Chemical production energy consumption prediction method and system based on artificial intelligence

Abstract

The present application belongs to the technical field of industrial process control and energy management, and particularly relates to a chemical production energy consumption prediction method and system based on artificial intelligence, which comprises the following steps: obtaining production operation data and process mechanism knowledge, forming a high-quality time series data set through data cleaning and automatic working condition division; based on the data and knowledge, constructing a lightweight dynamic model for each working condition and each device, and integrating a whole-process dynamic relationship network; receiving operation instructions, performing quantitative safety assessment and generating a standardized control sequence; driving the network with the sequence to perform hierarchical dynamic simulation, and combining real-time feedback correction to obtain the future prediction trajectory of key variables; and finally converting the trajectory into a system comprehensive energy consumption prediction result according to the device-level energy consumption model. The present application is used to solve the problem that the prior art is difficult to predict the energy consumption dynamic change process caused by operation adjustment with high precision, and realizes the transition from static result prediction to dynamic process simulation.

Description

Technical Field

[0001] This invention belongs to the field of industrial process control and energy management technology, and in particular relates to an artificial intelligence-based method and system for predicting energy consumption in chemical production. Background Technology

[0002] Chemical production processes are typical high-energy-consuming process industries, and accurate prediction of their energy consumption is crucial for achieving energy conservation, emission reduction, and optimized production operations. Currently, relevant prediction methods mainly rely on two technical approaches: one is machine learning models based on historical data (such as neural networks and regression analysis), and the other is rigorous mechanistic models based on first principles (such as process simulation software). Data-driven methods are relatively simple to model, but they struggle to effectively depict the inherent dynamic characteristics of chemical processes and the complex coupling relationships between variables. Their predictions are essentially static correlations, failing to accurately reflect the dynamic changes in energy consumption after operational adjustments, and lacking the ability to assess the safety of operational instructions. While mechanistic models can describe the essence of the process, they are complex to model, computationally expensive, and dependent on ideal parameters, making it difficult to adapt to real-time changing operating conditions and perform rapid online prediction and correction.

[0003] The common shortcomings of existing technologies lie in their failure to effectively integrate process mechanism knowledge with real-time operational data to construct a predictive framework that can both characterize the dynamic coupling relationships of the entire process and perform fast, safe, and interpretable online simulations. Specifically, this manifests in three ways: first, the predictive models are mostly "black boxes," unable to explain the causal paths of energy consumption changes, which is detrimental to optimization decisions; second, they cannot proactively assess the feasibility of operational instructions and generate safe control sequences upon receiving them; and third, they lack efficient hierarchical decoupling simulation mechanisms and real-time feedback correction, making it difficult to achieve high-precision prediction of dynamic process energy consumption trajectories in engineering practice. Therefore, the industry urgently needs a new energy consumption prediction method that can solve the above problems and truly meet the dynamic characteristics of chemical production and the needs of engineering practice. Summary of the Invention

[0004] This invention provides an artificial intelligence-based method and system for predicting energy consumption in chemical production, which solves the problems of existing prediction methods being unable to accurately characterize the dynamic coupling characteristics of chemical processes, thus failing to conduct forward-looking safety assessments of operating instructions and failing to predict the dynamic trajectory of energy consumption caused by operational adjustments with high fidelity.

[0005] In a first aspect, the present invention provides an artificial intelligence-based method for predicting energy consumption in chemical production, comprising:

[0006] The operation data and process mechanism knowledge of the target chemical production process are obtained, and the operation data is preprocessed and divided into operating conditions to obtain a high-quality time series data set. The data of each steady-state operating condition stage is labeled with the corresponding steady-state operating condition label.

[0007] Based on the high-quality time-series data set and process mechanism knowledge, dynamic influence relationship models between key process variables are constructed for different equipment unit types, and integrated into a full-process dynamic relationship network stored and managed in a hierarchical directed graph structure. The top layer is the process unit level, the bottom layer is the variable level, and the edges represent the influence relationship and are associated with the corresponding dynamic model.

[0008] The system receives external input operation instructions, parses the instructions into a target variable, adjustment range, and expected duration, queries the safety knowledge base in the dynamic relationship network of the entire process, obtains the safe operating range and maximum allowable rate of change of the target variable under the current operating conditions, determines whether the adjustment range exceeds the safe operating range, and whether the average rate of change calculated based on the expected duration exceeds the maximum allowable rate of change; if neither exceeds, a standardized control sequence containing a safe change path is generated; if any exceeds, a constraint out-of-bounds alarm is output.

[0009] Using the standardized control sequence as input, the dynamic process network of the entire process is driven to perform dynamic process simulation with a finite time window and hierarchical decoupling. The dynamic process simulation is based on the hierarchical directed graph structure, and is scheduled and calculated from the bottom variable level to the top process unit level according to the process dependency order between the equipment. The simulated values ​​of key state variables are compared with the real-time measured values. When the deviation exceeds the set threshold, the state observer is activated to perform feedback correction on the simulated state and obtain the predicted trajectory of key energy consumption related variables.

[0010] Based on the equipment-level energy consumption model, the predicted trajectory is converted into the instantaneous energy consumption of the equipment, and the instantaneous energy consumption of all equipment is accumulated to obtain the total instantaneous energy consumption sequence of the system. The total instantaneous energy consumption sequence of the system is then numerically integrated over a future set period to obtain the comprehensive energy consumption prediction result of the system.

[0011] Secondly, the present invention provides an artificial intelligence-based chemical production energy consumption prediction system, applied to the artificial intelligence-based chemical production energy consumption prediction method described in the first aspect; the artificial intelligence-based chemical production energy consumption prediction system includes:

[0012] The data preprocessing and operating condition identification module is used to acquire the operating data and process mechanism knowledge of the target chemical production process, and to preprocess and classify the operating data to obtain a high-quality time series data set, and to label the data of each steady-state operating condition stage with the corresponding steady-state operating condition label.

[0013] The dynamic model construction and management module is used to construct dynamic influence relationship models between key process variables for different equipment units based on the high-quality time series data set and process mechanism knowledge, and integrate them into a full-process dynamic relationship network stored and managed in a hierarchical directed graph structure. The top layer is the process unit level, the bottom layer is the variable level, and the edges represent the influence relationship and are associated with the corresponding dynamic model.

