacities on the nameplates of all production equipment. k3 is the full-load operation coefficient of the equipment, taken as 1.0 for full production without faults, and 0.98-1.0 for minor maintenance allowance. C XSTo ensure the maximum capacity of the supporting system, the maximum supporting capacity of the water supply system is taken first. k4 is the aggressive operating condition correction coefficient and the comprehensive adaptation coefficient of the aggressive production scheduling. Without additional constraints, it is taken as 1.0.

[0269] Based on the calculated conservative capacity ceiling, aggressive capacity ceiling, and production restriction priority, a multi-level robust production scheduling scheme adapted to different water supply scenarios is generated to ensure the scientific validity and robustness of the scheme. First, the aggressive capacity ceiling C is determined. J max This upper limit represents the maximum feasible production capacity under a scenario of sufficient water supply. It is determined by comprehensively considering factors such as the historical maximum production capacity of the study area, total market demand, raw material supply capacity, and rated equipment capacity. For example, if the historical maximum production capacity is 230,000 tons / month, and market demand is sufficient and raw material supply is stable, then the aggressive production capacity upper limit is 230,000 tons / month. Based on different production capacity upper limits and production restriction priorities, three core production scheduling schemes are generated:

[0270] Conservative production scheduling plan: Adaptable to severe water shortage scenarios (water supply ≤ Q), the production capacity is set to a high-confidence conservative production capacity upper limit Q (e.g., 195,000 t / month), triggering production restrictions for processes with priority 1-3 in the production restriction priority list (executed according to the optimal production reduction ratio), while implementing process optimization plans across the entire production line to maximize water saving effect and ensure water supply security;

[0271] Aggressive production scheduling plan: Suitable for mild water shortage or abundant water scenarios (water supply ≥ C J max The capacity is set at an aggressive capacity cap C. J max (e.g., 230,000 tons / month) No production limit is required; process optimization is only implemented in core high water-consuming processes to balance production capacity efficiency and water conservation needs.

[0272] A phased production restriction and process adjustment combination plan: suitable for moderate water shortage scenarios (Q < water supply < C) J max The production capacity is set between Q and C. J max The intermediate value (e.g., 205,000 tons / month) triggers production restrictions for processes with priority 1-2 in the production restriction priority list (executed according to the optimal reduction ratio), and the process optimization plan is implemented on the core production line; at the same time, capacity adjustment gradients are set (e.g., 205,000 → 218,000 → 230,000 tons / month), each gradient corresponding to a different water supply threshold. When the real-time water supply reaches the corresponding threshold, the capacity is smoothly increased to avoid insufficient water supply or production disorder caused by a sudden increase in capacity.

[0273] In addition, to improve the operability of the plan, each production scheduling plan must specify the specific capacity allocation, production load rate, process parameter settings, and water supply allocation plan for each process.

[0274] Step S43: Link the production scheduling plan with the production scheduling system and process control system to automatically switch or smoothly adjust the plan according to the real-time water supply situation, and trigger the emergency adjustment plan in extreme water shortage scenarios.

[0275] First, complete the interface development and data connection between the production scheduling plan and the production scheduling system, process control system and water supply monitoring system to achieve interconnection and interoperability of plan parameters, real-time production data and real-time water supply data.

[0276] Set real-time water supply monitoring thresholds that correspond to water resource constraint levels: Severe water shortage threshold (water supply ≤ Q, e.g., ≤ 580,000 m³). 3 / month), moderate water shortage threshold (Q < water supply < C) J max For example, 580,000-650,000 m³ 3 / month), mild water shortage threshold (water supply ≥C) J max For example, ≥650,000 m 3 / moon);

[0277] The water supply monitoring system collects water supply data in real time. When the real-time water supply reaches a certain threshold, the MES system automatically triggers the corresponding production scheduling plan. The DCS system automatically receives and executes the process adjustment and capacity control instructions in the plan, realizing automatic switching of the plan. The adjustment lag time is controlled within 30 minutes. The determination of this lag time indicator is based on the following: In a typical petrochemical production scenario, the scheduling cycle of the water supply system is usually 1 hour, that is, significant changes in water supply occur at the hourly level. At the same time, the adjustment of production line process parameters and capacity switching require a certain transition time (such as heating / cooling, equipment start-up and shutdown, etc.), with a typical transition time of 15-20 minutes. Controlling the lag time within 30 minutes ensures that the production scheduling plan switch is completed before the start of the next scheduling cycle after the water supply change, and also reserves sufficient buffer time for the process adjustment process to avoid production disorder caused by too rapid switching. This indicator can be adjusted according to the water supply scheduling cycle and process transition time of the actual production scenario.

