Continuous hydrogenation interconnection equipment regulation and control system based on hybrid bat optimization algorithm

Abstract

The invention provides a continuous hydrogenation interconnection equipment regulation and control system based on a hybrid bat optimization algorithm, belongs to the technical field of chemical industry, and solves the problems that an existing hydrogenation equipment regulation and control system is low in intelligent degree, insufficient in real-time monitoring capability, lack of an intelligent production decision-making function and the like. Comprising a data acquisition module, a data preprocessing module, a hybrid optimization calculation module, a regulation and control execution module, a monitoring display module, a man-machine collaborative decision-making module, a database module, a federal learning module and a digital twinning module. An intelligent system integrating hydrogenation production system real-time monitoring, production scheduling guidance and production data analysis management is constructed, the production state is monitored and pre-judged in real time by achieving core functions of intelligent optimization, production decision making and the like, unmanned workshops are pushed to land in combination with an optimization strategy, and the production efficiency is improved. And meanwhile, real-time front-line production data is provided for production managers, efficient command and scientific scheduling are assisted, and the production intelligent level and the management efficiency are comprehensively improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of chemical industry, and relates to a continuous hydrogenation interconnected equipment regulation system, in particular to a continuous hydrogenation interconnected equipment regulation system based on a hybrid bat optimization algorithm. BACKGROUND

[0002] Hydrogenation reaction is one of common unit reactions, and is widely applied to the fields of petroleum chemical industry, fine chemical industry, polymer synthesis and the like. At present, most of the hydrogenation processes of fine chemical enterprises adopt the traditional production mode of intermittent hydrogenation. With the increasingly fierce competition in the fine chemical industry, the production enterprises continuously improve the requirements for production efficiency, production capacity, safety, environmental protection and the like, and the intermittent hydrogenation cannot meet the development of the industry. Based on the existing intermittent production process, Qiancai Chemical designs and develops a set of continuous hydrogenation engineering experimental device through process flow optimization and equipment development, realizes continuous hydrogenation, reduces reaction residence time, realizes product index optimization, and finally achieves the purpose of industrialization.

[0003] The continuous hydrogenation expert system relies on big data, artificial intelligence, industrial internet and communication technology, realizes real-time monitoring, production scheduling command, production data analysis management and the like of the hydrogenation production system. Through intelligent upgrading, the functions of intelligent optimization, equipment management and production decision are realized. The continuous hydrogenation expert system of Zhongxin can realize real-time monitoring and state prediction of the production situation, realizes the function of unmanned workshop in combination with the optimization strategy. Meanwhile, the production managers can also timely grasp the first-line production data, reasonably command and schedule the production work.

[0004] Based on this, we propose a continuous hydrogenation interconnected equipment regulation system based on a hybrid bat optimization algorithm, so that the application aims to build an intelligent system integrating real-time monitoring, production scheduling guidance and production data analysis management of the hydrogenation production system, realizes core functions such as intelligent optimization and production decision, realizes real-time monitoring and prediction of the production state in combination with the optimization strategy, promotes the landing of unmanned workshop, simultaneously provides real-time first-line production data for the production managers, helps efficient command and scientific scheduling, and comprehensively improves the production intelligent level and management efficiency. SUMMARY

[0005] The purpose of the present application is to solve the above-mentioned problems existing in the prior art, and a continuous hydrogenation interconnected equipment regulation and control system based on a hybrid bat optimization algorithm is proposed.

[0006] The purpose of the present application can be achieved by the following technical solutions: A continuous hydrogenation interconnected equipment regulation and control system based on a hybrid bat optimization algorithm, comprising a data acquisition module, a data preprocessing module, a hybrid optimization calculation module, a regulation and control execution module, a monitoring and display module, a man-machine collaborative decision-making module, a database module, a federated learning module, and a digital twin module; the output end of the data acquisition module is connected to the input end of the data preprocessing module, which is used to acquire the operating parameters of each device in the hydrogenation production system in real time and transmit them to the data preprocessing module; the module supports the access of multiple source heterogeneous devices such as PLC and DCS, and realizes real-time data transmission and standardized processing through Modbus and OPCUA protocols; the output end of the data preprocessing module is connected to the input end of the hybrid optimization calculation module, which is used to process more accurate data and transmit them to the hybrid optimization calculation module; the hybrid optimization calculation module is used to run the hybrid bat optimization algorithm program, optimize the real-time data after preprocessing, and output device regulation and control instructions; the regulation and control execution module receives the instructions of the hybrid optimization calculation module and executes device parameter regulation; the monitoring and display module is bidirectionally connected to the hybrid optimization calculation module and the regulation and control execution module, respectively, to real-time feedback operating data and regulation and control state, and is used for man-machine interaction; the man-machine collaborative decision-making module uses explainable AI technology to generate a decision basis report for the regulation and control instructions output by the hybrid optimization calculation module, and receives feedback instructions from the operator, which are used to continuously optimize the algorithm model in the hybrid optimization calculation module; the database module is bidirectionally connected to the data acquisition module and the hybrid optimization calculation module, respectively, to persistently store historical data and algorithm operation data, and support data query and model training; the federated learning module aggregates and shares model parameters of local models of multiple production bases in an encrypted state to collaboratively improve algorithm performance; the digital twin module constructs a virtual model that is a complete mirror image of the physical workshop based on historical data, mechanism models, and real-time data streams, uses the model to predict the remaining useful life of key components, and generates maintenance warnings in advance.

[0007] Working principle of the present application: The system realizes intelligent control of the hydrogenation production process through the cooperative work of multiple modules. First, the data acquisition module collects real-time data of equipment operation parameters, catalyst state, raw material composition and carbon emission data through a multi-source sensor network, and ensures the spatio-temporal consistency through the "5G+NTP" dual synchronization mechanism. The data preprocessing module performs weight analysis, outlier processing and standardization formatting on the original data, providing a high-quality data foundation for subsequent optimization calculation.

[0008] The hybrid optimization calculation module adopts a hybrid bat optimization algorithm, which combines the advantages of BA, PSO, SA and WOA algorithms, and performs multi-objective optimization under the reinforcement learning framework. Through an adaptive parameter adjustment mechanism, the algorithm dynamically balances global exploration and local development capabilities, taking product quality, energy consumption, equipment wear rate and carbon emission intensity as the comprehensive objective function, and outputting the optimal process parameter combination.

[0009] The digital twin module constructs a virtual production line based on real-time data and mechanism models, pre-executes and verifies the optimization instructions, and realizes equipment health state prediction. The man-machine collaborative decision-making module generates a decision basis report through interpretable AI technology for operators to review and confirm. The control execution module accurately controls the reactant flow and equipment power according to the final instruction, and realizes energy optimization through photovoltaic- energy storage linkage.

[0010] The monitoring and display module displays the whole process running state and early warning information in real time. The database module persistently stores all process data, supporting historical tracing and model training. The federated learning module realizes encrypted parameter sharing of multiple production bases, continuously improving the model generalization ability. Through the closed-loop feedback mechanism, the modules form a continuously optimized intelligent control system, and finally realize the safe, efficient and low-carbon operation of the hydrogenation production process.

[0011] The data acquisition module comprises a sensor monitoring unit, a catalyst state monitoring unit, a raw material microanalysis unit, a carbon footprint monitoring unit and a time synchronization unit, the sensor monitoring unit comprises a voltage sensor, a current sensor, a temperature sensor, a pressure sensor and a power sensor; the voltage sensor acquires the real-time power supply voltage of the hydrogenation equipment, the measurement range is 0-10kV, and the accuracy is ±0.5%; the current sensor acquires the input / output current of the equipment, the measurement range is 0-500A, and the accuracy is ±0.3%; the power sensor acquires the active power, the measurement range is 0-1000kW, and the accuracy is ±0.2%; the temperature sensor is arranged at the inlet, outlet and reaction section of each reaction kettle, and acquires temperature data, the measurement range is-50℃~500℃, and the accuracy is ±0.1℃; the pressure sensor acquires the pressure in each reaction kettle and the pressure difference of each section of the reaction tower, the measurement range is 0-10MPa, and the accuracy is ±0.05MPa; the catalyst state monitoring unit is provided with an X-ray fluorescence spectrum sensor, the measurement range is 0-100%, the accuracy is ±0.1%, the real-time acquisition of catalyst active components and crystal form parameters is realized, and the data sampling frequency is 1 / minute; the raw material microanalysis unit integrates a near-infrared spectrometer, completes the detection of raw material purity and impurity content within 10 seconds, supports the rapid analysis of raw materials such as methanol and toluene; the carbon footprint monitoring unit is provided with a carbon emission sensor, the measurement range is 0-50kgCO2 / h, the accuracy is ±2%, and is used for acquiring the carbon emission data related to the tail gas of the reaction kettle and power consumption, and calculating the total carbon emission C in real time, the formula is: In the formula, Mi is the amount of raw material, Fi is the carbon emission factor of raw material, E is the power consumption, G is the carbon emission factor of power, Wj is the waste emission amount, and Hj is the treatment carbon emission factor, the index is integrated into the optimization algorithm decision process. The time synchronization unit is based on NTP protocol, adopts a "5G+NTP" double synchronization mechanism, guarantees the space-time consistency of micro data and macro data, realizes the timestamp calibration of each sensor data, ensures that all sensor data have unified timestamps, the time synchronization error is ≤0.5ms, and guarantees the space-time consistency of micro data and macro data.