[0014] The instruction safety assessment and serialization module is used to receive externally input operation instructions, parse the operation instructions into target variables, adjustment ranges, and expected durations, query the safety knowledge base in the dynamic relationship network of the entire process, obtain the safe operating range and maximum permissible rate of change of the target variable under the current operating conditions, determine whether the adjustment range exceeds the safe operating range, and whether the average rate of change calculated based on the expected duration exceeds the maximum permissible rate of change; if neither exceeds, a standardized control sequence containing safe change paths is generated; if any exceeds, a constraint out-of-bounds alarm is output.

[0015] The online simulation and correction engine module is used to drive the dynamic process simulation of the whole process dynamic relationship network with a finite time window and hierarchical decoupling, taking the standardized control sequence as input. The dynamic process simulation is based on the hierarchical directed graph structure, from the bottom variable level to the top process unit level, and is scheduled and calculated according to the process dependency order between the equipment. The simulated values ​​of key state variables are compared with the real-time measured values. When the deviation exceeds the set threshold, the state observer is activated to perform feedback correction on the simulated state and obtain the predicted trajectory of key energy consumption related variables.

[0016] The energy consumption calculation and report generation module is used to convert the predicted trajectory into instantaneous energy consumption of the equipment according to the equipment-level energy consumption model, accumulate the instantaneous energy consumption of all equipment to obtain the total instantaneous energy consumption sequence of the system, and perform numerical integration on the total instantaneous energy consumption sequence of the system over a future set period to obtain the comprehensive energy consumption prediction result of the system.

[0017] Thirdly, the present invention also provides an electronic device, comprising: a memory for storing computer software programs; and a processor for reading and executing the computer software programs, thereby realizing the artificial intelligence-based chemical production energy consumption prediction method as described above.

[0018] Fourthly, the present invention also provides a non-transitory computer-readable storage medium storing a computer software program, which, when executed by a processor, implements the artificial intelligence-based chemical production energy consumption prediction method described above.

[0019] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the artificial intelligence-based chemical production energy consumption prediction method as described above.

[0020] This invention integrates process mechanism knowledge with real-time operation data to construct a full-process dynamic relationship network that can accurately characterize the dynamic coupling relationship between variables, thereby effectively overcoming the shortcomings of traditional data-driven models that cannot depict process dynamics and dynamic influence relationships.

[0021] Building upon this foundation, a pre-quantitative safety assessment of operational commands is introduced to generate standardized control sequences, addressing the lack of safety boundary warnings in traditional methods. Furthermore, this sequence drives hierarchical decoupled dynamic process simulation and introduces real-time feedback correction, enabling efficient and high-fidelity prediction of energy consumption dynamic changes over a finite future timeframe. This overcomes the bottleneck of traditional mechanistic models being computationally complex and difficult to apply online. Finally, energy consumption influencing factor analysis based on simulation results clearly reveals key influencing paths, providing direct and interpretable decision-making basis for energy-saving optimization, achieving a leap from static result prediction to dynamic process optimization guidance. Attached Figure Description

[0022] Figure 1 This is a schematic flowchart of the artificial intelligence-based chemical production energy consumption prediction method provided in an embodiment of the present invention;

[0023] Figure 2 This is a schematic diagram of the structure of the artificial intelligence-based chemical production energy consumption prediction system provided in an embodiment of the present invention;

[0024] Figure 3 An embodiment diagram of the electronic device provided in this invention;

[0025] Figure 4 An embodiment diagram of a computer-readable medium provided for an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0028] In the description of this invention, the term "such as" is used to mean "used as an example, illustration, or description." Any embodiment described as "such as" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. In the following description, for purposes of explanation...

[0029] Details are listed. It should be understood that those skilled in the art will recognize that the invention can be implemented without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but rather to be consistent with...

[0030] This is consistent with the broadest scope of the principles and features disclosed in this invention.

[0031] See Figure 1 , Figure 1 This is a flowchart illustrating the AI-based chemical production energy consumption prediction method provided by the present invention. In this embodiment, the executing entity of the AI-based chemical production energy consumption prediction method is an AI-based chemical production energy consumption prediction system. Therefore, the AI-based chemical production energy consumption prediction method includes:

[0032] Step 10: Obtain the operating data and process mechanism knowledge of the target chemical production process, preprocess the operating data and divide it into operating conditions to obtain a high-quality time series data set, and label the data of each steady-state operating condition stage with the corresponding steady-state operating condition label.

[0033] Furthermore, the operational data includes historical and real-time data from the distributed control system, while the process mechanism knowledge defines the process flow and the relationship between variables.

[0034] This step aims to transform the raw data into clean, structured data that can be used for modeling. The specific implementation process begins with data cleaning, including outlier removal, missing value imputation, and noise filtering to suppress noise.

[0035] Subsequently, by analyzing the changes in statistical characteristics within the sliding window, different steady-state operation stages in the production process are automatically identified, and operating condition labels are assigned to the data of each stage. This is illustrated in steps 101 to 103.

[0036] Step 20: Based on high-quality time series data sets and process mechanism knowledge, construct dynamic influence relationship models between key process variables for different equipment unit types, and integrate them into a dynamic relationship network for the entire process.

[0037] Furthermore, the key design idea for building this network is to integrate mechanisms and data, and to perform decoupled modeling for different working conditions and different equipment.

[0038] First, the input, state, and output variables of each equipment unit are determined based on the process mechanism. Then, according to the differences in the dynamic characteristics of the equipment, a transfer function model is used for reactors, heat exchangers, etc., while a simplified state-space model is used for distillation columns, compressors, etc.

[0039] Finally, using historical data for the corresponding operating conditions, the parameters of each model are determined through parameter identification, and all unit models are connected according to the process flow to form a network. This network is stored and managed using a hierarchical directed graph structure, where the top layer is the process unit level, the bottom layer is the variable level, and edges represent influence relationships and are associated with the corresponding dynamic models. Dynamic process simulation is based on this hierarchical directed graph structure, scheduling calculations from the bottom variable level to the top process unit level according to the process dependency order between equipment, providing a clear basis for simulation scheduling. See steps 201 to 203 for details.

