Massive distribution network industrial resource coordination method based on load interaction and response analysis

Abstract

The present application relates to data processing technology, provide a kind of based on load interaction and response analysis's mass distribution network industrial resource coordination method, comprising: first, build real-time synchronous data acquisition processing system, fusion multi-source data, form unified data base;Second, realize flexibility dynamic identification by mechanism modeling and data driving, introduce flexibility label quantitative adjustable power, duration and response rate;Again, based on flexibility label, build interactive network, depict space-time coupling effect and group response characteristics, and again through multi-objective hierarchical optimization model, combine market electricity price and renewable power output prediction, generate equipment, station and regional level dispatching scheme;Finally, compare plan and actual response in actual scene, and correct;The present application can improve the ability of distribution network to mass industrial load coordination scheduling and fine management, improve flexibility utilization level, enhance the economy and robustness of system.

Description

Technical Field

[0001] This invention relates to data processing technology, and more particularly to a method for coordinating massive industrial resources in power distribution networks based on load interaction and response analysis. Background Technology

[0002] With the continuous connection of large-scale industrial users to the power distribution network, the operating environment faced by the power system is becoming increasingly complex. The massive industrial flexibility resources (such as motors, compressors, kilns, refrigeration units, and pump loads) exhibit significant differences in operating conditions, start-stop characteristics, response speed, and adjustability. These loads demonstrate strong heterogeneity in terms of process constraints, energy consumption characteristics, production continuity, and adjustment boundaries, making it difficult to uniformly measure and effectively utilize their flexibility.

[0003] Existing demand response and load management methods often focus on individual loads or small groups, achieving load regulation within a localized scope. However, they struggle to simultaneously address refined modeling, global collaborative optimization, and real-time response control in scenarios with massive industrial users. Furthermore, existing solutions generally rely on single mechanistic modeling or empirical statistical methods, lacking an accurate characterization of the dynamic evolution of flexibility under complex operating conditions, and are ill-equipped to handle rapid fluctuations and multiple uncertainties across different time scales.

[0004] Therefore, how to achieve refined management and coordinated scheduling of large-scale industrial users and improve the overall utilization level of flexible resources in the power distribution network has become a key issue that urgently needs to be addressed. Summary of the Invention

[0005] This invention provides a method for coordinating industrial resources in a massive distribution network based on load interaction and response analysis. This method can achieve refined management and coordinated scheduling of large-scale industrial users, and improve the overall utilization level of flexible resources in the distribution network.

[0006] A first aspect of the present invention provides a method for coordinating massive industrial resources in power distribution networks based on load interaction and response analysis, comprising:

[0007] The system collects multi-source data and performs standardization processing on the multi-source data to obtain the data base corresponding to the multi-source data.

[0008] Based on the data base, the mechanism model corresponding to each industrial load is determined, and individual flexibility labels are obtained based on the mechanism model and data-driven approach.

[0009] A directed load interaction network is constructed based on the individual flexibility labels, a constraint function is determined based on the directed load interaction network, and the group response capability is obtained based on the directed load interaction network and the constraint function.

[0010] A multi-objective optimization model is constructed based on the group response capability, and a hierarchical response scheduling scheme is generated based on the multi-objective optimization objective model;

[0011] The scheduling scheme is applied to a real-world scenario, and the actual response results are collected and compared with the prediction benchmark to obtain the comparison deviation. Based on the comparison deviation, the individual flexibility label and the group response capability are corrected.

[0012] Optionally, in one possible implementation of the first aspect, the step of collecting multi-source data according to the data acquisition and processing system and standardizing the multi-source data to obtain a data base corresponding to the multi-source data includes:

[0013] The time base of multi-source data is unified to obtain a unified multi-source dataset. The multi-source data includes the operation data of the power distribution network, the operation data and process data of industrial load, the environmental data of the external environment, and the market data of the external market.

