A petrochemical plant energy consumption big data analysis optimization energy-saving intelligent management and control system

Abstract

The application discloses a petrochemical plant area energy consumption big data analysis optimization energy-saving intelligent management and control system, belongs to the field of petroleum chemical energy management, and is characterized in that: through accurate collection in different domains, four-layer normalized data packets are independently constructed for each energy consumption monitoring sub-domain, multi-source heterogeneous data structured fusion is realized, and based on single-domain dynamic benchmark calibration and cross-domain coupling quantitative double analysis mechanisms, fine description of energy consumption states and quantitative capture of cross-sub-domain influences are realized, the problems of static benchmark misalignment and difficult quantification of coupling relationships are effectively solved, and then energy efficiency prediction is performed by taking time delay and coupling strength as constraints, a mechanism-data grey box model is fused to iteratively generate optimal operation parameters, intelligent management and control of prediction-optimization-closed loop are realized, the closed loop is issued to a control system, energy efficiency benchmark self-adaptation, multi-energy flow collaborative optimization and real-time closed-loop regulation and control are realized, and the comprehensive energy consumption is significantly reduced, and the adaptability and intelligent level of the system to dynamic working conditions are improved.

Description

Technical Field

[0001] This invention relates to the field of petrochemical energy management, and more specifically, to a petrochemical plant energy consumption big data analysis and optimization intelligent control system for energy conservation. Background Technology

[0002] The petrochemical industry is a typical high-energy-consuming and high-carbon-emission process industry, with energy costs accounting for 30%-60% of total production costs. Currently, most petrochemical plants have deployed energy management systems (EMS), but these are mainly limited to basic energy consumption data collection, statistical reporting, and single-point energy-saving technologies. In actual operation, the following technical problems still exist:

[0003] I. Data Acquisition and Processing Level: The system mostly adopts a single, unstructured data acquisition method, only collecting macro-level energy consumption data such as electricity and steam, ignoring the differentiated characteristics between different devices and different energy flows, and lacking effective means of aligning and integrating multi-source heterogeneous data (such as process parameters, equipment status, and environmental data), resulting in shallow data utilization and serious information silos. At the same time, traditional EMS uses fixed energy efficiency benchmarks or single mechanism models, which cannot adapt to dynamic operating conditions such as changes in raw materials, load fluctuations, and equipment aging, resulting in distorted energy efficiency evaluations and optimization suggestions that deviate from reality.

[0004] II. Energy Flow Coupling Analysis: Petrochemical plants involve multiple energy flows such as electricity, steam, fuel gas, and water. There are complex coupling relationships and transmission delays between these energy flows and between different production units. There is a lack of quantitative identification methods for the coupling relationships of multiple energy flows, especially a lack of methods for estimating dynamic transmission delays. This makes it difficult to implement collaborative optimization across units and media, often resulting in local energy saving but overall suboptimal performance.

[0005] III. Anomaly Diagnosis and Closed-Loop Management: Current anomaly alarm mechanisms are mostly based on fixed thresholds. When energy consumption exceeds the standard, actual losses have already occurred, which is an ex-post analysis. There is a lack of data-driven predictive root cause localization methods. At the same time, there is a lack of closed-loop linkage between optimization decision-making and control execution, making it impossible to achieve an automated cycle of "perception-analysis-optimization-execution".

[0006] In summary, there is a need for an intelligent control system capable of multi-source heterogeneous data fusion, dynamic energy efficiency benchmark adaptation, quantitative identification of multi-energy flow coupling relationships, and collaborative closed-loop prediction and optimization to address the aforementioned shortcomings of existing technologies. Summary of the Invention

[0007] The purpose of this invention is to address practical shortcomings, and a petrochemical plant energy consumption big data analysis and optimization intelligent management and control system is provided.

[0008] The objective of this invention can be achieved through the following technical solution: a petrochemical plant energy consumption big data analysis and optimization intelligent control system, comprising:

[0009] The control platform is used to divide the petrochemical plant area into multiple energy consumption monitoring subdomains according to preset division rules, generate a subdomain list, and store the measurement point configuration table and energy flow transmission correlation matrix for the subdomain list.

[0010] The domain acquisition and processing module is used to vertically acquire differentiated raw energy flow parameter combinations that are strongly correlated with the energy efficiency assessment of each subdomain according to the dedicated measurement point configuration table for each subdomain, and independently construct four-layer normalized data packets for each subdomain.