[0040] Step 30: Receive externally input operation instructions, and combine the dynamic relationship network of the entire process with real-time process data to quantitatively evaluate the feasibility of the operation instructions and generate a standardized control sequence that includes safe change paths.

[0041] Furthermore, this step is a crucial hurdle to ensure that simulation predictions are conducted within safe boundaries.

[0042] The system first parses the instruction, extracting the target variable, adjustment range, and expected duration. Next, it queries the safety knowledge base within the dynamic relationship network to obtain the allowed operating range and maximum rate of change constraint for that variable.

[0043] Then, a dual quantitative judgment is made from the perspectives of safety and equipment protection: whether the adjustment range exceeds the limit, and whether the required average rate is too fast. Only instructions that pass the check will be converted into an S-shaped curve path that meets the requirements of gradual process change, serving as a standardized control sequence for safety. This is specifically illustrated in steps 301 to 304.

[0044] Step 40: Using the standardized control sequence as input, drive the dynamic relationship network of the entire process to perform dynamic process simulation with a finite time window and hierarchical decoupling, and introduce real-time data to provide feedback correction to the simulation state, thereby obtaining the predicted trajectory of key energy consumption-related variables.

[0045] Furthermore, this step is the core computational process for performing dynamic simulation. The simulation engine schedules computations based on the hierarchical directed graph structure of the network.

[0046] When solving the model, a weighted summation method is used to handle the combined effects of multiple variables. The simulation proceeds in discrete time steps, decomposing the safe path into step size settings, and solving each element model sequentially to simulate disturbance propagation.

[0047] To maintain prediction accuracy, a feedback correction mechanism based on real-time measurement is introduced. When the simulated value deviates from the measured value by more than a threshold, the internal state of the model is corrected using a state observer.

[0048] The simulation continues until the end of the future time window, outputting the complete predicted trajectory of key energy consumption-related variables. Based on this predicted trajectory and the dynamic relationship network of the entire process, following the energy consumption impact analysis steps, for different dynamic impact paths pointing to the total energy consumption node, the sensitivity coefficient of the unit change of variables on the total energy consumption is calculated for each path, identifying the key energy consumption dynamic impact paths. This is detailed in steps 401 to 404.

[0049] Step 50: Based on the equipment-level energy consumption model, the predicted trajectory is converted into the instantaneous energy consumption of the equipment, and the instantaneous energy consumption of all equipment is accumulated to obtain the total instantaneous energy consumption sequence of the system. The total instantaneous energy consumption sequence of the system is then numerically integrated over a future set period to obtain the comprehensive energy consumption prediction result of the system.

[0050] Furthermore, this step transforms the dynamic process variable predictions into intuitive energy consumption indicators. The system pre-configures energy consumption calculation sub-models based on the physical characteristics of each major energy-consuming device; for example, the power of a pump is related to its flow rate and head.

[0051] After obtaining the full-process variable prediction trajectory, the relevant variable values ​​are input into the energy consumption sub-model of the corresponding device to calculate the instantaneous energy consumption of each device at different future times.

[0052] At each moment, the instantaneous energy consumption of all devices is summed to obtain the total instantaneous energy consumption sequence of the system. Finally, this total instantaneous energy consumption sequence is numerically integrated over the entire prediction period to obtain the cumulative total energy consumption prediction value for that period. The specific process is as described in steps 501 to 504.

[0053] This invention integrates process mechanism knowledge with real-time operational data to construct a full-process dynamic relationship network that can accurately characterize the dynamic coupling relationship between variables, thereby effectively overcoming the shortcomings of traditional data-driven models that are difficult to depict process dynamics and dynamic influence relationships.

[0054] Based on this, a pre-quantitative safety assessment of the operation instructions is introduced to generate a standardized control sequence, which solves the hidden danger of the lack of safety boundary warning in traditional methods. Furthermore, this sequence is used to drive the simulation of the hierarchical decoupled dynamic process and introduce real-time feedback correction, which realizes efficient and high-fidelity prediction of the dynamic change trajectory of energy consumption in the future within a limited time period. This breaks through the bottleneck of the traditional mechanism model being computationally complex and difficult to apply online.

[0055] Ultimately, the energy consumption influencing factor analysis based on simulation results can clearly reveal key influencing paths, providing direct and interpretable decision-making basis for energy-saving optimization, and realizing a leap from static result prediction to dynamic process optimization guidance.

[0056] In one embodiment, steps 101-103 are described as follows:

[0057] Step 101: Perform outlier removal, noise filtering, and missing value imputation on the original running data.

[0058] This step is the first stage of data preprocessing, and its purpose is to ensure the integrity and reliability of the data.

[0059] In practice, outlier removal can be done using methods based on statistical principles, such as calculating the mean and standard deviation of the data, and identifying values ​​that deviate from the mean by more than three times the standard deviation as outliers and removing them.

[0060] Noise filtering aims to smooth high-frequency random fluctuations in data. It can be achieved by moving average filtering, which takes the average value of the data at the current time and several consecutive time points before it as the filtered value at the current time.

[0061] Missing value imputation is used to handle data gaps caused by signal interruption. A linear interpolation method can be used, which uses the two most recent valid data before and after the missing point, assuming that their change is linear, to calculate the estimated value of the missing point.

[0062] This process eliminates significant errors and irrelevant interference in the original data, transforming the original running data sequence into a continuous, smooth, high-quality data sequence, laying the foundation for subsequent accurate analysis.

[0063] Step 102: Calculate the statistical characteristics (including but not limited to mean, variance, and range) of key process variables within the sliding window of the data. Based on whether the fluctuation amplitude of the characteristics exceeds the preset threshold, automatically identify and classify different steady-state production conditions.

[0064] This step is the core of automatic condition segmentation, and its purpose is to divide continuous production operation data into steady-state stages with different statistical characteristics.

[0065] In practice, the system first sets a sliding window of a fixed time length. Then, the system slides the window along the time axis and calculates the statistical characteristics of one or more key process variables within the window at each window position, such as mean, variance, or range.

[0066] When production is in a stable state, these statistical characteristics will remain relatively stable in one sliding window after another, with fluctuations below a preset stability threshold.