[0014] The data from the multi-source dataset is processed to obtain a standard dataset;

[0015] The data in the standard dataset is normalized to obtain a data base. The data in the data base is then uniformly stored and interfaced, and updated in real time.

[0016] Optionally, in one possible implementation of the first aspect, the unification of the time base of the multi-source data to obtain a unified multi-source dataset includes:

[0017] A unified time step is set, and low-frequency data with a frequency lower than the unified time step and high-frequency data with a frequency higher than the unified time step are collected. The low-frequency data in the multi-source data is interpolated, and the high-frequency data in the multi-source data is downsampled to obtain data with a unified time series. The multi-source dataset is determined based on the data with a unified time series.

[0018] Optionally, in one possible implementation of the first aspect, the step of performing detection processing on the data in the multi-source dataset to obtain a standard dataset includes:

[0019] Retrieve 3 The principle is to perform anomaly detection on the data in the multi-source dataset, obtain abnormal data, correct or remove the abnormal data, and supplement the removed abnormal data to obtain a standard dataset.

[0020] Optionally, in one possible implementation of the first aspect, determining the mechanism model corresponding to each industrial load based on the data foundation, and obtaining individual flexibility labels based on the mechanism model and data-driven approach, includes:

[0021] The general mechanism equations are matched according to the equipment type of the industrial load. The equipment types include thermal inertial equipment and power equipment. The general mechanism equation for thermal inertial equipment is the energy conservation formula, and the general mechanism equation for power equipment is the first-order dynamic formula.

[0022] Based on the data base, the mechanism model parameters in each general mechanism equation are identified to obtain the mechanism model;

[0023] By using mechanistic models and data-driven methods, the flexibility of each industrial load is dynamically identified to obtain the actual adjustment capacity of each industrial load. A unique identifier and traceability information are added to the actual adjustment capacity of each industrial load to obtain an individual flexibility label.

[0024] Optionally, in one possible implementation of the first aspect, the step of constructing a directed load interaction network based on the individual flexibility labels, determining a constraint function based on the directed load interaction network, and obtaining the group response capability based on the directed load interaction network and the constraint function includes:

[0025] Using industrial loads as nodes and connecting them to each other, the actual adjustment capacity parameter in the individual flexibility label is used as the node attribute, the quantitative constraint information carried by the directed edges connecting the nodes is obtained as the edge attribute, and the directionality of the edges between nodes is set according to the direction of the coupling relationship to construct a directed load interaction network.

[0026] Based on the node and edge attributes of the directed load interaction network, the constraint function is determined.

[0027] The adjustable power capacity of the group equipment is calculated based on the directed load interaction network and constraint functions, and the group response capability is obtained based on the adjustable power capacity.

[0028] Optionally, in one possible implementation of the first aspect, the step of constructing a multi-objective optimization model based on the group response capability and generating a hierarchical response scheduling scheme according to the multi-objective optimization objective model includes:

[0029] A multi-objective optimization model is constructed by using a weighted summation method, with the group response capability as the constraint boundary.

[0030] By integrating a multi-objective optimization model and employing robust optimization to handle renewable energy output and opportunity constraints to handle electricity prices, an optimization model with uncertainty handling is formed.

[0031] The optimization model is solved hierarchically to generate a hierarchical response scheduling scheme.

[0032] Optionally, in one possible implementation of the first aspect, constructing the multi-objective optimization model through a weighted summation includes:

[0033] The multi-objective optimization model is obtained through the following formula.

[0034] ,

[0035] in, For operating costs, For emission costs, For production, for Target weights for Target weights for The target weight, The actual power of the node. For the charging and discharging power of the energy storage system, To optimize the objective.