[0011] The energy consumption big data analysis module receives four-layer normalized data packets from each subdomain and performs single-domain dynamic benchmark calibration and cross-domain coupled quantitative analysis, wherein:

[0012] The single-domain dynamic benchmark calibration analysis calculates multiple energy efficiency sub-indices based on the original energy flow parameters of the subdomain to form an energy efficiency feature vector. A dynamic energy efficiency benchmark is established by clustering and classifying the historical energy efficiency feature vector sample set.

[0013] Cross-domain coupling quantization analysis identifies source and charge subdomains based on the energy flow transmission correlation matrix, estimates dynamic transmission delay through cross-correlation analysis, mines the causal relationship to quantify the coupling strength, and generates cross-domain associated data packets after aligning them according to the delay.

[0014] The hybrid-driven optimization decision module receives four-layer normalized data packets, dynamic energy efficiency benchmarks, and cross-domain related data packets. It uses dynamic transmission delay and coupling strength as constraints, and solves for the optimal operating parameter settings within the feasible domain of the decision variables based on the hybrid-driven model and multi-objective optimization algorithm.

[0015] The closed-loop control module is used to receive the optimal operating parameter settings and send them to the distributed control system or programmable logic controller for execution.

[0016] Furthermore, the preset partitioning rules include process zoning rules, energy flow boundary rules, and equipment clustering rules. The energy consumption monitoring subdomain includes the process unit subdomain, boiler island subdomain, and circulating water field subdomain.

[0017] Furthermore, the domain acquisition and processing module configures an independent edge acquisition gateway and a measurement point configuration table for each energy consumption monitoring subdomain. The measurement point configuration table defines the measurement point identifier, the energy flow type, the acquisition frequency, the instrument accuracy, and the initial value of the normalized parameter.

[0018] Furthermore, the four-layer normalized data packet specifically refers to:

[0019] Metadata layer: at least includes energy consumption monitoring subdomain identifier, device identifier, data type encoding, sampling timestamp, and instrument accuracy level;

[0020] Raw data buffer layer: used to store the raw sampled values ​​within the most recent preset time period, and the time series of all raw energy flow parameters;

[0021] Normalized feature vector layer: The original energy flow parameters are standardized and mapped using dynamic normalization parameters specific to each subdomain. The dynamic normalization parameters are the mean and standard deviation of the subdomain under normal operating conditions, and are updated according to a preset period as the operating conditions drift.

[0022] Time alignment layer: Unifies data with different sampling frequencies into the same time grid through multi-scale time alignment technology, and adds flag bits to the interpolated or padded data.

[0023] Furthermore, the specific process for performing single-domain dynamic benchmark calibration analysis is as follows:

[0024] Extract multiple raw sample values ​​within a preset time window from the raw data buffer layer of this subdomain, calculate multiple energy efficiency sub-indicators according to the preset energy efficiency sub-indicator calculation formula of this subdomain, and form an energy efficiency feature vector;

[0025] Clustering is performed on the sample set composed of historical energy efficiency feature vectors to obtain multiple energy efficiency model clusters. The cluster with the best energy efficiency is selected as the benchmark cluster. The dynamic energy efficiency benchmark of the subdomain is established based on the statistical values ​​of the samples in the benchmark cluster. That is, the dynamic energy efficiency benchmark is the P85 quantile or the linear combination of the mean and standard deviation of the energy efficiency sub-indicators of all samples in the benchmark cluster.

[0026] Furthermore, clustering classification in single-domain dynamic benchmark calibration analysis specifically includes:

[0027] The mean or standard deviation of all energy efficiency feature vectors in the sample set are calculated according to the feature dimension, and Z-score standardization is performed. The number of clusters K is determined based on the elbow rule / profile coefficient. The K-means algorithm is used to divide the standardized sample set into K energy efficiency pattern clusters. The cluster center of each cluster is calculated. The cluster center is the mean vector of all energy efficiency feature vectors in the cluster. The values ​​of each energy efficiency sub-indicator in the cluster center are compared and sorted according to the preset priority rules. The cluster with the best overall energy efficiency is selected as the benchmark cluster.

[0028] Furthermore, in the cross-domain coupling quantitative analysis:

[0029] The definitions of source and charge domains are based on the direction of energy flow transmission. The source domain is the subdomain that outputs energy or materials, and the charge domain is the subdomain that receives and consumes energy or materials. The energy flow transmission correlation matrix is ​​a directed matrix, with rows corresponding to source domains and columns corresponding to charge domains. The matrix elements indicate whether energy flow transmission exists and the type of energy flow.

[0030] The original time series of key energy flow variables were extracted from the original data buffer layers of the source domain and the charge domain, respectively. The dynamic transmission delay τ was estimated by normalized cross-correlation analysis. The key energy flow variables are process parameters that are selected from the original energy flow parameters at specific measurement points and can characterize the energy flow transmission state.