[0278] For example, in mid-April 2024, the real-time water supply dropped to 590,000 m³. 3 When the water shortage threshold is reached, the system automatically switches from the aggressive solution to a phased combined solution and issues process adjustment and production restriction instructions through the DCS system.

[0279] At the same time, an emergency threshold for extreme water shortages (such as a water supply of ≤550,000 m³) is set. 3 / month, far below the high-confidence conservative lower limit), when the real-time water supply reaches the emergency threshold, the emergency adjustment plan is triggered: non-core production lines with priority 8-10 in the production restriction priority list are suspended, core production lines are reduced by 30%, priority is given to ensuring the production capacity and water supply required for core orders, and emergency water supply measures are initiated (such as activating backup water sources and increasing the proportion of recycled water reuse) to ensure the basic stability of the production system and the delivery of core orders.

[0280] In addition, a monitoring mechanism for the implementation effect of the plan was established to track the completion of production capacity, the effect of water consumption control, and the progress of order delivery in real time. When deviations occur (such as actual water consumption exceeding the predicted value by more than 5%), the plan parameters are adjusted in a timely manner.

[0281] According to another aspect of this application, a "production determined by water availability" system is provided, comprising:

[0282] At least one processor; and

[0283] A memory communicatively connected to at least one of the processors; wherein,

[0284] The memory stores instructions that can be executed by the processor to implement the "production based on water availability" method described above.

[0285] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. The method of "production determined by water availability" is characterized by, Includes the following steps: Step S1: Collect data of the study area, including historical data of upstream water sources and water supply systems, and construct a multi-scenario set of water volume for different time periods in the study area, and calculate the confidence water volume envelope. Step S2: Based on the historical operating data of the production line under different operating conditions, construct a process-water consumption response surface model and generate process optimization coefficients; Step S3: Construct water network diagram and production network diagram respectively. Based on the confidence water volume envelope band, couple water network diagram and production network diagram to construct a two-layer structure of water network-production network and calculate production restriction priority. Step S4: Calculate the expected real-time water consumption coefficient based on the real-time operation data of the production line; determine the aggressive capacity limit based on the historical maximum capacity of the study area, market demand, and rated capacity of the equipment; calculate the conservative capacity limit based on different confidence lower limits in the confidence water volume envelope, the expected real-time water consumption coefficient, and the process optimization coefficient; and generate a multi-level robust production scheduling scheme and link it for control based on the conservative capacity limit, the aggressive capacity limit, and the production restriction priority.

2. The method of "production determined by water availability" as described in claim 1, characterized in that, Step S1 further comprises: Step S11: Collect historical data of upstream water sources, historical data of water supply systems, basic data of processes, and historical operating data of production lines under different operating conditions in the study area. The historical data of upstream water sources and historical data of water supply systems include time series of water inflow and supply in the same season, statistical data of water supply facility failures, and data on fluctuations in unconventional water production. Step S12: Based on historical data of upstream water sources and historical data of water supply systems, construct a set of multiple water volume scenarios for each future time period, and calculate the probability of each scenario. Step S13: For each future time period, based on its corresponding water volume multi-scenario set and scenario probability, calculate at least two lower limits for water volume at different confidence levels, including the lower limit for high confidence level and the lower limit for low confidence level, to form the confidence water volume envelope for that time period.

3. The method of "production determined by water availability" as described in claim 2, characterized in that, Step S12 further comprises: Step S12a: Based on the pre-constructed correlation model between historical precipitation and inflow, divide the water into three levels: abundant water, normal water, and dry water, and set the natural inflow fluctuation range for each level. Step S12b: Based on fault statistics, classify fault levels and set the reduction range for water supply for each level; Step S12c: Based on the historical production data of the reclaimed water station, the fluctuation level is divided, and the natural water inflow fluctuation scenario, the water supply facility failure scenario and the unconventional water production fluctuation scenario are combined in all dimensions to generate a basic scenario set, and the water volume multi-scenario set corresponding to each future time period is obtained respectively. Step S12d: Based on historical data, the occurrence frequency of each basic scenario is statistically analyzed, and the industry prior probability is used as a benchmark constraint and the historical frequency is corrected to obtain the final scenario probability.

4. The method of "production determined by water availability" as described in claim 1, characterized in that, Step S2 further comprises: Step S21: Extract historical operating data of the production line under different operating conditions, construct an operating sample set, and use data-driven fitting to obtain the process-water consumption response surface model; Step S22: Continuously collect new operating data sample sets, correct the process-water consumption response surface model, identify whether the current operating conditions deviate from the training data range, and obtain the optimal process-water consumption response surface model. Step S23: Calculate the partial derivatives of the optimal process-water consumption response surface model with respect to each process parameter. Based on the feasible range of process parameters, production load constraints, and product quality constraints, construct a constrained optimization problem and solve it to obtain the optimal process parameters. Step S24: Based on the optimal process parameters and the unit water consumption corresponding to the current process parameters, calculate the process optimization coefficient, generate specific control instructions, and obtain the process optimization scheme.