[0036] Optionally, in one possible implementation of the first aspect, applying the scheduling scheme to a real-world scenario, collecting actual response results and comparing them with a prediction benchmark to obtain a comparison deviation, and correcting individual flexibility labels and group response capabilities based on the comparison deviation, includes:

[0037] Based on the pilot system or simulation platform, the scheduling scheme is applied to the actual load or virtual load cluster to obtain the actual response results;

[0038] The actual response results are compared with the prediction benchmark of the multi-objective optimization model to obtain the comparison bias, and then corrected according to the individual flexibility label and the group response capability.

[0039] The beneficial effects of this invention are as follows:

[0040] 1. This invention addresses the challenges of high heterogeneity and difficulty in quantifying the flexibility of massive industrial loads. On one hand, it constructs a data foundation encompassing distribution network operation, industrial processes, environmental data, and market data. Through time synchronization, anomaly detection, and normalization, it achieves the fusion and interoperability of multi-source data, providing high-quality data support for subsequent analysis. On the other hand, it innovatively combines mechanistic modeling and data-driven methods, matching specific mechanistic equations for thermally inertial equipment and power equipment. Through parameter identification and dynamic recognition, it generates individual flexibility labels containing dimensions such as adjustable power, response rate, and production impact costs, accurately characterizing the adjustment characteristics of individual loads. Based on this, a directed load interaction network is constructed using these labels as node attributes, quantifying the physical and technological coupling relationships between loads. Constraint functions integrate group response capabilities, laying the foundation for hierarchical collaborative scheduling and achieving refined collaboration at the equipment, plant, and regional levels, thereby improving the efficiency of the distribution network in the overall utilization of massive flexibility resources.

[0041] 2. This invention significantly improves the robustness and adaptability of scheduling schemes by integrating multi-objective optimization and uncertainty handling. At the optimization level, a multi-objective model is constructed with the collective response capability as the constraint boundary. Robust optimization is used to address fluctuations in renewable energy output, and opportunity constraints are used to handle electricity price uncertainty, achieving a balance between security constraints and economic efficiency. This invention decomposes the global problem into parallel solvable local subproblems through hierarchical solution. Furthermore, through dual-path verification via pilot systems and simulation platforms, actual response results are collected and compared with the predicted benchmark, and deviations are fed back to flexibility labels and collective response capability corrections. This mechanism can not only dynamically adapt to changes such as equipment aging and scenario fluctuations, but also reduce operating costs and carbon emissions while ensuring production continuity, providing reliable technical support for the large-scale absorption of renewable energy and improved operational stability of distribution networks. Attached Figure Description

[0042] Figure 1 The flowchart of the method for coordinating massive industrial resources in power distribution networks based on load interaction and response analysis is provided by the present invention.

[0043] Figure 2 This is a schematic diagram of the hardware structure of an electronic device provided by the present invention. Detailed Implementation

[0044] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0045] This invention provides a method for coordinating industrial resources in massive power distribution networks based on load interaction and response analysis, such as... Figure 1 As shown, it includes:

[0046] S1. Collect multi-source data according to the data acquisition and processing system, and perform standardization processing on the multi-source data to obtain the data base corresponding to the multi-source data.

[0047] Understandably, building a high-precision, real-time synchronized data acquisition and processing system is crucial. This system connects to SCADA systems, smart meters, process control systems, and environmental sensors across the power distribution network and massive industrial loads. It comprehensively collects multi-source data covering industrial loads, power distribution network operation, and the external environment, ensuring the integrity and real-time nature of this data. The system performs anomaly detection, time alignment, and completion on the collected multi-source data, integrating it into a unified dataset. This data foundation, capable of storage, retrieval, and real-time updates, provides complete and reliable input for subsequent flexible identification and interactive analysis.

[0048] Specifically, operational data such as voltage, current, power flow, and switch status are acquired through SCADA (Supervisory Control and Data Acquisition) systems. Industrial load data, including power consumption, load curves, and power factor, are collected through smart meters. Process data (such as temperature, speed, and output indicators) are input through the interfaces of DCS (Distributed Control System) and PLC (Programmable Logic Controller). In addition, environmental data and market data are also required. Environmental data includes temperature, humidity, wind speed, and light intensity, while market data includes external influencing factors such as electricity prices and carbon prices.