[0012] With the above structure, all-around parameter sensing: the module covers all key dimensions from power consumption, reaction environment, to material characteristics and even environmental protection indicators through five professional units, realizing synchronous tracking of "energy flow", "material flow" and "information flow". Micro and macro data fusion: it not only collects traditional macro process parameters, but also innovatively introduces advanced sensing technologies such as X-ray fluorescence spectrum and near-infrared spectrum, realizing online real-time analysis of catalyst activity, raw material composition and other micro-chemical properties. This enables data collection to go from macro appearance to the micro level that affects the nature of the reaction. Real-time carbon footprint quantification: through a dedicated carbon emission sensor and formulas, the module can online and real-time calculate and quantify the total carbon emissions in the production process, changing environmental protection indicators from offline accounting to online and optimized process variables, providing direct data support for green and low-carbon production. High-precision time and space synchronization: the "nerve center" of the module, the time synchronization unit, uses a "5G+NTP" dual synchronization mechanism to apply a unified timestamp to all sensor data with an extremely high precision of ≤0.5ms. This ensures that data collected from different sources and at different frequencies are completely aligned in the time dimension, eliminating analysis errors caused by data misplacement, and providing time consistency guarantee for subsequent digital twin modeling, causal analysis and precise control.

[0013] The data preprocessing module includes a weight analysis unit, an extreme data processing unit, a data deduplication unit and a standardization processing unit. The weight analysis unit analyzes the correlation between parameters and product quality based on a random forest algorithm, assigns a hydrogen flow weight of 0.3, a temperature weight of 0.25, and a pressure weight of 0.2, and is used for optimizing instruction priority sorting; the extreme data processing unit uses the 3σ principle to eliminate outliers and uses linear interpolation method to supplement missing data; the data deduplication unit retains the earliest repeated data with timestamp, deletes redundant data, and ensures data uniqueness; the standardization processing unit unifies the data format to timestamp-equipment ID-parameter type-value-unit, supports multiple source heterogeneous equipment access such as PLC and DCS.

[0014] With the above structure, through a standardized data pipeline, raw, heterogeneous, noisy field data is transformed into high-quality, structured, standardized information that can be directly used for intelligent decision-making. The workflow follows the logic chain of "empowerment → purification → regularization → unification": intelligent empowerment (weight analysis unit): based on the random forest algorithm, this unit quantitatively analyzes the relevance of input parameters to the quality of the final product. By giving different weights to key process parameters, it provides decision-making priority guidance for subsequent optimization algorithms, enabling optimization resources to be concentrated on the key variables that have the greatest impact on product quality. Data purification (extreme data processing unit): using the 3σ principle as a "filter", it automatically identifies and removes significantly abnormal data points caused by sensor failure or transmission interference. At the same time, it uses linear interpolation to reasonably repair missing data, effectively ensuring the integrity and reliability of the data set, providing a stable and reliable data foundation for the algorithm, preventing "garbage in, garbage out". Information deduplication (data deduplication unit): for repeated records that may occur during data transmission or collection, this unit performs deduplication by comparing timestamps. This ensures the uniqueness of data records, avoids bias caused by repeated data in statistical analysis and model training, and improves the efficiency and accuracy of the data set. Format unification (standardization processing unit): as the last step in data output, this unit converts heterogeneous data from different sources such as PLC and DCS into a unified structured format of "timestamp-device ID-parameter type-value-unit". This completely solves the problem of accessing multiple heterogeneous devices, allowing the downstream hybrid optimization calculation module to identify and process all data without discrimination, achieving standardization of the data language within the system.

[0015] The hybrid optimization calculation module includes a processor and a storage unit. The processor uses an Intel Core i7-12700K or an industrial processor with equivalent performance to call algorithm programs and perform operations on real-time data input by the data acquisition module, outputting control instructions such as reactant addition rate and equipment power. The storage unit stores a hybrid bat optimization algorithm program. The hybrid bat optimization algorithm is a composite optimization algorithm integrating whale optimization algorithm, bat algorithm, particle swarm optimization algorithm, and simulated annealing algorithm.

[0016] With the above structure, relying on high-performance hardware, a hybrid bat optimization algorithm that integrates the advantages of multiple intelligent algorithms is used to perform parallel and efficient global search on the precise data input from the front end, finally outputting control instructions that optimize the production target.

[0017] Its operation can be divided into two levels: 1. Hardware execution level: The module provides powerful computing capabilities through an Intel Core i7-12700K or equivalent industrial processor, ensuring real-time processing of massive amounts of data and execution of complex optimization algorithms. The storage unit serves as an "algorithm repository," carrying the core hybrid bat optimization algorithm program. The processor calls this program from the storage unit, performs high-speed calculations on the input data, and ultimately outputs the optimal control instruction set regarding reactant addition rate, device power, etc. 2. Algorithm core level (coordination mechanism of hybrid bat optimization algorithm): The soul of this module lies in its unique hybrid bat optimization algorithm, which is not a simple combination but an organic integration of the advantages of four algorithms to form a highly efficient search and decision engine: The bat algorithm serves as the main framework: simulating the echolocation behavior of bats, it provides a basic search mechanism by performing global exploration and local development through frequency adjustment and pulse emission. The particle swarm optimization algorithm injects "collective intelligence": by introducing "individual historical best" and "global best" information, it guides the entire bat population to converge quickly towards a better region, effectively solving the problem of slow convergence speed of traditional algorithms. Simulated annealing provides "breakthrough capability": its mechanism of probabilistically accepting worse solutions allows the algorithm to escape local optima, significantly enhancing its ability to find the global optimum. Whale optimization enhances the search strategy: its unique prey-encircling and spiral position update mechanism provides more diverse path choices throughout the search process, further improving the comprehensiveness and robustness of the search.

[0018] The specific steps for executing the hybrid bat optimization algorithm are as follows: Step S1, Initialization: Randomly initialize a group of bats in the solution space, define the position Xi and velocity vi of each bat, and set parameters such as initial frequency f0, loudness A0, and pulse emission rate r0; at the same time, initialize the inertial weight w, learning factors c1 and c2 of the particle swarm optimization algorithm, and the initial temperature T of the simulated annealing algorithm. Step S2, Fitness Assessment: Calculate the fitness value for each bat based on the objective function; Step S3: Location update and hybrid optimization; S3a) Exploring the Bat Algorithm: Updating the frequency, speed, and position of bats based on the standard bat algorithm; fi = fmin + (fmax - fmin) * β (β is a random number in [0,1]); vi(t+1)=vi(t)+(Xi(t)-Xglobal)*fi; Xi_new=Xi(t)+vi(t+1); S3b) PSO Collaborative Guidance: To improve the convergence speed, a PSO group collaboration mechanism is introduced to perform a secondary correction on the bat's speed; vi(t+1)=w*vi(t+1)+c1*r1(Xpbest-Xi(t))+c2*r2*(Xglobal-Xi(t)) Where Xpbest is the individual historical best position of the bat, and Xglobal is the current global best position; S3c)SA perturbation escapes local optima: To avoid premature convergence, the Metropolis criterion of the simulated annealing algorithm is used to accept new solutions; calculate the fitness difference ΔE between the new position Xi_new and the old position Xi(t); if ΔE<0, accept the new solution; if ΔE>=0, accept the new solution with probability P_acc=exp(-ΔE / T) (where T is the current temperature); Step S4, Local Search: Perform a random walk around the current optimal solution Xglobal to generate a new local solution; if the new solution is better than the current optimal solution and its random number is greater than the pulse emission rate ri, then accept the new local solution; Step S5, Update State: If the new solution is accepted, update the loudness Ai and pulse emission rate ri of the bat; Ai(t+1) = α*Ai(t); ri(t+1)=ri(0)*[1-exp(-γ*t)]; Step S6, Cooling: Reduce the current temperature T of the simulated annealing algorithm according to the cooling plan; Step S7, Termination Judgment: Repeat steps S2 to S6 until the maximum number of iterations is reached or the convergence accuracy requirement is met, and output the global optimal solution Xglobal, which is a set of optimal control parameter instructions.

[0019] The optimization process of the hybrid optimization computation module is constructed as a reinforcement learning environment, in which the algorithm acts as an agent, its actions are to propose adjustment parameters, and the reward function R is a multi-dimensional comprehensive index. R = α * product quality + β * (1 / energy consumption) + γ * (1 / equipment loss rate) + δ * (1 / carbon emission intensity). The intelligent agent (algorithm) learns dynamic control strategies through interaction with the environment.

[0020] With the above structure, the system acts as an "intelligent agent" with learning capabilities. Through continuous interaction with the "environment" and a "trial and error-reward" mechanism, it learns autonomously and dynamically optimizes a set of optimal control strategies, rather than simply searching for a set of static optimal parameters.

[0021] The mechanism by which this process works is as follows: Agent and Environment Setup: Intelligent agent: namely, the hybrid bat optimization algorithm.

[0022] Environment: i.e., a continuous hydrogen production system or its high-fidelity digital twin model.

[0023] Status: The set of all production parameters acquired by the data acquisition module at a certain moment.

[0024] Action: The agent's proposed control parameters based on the current state.

[0025] Core learning driver – reward function: The system defines a multi-dimensional comprehensive reward function R: R = α * product quality + β * (1 / energy consumption) + γ * (1 / equipment loss rate) + δ * (1 / carbon emission intensity) How it works: Whenever an agent performs an "action," the environment provides feedback with a new "state" and a reward value R. This R value is a comprehensive score that measures the overall effectiveness of the action across four key dimensions: economy, equipment lifespan, and environmental impact.