[0067] Once the system detects that the statistical characteristics of a certain window have undergone a significant jump exceeding a threshold or show a continuous trend change relative to the preceding window, it determines that a change in operating conditions has occurred.

[0068] The system merges all adjacent windows with stable statistical characteristics and marks them as a steady-state operating condition stage. By traversing the entire time series, the system automatically identifies and delineates the boundaries of all steady-state operating condition stages.

[0069] Step 103: Label the data for each steady-state operating condition stage to form a high-quality time series data set.

[0070] This step involves data organization and management, with the aim of establishing a mapping relationship between data and operating conditions to form a structured dataset.

[0071] In practice, the system assigns a unique identifier as a condition label to each steady-state operating condition stage identified in step 102.

[0072] Then, the system associates and binds all high-quality time-series data that have been preprocessed in step 101 within the time period corresponding to the working condition stage with the working condition label, and stores this association in a database or a specific data structure.

[0073] Ultimately, all the data were organized into categories according to their respective operating condition labels, forming a high-quality time-series data set with a clear structure and both time series and operating condition dimensions.

[0074] This step is the cornerstone of building a reliable prediction model. By systematically cleaning the raw data (removing anomalies, filtering noise, and imputing missing data), the quality of the input data is fundamentally guaranteed, avoiding the problem of "garbage in, garbage out".

[0075] Furthermore, through sliding window statistical analysis and automatic operating condition identification, massive and mixed continuous operating data are intelligently divided into different operating condition stages with clear statistical steady-state characteristics, and the data are labeled with operating condition tags.

[0076] This not only significantly reduces the data complexity of subsequent modeling, but more importantly, it ensures that the constructed dynamic model can accurately correspond to specific production states, providing a solid, clean, and structured data foundation for subsequent high-fidelity dynamic simulation prediction of different working conditions, thereby improving the engineering practicality and accuracy of the entire prediction method from the source.

[0077] In one embodiment, steps 201-203 are described as follows:

[0078] Step 201: Based on the knowledge of process mechanism, determine the core input variables, state variables and output variables of different equipment units to form a variable subset.

[0079] This step is the first task in building the device-level model, aiming to clarify the model's boundaries and interfaces, and to provide an important prerequisite for the subsequent establishment of a dynamic mathematical model.

[0080] In practice, it is necessary to conduct in-depth analysis of process piping, instrumentation diagrams, and material-energy balance relationships. For a specific equipment unit, such as a centrifugal pump, its core input variables may include the inlet valve opening and motor frequency; its internal state variables may include the pump speed; and its core output variables may include the outlet flow rate and outlet pressure.

[0081] Through this analysis, a unique set of input variables, state variables, and output variables is defined for each key equipment unit in the process.

[0082] This step transforms the abstract process equipment into a mathematical modeling object described by a set of explicit variables and their causal relationships, thus defining a clear scope for the subsequent establishment of dynamic mathematical models.

[0083] Step 202: For reactors and heat exchangers, a transfer function model with time delay is used to describe their dynamics; for distillation columns and compressors, a low-order state-space model based on mechanism simplification is used to describe their dynamics.

[0084] This step involves selecting the appropriate model type based on the differences in the dynamic characteristics of the equipment.

[0085] In practical implementation, for devices whose dynamic characteristics are mainly manifested as heat or mass transfer inertia and have pure time delay, such as most reactors and heat exchangers, using a transfer function model is an efficient and sufficiently accurate choice.

[0086] For example, the dynamic effect of feed flow rate changes on reactor temperature can be described by a mathematical expression that includes gain, time constant, and pure time delay.

[0087] For equipment with many internal variables, strong coupling, and complex dynamics, such as distillation columns and compressors, a low-order state-space model derived from the simplified derivation of their core mass and energy balance equations is used to describe the coupled dynamics among multiple variables.

[0088] This differentiated modeling strategy effectively reduces the overall complexity of the entire process model while ensuring that the core dynamics are depicted.

[0089] Step 203: Using a high-quality time series dataset under the corresponding working conditions, determine the parameters of each model through parameter identification methods.

[0090] This step is crucial for calibrating the mechanistic model framework using operational data.

[0091] In practice, for the various models established in step 202, their mathematical forms are fixed, but they contain unknown parameters. The system extracts historical data segments under specific working conditions corresponding to the model to be identified from a high-quality time series dataset. These data segments contain the actual change sequences of the model's input and output variables.

[0092] Then, a parameter identification algorithm is used to find a set of model parameters such that when the model is driven by the same input, the overall error between the simulation output change sequence and the actual historical output data sequence is minimized.

[0093] For different types of dynamic models, such as transfer function models and low-order state-space models, parameter identification algorithms adapted to the corresponding model types, such as recursive least squares method, can be used for parameter identification.

[0094] Through this process, each equipment model is calibrated into a personalized model that reflects the actual dynamic behavior of the equipment under specific operating conditions, and is ultimately integrated into a dynamic relationship network covering the entire process.

[0095] This step is the core engine for realizing dynamic process simulation in this invention. By combining process mechanism (defining a subset of variables and selecting a model structure) with historical operating data (for parameter identification), differentiated lightweight dynamic models are constructed for different equipment types and integrated into a full-process dynamic relationship network.

[0096] This integrated modeling approach of "mechanistic framework + data calibration" overcomes the weaknesses of pure data models, such as lack of physical interpretability and dynamic characterization, and avoids the drawbacks of rigorous mechanistic models being too complex and difficult to apply online.

[0097] The resulting network is a computationally efficient, parameter-accurate, and structurally transparent prototype of a "digital twin." It accurately represents the dynamic influence and coupling relationships between variables within the process, providing an indispensable and computable mathematical model foundation for subsequent safety assessments and driving high-fidelity dynamic process simulations. It is the key to achieving the leap from static correlation prediction to dynamic process prediction.

[0098] In one embodiment, steps 301-304 are described as follows:

[0099] Step 301: Parse the operation instructions into target variables, adjustment range, and expected duration.

[0100] This step involves understanding the operational intent and converting it into quantifiable information that a computer can process.

[0101] In practice, the system receives operation instructions from the human-machine interface or the upper-level system. These instructions may be input in the form of natural language, structured forms, or setpoint commands.