[0031] The coupling strength between key energy flow variables in the source subdomain and key energy flow variables in the charge subdomain is quantified by a causal relationship mining algorithm. When the coupling strength meets the preset conditions, the four-layer normalized data packets of the source subdomain at time t-τ and the data packets of the charge subdomain at time t are aligned by time delay and then fused to generate cross-domain associated data packets. These data packets contain at least one of the following as influencing variables: dynamic transmission delay τ, coupling strength, and key energy flow variables, energy efficiency sub-indicators, or normalized feature vectors extracted from the four-layer normalized data packets of the source subdomain at time t-τ.

[0032] Furthermore, the hybrid-driven optimization decision-making module adopts a model predictive control framework, and the specific processing flow is as follows:

[0033] S1. Energy efficiency prediction: Extract the original energy flow parameters of each subdomain at the current and historical time from the original data buffer layer, extract the influence variables of the source subdomain at time t-τ from the cross-domain associated data packets, input the original energy flow parameters, influence variables and candidate operation parameters provided by the multi-objective optimization algorithm into the hybrid driving model, and output the predicted values ​​of energy efficiency sub-indicators at multiple future time points.

[0034] S2. Optimization Iteration: Compare the predicted values ​​of energy efficiency sub-indicators with the dynamic energy efficiency benchmark, calculate the deviation, and take the minimum total energy consumption, carbon emission constraints and production plan completion rate as optimization objectives. Introduce the deviation as a penalty term, which together with the coupling relationship strength constitutes a constraint to construct a multi-objective function.

[0035] The multi-objective optimization algorithm updates candidate operation parameters based on the multi-objective function value, and repeats S1 and S2 until the convergence condition is met or the preset number of iterations is reached, and outputs the optimal candidate operation parameters as the final operation parameter settings.

[0036] Compared with the prior art, the advantages of this invention are:

[0037] 1. This invention achieves precise data acquisition by domain segmentation, independently forming a four-layer normalized data packet for each energy consumption monitoring subdomain, comprising a metadata layer, a raw data buffer layer, a normalized feature vector layer, and a time alignment layer. It also performs single-domain dynamic benchmark calibration analysis based on energy efficiency feature vector clustering for a single energy consumption monitoring subdomain to establish a dynamic energy efficiency benchmark, and performs cross-domain coupling correlation analysis based on cross-correlation delay estimation and causality testing for multiple subdomains with energy flow transmission relationships to generate cross-domain correlation data packets.

[0038] This technology enables the structured fusion of multi-source heterogeneous data, adaptive updating of dynamic energy efficiency benchmarks, and quantitative identification of multi-energy flow coupling relationships. It addresses the problems in existing technologies, such as shallow data utilization, static energy efficiency models that cannot track changes in operating conditions, and difficulty in quantifying multi-energy flow coupling relationships leading to inaccurate optimization. This lays a structured data foundation for subsequent high-precision analysis and optimization.

[0039] 2. Based on the dynamic energy efficiency benchmark obtained from the single-domain dynamic benchmark calibration analysis, the dynamic transmission delay and coupling strength obtained from the cross-domain coupling quantization analysis, and the original data buffer layer in the four-layer normalized data packet, the present invention adopts a hybrid driving model of fusion mechanism and data-driven approach to predict energy efficiency. With the optimization objectives of minimizing total energy consumption, carbon emission constraints, and production plan completion rate, the optimal operating parameter setting value is iteratively solved through a multi-objective optimization algorithm within the feasible domain of decision variables.

[0040] This technology enables closed-loop linkage between prediction and optimization, dynamic adaptive adjustment of operating parameters, and multi-objective collaborative optimization. It addresses the problems of disconnect between optimization decision-making and process control, lack of real-time closed-loop capability, and inability to balance multiple constraints such as energy consumption, carbon emissions, and production planning in existing technologies. This significantly reduces overall energy consumption and carbon emissions, and improves the system's intelligence and operability. Attached Figure Description

[0041] Figure 1 This is a diagram showing the module composition and internal data flow state of the present invention;

[0042] Figure 2 This diagram illustrates the processing mechanism of the energy consumption big data analysis module of the present invention, which performs single-domain dynamic benchmark calibration and cross-domain coupled quantitative analysis. Detailed Implementation

[0043] 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. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0044] Example 1: This invention discloses a petrochemical plant energy consumption big data analysis and optimization intelligent management and control system. Please refer to [link / reference]. Figures 1-2It includes a management and control platform, a domain-based data acquisition and processing module, an energy consumption big data analysis module, a hybrid-driven optimization decision-making module, and a closed-loop management and control module. The management and control platform is connected to the domain-based data acquisition and processing module and the energy consumption big data analysis module. The domain-based data acquisition and processing module is connected to the energy consumption big data analysis module. The energy consumption big data analysis module is connected to the hybrid-driven optimization decision-making module. The hybrid-driven optimization decision-making module is connected to the closed-loop management and control module.