5. The method of "production determined by water availability" as described in claim 4, characterized in that, Step S21 further comprises: Step S21a: Extract historical operating data of the production line under different operating conditions and construct an operating sample set; Step S21b: Construct process-water consumption response surface models based on different operating conditions using multinomial response surface model, random forest tree model, and BP neural network model respectively; Step S21c: Divide the running sample set into training set, validation set and test set in a ratio of 6:2:2, train the process-water consumption response surface model, calculate the goodness of fit, root mean square error and mean absolute error respectively, and verify the accuracy of the model to obtain the process-water consumption response surface model.

6. The method of "production determined by water availability" as described in claim 4, characterized in that, Step S22 further comprises: Step S22a: Collect new operational data; Step S22b: Set the scrolling window size and the corresponding window increment threshold. When the number of newly collected samples reaches the window increment threshold, remove the earliest group of samples in the window, add new samples, and obtain a new set of running data samples. Step S22c: Update the parameters of the process-water consumption response surface model and set an accuracy warning threshold. When the model accuracy is lower than the accuracy warning threshold, trigger a full retraining of the model. Step S22d: Collect the feature cluster centers of all historical normal operating conditions, construct the operating condition benchmark library, calculate the Euclidean distance between the current operating condition features and the cluster centers of the benchmark library, and determine the operating condition deviation when the distance is greater than the threshold. Correct the process-water consumption response surface model until no deviation occurs, and obtain the optimal process-water consumption response surface model.

7. The method of "production determined by water availability" as described in claim 1, characterized in that, Step S3 further comprises: Step S31: Construct a water network diagram based on the basic data of the process, with the process or device as the node and various water flows as the edge. The node attributes include water consumption, reuse amount, and water quality requirements, and the edge attributes include water quantity and water quality changes. Step S32: Construct a production network diagram based on process basic data, with processes or production lines as nodes and material dependencies as edges. Node attributes include output value, output, and order dependency degree. Step S33: Determine water resource constraints based on the confidence water volume envelope. For each candidate production restriction process, simulate the impact of partial or complete shutdown on the water network and production network. Calculate the changes in water consumption and output value of the entire system. Calculate the marginal water benefit index based on the ratio of the changes in water consumption to the changes in output value of the entire system. Determine the production restriction priority based on the ranking of marginal water benefit index values.

8. The method of "production determined by water availability" as described in claim 7, characterized in that, Step S33 further comprises: Step S33a: Determine different water resource constraint levels based on the confidence water volume envelope, design simulation scenarios for each constraint level, set different reduction ratio gradients for each candidate production restriction process, and simulate the system response under different reduction degrees. Step S33b: Simulate the difference between the total water consumption and total output value of the entire system before and after the production reduction, calculate the changes in water consumption and output value of the entire system, calculate the impact of core orders based on the ratio of the number of core orders affected by production restrictions to the total number of core orders, and calculate the production stability index based on the connectivity of the production network and the continuity of material flow. Step S33c: Calculate the marginal water benefit index using a weighted summation method based on changes in water consumption across the entire system, changes in output value across the entire system, the impact of core orders, and production stability indicators. Step S33d: Sort the marginal water benefit indicators of all candidate processes in descending order, generate a production restriction priority list, and obtain the production restriction order, optimal production reduction ratio and triggering conditions for each process.

9. The method of "production determined by water availability" as described in claim 1, characterized in that, Step S4 further comprises: Step S41: Calculate the expected real-time water consumption coefficient based on the real-time operation data of the production line; calculate the conservative upper limit of production capacity based on the different lower limits of confidence in the confidence water volume envelope, the expected real-time water consumption coefficient, and the process optimization coefficient. Step S42: Determine the aggressive production capacity limit based on the historical maximum production capacity of the study area, the total market demand, and the rated production capacity of the equipment. Based on the conservative production capacity limit, the aggressive production capacity limit, and the production restriction priority, generate a conservative production scheduling plan, an aggressive production scheduling plan, and a combination of phased production restriction and process adjustment. Step S43: Link the production scheduling plan with the production scheduling system and process control system to automatically switch or smoothly adjust the plan according to the real-time water supply situation, and trigger the emergency adjustment plan in extreme water shortage scenarios.

10. A system that "determines production based on water availability" is characterized by: include: At least one processor; as well as A memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor to implement the "production based on water availability" method as described in any one of claims 1 to 9.

Images