[0049] Among them, the data acquisition and processing system refers to an integrated hardware and software system that adapts to the multi-source data acquisition needs of power distribution network industrial scenarios. Standardized processing refers to a series of systematic data processing operations on multi-source data, including time synchronization, anomaly detection and correction, and dimension normalization. The data foundation refers to the data system built on standardized multi-source data, which is the unified data support platform for the entire technical solution and has the characteristics of centralized data storage, standardized access, real-time updates, and full-link traceability.

[0050] In some embodiments, step S1 (collecting multi-source data according to the data acquisition and processing system, and standardizing the multi-source data to obtain the data base corresponding to the multi-source data) includes S11-S13:

[0051] S11, unify the time base of the multi-source data to obtain a unified multi-source dataset. The multi-source data includes the operation data of the power distribution network, the operation data and process data of the industrial load, the environmental data of the external environment, and the market data of the external market.

[0052] Understandably, establishing a unified time standard is necessary to address issues related to different sampling frequencies, sampling delays, and data asynchrony. Millisecond-level time synchronization can be achieved using GPS (Global Positioning System) time synchronization or IEEE 1588 PTP (Precision Time Protocol) technology.

[0053] The time base refers to the reference time standard that all multi-source data follow during the multi-source data time synchronization process. Multi-source data refers to all-dimensional data supporting the coordinated scheduling of industrial resources in the distribution network, including distribution network operation data, industrial load operation data and process data, external environmental data, and external market data. The multi-source dataset refers to the dataset after time base unification. Industrial load represents equipment that consumes or converts electrical energy in industrial production activities. Distribution network operation data refers to data reflecting the operating status of the distribution network. Industrial load operation data refers to data reflecting the operating status of industrial equipment (including power consumption, load curves, power factor, etc.). Industrial load process data refers to data reflecting the execution status of process steps in the production process (such as temperature, speed, and output indicators). External environmental data refers to natural environmental data affecting industrial production efficiency and renewable energy output. External market data refers to data related to electricity market transactions and carbon emission accounting.

[0054] In some embodiments, step S11 (unifying the time base of the multi-source data to obtain a unified multi-source dataset) includes S111:

[0055] S111, set a unified time step, collect low-frequency data with a frequency lower than the unified time step and high-frequency data with a frequency higher than the unified time step, interpolate the low-frequency data in the multi-source data, downsample the high-frequency data in the multi-source data to obtain data with unified time sequence, and determine the multi-source dataset based on the data with unified time sequence.

[0056] Understandably, based on the real-time requirements and data characteristics of distribution network scheduling, a unique standardized time step is determined as the unified time series granularity for all data, providing a unified "time series benchmark" for high and low frequency data processing, and ensuring that all data is eventually normalized into time series data of this granularity.

[0057] The unified timing data can be obtained using the following formula.

[0058] ,

[0059] in, To unify the time-series data, These are the raw sampled values ​​from the multi-source data. For interpolation functions, An index for multi-source data sampling sequences. To standardize the time step, The total duration of the time range covered by the entire interpolation process. For a point in time, This serves as the reference point for interpolation.

[0060] Specifically, time step refers to time interval; low-frequency data refers to multi-source data with a collection frequency lower than the unified time step; high-frequency data refers to multi-source data with a collection frequency higher than the unified time step. Interpolation and downsampling are existing technologies and will not be elaborated here.

[0061] S12 performs detection processing on the data in the multi-source dataset to obtain the standard dataset.

[0062] It is understandable that anomaly detection and data cleaning are performed on the collected data to eliminate data anomalies caused by sensor failure, communication interruption or external interference, thereby ensuring data quality.