[0102] The system's built-in parser performs semantic analysis and extraction on the instructions. For example, for the instruction "Raise the reactor temperature by ten degrees Celsius within two hours", the parser will identify the target variable as "reactor temperature", the adjustment range as "positive ten degrees Celsius", and the expected duration as "two hours".

[0103] The three quantitative elements obtained from the analysis form the basic input for all subsequent security assessments and sequence generation logic.

[0104] Step 302: Query the dynamic relationship network of the entire process to obtain the current safe operating range and maximum allowable rate of change of the target variable.

[0105] This step involves retrieving safety constraints from the process knowledge base.

[0106] In practice, the entire process is dynamically linked to a knowledge base that stores various safety and process limitations. Once the system obtains the target variable, it immediately queries this knowledge base.

[0107] Safe operating ranges are typically determined by process cards, operating procedures, or equipment design parameters. The maximum permissible rate of change may stem from equipment mechanical stress limits, catalyst thermal shock limits, or process stability requirements.

[0108] Based on the current production conditions, the system retrieves the upper and lower limits of the safe operating range and the maximum allowable rate of change limit of the target variable from the knowledge base under these conditions.

[0109] Step 303: Determine whether the adjustment range exceeds the safe operating range and whether the average rate of change calculated based on the expected duration exceeds the maximum permissible rate of change.

[0110] This step involves performing specific security logic checks.

[0111] In practice, the system performs two independent calculations and comparisons. The first judgment is to calculate the target value required by the instruction and determine whether the target value falls within the safe operating range obtained from the query.

[0112] The second judgment: Calculate the average rate of change required to execute the instruction (the absolute value of the adjustment range divided by the expected duration), and determine whether the average rate exceeds the maximum allowable rate of change obtained from the query.

[0113] Both of these checks must be passed simultaneously for the instruction to be considered initially safe.

[0114] Step 304: If none of them exceed the limit, an S-shaped curve that smoothly transitions from the current value to the target value and meets the requirements of gradual process change is generated as a safe change path; if any one of them exceeds the limit, a constraint overrun alarm is output.

[0115] This step involves executing different branches based on the safety assessment results.

[0116] In practice, if both checks pass, the system will generate a standardized control sequence. This sequence is a trajectory that smoothly and continuously changes from the current value of the variable to the target value of the instruction.

[0117] To achieve the requirement of "gradual change," an S-shaped curve is usually used, such as a smooth transition curve constructed based on a sine function, whose mathematical expression is: .

[0118] in Indicates time The variable setting value at time, It is the current value of the variable. It is the adjustment range (the difference between the target value and the current value). It is the expected duration. It is a time variable, with a value range from 0 to... .

[0119] The curve changes at near-zero rates at the beginning and end, with a larger but smoother rate in the middle phase, and no abrupt changes overall. If any judgment fails, the system outputs a constraint violation alarm, clearly indicating the violation type (adjustment amplitude violation or average rate of change violation) and the corresponding safety constraint requirements.

[0120] This step embeds a proactive engineering safety layer into the entire prediction process, serving as a crucial bridge connecting operational intent with digital simulation.

[0121] By parsing instructions, querying the safety knowledge base, and performing dual quantitative judgments on amplitude and rate, the system can proactively assess the feasibility of operational instructions, thus preventing unsafe instructions from misleading simulation predictions at the source.

[0122] For safety instructions, by generating a standardized control sequence of S-shaped curves that meet the requirements of gradual process changes, the abstract operational objectives are transformed into specific, smooth, and safe time-domain drive signals.

[0123] This process not only ensures that the prediction itself is carried out within safety constraints, greatly improving the engineering credibility and decision-making value of the prediction results, but more importantly, it imitates the cautious operating logic of excellent operators, giving the prediction system an engineering mindset of "safety first," which is a necessary prerequisite for the reliable application of this method in the field.

[0124] In one embodiment, steps 401-404 are described as follows:

[0125] Step 401: Decompose the standardized control sequence into input values ​​at discrete time steps.

[0126] This step is the process of converting a continuous-time safety control path into a discrete instruction sequence that can be used for simulation calculations.

[0127] In practice, the system first sets a fixed simulation calculation step size, forming equally spaced discrete time steps. Then, the system samples the continuous S-shaped curve path along the time axis, starting from the current moment, using the equally spaced step size as the sampling interval.

[0128] At each sampling time point, the variable value corresponding to the S-curve is read and used as the input setpoint at the start of that simulation step. This discrete sequence precisely represents the specific execution steps of the standardized control command in the time domain.

[0129] Step 402: At each time step, solve the dynamic model of each unit in the order of device unit dependencies, and update its state variables and output variables.

[0130] This step is the core loop of the dynamic simulation calculation.

[0131] In practice, the simulation engine, based on the hierarchical directed graph structure of the dynamic relationship network of the entire process, determines the calculation order from the bottom-level variables to the top-level process units, according to the process dependencies between equipment. Within a simulation step, the engine sequentially calls the dynamic model solver corresponding to each equipment unit for calculation in this order.

[0132] During calculation, the output of upstream equipment serves as the input of downstream equipment, and the state and output of each unit are solved and updated sequentially. If multiple upstream variables jointly affect a downstream variable, a weighted summation method is used to calculate the total impact; the weighting coefficients are pre-set based on the importance of inter-variable coupling defined in the process mechanism knowledge, or determined by analyzing historical data under the corresponding operating conditions.

[0133] In this way, the disturbances introduced by the operating instructions are propagated step by step and sequentially throughout the entire system.

[0134] Step 403: Compare the simulated values ​​of key state variables with the real-time measured values. When the deviation exceeds the set threshold, start the state observer to correct the simulated state.

[0135] This step introduces closed-loop feedback correction to the open-loop dynamic simulation.

[0136] In practice, the system continuously receives real-time measurements of key process variables from the field while the simulation is running. During the simulation, the system compares the model-simulated values ​​of the corresponding variables with the real-time measured values ​​at the same time and calculates the deviation between the two.

[0137] When the deviation exceeds the preset tolerance threshold, it indicates that the open-loop simulation has deviated significantly from the actual process. At this point, the system will activate a state observer designed in conjunction with the dynamic influence relationship model, taking real-time measurements as input, and dynamically re-estimate and correct the model's internal state variables based on the model.