[0026] Coefficient adjustment (α, β, γ, δ): These weighted coefficients act like "command sticks," allowing managers to dynamically adjust the direction of optimization based on production strategies.

[0027] Interaction and Learning Loop: The agent's goal is to maximize long-term cumulative rewards. Its learning process is a continuous closed loop: Observe the state → Generate an action → The environment executes the action and provides a reward → Update the strategy → Observe the new state again...

[0028] Through thousands of such interactions, the algorithm no longer merely seeks a fixed optimal solution, but learns a set of "experiences": under what production conditions, what control actions should be taken to obtain the highest long-term overall benefits. This enables the system to adapt to dynamic operating conditions such as changes in raw materials, catalyst decay, and fluctuations in market demand, achieving adaptive and forward-looking intelligent control.

[0029] The control and execution module includes a reaction material flow controller, a power adjustment unit, a photovoltaic-energy storage linkage unit, and a voltage stabilization unit. The reaction material flow controller uses a piezoelectric ceramic proportional valve with a flow accuracy of ±0.05%. It receives control commands from the mixing optimization calculation module and adjusts the switching frequency of the reaction materials. The switching frequency is 1-10Hz, achieving precise flow control. The power adjustment unit is based on a PID control algorithm and controls the rate at which the reaction materials are added according to commands, with a response time ≤0.5s. The photovoltaic-energy storage linkage unit increases the peak load when the photovoltaic power supply accounts for ≥60%. The voltage stabilization unit uses a UPS power supply. When the input voltage fluctuation exceeds ±5%, it automatically switches to regulated output to ensure stable equipment operation.

[0030] The above structure enables precise material control (reactant flow controller): employing a piezoelectric ceramic proportional valve as the core actuator, it boasts an extremely high flow accuracy of ±0.05% and a switching frequency adjustment capability of 1-10Hz. This translates the raw material addition rate command calculated by the optimization algorithm into precise material flow without hysteresis or overshoot, ensuring optimal reactant ratios from the source. Rapid power regulation (power regulation unit): Based on a PID control algorithm, this unit performs closed-loop control of the actuator's power. Its ≤0.5s rapid response time ensures the system can promptly track and execute dynamically changing optimization commands, overcoming equipment inertia and enabling process parameters to quickly stabilize at target setpoints, coping with fluctuations in production load. Green energy synergy (photovoltaic-energy storage linkage unit): This unit embodies the system's energy efficiency optimization strategy. When the photovoltaic power supply ratio is ≥60%, it automatically identifies a period of sufficient green energy and proactively increases the peak load of electrical equipment. This mechanism achieves intelligent matching between production electricity consumption and renewable energy generation curves, effectively reducing purchased electricity costs and the system's carbon footprint. Safety Cornerstone (Voltage Stabilization Unit): A power buffer and protection layer is constructed using a UPS power supply. When the mains voltage fluctuation exceeds the safety threshold of ±5%, the unit immediately and automatically switches to regulated output mode. This provides a stable, interference-free operating environment for front-end precision controllers and sensitive equipment, and is the fundamental guarantee for the reliable and continuous operation of the entire system.

[0031] The monitoring and display module includes a data visualization unit, an alarm unit, a carbon footprint dashboard, and an equipment health warning interface. The data visualization unit uses an industrial touchscreen to display real-time and historical curves of temperature, pressure parameters, catalyst loading, and product quality at each stage of the reactor, as well as predicted product quality, in graphical form, and supports data export. The alarm unit issues an audible and visual alarm when the operating parameters of the continuous hydrogenation system exceed safety thresholds, displays the alarm location and parameter deviation value on the screen, and triggers an emergency response procedure. The carbon footprint dashboard displays the carbon emission intensity per unit of product. The equipment health warning interface displays the probability of failure using three-color indicator lights.

[0032] Adopting the above structure, the system features comprehensive visual monitoring (data visualization unit): using an industrial touchscreen as the carrier, it integrates and displays key parameters such as temperature, pressure, and catalyst loading at each stage of the reactor through various visualization methods, including trend curves, real-time data dashboards, and historical data comparisons. It not only displays the current status but also provides predicted product quality curves, realizing a shift from "post-event analysis" to "pre-event prediction." It supports data export, providing complete data support for process optimization and problem tracing. The proactive safety protection system (alarm unit) establishes a tiered early warning mechanism that activates immediately when parameters exceed safety thresholds. It employs multi-modal alarm methods: audible and visual alarms (volume ≥ 85dB) ensure timely detection by on-site personnel, and the screen accurately displays the specific alarm location and deviation value. This system enables alarm linkage mechanisms, automatically triggering corresponding emergency response procedures to form a complete "monitoring-alarm-response" closed loop; transparent environmental performance management (carbon footprint dashboard): pushing carbon emission data from backend calculations to the frontend display, quantifying and displaying the carbon emission intensity of unit products in real time, providing production managers with an intuitive environmental performance dashboard, supporting green production decisions, and transforming environmental indicators from abstract concepts into specific and optimizable production parameters; predictive maintenance support (equipment health early warning interface): using a three-color indicator system (green - normal, yellow - attention, red - warning) to intuitively display the probability of equipment failure, and based on the analysis results of the digital twin module, transforming complex equipment health status into easily understandable visual signals, realizing the transformation from "reactive maintenance" to "predictive maintenance" operation and maintenance mode.

[0033] The database module includes a historical data storage unit and a data query unit. The historical data storage unit uses a MySQL database to store real-time data from the data acquisition module, calculation results from the hybrid optimization calculation module, and instruction data from the control execution module in a "year-month-day-hour" time dimension, with a storage period of ≥5 years. The data query unit supports users in querying historical data by time range, accurate to the minute, parameter type, and device ID, with a query response time of ≤1 second, and generates statistical reports.

[0034] Adopting the above structure, the structured data lifecycle management (historical data storage unit) utilizes a mature MySQL relational database as the storage core to ensure data consistency, integrity, and transaction reliability. A four-level time-dimensional index system (year-month-day-hour) is established to achieve refined data management along the timeline. A full-link data archiving mechanism is constructed to completely store the entire data chain from raw data acquisition and optimization processes to execution command results. A long-term data retention strategy of ≥5 years is set to meet the long-term needs of process research, quality traceability, and compliance auditing. Intelligent data services and decision support (data query unit) provide multi-dimensional... The system features combined query capabilities, supporting flexible searches based on time, parameter type, device ID, and other criteria, achieving a rapid response time of ≤1 second. This ensures a smooth interactive data analysis experience, supports real-time decision-making, and includes a built-in intelligent report generation engine that automatically calculates key performance indicators. It provides a data visualization interface, offering standardized data services for upper-layer applications. Furthermore, it facilitates data value mining and knowledge accumulation by fully recording the calculation process and results of the optimization module to form a "decision-result" related knowledge base. A data quality monitoring mechanism is established to ensure the accuracy and availability of stored data. It also supports time-series data mining and analysis, providing a data foundation for process optimization and predictive maintenance.

[0035] The federated learning module is deployed on the database module and uses the SM4 encryption algorithm and federated averaging algorithm to achieve encrypted sharing of model parameters across multiple plants, improving the model's generalization ability by more than 40%. The federated learning mode supports collaborative optimization across multiple plants. Based on the federated learning mode, the optimization algorithm model is updated and shared regularly.

[0036] Using the above structure, local knowledge extraction (independent computation in each plant area): Each production base independently trains its own hybrid optimization computation model on a local server using pre-processed data. Each local model incorporates tacit knowledge specific to that plant area, such as production processes, equipment characteristics, and raw material features. During training, the original data remains locally, eliminating the risk of privacy leaks. Secure parameter transmission (encrypted upload mechanism): Each plant area only encrypts the trained model parameters (weights and biases) using the SM4 national cryptographic algorithm. The encrypted parameters are then uploaded to the federated learning platform through a secure channel, ensuring information security during transmission. This step achieves the separation of "knowledge" and "data," sharing only the knowledge structure. Crystallization without exposing raw data; Global intelligence fusion (federated average aggregation): After receiving encrypted parameters uploaded by each node, the federated platform uses a federated average algorithm for aggregation. The algorithm integrates the local knowledge of each plant into a more comprehensive and robust global model through weighted averaging and other methods. This fusion mechanism enables the model to learn the general rules under different working conditions, significantly improving the generalization ability by more than 40%; Co-evolutionary loop (model update and distribution): The enhanced global model obtained by aggregation is encrypted and distributed to each participating plant to update its local system. While obtaining global intelligence, each plant can make personalized fine-tuning based on local data, forming a continuous evolutionary closed loop of "local training → encrypted upload → global aggregation → distribution and update".

[0037] The digital twin module includes a process simulation unit, a predictive maintenance unit, and an operation simulation unit. The process simulation unit pre-executes control commands in a virtual space to predict the production status and product quality in the future, achieving forward-looking optimization. The predictive maintenance unit analyzes the digital mapping of equipment operation data, uses machine learning models to predict the remaining service life of key components, and generates maintenance work orders in advance. The operation simulation unit provides a high-fidelity simulation environment for new employee training and operation plan drills.