[0045] The control platform is used to divide the petrochemical plant area into multiple energy consumption monitoring subdomains according to preset division rules, generate a subdomain list, and store the measurement point configuration table and energy flow transmission correlation matrix for the subdomain list. The control platform reads the P&ID map of the plant area and the equipment list in the DCS, and automatically generates the energy consumption monitoring subdomain list in combination with the preset division rules. The preset division rules include process zoning rules, energy flow boundary rules, and equipment clustering rules. The energy consumption monitoring subdomains include process unit subdomains, boiler island subdomains, and circulating water field subdomains. The energy flow transmission correlation matrix represents the energy flow transmission relationship of each subdomain.

[0046] The domain acquisition and processing module is used to collect raw energy flow parameters related to the energy efficiency assessment of each energy consumption monitoring subdomain, and independently generate data packets for each energy consumption monitoring subdomain, including a metadata layer, a raw data buffer layer, a normalized feature vector layer, and a time alignment layer.

[0047] The domain acquisition and processing module adopts a domain acquisition architecture, configuring an independent edge acquisition gateway and a measurement point configuration table for each energy consumption monitoring subdomain. The measurement point configuration table defines the measurement point identifier, the energy flow type, the acquisition frequency, the instrument accuracy, and the initial value of the normalized parameter.

[0048] The raw energy flow parameters of the process unit subdomain include fuel gas volumetric flow rate, calorific value, pressure, and the active power and current of the drive motor.

[0049] The raw energy flow parameters for the boiler island sub-domain include the calorific value and flow rate of fuel gas, the flow rate and temperature of boiler feedwater, and the pressure, temperature, and flow rate of main steam.

[0050] The original energy flow parameters for the circulating water field subdomain include the active power of the circulating water pump, the inlet water temperature, the return water temperature, and the circulating water flow rate.

[0051] Specifically, the four-layer normalized data packets are as follows:

[0052] Metadata layer: at least includes energy consumption monitoring subdomain identifier, device identifier, data type encoding, sampling timestamp, and instrument accuracy level;

[0053] Raw data buffer layer: used to store the raw sampled values ​​within the most recent preset time period, and the time series of all raw energy flow parameters;

[0054] Normalized feature vector layer: The original sampled values ​​in the original energy flow parameters are standardized and mapped using dynamic normalization parameters specific to each subdomain. The dynamic normalization parameters are the mean and standard deviation of the subdomain under normal operating conditions, and are updated according to a preset period as the operating conditions drift.

[0055] Time alignment layer: Unifies data with different sampling frequencies into the same time grid through multi-scale time alignment technology, and adds flag bits to the interpolated or padded data.

[0056] The energy consumption big data analysis module receives data packets from various energy consumption monitoring subdomains and performs single-domain dynamic benchmark calibration and cross-domain coupled quantitative analysis, including:

[0057] The single-domain dynamic benchmark calibration analysis calculates multiple energy efficiency sub-indices based on the original energy flow parameters within the subdomain to form an energy efficiency feature vector. A dynamic energy efficiency benchmark is established by clustering and classifying the historical energy efficiency feature vector sample set.

[0058] Cross-domain coupling quantization analysis identifies source and charge subdomains based on the energy flow transmission correlation matrix, estimates dynamic transmission delay through cross-correlation analysis, mines the causal relationship to quantify the coupling strength, and generates cross-domain associated data packets after aligning them according to the delay.

[0059] Single-domain optimization is used to locate inefficient equipment or operations and perform local optimization; cross-domain optimization is used to discover global and dynamic sources of energy waste (such as regulation lag and unreasonable energy cascade utilization) and achieve more advanced multi-energy flow collaborative optimization.