[0138] The observer takes real-time measurements as input and, based on model dynamics, re-estimates and corrects the model's internal state variables. The corrected state replaces the original simulated state and serves as the initial condition for the next simulation step.

[0139] Step 404: Repeat the iteration until the end of the set future time window, and output the predicted trajectories of all variables.

[0140] This step controls the simulation loop and outputs the final result.

[0141] In practice, the system sets the desired future prediction time window length. The simulation engine starts from the current moment and cyclically executes each step of the simulation and correction calculation.

[0142] This loop continues until the simulation time reaches or exceeds the end of the set future time window. During the loop, the system records the values ​​of all relevant process and state variables at each simulation step.

[0143] When the simulation ends, the data recorded in chronological order constitutes a complete dataset of predicted trajectories from the current moment to some future moment.

[0144] This step is the core execution link for achieving "dynamic process prediction" in this invention. Driven by a safe and standardized control sequence, it efficiently drives the forward simulation of the entire process dynamic relationship network through discretization, hierarchical scheduling, and sequential solving, accurately simulating the dynamic response process triggered by operation commands and propagating throughout the entire process.

[0145] In particular, the introduced feedback correction mechanism (state observer) based on real-time measurement transforms the open-loop simulation into a closed-loop prediction-correction system, which can effectively compensate for model mismatch and unknown disturbances, ensuring that the long-term predicted trajectory always fits the actual process, and greatly enhancing the robustness and practicality of the prediction.

[0146] The final output of the key variable prediction trajectory fully depicts the entire transition process from the current state to the new steady state in the future, providing an accurate time-domain data foundation for the accurate calculation of transient energy consumption and effectively breaking through the static prediction mode that only gives the final steady-state result.

[0147] In one embodiment, steps 501-504 are described as follows:

[0148] Step 501: Configure instantaneous energy consumption calculation sub-models based on equipment characteristics for pumps, compressors, heating furnaces, and reactors respectively.

[0149] This step forms the basis for establishing the relationship between process variables and energy consumption indicators.

[0150] In practice, for each type of major energy-consuming equipment in the process, a static or quasi-static energy consumption calculation model is established based on its energy conversion principle and the performance curves provided by the manufacturer.

[0151] For example, for a centrifugal pump, its shaft power is related to flow rate, head, and efficiency; for a compressor, its power is related to intake flow rate, compression ratio, intake temperature, and efficiency; and for a heating furnace, its fuel consumption is related to heat load and thermal efficiency.

[0152] These sub-models constitute a dictionary that "translates" the variable trajectories generated by dynamic simulation into energy consumption data.

[0153] Step 502: Input the relevant variables in the predicted trajectory into the energy consumption calculation sub-model of the corresponding device to obtain the instantaneous energy consumption of each device at different times.

[0154] This step involves applying an energy consumption calculation sub-model to calculate the equipment's energy consumption time-by-time.

[0155] In practice, the system reads the entire process variable prediction trajectory dataset. For each major energy-consuming device in the process, the system extracts the predicted values ​​of the input variables required by the energy consumption calculation sub-model of that device at each future simulation time point from this dataset.

[0156] Then, these variable values ​​are substituted into the energy consumption calculation sub-model formula corresponding to the device to calculate the instantaneous energy consumption value of the device at that moment.

[0157] The system repeats this calculation for each simulation time point, thereby generating an instantaneous energy consumption prediction sequence for each device.

[0158] Step 503: Sum the instantaneous energy consumption of all devices to obtain the total instantaneous energy consumption sequence of the system.

[0159] This step involves system-level integration based on device-level energy consumption calculations.

[0160] In practice, at each identical simulation time point, the system algebraically sums the instantaneous energy consumption values ​​of all online operating devices.

[0161] The system performs this summation operation for each time point within the prediction time window. Ultimately, a new sequence is obtained that corresponds perfectly to the prediction time axis, which is the prediction sequence of the system's total instantaneous energy consumption.

[0162] Step 504: Perform numerical integration on the total instantaneous energy consumption sequence of the system over the prediction period to obtain the cumulative energy consumption prediction value.

[0163] This step is the final conversion from instantaneous power to accumulated energy.

[0164] In practice, the system acquires the total instantaneous energy consumption sequence. The cumulative energy consumption is the integral of this power sequence curve over the prediction period, which physically represents the total energy consumed during this period.

[0165] Because the data is discrete, the system uses numerical integration methods for calculation, such as the trapezoidal rule: .

[0166] in This represents the cumulative total energy consumption during the forecast period. Indicates the first Total instantaneous energy consumption of the system at each point in time. For the simulation time step, The summation symbol ∑ represents the total number of steps. arrive Add up all the items.

[0167] The results are expressed in units of energy, such as kilowatt-hours or gigajoules.

[0168] This step is the final value output stage, which transforms the high-dimensional variable trajectory generated by dynamic simulation into energy consumption indicators that decision-makers can directly understand and use.

[0169] By configuring a physical characteristic-based energy consumption calculation sub-model for each major energy-consuming device and substituting the predicted process variables into the calculation step by step, a precise mapping from "process dynamics" to "energy consumption" is achieved.

[0170] Its core value lies in providing prediction results in two dimensions: one is the total instantaneous energy consumption sequence of the system, which fully reveals the dynamic changes in energy consumption during the entire operation transition (such as peaks, valleys, and trends), providing unprecedented insights for real-time load management and optimization of operation rhythm;

[0171] Secondly, the cumulative total energy consumption forecast obtained through numerical integration provides a crucial quantitative basis for evaluating the total energy cost of different operating schemes, comparing energy efficiency, and making energy-saving decisions. This enables a complete and refined prediction of energy consumption in chemical production, from static results to dynamic processes.

[0172] Optionally, refer to Figure 2 , Figure 2 This is a schematic diagram of the structure of the AI-based chemical production energy consumption prediction system provided by the present invention. The AI-based chemical production energy consumption prediction system includes...

[0173] The data preprocessing and operating condition identification module 210 is used to acquire the operating data and process mechanism knowledge of the target chemical production process, preprocess the operating data and divide the operating conditions to obtain a high-quality time series data set, and label the data of each steady-state operating condition stage with the corresponding steady-state operating condition label.