[0038] Using the above structure, the process forward simulation (process simulation unit) pre-executes and verifies the effects of control commands generated by the optimization calculation module in a virtual space. Based on mechanistic models and real-time data, it dynamically predicts the production status and product quality over a future period, realizing a shift from ex-post control of "execution-observation-adjustment" to forward-looking control of "prediction-optimization-execution." This effectively avoids production fluctuations and quality risks caused by improper control, improving the safety and reliability of optimization decisions. The equipment health prediction (predictive maintenance unit) constructs degradation models for key components through digital mapping of equipment operating data. By using machine learning algorithms to predict remaining service life and generate probabilistic fault warnings, maintenance work orders can be generated in advance, realizing the transformation from "planned maintenance" to "on-demand maintenance," significantly reducing unplanned downtime and improving the overall utilization rate and service life of equipment. Operation safety pre-drill (operation simulation unit): provides a high-fidelity simulation environment that accurately reproduces actual production scenarios and operation interfaces, supports new employees to conduct system operation and fault handling training in a zero-risk environment, and allows experienced operators to repeatedly practice and optimize abnormal operating conditions and emergency plans, reducing practical training costs and safety risks, and improving personnel's emergency response capabilities.

[0039] A method for intelligent control of continuous hydrogenation using the above system includes the following control steps: A1: Real-time acquisition of data from the entire hydrogenation production process via the data acquisition module; A2: Preprocess the collected data, including weight analysis, extreme data processing, and data deduplication; among which, weight analysis is used to determine the correlation between the operating variables and product quality and to provide a suggested value for hydrogen flow rate; A3: The hybrid bat optimization algorithm in the hybrid optimization calculation module is used to calculate the preprocessed data and generate the optimal equipment control instructions; A4: The instructions are executed by the control and execution module to control the reactant addition rate and related equipment parameters; A5: Real-time visual monitoring and alarms are provided for the entire process through the monitoring and display module; A6: Store all data through the database module and establish spatiotemporal relationships between production data to support full-chain traceability; In step A2 or A3, the method further includes using a depth-first search algorithm to traverse the correlation data between the amount of raw materials added and the degree of reaction and product quality, in order to optimize the strategy formulation; and / or using an RBF neural network to establish a soft measurement model with multiple parameters, including temperature.

[0040] A method for calculating product qualification rate based on the above system includes the following steps: 1) Threshold setting: Users can set the pass threshold θ for key product quality indicators; 2) Pass rate calculation, the formula is as follows: ; In the formula, si represents the number of qualified products identified in the i-th case, n represents the number of individual data groups identified, S is the total amount of data stream, and Y represents the product prediction pass rate.

[0041] Compared with existing technologies, the continuous hydrogenation interconnection equipment control system based on the hybrid bat optimization algorithm has the following advantages: The system constructs a complete closed loop of "perception-decision-execution-learning". Through the collaboration of multi-source data perception, intelligent algorithm decision-making and high-precision execution mechanisms, it upgrades traditional experience-driven management into automated and self-optimizing intelligent control.

[0042] The core algorithm integrates the advantages of multiple intelligent optimization algorithms and combines reinforcement learning and federated learning. It can optimize dynamically in real time and continuously iterate through multi-base knowledge sharing to improve generalization and adaptability.

[0043] The plan incorporates carbon footprint into the optimization objectives and uses digital twins to achieve safety early warning and predictive maintenance, thereby simultaneously improving green and low-carbon practices, production safety, and economic benefits.

[0044] Edge-cloud collaboration and modular design not only ensure real-time control and system reliability, but also enable the system to flexibly adapt to future changes in demand and technological upgrades, supporting continuous evolution. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating the present invention.

[0046] Figure 2 This is a correlation graph between concentration (product concentration) and other variables in this invention.

[0047] Figure 3 This is a heatmap showing the correlation between concentration and related variables in this invention.

[0048] Figure 4 This is a graph of weighted analysis coefficients in this invention.

[0049] Figure 5 This is a graph showing the product qualification rate calculation in this invention. Detailed Implementation

[0050] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings to further illustrate the technical solutions of the present invention. However, the present invention does not include, but is not limited to, these embodiments.

[0051] like Figure 1As shown, this continuous hydrogenation interconnection equipment control system based on the hybrid bat optimization algorithm includes a data acquisition module, a data preprocessing module, a hybrid optimization calculation module, a control execution module, a monitoring and display module, a human-machine collaborative decision-making module, a database module, a federated learning module, and a digital twin module. The output of the data acquisition module is connected to the input of the data preprocessing module, used to collect the operating parameters of each device in the hydrogenation production system in real time and transmit them to the data preprocessing module. This module supports the access of multi-source heterogeneous devices such as PLCs and DCSs, and realizes real-time data transmission and standardized processing through Modbus and OPCUA protocols. The output of the data preprocessing module is connected to the input of the hybrid optimization calculation module, used to process the data to obtain more accurate data and transmit it to the hybrid optimization calculation module. The hybrid optimization calculation module is used to run the hybrid bat optimization algorithm program, optimize the preprocessed real-time data, and output equipment control commands. The control execution module receives commands from the hybrid optimization calculation module. The system commands and executes equipment parameter adjustments; the monitoring and display module is bidirectionally connected to both the hybrid optimization calculation module and the control execution module, providing real-time feedback on operating data and control status for human-machine interaction; the human-machine collaborative decision-making module uses interpretable AI technology to generate decision-based reports from the control commands output by the hybrid optimization calculation module and receives feedback commands from the operator, which are used to continuously optimize the algorithm model in the hybrid optimization calculation module; the database module is bidirectionally connected to the data acquisition module and the hybrid optimization calculation module, persistently storing historical data and algorithm operation data to support data querying and model training; the federated learning module aggregates and shares model parameters for local models from multiple production bases in an encrypted state to collaboratively improve algorithm performance; the digital twin module constructs a virtual model that is a complete mirror image of the physical workshop based on historical data, mechanistic models (such as the Arrhenius equation), and real-time data streams, using the model to predict the remaining service life of key components and generate maintenance warnings in advance.

[0052] This system achieves intelligent control of the hydrogenation production process through the collaborative work of multiple modules. First, the data acquisition module collects real-time data on equipment operating parameters, catalyst status, feedstock composition, and carbon emissions via a multi-source sensor network, ensuring spatiotemporal consistency through a "5G+NTP" dual synchronization mechanism. The data preprocessing module performs weight analysis, outlier handling, and standardization on the raw data, providing a high-quality data foundation for subsequent optimization calculations.

[0053] The hybrid optimization computation module employs the Hybrid Bat Optimization Algorithm, which integrates the advantages of BA, PSO, SA, and WOA algorithms to perform multi-objective optimization within a reinforcement learning framework. This algorithm dynamically balances global exploration and local exploitation capabilities through an adaptive parameter adjustment mechanism, using product quality, energy consumption, equipment loss rate, and carbon emission intensity as comprehensive objective functions to output the optimal combination of process parameters.

[0054] The digital twin module constructs a virtual production line based on real-time data and mechanistic models, pre-executing and verifying optimization instructions to predict equipment health status. The human-machine collaborative decision-making module generates decision-making reports using interpretable AI technology for operator review and confirmation. The control and execution module precisely controls the flow rate of reactants and equipment power according to final instructions, and achieves energy optimization through photovoltaic-energy storage linkage.

[0055] The monitoring and display module shows the real-time operating status and early warning information of the entire process. The database module persistently stores all process data, supporting historical traceability and model training. The federated learning module enables encrypted parameter sharing across multiple production sites, continuously improving the model's generalization ability. All modules form a continuously optimized intelligent control system through a closed-loop feedback mechanism, ultimately achieving safe, efficient, and low-carbon operation of the hydrogenation production process.

[0056] The data acquisition module includes a sensor monitoring unit, a catalyst status monitoring unit, a raw material microscopic analysis unit, a carbon footprint monitoring unit, and a time synchronization unit. The sensor monitoring unit includes voltage sensors, current sensors, temperature sensors, pressure sensors, and power sensors. Voltage sensors acquire the real-time power supply voltage of the hydrogenation equipment, with a measurement range of 0-10kV and an accuracy of ±0.5%. Current sensors acquire the input / output current of the equipment, with a measurement range of 0-500A and an accuracy of ±0.3%. Power sensors acquire active power, with a measurement range of 0-1000kW and an accuracy of ±0.2%. Temperature sensors are located at the inlet, outlet, and reaction section of each reactor to acquire temperature data, with a measurement range of -50℃ to 500℃ and an accuracy of ±0.1℃. Pressure sensors acquire the pressure inside each reactor and the pressure difference between different sections of the reaction tower, with a measurement range of 0-10MPa and an accuracy of ±0.05MPa. Catalyst status monitoring... The unit is equipped with an X-ray fluorescence spectrometer (model: Thermo Scientific ARL Perform'X), with a measurement range of 0-100% and an accuracy of ±0.1%, which collects real-time data on catalyst active components (such as Ni content in Raney nickel) and crystal structure parameters, with a data sampling frequency of 1 time / minute. The raw material microstructure analysis unit integrates a near-infrared spectrometer (model: Bruker Matrix-F), which completes the detection of raw material purity (accuracy ±0.05%) and impurity content (detection limit 0.01%) within 10 seconds, supporting rapid analysis of raw materials such as methanol and toluene. The carbon footprint monitoring unit is equipped with a carbon emission sensor (model: Vaisala CARBOCAP® GMP343), with a measurement range of 0-50 kg CO2 / h and an accuracy of ±2%, used to collect carbon emission data related to reactor tail gas and power consumption, and to calculate the total carbon emissions C in real time using the formula: Where Mi is the amount of raw materials used, Fi is the carbon emission factor of raw materials, E is the power consumption, G is the carbon emission factor of electricity, Wj is the amount of waste discharged, and Hj is the carbon emission factor of treatment. This indicator is incorporated into the decision-making process of the optimization algorithm. The time synchronization unit is based on the NTP protocol and adopts a "5G+NTP" dual synchronization mechanism to ensure the spatiotemporal consistency of micro and macro data, realize the timestamp calibration of data from each sensor, ensure that all sensor data have a unified timestamp, and the time synchronization error is ≤0.5ms, thus ensuring the spatiotemporal consistency of micro and macro data.