[0060] Specifically:

[0061] The specific process for performing single-domain dynamic benchmark calibration analysis is as follows:

[0062] Extract multiple raw sample values ​​within a preset time window from the raw data buffer layer of this subdomain, calculate multiple energy efficiency sub-indicators according to the preset energy efficiency sub-indicator calculation formula of this subdomain, and form an energy efficiency feature vector;

[0063] Clustering is performed on the sample set composed of historical energy efficiency feature vectors. Each sample is a three-dimensional vector. For example, the three-dimensional vector of the process unit subdomain is [thermal efficiency, power consumption, steam consumption]. Multiple energy efficiency model clusters are obtained. The cluster with the best energy efficiency is selected as the benchmark cluster. The dynamic energy efficiency benchmark of the subdomain is established based on the statistical values ​​of the samples in the benchmark cluster. That is, the dynamic energy efficiency benchmark is the P85 quantile or the linear combination of the mean and standard deviation of the energy efficiency sub-indicators of all samples in the benchmark cluster.

[0064] Different energy efficiency sub-indicators and calculation formulas are predefined for different types of energy consumption monitoring sub-domains, including: for the process unit sub-domain, energy efficiency sub-indicators include furnace thermal efficiency, unit product power consumption, and unit product steam consumption; for the boiler island sub-domain, energy efficiency sub-indicators include boiler thermal efficiency, flue gas temperature, and excess air coefficient; for the circulating water field sub-domain, energy efficiency sub-indicators include cooling efficiency and unit power consumption.

[0065] Cluster typing in single-domain dynamic benchmark calibration analysis specifically includes the following sub-steps:

[0066] The mean or standard deviation of all energy efficiency feature vectors in the sample set are calculated according to the feature dimension, and Z-score standardization is performed to make each feature dimension have zero mean and unit variance, thus eliminating the influence of dimensions.

[0067] The number of clusters K is determined based on the elbow rule / profile coefficient. The K-means algorithm is used to divide the standardized sample set into K energy efficiency pattern clusters. The cluster center of each cluster is calculated. The cluster center of each cluster is the mean vector of all energy efficiency feature vectors in the cluster. The values ​​of each energy efficiency sub-indicator in the cluster center are compared and sorted according to the preset priority rules (thermal efficiency first or unit product energy consumption first). The cluster with the best comprehensive energy efficiency is selected as the benchmark cluster.

[0068] The dynamic energy efficiency benchmark is updated in two ways: periodic updates and event-triggered updates. Periodic updates re-execute the single-domain dynamic benchmark calibration analysis and calculate the benchmark every 24 hours using the historical data of the most recent 7 days. Event-triggered updates are triggered immediately when a raw material switch, equipment overhaul, or load change exceeds a preset threshold, and a recalculation is performed.

[0069] In cross-domain coupled quantitative analysis:

[0070] The definitions of source subdomain and charge subdomain are based on the direction of energy flow transmission. The source subdomain is the subdomain that outputs energy or material, and the charge subdomain is the subdomain that receives and consumes energy or material.

[0071] The energy flow transmission correlation matrix is ​​a directed matrix, with rows corresponding to source subdomains and columns corresponding to charge subdomains. Matrix elements indicate whether energy flow transmission exists and the type of energy flow. This matrix is ​​pre-configured and stored in the monitoring platform based on the plant process flow diagram and energy balance diagram.

[0072] The original time series of key energy flow variables were extracted from the original data buffer layers of the source and charge domains, respectively. The dynamic transmission delay τ was estimated by normalized cross-correlation analysis. The key energy flow variables are process parameters selected from the original energy flow parameters at specific measurement points that can characterize the energy flow transmission state. For steam transmission, the key energy flow variables include the outlet steam pressure of the source domain and the inlet steam pressure of the charge domain; for circulating water transmission, the key energy flow variables include the supply water temperature of the source domain and the return water temperature of the charge domain.

[0073] The coupling strength between key variables of source subdomain energy flow and key variables of charge subdomain energy flow is quantified by causal relationship mining algorithm;

[0074] When the coupling strength meets the preset conditions, the four-layer normalized data packets of the source subdomain at time t-τ and the data packets of the charge subdomain at time t are aligned by time delay and then fused to generate cross-domain associated data packets. The cross-domain associated data packets contain at least one of the following as influencing variables: dynamic transmission delay τ, coupling strength, and energy flow key variables, energy efficiency sub-indicators, or normalized feature vectors extracted from the four-layer normalized data packets of the source subdomain at time t-τ.

[0075] It should be added here that the dynamic transmission delay is obtained through cross-correlation analysis: the cross-correlation function of the original time series of the key variables of the source subdomain energy flow and the original time series of the key variables of the charge subdomain energy flow is calculated, the calculation method of Pearson correlation coefficient is used, and the lag time corresponding to the maximum value of the cross-correlation function is taken as the dynamic transmission delay τ.