[0174] The dynamic model building and management module 220 is used to build dynamic influence relationship models between key process variables for different equipment units based on high-quality time series data sets and process mechanism knowledge, and integrate them into a dynamic relationship network for the entire process.

[0175] The instruction safety assessment and serialization module 230 is used to receive externally input operation instructions and, in conjunction with the dynamic relationship network of the entire process and real-time process data, to quantitatively assess the feasibility of the operation instructions and generate a standardized control sequence containing safe change paths.

[0176] The online simulation and correction engine module 240 is used to drive the dynamic relationship network of the entire process to perform dynamic process simulation with a finite time window and hierarchical decoupling, using a standardized control sequence as input, and to introduce real-time data to provide feedback correction to the simulation state, thereby obtaining the predicted trajectory of key energy consumption-related variables.

[0177] The energy consumption calculation and report generation module 250 is used to convert the predicted trajectory into the instantaneous energy consumption of the equipment according to the equipment-level energy consumption model, and to accumulate the instantaneous energy consumption of all equipment to obtain the total instantaneous energy consumption sequence of the system. The module then performs numerical integration on the total instantaneous energy consumption sequence of the system over a future set period to obtain the comprehensive energy consumption prediction result of the system.

[0178] This invention integrates process mechanism knowledge with real-time operational data to construct a full-process dynamic relationship network that can accurately characterize the dynamic coupling relationship between variables, thereby effectively overcoming the shortcomings of traditional data-driven models that cannot depict process dynamics and dynamic influence relationships.

[0179] Building upon this foundation, a pre-quantitative safety assessment of operational commands is introduced to generate standardized control sequences, addressing the lack of safety boundary warnings in traditional methods. Furthermore, this sequence drives hierarchical decoupled dynamic process simulation and introduces real-time feedback correction, enabling efficient and high-fidelity prediction of energy consumption dynamic changes over a finite future timeframe. This overcomes the bottleneck of traditional mechanistic models being computationally complex and difficult to apply online. Finally, energy consumption influencing factor analysis based on simulation results clearly reveals key influencing paths, providing direct and interpretable decision-making basis for energy-saving optimization, achieving a leap from static result prediction to dynamic process optimization guidance.

[0180] Please see Figure 3 , Figure 3 An embodiment diagram of an electronic device provided in accordance with the present invention. For example... Figure 3 As shown, this embodiment of the invention provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, it performs the following steps:

[0181] The system acquires operational data and process mechanism knowledge of the target chemical production process, preprocesses and classifies the operational data to obtain a high-quality time series data set, and labels the data of each steady-state operating stage with the corresponding steady-state operating condition label.

[0182] Based on high-quality time series data sets and knowledge of process mechanisms, dynamic influence relationship models among key process variables are constructed for different types of equipment units, and integrated into a dynamic relationship network for the entire process.

[0183] It receives external input operation instructions and combines them with the dynamic relationship network of the entire process and real-time process data to quantitatively evaluate the feasibility of the operation instructions and generate a standardized control sequence that includes safe change paths.

[0184] Using standardized control sequences as input, the dynamic relationship network of the entire process is driven to perform dynamic process simulation with finite time window and hierarchical decoupling, and real-time data is introduced to provide feedback correction to the simulation state, thereby obtaining the predicted trajectory of key energy consumption-related variables.

[0185] Based on the equipment-level energy consumption model, the predicted trajectory is converted into the instantaneous energy consumption of the equipment, and the instantaneous energy consumption of all equipment is accumulated to obtain the total instantaneous energy consumption sequence of the system. The total instantaneous energy consumption sequence of the system is then numerically integrated over a future set period to obtain the comprehensive energy consumption prediction result of the system.

[0186] Please see Figure 4 , Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with an embodiment of the present invention is shown. Figure 4 As shown, this embodiment provides a computer-readable storage medium 400 on which a computer program 311 is stored. When the computer program 311 is executed by a processor, it performs the following steps:

[0187] The system acquires operational data and process mechanism knowledge of the target chemical production process, preprocesses and classifies the operational data to obtain a high-quality time series data set, and labels the data of each steady-state operating stage with the corresponding steady-state operating condition label.

[0188] Based on high-quality time series data sets and knowledge of process mechanisms, dynamic influence relationship models among key process variables are constructed for different types of equipment units, and integrated into a dynamic relationship network for the entire process.

[0189] It receives external input operation instructions and combines them with the dynamic relationship network of the entire process and real-time process data to quantitatively evaluate the feasibility of the operation instructions and generate a standardized control sequence that includes safe change paths.

[0190] Using standardized control sequences as input, the dynamic relationship network of the entire process is driven to perform dynamic process simulation with finite time window and hierarchical decoupling, and real-time data is introduced to provide feedback correction to the simulation state, thereby obtaining the predicted trajectory of key energy consumption-related variables.

[0191] Based on the equipment-level energy consumption model, the predicted trajectory is converted into the instantaneous energy consumption of the equipment, and the instantaneous energy consumption of all equipment is accumulated to obtain the total instantaneous energy consumption sequence of the system. The total instantaneous energy consumption sequence of the system is then numerically integrated over a future set period to obtain the comprehensive energy consumption prediction result of the system.