[0057] Comprehensive parameter sensing: The module covers all key dimensions, from power consumption (voltage, current, power), reaction environment (temperature, pressure), to material characteristics (raw material purity, catalyst activity), and even environmental indicators (carbon emissions), through five specialized units. This enables simultaneous tracking of "energy flow," "material flow," and "information flow."

[0058] The fusion of microscopic and macroscopic data: It not only collects traditional macroscopic process parameters (such as temperature and pressure), but also innovatively introduces advanced sensing technologies such as X-ray fluorescence spectroscopy and near-infrared spectroscopy to achieve online real-time analysis of microscopic chemical properties such as catalyst activity and raw material composition. This allows data acquisition to delve from macroscopic appearances to the microscopic level that affects the essence of the reaction.

[0059] Real-time quantification of carbon footprint: using dedicated carbon emission sensors combined with formulas The module can calculate and quantify the total carbon emissions in the production process online in real time, transforming environmental indicators from offline accounting into online optimizable process variables, and providing direct data support for green and low-carbon production.

[0060] High-precision spatiotemporal synchronization: The module's "nerve center"—the time synchronization unit—employs a "5G+NTP" dual synchronization mechanism, assigning a unified timestamp to all sensor data with an extremely high precision of ≤0.5ms. This ensures that data collected from different sources and at different frequencies are completely aligned in the time dimension, eliminating analytical errors caused by data misalignment and providing temporal consistency guarantees for subsequent digital twin modeling, causal analysis, and precise control.

[0061] The data preprocessing module includes a weight analysis unit, an extreme data processing unit, a data deduplication unit, and a standardization unit. The weight analysis unit, based on the random forest algorithm, analyzes the correlation between parameters and product quality, assigning a weight of 0.3 to hydrogen flow rate, 0.25 to temperature, and 0.2 to pressure to optimize instruction priority sorting. The extreme data processing unit uses the 3σ principle to remove outliers (such as temperatures exceeding the mean ± 3 standard deviations) and uses linear interpolation to supplement missing data. The data deduplication unit retains the earliest timestamp of duplicate data and deletes redundant data to ensure data uniqueness. The standardization unit uses a unified data format of timestamp-device ID-parameter type-value-unit, supporting access from multiple heterogeneous devices such as PLCs and DCS.

[0062] A standardized data pipeline transforms raw, heterogeneous, and noisy field data into high-quality, structured, and directly usable standardized information for intelligent decision-making. Its workflow follows a logical chain of "empowerment → purification → standardization → unification." Intelligent Weighting (Weight Analysis Unit): Based on the random forest algorithm, this unit quantifies the correlation between input parameters and final product quality. By assigning different weights to key process parameters (such as hydrogen flow rate 0.3, temperature 0.25, and pressure 0.2), the weight analysis coefficient diagram is shown below. Figure 4 As shown, it provides decision-priority guidance for subsequent optimization algorithms, enabling optimization resources to be more concentrated on the key variables that have the greatest impact on product quality. The correlation plot between concentration (product concentration) and other variables is shown in the figure. Figure 2 As shown, the heatmap shows the correlation between concentration and related variables. Figure 3 As shown.

[0063] Data Cleansing (Extreme Data Processing Unit): This unit uses the 3σ principle as a "filter" to automatically identify and remove significant abnormal data points caused by sensor malfunctions or transmission interference. Simultaneously, it uses linear interpolation to reasonably repair missing data, effectively ensuring the integrity and reliability of the dataset and providing a stable and trustworthy data foundation for the algorithm, preventing "garbage in, garbage out."

[0064] Information deduplication (data deduplication unit): This unit performs deduplication (retaining the earliest record) by comparing timestamps to address duplicate records that may occur during data transmission or collection. This ensures the uniqueness of data records, avoids biases caused by duplicate data in statistical analysis and model training, and improves the efficiency and accuracy of the dataset.

[0065] Standardized Formatting Unit: As the final gate for data output, this unit forcibly converts heterogeneous data from different sources such as PLCs and DCS into a unified structured format of "timestamp-device ID-parameter type-value-unit". This completely solves the access problem of multi-source heterogeneous devices, enabling downstream hybrid optimization calculation modules to identify and process all data without discrimination, and achieving standardization of the data language within the system.

[0066] The hybrid optimization computing module includes a processor and a storage unit. The processor uses an Intel Core i7-12700K or an industrial processor of equivalent performance. It calls the algorithm program to perform calculations on the real-time data input from the data acquisition module and outputs control instructions such as the addition rate of reactants (methanol, toluene, hydrogen, catalyst) and equipment power. The storage unit stores the hybrid bat optimization algorithm program, which is a composite optimization algorithm that integrates the whale optimization algorithm (WOA), bat algorithm (BA), particle swarm optimization algorithm (PSO), and simulated annealing algorithm (SA).

[0067] Relying on high-performance hardware, a hybrid bat optimization algorithm that integrates the advantages of multiple intelligent algorithms is run to perform parallel and efficient global search on the accurate data input from the front end, and finally output control instructions that optimize the production target.

[0068] Its working process can be divided into two levels: 1. Hardware execution level: The module provides powerful computing capabilities through an Intel Core i7-12700K or equivalent industrial processor, ensuring real-time processing of massive amounts of data and execution of complex optimization algorithms. The storage unit serves as an "algorithm repository," hosting the core hybrid bat optimization algorithm program. The processor retrieves this program from the storage unit, performs high-speed calculations on the input data, and ultimately outputs an optimal control instruction set regarding reactant addition rates, device power, and other parameters.

[0069] 2. Core Algorithm Level (Collaborative Mechanism of the Hybrid Bat Optimization Algorithm): The core of this module lies in its unique hybrid bat optimization algorithm, which is not a simple combination but an organic integration of the advantages of four algorithms to form a highly efficient search and decision engine: The bat algorithm serves as the main framework: it simulates the echolocation behavior of bats, and provides a basic search mechanism by using frequency modulation and pulse emission for global exploration and local exploitation.

[0070] The Particle Swarm Optimization algorithm incorporates "collective intelligence": by introducing "individual historical best" and "global best" information, it guides the entire bat population to converge quickly to a better region, effectively solving the problem of slow convergence speed in traditional algorithms.

[0071] Simulated annealing provides "breakthrough capability": its mechanism of accepting poor solutions with probability allows the algorithm to jump out when it gets stuck in local optima, significantly enhancing its ability to escape local optima and find the global optimum.

[0072] The Whale Optimization Algorithm enhances the search strategy: its unique prey encirclement and spiral position update mechanism provides more diverse path choices for the entire search process, further improving the comprehensiveness and robustness of the search.

[0073] The specific steps for implementing the hybrid bat optimization algorithm are as follows: Step S1, Initialization: Randomly initialize a group of bats (i.e., candidate solutions) in the solution space, define the position Xi and velocity vi of each bat, and set parameters such as initial frequency f0, loudness A0, and pulse emission rate r0; at the same time, initialize the inertial weight w, learning factors c1 and c2 of the particle swarm optimization algorithm, and the initial temperature T of the simulated annealing algorithm. Step S2, Fitness Assessment: Calculate the fitness value of each bat based on the objective function (e.g., highest product quality, lowest energy consumption, minimum pressure fluctuation, etc.); Step S3: Location update and hybrid optimization; S3a) Exploring the Bat Algorithm: Updating the frequency, speed, and position of bats based on the standard bat algorithm; fi = fmin + (fmax - fmin) * β (β is a random number in [0,1]); vi(t+1)=vi(t)+(Xi(t)-Xglobal)*fi; Xi_new=Xi(t)+vi(t+1); S3b) PSO Collaborative Guidance: To improve the convergence speed, a PSO group collaboration mechanism is introduced to perform a secondary correction on the bat's speed; vi(t+1)=w*vi(t+1)+c1*r1(Xpbest-Xi(t))+c2*r2*(Xglobal-Xi(t)) Where Xpbest is the individual historical best position of the bat, and Xglobal is the current global best position; S3c)SA perturbation escapes local optima: To avoid premature convergence, the Metropolis criterion of the simulated annealing algorithm is used to accept new solutions; calculate the fitness difference ΔE between the new position Xi_new and the old position Xi(t); if ΔE<0, accept the new solution; if ΔE>=0, accept the new solution with probability P_acc=exp(-ΔE / T) (where T is the current temperature); Step S4, Local Search: Perform a random walk around the current optimal solution Xglobal to generate a new local solution; if the new solution is better than the current optimal solution and its random number is greater than the pulse emission rate ri, then accept the new local solution; Step S5, Update State: If the new solution is accepted, update the loudness Ai and pulse emission rate ri of the bat; Ai(t+1) = α*Ai(t); ri(t+1)=ri(0)*[1-exp(-γ*t)]; Step S6, Cooling: Reduce the current temperature T of the simulated annealing algorithm according to the cooling plan, for example, T(t+1)=0.95*T(t); Step S7, Termination Judgment: Repeat steps S2 to S6 until the maximum number of iterations is reached or the convergence accuracy requirement is met, and output the global optimal solution Xglobal, which is a set of optimal control parameter instructions.