[0076] The coupling relationship quantification index adopts the Granger causality test algorithm, and outputs the p-value to characterize the coupling strength. The coupling strength is to test whether the source variable makes a significant contribution to the prediction of the load variable. The core idea of ​​the Granger causality test is: if the historical information of variable X helps to predict the future value of variable Y (provided that the historical information of Y itself has been used), then X is called a Granger cause of Y.

[0077] The method for generating cross-domain association data packets is as follows: when the Granger causality test is significant (p<0.05), it is considered that the source subdomain has a significant coupling effect on the charge subdomain. The four-layer normalized data packets of the source subdomain at time t-τ and the four-layer normalized data packets of the charge subdomain at time t are concatenated by feature dimensions to form cross-domain association data packets. Otherwise, it is skipped, indicating that there is no significant association in the current time period. It should be noted that when concatenating the two by feature dimensions: for example, the feature vector dimension of the source subdomain is D_s and that of the charge subdomain is D_h, the concatenation results in a cross-domain association feature vector with dimension D_s+D_h.

[0078] Example 2: Hybrid-driven optimization decision module, used to receive the original data buffer layer, dynamic energy efficiency benchmark and cross-domain related data packets in the four-layer normalized data packets, with dynamic transmission delay and coupling strength as constraints, adopts hybrid-driven model and multi-objective optimization algorithm, with the optimization objectives of minimizing total energy consumption, carbon emission constraints and production plan completion rate, and solves the optimal operating parameter setting value within the feasible domain of decision variables;

[0079] The hybrid-driven optimization decision-making module adopts a model predictive control framework, and the specific processing flow is as follows:

[0080] S1, Energy Efficiency Prediction

[0081] The original energy flow parameters of each subdomain at the current and historical time are extracted from the original data buffer layer. The influence variables of the source subdomain at time t-τ are extracted from the cross-domain associated data packets. The original energy flow parameters, influence variables and candidate operation parameters provided by the multi-objective optimization algorithm are input into the hybrid driving model, and the predicted values ​​of energy efficiency sub-indicators at multiple future time points are output.

[0082] The hybrid-driven model is a gray-box modeling method that integrates first-principles (mechanistic modeling) and data-driven (artificial intelligence / machine learning) approaches. It is used to predict future energy efficiency sub-indicators of key equipment (such as heaters, boilers, and compressors) in petrochemical plants. It is retrained based on events, such as when major changes in operating conditions are detected, such as raw material switching, equipment overhaul, or load exceeding a preset threshold, the data-driven model is retrained using historical data from the past 7 days (including actual energy efficiency sub-indicators, operating parameters, and operating condition labels). Through this hybrid-driven model, the system can more accurately predict future energy efficiency under different combinations of operating parameters, providing a reliable evaluation basis for multi-objective optimization algorithms.

[0083] It should be noted that during the first model prediction, the candidate operating parameters use the current actual operating parameters as initial values, and in subsequent model prediction processes, updated candidate operating parameters are used.

[0084] S2, Optimization Iteration

[0085] The predicted values ​​of energy efficiency sub-indicators are compared with the dynamic energy efficiency benchmark to calculate the deviation. The optimization objectives are to minimize total energy consumption, constrain carbon emissions, and achieve the production plan completion rate. The deviation is introduced as a penalty term, which together with the coupling strength (p<0.05) constitutes a constraint to construct a multi-objective function.

[0086] The multi-objective optimization algorithm updates the candidate operation parameters based on the multi-objective function value and repeats S1 and S2 until the convergence condition is met or the preset number of iterations is reached.

[0087] The optimal candidate operating parameters are output as the final operating parameter settings, including the load distribution of each energy consumption monitoring subdomain, steam pressure setpoint, fuel gas ratio, and circulating water valve opening.

[0088] For example, taking steam transfer between the boiler island (source domain) and the atmospheric and vacuum distillation unit (charge domain) as an example, the cross-domain correlation data package generated by cross-domain coupling quantitative analysis contains the following influencing variables:

[0089] The outlet steam pressure (2.5 MPa), temperature (400℃), and flow rate (80 t / h) at time t-τ were extracted from the raw data buffer layer of the boiler island.

[0090] The thermal efficiency (0.92) and its normalized value (1.2) at this moment were extracted from the normalized feature vector layer of the boiler island.

[0091] Operating condition tags (high load) extracted from the metadata layer.

[0092] The closed-loop control module is used to receive the optimal operating parameter settings and send them to the distributed control system or programmable logic controller for execution.