[0192] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0193] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0194] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An artificial intelligence-based method for predicting energy consumption in chemical production, characterized in that, Includes the following steps: The operation data and process mechanism knowledge of the target chemical production process are obtained, and the operation data is preprocessed and divided into operating conditions to obtain a high-quality time series data set. The data of each steady-state operating condition stage is labeled with the corresponding steady-state operating condition label. Based on the high-quality time-series data set and process mechanism knowledge, dynamic influence relationship models between key process variables are constructed for different equipment unit types, and integrated into a full-process dynamic relationship network stored and managed in a hierarchical directed graph structure. The top layer is the process unit level, the bottom layer is the variable level, and the edges represent the influence relationship and are associated with the corresponding dynamic model. The system receives external input operation instructions, parses the instructions into a target variable, adjustment range, and expected duration, queries the safety knowledge base in the dynamic relationship network of the entire process, obtains the safe operating range and maximum allowable rate of change of the target variable under the current operating conditions, determines whether the adjustment range exceeds the safe operating range, and whether the average rate of change calculated based on the expected duration exceeds the maximum allowable rate of change; if neither exceeds, a standardized control sequence containing a safe change path is generated; if any exceeds, a constraint out-of-bounds alarm is output. Using the standardized control sequence as input, the dynamic process network of the entire process is driven to perform dynamic process simulation with a finite time window and hierarchical decoupling. The dynamic process simulation is based on the hierarchical directed graph structure, and is scheduled and calculated from the bottom variable level to the top process unit level according to the process dependency order between the equipment. The simulated values ​​of key state variables are compared with the real-time measured values. When the deviation exceeds the set threshold, the state observer is activated to perform feedback correction on the simulated state and obtain the predicted trajectory of key energy consumption related variables. Based on the equipment-level energy consumption model, the predicted trajectory is converted into the instantaneous energy consumption of the equipment, and the instantaneous energy consumption of all equipment is accumulated to obtain the total instantaneous energy consumption sequence of the system. The total instantaneous energy consumption sequence of the system is then numerically integrated over a future set period to obtain the comprehensive energy consumption prediction result of the system.

2. The method for predicting energy consumption in chemical production based on artificial intelligence according to claim 1, characterized in that, The preprocessing and working condition division include: The raw operating data is subjected to outlier removal, noise filtering, and missing value imputation. The statistical characteristics of key process variables within the sliding window of the calculation data are used to automatically identify and classify different steady-state production stages based on whether the fluctuation amplitude of the characteristics exceeds a preset threshold. The data for each steady-state operating condition stage are labeled to form the high-quality time-series data set.

3. The method for predicting energy consumption in chemical production based on artificial intelligence according to claim 1, characterized in that, The construction of the dynamic influence relationship model among its key process variables includes: Based on the knowledge of process mechanism, the core input variables, state variables and output variables of different equipment units are determined to form a variable subset; For reactors and heat exchangers, a transfer function model with time delay is used to describe their dynamics; for distillation columns and compressors, a low-order state-space model based on mechanism simplification is used to describe their dynamics. Using the high-quality time-series data set under the corresponding working conditions, the parameters of each model are determined by a parameter identification method adapted to the corresponding model type.

4. The method for predicting energy consumption in chemical production based on artificial intelligence according to claim 1, characterized in that, The obtained system comprehensive energy consumption prediction results include: Configure instantaneous energy consumption calculation sub-models based on equipment characteristics for pumps, compressors, heaters, and reactors respectively; Input the relevant variables in the predicted trajectory into the energy consumption calculation sub-model of the corresponding device to obtain the instantaneous energy consumption of each device at different times; The instantaneous energy consumption of all devices is summed to obtain the total instantaneous energy consumption sequence of the system; The total instantaneous energy consumption sequence of the system is numerically integrated over the prediction period to obtain the cumulative energy consumption prediction value.

5. The method for predicting energy consumption in chemical production based on artificial intelligence according to claim 1, characterized in that, After obtaining the predicted trajectories of the key energy consumption-related variables, the method further includes an energy consumption influencing factor analysis step: Based on the predicted trajectory and the dynamic relationship network of the entire process, for different dynamic influence paths pointing to the total energy consumption node, the sensitivity coefficient of the unit change of variables on the total energy consumption is calculated. The dynamic impact paths are sorted according to the sensitivity coefficients, and the path with the highest sensitivity is identified as the key energy consumption dynamic impact path.

6. The method for predicting energy consumption in chemical production based on artificial intelligence according to claim 1, characterized in that, When solving the dynamic model of each unit in sequence, if multiple upstream variables jointly affect a downstream variable, the total impact is calculated by weighted summation. The weighting coefficients are pre-set based on the importance of the coupling between variables as defined in the process mechanism knowledge, or determined by analyzing historical data under the corresponding operating conditions.

7. An artificial intelligence-based energy consumption prediction system for chemical production, characterized in that, The system, applied to the artificial intelligence-based chemical production energy consumption prediction method as described in any one of claims 1 to 6, comprises: The data preprocessing and operating condition identification module is used to acquire the operating data and process mechanism knowledge of the target chemical production process, and to preprocess and classify the operating data to obtain a high-quality time series data set, and to label the data of each steady-state operating condition stage with the corresponding steady-state operating condition label. The dynamic model construction and management module is used to construct dynamic influence relationship models between key process variables for different equipment units based on the high-quality time series data set and process mechanism knowledge, and integrate them into a full-process dynamic relationship network stored and managed in a hierarchical directed graph structure. The top layer is the process unit level, the bottom layer is the variable level, and the edges represent the influence relationship and are associated with the corresponding dynamic model. The instruction safety assessment and serialization module is used to receive externally input operation instructions, parse the operation instructions into target variables, adjustment ranges, and expected durations, query the safety knowledge base in the dynamic relationship network of the entire process, obtain the safe operating range and maximum permissible rate of change of the target variable under the current operating conditions, determine whether the adjustment range exceeds the safe operating range, and whether the average rate of change calculated based on the expected duration exceeds the maximum permissible rate of change; if neither exceeds, a standardized control sequence containing safe change paths is generated; if any exceeds, a constraint out-of-bounds alarm is output. The online simulation and correction engine module is used to drive the dynamic process simulation of the whole process dynamic relationship network with a finite time window and hierarchical decoupling, taking the standardized control sequence as input. The dynamic process simulation is based on the hierarchical directed graph structure, from the bottom variable level to the top process unit level, and is scheduled and calculated according to the process dependency order between the equipment. The simulated values ​​of key state variables are compared with the real-time measured values. When the deviation exceeds the set threshold, the state observer is activated to perform feedback correction on the simulated state and obtain the predicted trajectory of key energy consumption related variables. The energy consumption calculation and report generation module is used to convert the predicted trajectory into instantaneous energy consumption of the equipment according to the equipment-level energy consumption model, accumulate the instantaneous energy consumption of all equipment to obtain the total instantaneous energy consumption sequence of the system, and perform numerical integration on the total instantaneous energy consumption sequence of the system over a future set period to obtain the comprehensive energy consumption prediction result of the system.

Images