[0074] The optimization process of the hybrid optimization computation module is constructed as a reinforcement learning environment, where the algorithm acts as an agent, its actions are to propose parameter adjustments, and the reward function R is a multi-dimensional comprehensive index. R = α * product quality + β * (1 / energy consumption) + γ * (1 / equipment loss rate) + δ * (1 / carbon emission intensity). The intelligent agent (algorithm) learns dynamic control strategies through interaction with the environment.

[0075] As an intelligent agent with learning capabilities, it continuously interacts with the environment (the actual production system or digital twin model) and learns autonomously and dynamically optimizes a set of best control strategies through a trial-and-error-reward mechanism, rather than simply searching for a set of static optimal parameters.

[0076] The mechanism by which this process works is as follows: Agent and Environment Setup: Intelligent agent: namely, the hybrid bat optimization algorithm.

[0077] Environment: i.e., a continuous hydrogen production system or its high-fidelity digital twin model.

[0078] Status: The set of all production parameters (such as temperature, pressure, flow rate, etc.) acquired by the data acquisition module at a certain moment.

[0079] Action: The agent outputs control parameter suggestions based on the current state (such as adjusting the hydrogen feed rate, changing the reaction temperature, etc.).

[0080] Core learning driver – reward function: The system defines a multi-dimensional comprehensive reward function R: R = α * product quality + β * (1 / energy consumption) + γ * (1 / equipment loss rate) + δ * (1 / carbon emission intensity) How it works: Whenever an agent performs an "action", the environment will provide a new "state" and a reward value R. This R value is a comprehensive score that measures the overall effect of the action in four key dimensions: economy (quality, energy consumption), equipment lifespan, and environmental protection (carbon emissions).

[0081] Coefficient adjustment (α, β, γ, δ): These weighted coefficients act like "command sticks," allowing managers to dynamically adjust the direction of optimization based on production strategies (such as pursuing the highest quality, the lowest energy consumption, or the maximum emission reduction).

[0082] Interaction and Learning Loop: The goal of the intelligent agent (algorithm) is to maximize long-term cumulative rewards. Its learning process is a continuous closed loop: Observe the state → Generate an action → The environment executes the action and provides a reward → Update the strategy → Observe the new state again...

[0083] Through thousands of such interactions, the algorithm no longer merely seeks a fixed optimal solution, but learns a set of "experiences": under what production conditions, what control actions should be taken to obtain the highest long-term overall benefits. This enables the system to adapt to dynamic operating conditions such as changes in raw materials, catalyst decay, and fluctuations in market demand, achieving adaptive and forward-looking intelligent control.

[0084] The control and execution module includes a reaction material flow controller, a power regulation unit, a photovoltaic-energy storage linkage unit, and a voltage stabilization unit. The reaction material flow controller uses a piezoelectric ceramic proportional valve with a flow accuracy of ±0.05%. It receives control commands from the mixing optimization calculation module and adjusts the switching frequency of the reaction materials. The switching frequency is 1-10Hz, achieving precise flow control. The power regulation unit is based on a PID control algorithm and controls the rate at which the reaction materials are added according to the commands, with a response time ≤0.5s. The photovoltaic-energy storage linkage unit increases the peak load when the photovoltaic power supply accounts for ≥60%. The voltage stabilization unit uses a UPS power supply. When the input voltage fluctuation exceeds ±5%, it automatically switches to regulated output to ensure stable equipment operation.

[0085] Precise material control (reaction feedstock flow controller): Using a piezoelectric ceramic proportional valve as the core actuator, with an extremely high flow accuracy of ±0.05% and a switching frequency adjustment capability of 1-10Hz, the feedstock (such as methanol and hydrogen) calculated by the optimization algorithm is added to the rate command, and converted into precise material flow without hysteresis or overshoot, ensuring the optimization of reactant ratio from the source.

[0086] Rapid power regulation (power regulation unit): Based on the PID control algorithm, this unit performs closed-loop control of the actuator's power. Its rapid response time of ≤0.5s ensures that the system can promptly track and execute dynamically changing optimization commands, overcome equipment inertia, and enable process parameters to quickly stabilize at the target setpoint, thus coping with fluctuations in production load.

[0087] Green Energy Synergy (Photovoltaic-Energy Storage Linkage Unit): This unit embodies the system's energy efficiency optimization strategy. When the photovoltaic power supply ratio is ≥60%, it automatically determines that it is a period of sufficient green energy and proactively increases the peak load of electrical equipment. This mechanism achieves intelligent matching between production electricity consumption and renewable energy generation curves, effectively reducing the cost of purchased electricity and the system's carbon footprint.

[0088] Safety Cornerstone (Voltage Stabilization Unit): A power buffer and protection layer is constructed using a UPS power supply. When the mains voltage fluctuation exceeds the safety threshold of ±5%, the unit immediately and automatically switches to regulated output mode. This provides a stable, interference-free operating environment for front-end precision controllers and sensitive equipment, and is the fundamental guarantee for the reliable and continuous operation of the entire system.

[0089] The monitoring and display module includes a data visualization unit, an alarm unit, a carbon footprint dashboard, and an equipment health warning interface. The data visualization unit uses an industrial touch screen to display real-time and historical curves of temperature, pressure parameters, catalyst loading, and product quality at each stage of the reactor, as well as predicted product quality, in graphical form, and supports data export. The alarm unit issues an audible and visual alarm (e.g., alarm volume ≥ 85dB) when the operating parameters of the continuous hydrogenation system exceed safety thresholds (e.g., temperature ≥ 400℃, pressure ≥ 8MPa), displays the alarm location and parameter deviation value on the screen, and triggers the emergency handling procedure. The carbon footprint dashboard displays the carbon emission intensity per unit of product. The equipment health warning interface displays the probability of failure through three-color indicator lights.

[0090] Full-element visual monitoring (data visualization unit): Using an industrial touch screen as the carrier, it integrates and displays key parameters such as temperature, pressure, and catalyst loading at each stage of the reactor in a spatiotemporal dimension through various visualization forms such as trend curves, real-time data dashboards, and historical data comparisons. It not only displays the current status but also provides a prediction of product quality curves, realizing the transformation from "post-event analysis" to "pre-event prediction". It supports data export function, providing complete data support for process optimization and problem tracing. Active safety protection system (alarm unit): Establish a graded early warning mechanism, which will be activated immediately when the parameters exceed the safety threshold (such as temperature ≥400℃, pressure ≥8MPa). It adopts a multi-modal alarm method: audible and visual alarm (volume ≥85dB) to ensure that on-site personnel can perceive it in time. The screen accurately locates and displays the specific alarm location and deviation value, realizes the alarm linkage mechanism, automatically triggers the corresponding emergency handling process, and forms a complete "monitoring-alarm-response" closed loop. Transparent Environmental Performance Management (Carbon Footprint Dashboard): This system moves carbon emission data from back-end calculations to the front-end display, providing real-time quantitative display of carbon emission intensity per unit of product. It offers production managers an intuitive environmental performance dashboard, supports green production decisions, and transforms environmental indicators from abstract concepts into specific, optimizable production parameters. Predictive maintenance support (equipment health warning interface): The three-color indicator system (green - normal, yellow - attention, red - warning) intuitively displays the probability of equipment failure. Based on the analysis results of the digital twin module, the complex equipment health status is transformed into an easy-to-understand visual signal, realizing the transformation of the operation and maintenance mode from "reactive maintenance" to "predictive maintenance".

[0091] The database module includes a historical data storage unit and a data query unit. The historical data storage unit uses a MySQL database to store real-time data from the data acquisition module, calculation results from the hybrid optimization calculation module, and instruction data from the control execution module in a "year-month-day-hour" time dimension, with a storage period of ≥5 years. The data query unit supports users in querying historical data by time range, accurate to the minute, parameter type (such as temperature, pressure), and device ID, with a query response time of ≤1 second, and generates statistical reports (such as daily average pass rate, raw material consumption statistics).

[0092] Structured data lifecycle management (historical data storage unit): Adopting a mature MySQL relational database as the storage core, ensuring data consistency, integrity and transaction reliability, establishing a four-level time dimension index system of "year-month-day-hour" to achieve fine-grained management of data on the timeline, building a full-link data archiving mechanism, fully storing the complete data chain from raw data collection, optimization calculation process to execution command results, and setting a long-term data retention strategy of ≥5 years to meet the long-term needs of process research, quality traceability and compliance audit; Intelligent Data Services and Decision Support (Data Query Unit): Provides multi-dimensional combined query capabilities, supports flexible retrieval by time (accurate to the minute), parameter type, device ID, etc., achieves a fast response performance of ≤1 second, ensures a smooth interactive data analysis experience, supports real-time decision-making, has a built-in intelligent report generation engine, automatically calculates key performance indicators (such as daily average pass rate, raw material consumption statistics), provides data visualization interfaces, provides standardized data services for upper-level applications, data value mining and knowledge accumulation, forms a "decision-result" related knowledge base by fully recording the calculation process and results of the optimization calculation module, establishes a data quality monitoring mechanism to ensure the accuracy and availability of stored data, supports time series data mining and analysis, and provides a data foundation for process optimization and predictive maintenance.