[0093] It should be added that the article involves multiple types of threshold comparisons. Thresholds, preset values, preset ranges, etc. are set for result comparison and analysis in order to determine good or bad. The magnitude of these values ​​is set, entered, and stored based on a combination of large-scale model analysis of sample data and human experience. They can also be appropriately adjusted based on seasonal or common-sense influencing conditions.

[0094] In summary, the unified representation of multi-source heterogeneous energy flow data is achieved through domain-specific differentiated acquisition and four-layer structured data packets. Furthermore, a dual analysis mechanism of single-domain dynamic clustering benchmark calibration and cross-domain delay causality quantification is employed. This dual analysis mechanism addresses the technical challenges in existing technologies, such as the inability to dynamically track energy efficiency benchmarks and the difficulty in quantifying the transmission delay and causal coupling relationship of multi-energy flow, providing accurate constraints for collaborative optimization.

[0095] Furthermore, with time delay and coupling strength as constraints, a mechanism-data gray box hybrid driving model is adopted to perform high-precision energy efficiency prediction, and the optimal operating parameters are iteratively solved to form intelligent control of prediction-optimization-closed loop;

[0096] Compared with traditional technologies, this invention adapts to changes in operating conditions, quantifies the time-delay coupling of multiple energy flows, and achieves real-time closed-loop optimization, significantly reducing overall energy consumption and improving the intelligence level and energy-saving and carbon-reducing effect of the petrochemical plant's energy system.

[0097] The above description is merely a preferred embodiment of the present invention; however, the scope of protection of the present invention is not limited thereto; any equivalent substitutions or modifications made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and its improved concept, should be covered within the scope of protection of the present invention.

Claims

1. A petrochemical plant energy consumption big data analysis and optimization intelligent control system, characterized in that: include: The control platform is used to divide the petrochemical plant area into multiple energy consumption monitoring subdomains according to preset division rules, generate a subdomain list, and store the measurement point configuration table and energy flow transmission correlation matrix for the subdomain list. The domain acquisition and processing module is used to vertically acquire differentiated raw energy flow parameter combinations that are strongly correlated with the energy efficiency assessment of each subdomain according to the dedicated measurement point configuration table for each subdomain, and independently construct four-layer normalized data packets for each subdomain. The energy consumption big data analysis module receives four-layer normalized data packets from each subdomain and performs single-domain dynamic benchmark calibration and cross-domain coupled quantitative analysis, wherein: The single-domain dynamic benchmark calibration analysis calculates multiple energy efficiency sub-indices based on the original energy flow parameters of the subdomain to form an energy efficiency feature vector. A dynamic energy efficiency benchmark is established by clustering and classifying the historical energy efficiency feature vector sample set. Cross-domain coupling quantization analysis identifies source and charge subdomains based on the energy flow transmission correlation matrix, estimates dynamic transmission delay through cross-correlation analysis, mines the causal relationship to quantify the coupling strength, and generates cross-domain associated data packets after aligning them according to the delay. The hybrid-driven optimization decision module receives four-layer normalized data packets, dynamic energy efficiency benchmarks, and cross-domain related data packets. It uses dynamic transmission delay and coupling strength as constraints, and solves for the optimal operating parameter settings within the feasible domain of the decision variables based on the hybrid-driven model and multi-objective optimization algorithm. The closed-loop control module is used to receive the optimal operating parameter settings and send them to the distributed control system or programmable logic controller for execution.

2. The intelligent energy-saving control system for petrochemical plant area based on big data analysis and optimization according to claim 1, characterized in that: The preset partitioning rules include process zoning rules, energy flow boundary rules, and equipment clustering rules. The energy consumption monitoring subdomain includes the process unit subdomain, boiler island subdomain, and circulating water field subdomain.

3. The intelligent energy-saving control system for petrochemical plant area based on big data analysis and optimization according to claim 2, characterized in that: The domain acquisition and processing module configures an independent edge acquisition gateway and a measurement point configuration table for each energy consumption monitoring subdomain. The measurement point configuration table defines the measurement point identifier, the energy flow type, the acquisition frequency, the instrument accuracy, and the initial value of the normalized parameter.

4. The intelligent control system for energy consumption analysis and optimization in petrochemical plants according to claim 3, characterized in that: The four-layer normalized data packet is specifically as follows: Metadata layer: at least includes energy consumption monitoring subdomain identifier, device identifier, data type encoding, sampling timestamp, and meter accuracy level; Raw data buffer layer: used to store the raw sampled values ​​within the most recent preset time period, and the time series of all raw energy flow parameters; Normalized feature vector layer: The original energy flow parameters are standardized and mapped using dynamic normalization parameters specific to each subdomain. The dynamic normalization parameters are the mean and standard deviation of the subdomain under normal operating conditions, and are updated according to a preset period as the operating conditions drift. Time alignment layer: Unifies data with different sampling frequencies into the same time grid through multi-scale time alignment technology, and adds flag bits to the interpolated or padded data.

5. The intelligent control system for energy consumption analysis and optimization in petrochemical plants according to claim 4, characterized in that: The specific process for performing single-domain dynamic benchmark calibration analysis is as follows: Extract multiple raw sample values ​​within a preset time window from the raw data buffer layer of this subdomain, calculate multiple energy efficiency sub-indicators according to the preset energy efficiency sub-indicator calculation formula of this subdomain, and form an energy efficiency feature vector; Clustering is performed on the sample set composed of historical energy efficiency feature vectors to obtain multiple energy efficiency model clusters. The cluster with the best energy efficiency is selected as the benchmark cluster. The dynamic energy efficiency benchmark of the subdomain is established based on the statistical values ​​of the samples in the benchmark cluster. That is, the dynamic energy efficiency benchmark is the P85 quantile or the linear combination of the mean and standard deviation of the energy efficiency sub-indicators of all samples in the benchmark cluster.

6. The intelligent energy-saving control system for petrochemical plant area based on big data analysis and optimization according to claim 5, characterized in that: Cluster typing in single-domain dynamic benchmark calibration analysis specifically includes: The mean or standard deviation of all energy efficiency feature vectors in the sample set are calculated according to the feature dimension, and Z-score standardization is performed. The number of clusters K is determined based on the elbow rule / profile coefficient. The K-means algorithm is used to divide the standardized sample set into K energy efficiency pattern clusters. The cluster center of each cluster is calculated. The cluster center is the mean vector of all energy efficiency feature vectors in the cluster. The values ​​of each energy efficiency sub-indicator in the cluster center are compared and sorted according to the preset priority rules. The cluster with the best overall energy efficiency is selected as the benchmark cluster.

7. The intelligent energy-saving control system for petrochemical plant area based on big data analysis and optimization according to claim 6, characterized in that: In the cross-domain coupling quantitative analysis: The definitions of source and charge domains are based on the direction of energy flow transmission. The source domain is the subdomain that outputs energy or materials, and the charge domain is the subdomain that receives and consumes energy or materials. The energy flow transmission correlation matrix is ​​a directed matrix, with rows corresponding to source domains and columns corresponding to charge domains. The matrix elements indicate whether energy flow transmission exists and the type of energy flow. The original time series of key energy flow variables were extracted from the original data buffer layers of the source domain and the charge domain, respectively. The dynamic transmission delay τ was estimated by normalized cross-correlation analysis. The key energy flow variables are process parameters that are selected from the original energy flow parameters at specific measurement points and can characterize the energy flow transmission state. The coupling strength between key energy flow variables in the source subdomain and key energy flow variables in the charge subdomain is quantified by a causal relationship mining algorithm. When the coupling strength meets the preset conditions, the four-layer normalized data packets of the source subdomain at time t-τ and the data packets of the charge subdomain at time t are aligned by time delay and then fused to generate cross-domain associated data packets. These data packets contain at least one of the following as influencing variables: dynamic transmission delay τ, coupling strength, and key energy flow variables, energy efficiency sub-indicators, or normalized feature vectors extracted from the four-layer normalized data packets of the source subdomain at time t-τ.

8. The intelligent control system for energy consumption big data analysis and optimization in petrochemical plants according to claim 7, characterized in that: The hybrid-driven optimization decision-making module adopts a model predictive control framework, and the specific processing flow is as follows: S1. Energy efficiency prediction: Extract the original energy flow parameters of each subdomain at the current and historical time from the original data buffer layer, extract the influence variables of the source subdomain at time t-τ from the cross-domain associated data packets, input the original energy flow parameters, influence variables and candidate operation parameters provided by the multi-objective optimization algorithm into the hybrid driving model, and output the predicted values ​​of energy efficiency sub-indicators at multiple future time points. S2. Optimization Iteration: Compare the predicted values ​​of energy efficiency sub-indicators with the dynamic energy efficiency benchmark, calculate the deviation, and take the minimum total energy consumption, carbon emission constraints and production plan completion rate as optimization objectives. Introduce the deviation as a penalty term, which together with the coupling relationship strength constitutes a constraint to construct a multi-objective function. The multi-objective optimization algorithm updates candidate operation parameters based on the multi-objective function value, and repeats S1 and S2 until the convergence condition is met or the preset number of iterations is reached, and outputs the optimal candidate operation parameters as the final operation parameter settings.

Images