[0093] The federated learning module is deployed on the database module and uses the SM4 encryption algorithm and federated averaging algorithm to achieve encrypted sharing of model parameters across multiple plants, improving the model's generalization ability by more than 40%. The federated learning mode supports collaborative optimization across multiple plants (local preprocessing of data in each plant, encrypted uploading of model parameters, aggregation of parameters on the federated platform, and downloading of the global model to update the local system). Based on the federated learning mode, the optimization algorithm model is updated and shared regularly.

[0094] Local knowledge extraction (independent calculation in each plant): Each production base uses the pre-processed data to independently train its own hybrid optimization calculation model on the local server. Each local model absorbs the implicit knowledge of the plant's unique production process, equipment characteristics and raw material characteristics. During the training process, the original data is always kept locally, so there is no risk of privacy leakage. Secure parameter transmission (encrypted upload mechanism): Each plant only encrypts the trained model parameters (weights and biases) using the SM4 national cryptographic algorithm. The encrypted parameters are then uploaded to the federated learning platform through a secure channel, ensuring information security during transmission. This step achieves the separation of "knowledge" and "data," sharing only the knowledge crystallization without exposing the original data. Global intelligent fusion (federated average aggregation): After receiving the encrypted parameters uploaded by each node, the federated platform uses the federated average algorithm to aggregate them. The algorithm integrates the local knowledge of each plant area into a more comprehensive and robust global model through weighted averaging and other methods. This fusion mechanism enables the model to learn the general rules under different working conditions, significantly improving the generalization ability by more than 40%. Co-evolutionary loop (model update and distribution): The enhanced global model obtained by aggregation is encrypted and distributed to each participating plant area to update their local systems. While gaining global wisdom, each plant area can still make personalized fine-tuning based on local data, forming a continuous evolutionary closed loop of "local training → encrypted upload → global aggregation → distribution and update".

[0095] The digital twin module includes a process simulation unit, a predictive maintenance unit, and an operation simulation unit. The process simulation unit pre-executes control commands in a virtual space to predict the production status and product quality in the future, achieving forward-looking optimization. The predictive maintenance unit analyzes the digital mapping of equipment operation data, uses machine learning models to predict the remaining service life of key components (such as feed pumps and compressors), and generates maintenance work orders in advance. The operation simulation unit provides a high-fidelity simulation environment for new employee training and operation plan drills.

[0096] Using the above structure, the process forward simulation (process simulation unit) pre-executes and verifies the effects of the control instructions generated by the optimization calculation module in a virtual space. Based on the mechanism model and real-time data, it dynamically predicts the production status and product quality in the future, realizing the transformation from ex-post control of "execution-observation-adjustment" to forward-looking control of "prediction-optimization-execution". This effectively avoids production fluctuations and quality risks caused by improper control, and improves the safety and reliability of optimization decisions. Equipment health prediction (predictive maintenance unit): By digitally mapping equipment operating data, a degradation model of key components (such as feed pumps and compressors) is constructed. Machine learning algorithms are used to predict the remaining service life, generate probabilistic fault warnings, and generate maintenance work orders in advance, realizing the transformation from "planned maintenance" to "on-demand maintenance", significantly reducing unplanned downtime and improving the overall utilization rate and service life of equipment. Operational safety rehearsal (operational simulation unit): Provides a high-fidelity simulation environment that accurately reproduces actual production scenarios and operating interfaces. It supports new employees to conduct system operation and fault handling training in a zero-risk environment, and allows experienced operators to repeatedly practice and optimize abnormal operating conditions and emergency plans. This significantly reduces the cost of practical training and safety risks, and improves personnel's emergency response capabilities.

[0097] A method for intelligent control of continuous hydrogenation using the above system includes the following control steps: A1: Real-time acquisition of data from the entire hydrogenation production process via the data acquisition module; A2: Preprocess the collected data, including weight analysis, extreme data processing, and data deduplication; among which, weight analysis is used to determine the correlation between the operating variables and product quality and to provide a suggested value for hydrogen flow rate; A3: The hybrid bat optimization algorithm in the hybrid optimization calculation module is used to calculate the preprocessed data and generate the optimal equipment control instructions; A4: By controlling the execution module to execute instructions, the reactant addition rate and related equipment parameters are controlled; A5: Real-time visual monitoring and alarms are provided for the entire process through the monitoring and display module; A6: Store all data through the database module and establish spatiotemporal relationships between production data to support full-chain traceability; In step A2 or A3, the method further includes using a depth-first search algorithm to traverse the correlation data between the amount of raw materials added and the degree of reaction and product quality, in order to optimize the strategy formulation; and / or using an RBF neural network to establish a soft measurement model with multiple parameters, including temperature.

[0098] A method for calculating product qualification rate based on the above system includes the following steps: 1) Threshold setting: Users can set the pass threshold θ for key product quality indicators; 2) Pass rate calculation, the formula is as follows: ; In the formula, si represents the number of qualified products identified in the i-th case, n represents the number of individual data groups identified, S is the total amount of data stream, and Y represents the product prediction pass rate.

[0099] Product qualification rate calculation chart, as shown Figure 5 As shown.

[0100] In summary, the system constructs a complete closed loop of "perception-decision-execution-learning". Through the collaboration of multi-source data perception, intelligent algorithm decision-making and high-precision execution mechanisms, it upgrades traditional experience-driven systems into automated and self-optimizing intelligent control.

[0101] The core algorithm integrates the advantages of multiple intelligent optimization algorithms and combines reinforcement learning and federated learning. It can optimize dynamically in real time and continuously iterate through multi-base knowledge sharing to improve generalization and adaptability.

[0102] The plan incorporates carbon footprint into the optimization objectives and uses digital twins to achieve safety early warning and predictive maintenance, thereby simultaneously improving green and low-carbon practices, production safety, and economic benefits.

[0103] Edge-cloud collaboration and modular design not only ensure real-time control and system reliability, but also enable the system to flexibly adapt to future changes in demand and technological upgrades, supporting continuous evolution.

[0104] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A continuous hydrogenation interconnection equipment regulation system based on a hybrid bat optimization algorithm, comprising a data acquisition module, a data preprocessing module, a hybrid optimization calculation module, a regulation execution module, a monitoring display module, a man-machine collaborative decision module, a database module, a federated learning module and a digital twin module, characterized in that, The output end of the data acquisition module is connected with the input end of the data preprocessing module, for real-time acquisition of the operating parameters of each device in the hydrogen production system and transmission to the data preprocessing module, which supports access of multiple source heterogeneous devices including but not limited to PLC and DCS, and realizes real-time data transmission and standardized processing through Modbus and OPCUA protocols; The output end of the data preprocessing module is connected with the input end of the hybrid optimization calculation module, for processing of more accurate data and transmission to the hybrid optimization calculation module; the hybrid optimization calculation module is used for running a hybrid bat optimization algorithm program, optimizing calculation of the preprocessed real-time data, and outputting device control instructions; the control execution module receives the instructions of the hybrid optimization calculation module and executes device parameter control; The monitoring display module is bidirectionally connected with the hybrid optimization calculation module and the control execution module, for real-time feedback of operating data and control state, and for human-computer interaction; the human-computer collaborative decision-making module generates a decision basis report for the control instructions output by the hybrid optimization calculation module by using an interpretable AI technology, and receives feedback instructions of an operator, the feedback instructions being used for continuous optimization of the algorithm model in the hybrid optimization calculation module; the database module is bidirectionally connected with the data acquisition module and the hybrid optimization calculation module, for persistent storage of historical data and algorithm operation data, and for support of data query and model training; the federated learning module aggregates and shares model parameters of local models of multiple production bases in an encrypted state, so as to collaboratively improve the algorithm performance; the digital twin module constructs a virtual model completely mirroring a physical workshop based on historical data, mechanism models and real-time data streams, predicts the remaining service life of key components by using the model, and generates a maintenance warning in advance.

2. The continuous hydrogenation interconnection equipment regulation system based on hybrid bat optimization algorithm according to claim 1, characterized in that, The data acquisition module includes a sensor monitoring unit, a catalyst state monitoring unit, a raw material microanalysis unit, a carbon footprint monitoring unit and a time synchronization unit. The sensor monitoring unit includes a voltage sensor, a current sensor, a temperature sensor, a pressure sensor and a power sensor. The voltage sensor acquires real-time power supply voltage of the hydrogenation equipment, with a measurement range of 0-10 kV and an accuracy of ±0.5%. The current sensor acquires equipment input / output current, with a measurement range of 0-500 A and an accuracy of ±0.3%. The power sensor acquires active power, with a measurement range of 0-1000 kW and an accuracy of ±0.2%. The temperature sensor is arranged at the inlet, outlet and reaction section of each reaction kettle to acquire temperature data, with a measurement range of -50℃-500℃ and an accuracy of ±0.1℃. The pressure sensor acquires the pressure in each reaction kettle and the pressure difference of each section of the reaction tower, with a measurement range of 0-10 MPa and an accuracy of ±0.05 MPa. The catalyst state monitoring unit deploys an X-ray fluorescence spectrum sensor to measure the catalyst active component and crystal form parameters in real time, with a measurement range of 0-100% and an accuracy of ±0.1%, and a data sampling frequency of 1 time / minute. The raw material microanalysis unit integrates a near-infrared spectrometer to complete the detection of raw material purity and impurity content within 10 seconds, supporting rapid analysis of raw materials including but not limited to methanol and toluene. The carbon footprint monitoring unit installs a carbon emission sensor with a measurement range of 0-50 kg CO2 / h and an accuracy of ±2% to acquire carbon emission data related to reaction kettle tail gas and power consumption and calculate the total carbon emission C in real time, with the formula: ； wherein Mi is the raw material usage, Fi is the raw material carbon emission factor, E is the power consumption, G is the power carbon emission factor, Wj is the waste emission, and Hj is the treatment carbon emission factor. This index is integrated into the optimization algorithm decision-making process. The time synchronization unit is based on the NTP protocol and adopts a 5G+NTP dual synchronization mechanism to ensure the spatiotemporal consistency of micro and macro data, realize timestamp calibration of sensor data, ensure that all sensor data have a unified timestamp, and ensure the spatiotemporal consistency of micro and macro data with a time synchronization error of ≤0.5 ms.

3. The continuous hydrogenation interconnection equipment regulation system based on hybrid bat optimization algorithm according to claim 2, characterized in that, The data preprocessing module includes a weight analysis unit, an extreme data processing unit, a data deduplication unit and a standardization processing unit. The weight analysis unit is based on a random forest algorithm to analyze the correlation between parameters and product quality, assigns a hydrogen flow weight of 0.3, a temperature weight of 0.25 and a pressure weight of 0.2, and is used for optimizing instruction priority sorting. The extreme data processing unit adopts a 3σ principle to eliminate outliers and uses a linear interpolation method to supplement missing data. The data deduplication unit retains the earliest repeated data with a timestamp and deletes redundant data to ensure data uniqueness. The standardization processing unit unifies the data format to timestamp-equipment ID-parameter type-value-unit, supports multi-source heterogeneous device access including but not limited to PLC and DCS.

4. The continuous hydrogenation interconnection equipment regulation system based on hybrid bat optimization algorithm according to claim 3, characterized in that, The hybrid optimization calculation module comprises a processor and a storage unit, the processor adopts an Intel Core i7-12700K or an industrial processor with equivalent performance, calls an algorithm program, and performs operation on real-time data input by the data acquisition module to output a reactant addition rate and an equipment power control instruction; the storage unit stores a hybrid bat optimization algorithm program, and the hybrid bat optimization algorithm is a composite optimization algorithm integrating a whale optimization algorithm, a bat algorithm, a particle swarm optimization algorithm and a simulated annealing algorithm.

5. The continuous hydrogenation interconnection equipment regulation system based on hybrid bat optimization algorithm according to claim 4, characterized in that, The hybrid bat optimization algorithm specifically comprises the following steps: Step S1, initialization: a group of bats are randomly initialized in a solution space, the position Xi and the speed vi of each bat are defined, the initial frequency f0, the loudness A0 and the pulse emission rate r0 parameters are set, the inertia weight w, the learning factors c1 and c2 of the particle swarm optimization algorithm are initialized, and the initial temperature T of the simulated annealing algorithm is initialized; Step S2, fitness evaluation: the fitness value of each bat is calculated according to a target function; Step S3, position updating and hybrid optimization; S3a) bat algorithm exploration: the frequency, the speed and the position of the bat are updated according to the standard bat algorithm; fi=fmin+(fmax-fmin)*β (β is a random number in [0, 1]); vi(t+1)=vi(t)+(Xi(t)-Xglobal)*fi; Xi_new=Xi(t)+vi(t+1); S3b) PSO collaborative guidance: in order to improve the convergence speed, the group cooperation mechanism of PSO is introduced to correct the speed of the bat twice; vi(t+1)=w*vi(t+1)+c1*r1(Xpbest-Xi(t))+c2*r2*(Xglobal-Xi(t)) Wherein, Xpbest is the individual historical optimal position of the bat, and Xglobal is the current global optimal position; S3c) SA disturbance jump out of local optimum: in order to avoid premature convergence, the Metropolis criterion of the simulated annealing algorithm is used to accept a new solution; the fitness difference ΔE of the new position Xi_new and the old position Xi(t) is calculated; if ΔE<0, the new solution is accepted; if ΔE>=0, the new solution is accepted with a probability P_acc=exp(-ΔE / T) (wherein T is the current temperature); Step S4, local search: random walking is performed near the current optimal solution Xglobal to generate a local new solution; if the new solution is better than the current optimal solution and its random number is greater than the pulse emission rate ri, the local new solution is accepted; Step S5, update state: if the new solution is accepted, the loudness Ai and the pulse emission rate ri of the bat are updated; Ai(t+1)=α*Ai(t); ri(t+1)=ri(0)*[1-exp(-γ*t)]; Step S6, cooling: the current temperature T of the simulated annealing algorithm is reduced according to a cooling plan, T(t+1)=0.95*T(t). Step S7, termination judgment: repeat steps S2 to S6 until the maximum number of iterations is reached or the convergence accuracy requirement is met, and output the global optimal solution Xglobal, which is a set of optimal control parameter instructions.

6. The continuous hydrogenation interconnection equipment regulation system based on hybrid bat optimization algorithm according to claim 5, characterized in that, The optimization process of the mixed optimization calculation module is constructed as a reinforcement learning environment, where the algorithm acts as an agent, the action is the control parameter suggestion, and the reward function R is a multi-dimensional comprehensive index: R = α * product quality + β * (1 / energy consumption) + γ * (1 / equipment loss rate) + δ * (1 / carbon emission intensity), the agent learns the dynamic control strategy by interacting with the environment; The control execution module includes a reaction raw material flow controller, a power adjustment unit, a photovoltaic- energy storage linkage unit, and a voltage stabilization unit; the reaction raw material flow controller uses a piezoelectric ceramic proportional valve with a flow accuracy of ± 0.05%, which is used to receive the control instructions output by the mixed optimization calculation module, adjust the switching frequency of the reaction raw material, and achieve precise flow control with a switching frequency of 1-10 Hz; the power adjustment unit is based on the PID control algorithm and is used to control the rate of reaction raw material addition according to the instructions, with a response time ≤0.5s; the photovoltaic- energy storage linkage unit adjusts the peak load when the photovoltaic power supply ratio is ≥60%; the voltage stabilization unit uses a UPS power supply, which automatically switches to stabilized output when the input voltage fluctuation exceeds ± 5%, ensuring stable operation of the equipment.

7. The continuous hydrogenation interconnection equipment regulation system based on hybrid bat optimization algorithm according to claim 6, characterized in that, The monitoring and display module includes a data visualization unit, an alarm unit, a carbon footprint dashboard, and an equipment health warning interface; the data visualization unit uses an industrial touch screen to display the real-time, historical curves and predicted product quality of the reaction kettle in the form of charts, including temperature, pressure parameters, catalyst filling amount, and product quality, and supports data export; the alarm unit issues an audible and visual alarm when the continuous hydrogenation system operating parameters exceed the safety threshold, displays the alarm location and parameter deviation on the screen, and triggers the emergency handling process; the carbon footprint dashboard displays the carbon emission intensity per unit product; the equipment health warning interface displays the failure probability through a three-color indicator light; The database module includes a historical data storage unit and a data query unit; the historical data storage unit uses a MySQL database to store real-time data from the data acquisition module, operation results from the mixed optimization calculation module, and instruction data from the control execution module in the year-month-day-hour time dimension, with a storage period ≥5 years; the data query unit supports user queries of historical data by time range, parameter type, and equipment ID, with a query response time ≤1s, and generates statistical reports.

8. The continuous hydrogenation interconnection equipment regulation system based on hybrid bat optimization algorithm according to claim 7, characterized in that, The federal learning module is deployed in the database module, uses the SM4 encryption algorithm and the federal average algorithm, realizes encrypted sharing of multi-factory model parameters, improves the model generalization ability by more than 40%, and supports multi-factory collaborative optimization in the federal learning mode; based on the federal learning mode, the optimization algorithm model is updated and shared regularly; The digital twin module includes a process deduction unit, a predictive maintenance unit and an operation simulation unit, the process deduction unit pre-executes the control instruction in the virtual space, predicts the production state and product quality in the future period of time, and realizes forward-looking optimization; the predictive maintenance unit predicts the remaining service life of the key components by analyzing the digital mapping of the equipment operation data, and generates maintenance work orders in advance; the operation simulation unit provides a high-fidelity simulation environment for new employee training and operation pre-plan drilling.

9. A continuous hydrogenation intelligent control method using the system of any one of claims 1-8, characterized in that, The method comprises the following control steps: A1: Real-time acquisition of hydrogen production full-process data through a data acquisition module; A2: Preprocessing of the collected data, including weight analysis, extreme data processing and data deduplication processing; wherein the weight analysis is used to determine the correlation between the operation variables and the product quality and to give a hydrogen flow suggestion value; A3: Using hybrid bat optimization algorithm in the hybrid optimization calculation module to operate the preprocessed data, to generate the optimal equipment control instruction; A4: Executing the instruction through the control execution module to control the reactant addition rate and related equipment parameters; A5: Real-time visual monitoring and alarm of the whole process through the monitoring display module; A6: Storing all data through the database module and establishing the space-time correlation of production data to support full-link traceability; In step A2 or A3, a depth-first search algorithm is further used to traverse the associated data of the raw material addition amount, reaction degree and product quality for optimization strategy formulation; an RBF neural network is used to establish a soft measurement model of multiple parameters including temperature.

10. A product yield calculation method based on the system of any one of claims 1-8, characterized in that, The method comprises the following steps: 1) Threshold setting: the user sets the qualified threshold θ of the key quality indicators of the product by himself / herself; 2) Qualified rate calculation, the calculation formula is as follows: ； In the formula, si represents the qualified number of the i th identified, n represents the single data group identified, S is the total amount of data flow, and Y represents the product predicted qualified rate.

Images