[0063] In some embodiments, step S12 (performing detection processing on the data in the multi-source dataset to obtain a standard dataset) includes S121:

[0064] S121, retrieve 3 The principle is to perform anomaly detection on the data in the multi-source dataset, obtain abnormal data, correct or remove the abnormal data, and supplement the removed abnormal data to obtain a standard dataset.

[0065] It is understandable that the three statistical methods are used. The principle is to detect outlier data, meaning that data exceeding the mean ± 3 standard deviations are considered outliers.

[0066] ,

[0067] in, The mean, The standard deviation is defined as the absolute deviation of the mean. Outlier data refers to data whose absolute deviation exceeds the mean by ±3 times the standard deviation. The standard dataset refers to a new dataset obtained after processing and detecting multiple source datasets.

[0068] In addition, isolated forests or autoencoders can be used for anomaly detection in high-dimensional data. For detected outliers, nearest neighbor interpolation or backfilling methods based on historical patterns can be used for correction.

[0069] S13 normalizes the data in the standard dataset to obtain a data base, and stores and interfaces the data in the data base in a unified manner, and updates it in real time.

[0070] Understandably, normalizing and feature extraction of data addresses the difficulty of directly comparing and integrating data of different scales and at different levels, and extracts key features for flexibility analysis. The normalization method uses Min-Max normalization to scale the data to the [0,1] interval:

[0071] ,

[0072] in, This represents the original data (i.e., the data in the standard dataset). This represents the data obtained after normalization. express The minimum value in the standard dataset. express The maximum value in the standard dataset.

[0073] For the operation of industrial loads, features such as average power, peak-to-valley difference, and load fluctuation rate are extracted; for the distribution network, features such as voltage deviation and branch load rate are extracted; and for the industrial load's technological process, the production power elasticity coefficient is extracted. Load fluctuation rate is defined as:

[0074] ,

[0075] in, Indicates the standard deviation of load power. Indicates average power. This refers to the load fluctuation rate.

[0076] Furthermore, the normalized dataset is stored to obtain a data base, and the data in the data base is uniformly stored and interfaced, and updated in real time to provide consistent data services for subsequent modules.

[0077] Specifically, a time-series database (TSDB) supporting hybrid storage of time-series and relational data is constructed. An API (Application Programming Interface) is designed to enable standardized data access and support flexible tag generation and interactive analysis. The Kafka stream processing framework is used to achieve real-time streaming updates of data, ensuring that the decision-making module always accesses the latest data.

[0078] Normalization refers to the process of eliminating the differences in dimensions and magnitudes of different data and transforming them into dimensionless data within a unified range. The data foundation is a data support platform built on the normalization of data in a standard dataset. Interface-based refers to the development of standardized API interfaces, allowing subsequent steps to call data in the data foundation in a unified format without having to worry about the underlying storage details.

[0079] S2, based on the data base, determine the mechanism model corresponding to each industrial load, and obtain individual flexibility labels based on the mechanism model and data-driven approach.

[0080] Understandably, using the unified data foundation built in step S1 as the data source, the flexibility of each industrial load is dynamically identified through mechanistic models and data-driven methods. The adjustable power, duration, and response rate of individual and group loads are quantified. The "flexibility tag" technology is introduced to mark the adjustment source and constraint boundary of each unit load / energy, providing a comparable and traceable flexibility measurement for collaborative decision-making.

[0081] Among them, the mechanism model refers to the mathematical model built based on physical laws and process principles; data-driven refers to the use of statistical analysis, machine learning, deep learning and other methods to explore the actual operating rules and regulation characteristics of industrial loads based on data in the data base; and individual flexibility label refers to the data carrier after the regulation capacity and attribute information of a single industrial load are standardized and packaged.

[0082] In some embodiments, step S2 (determining the mechanism model corresponding to each industrial load based on the data base, and obtaining individual flexibility labels based on the mechanism model and data-driven approach) includes S21-S23:

[0083] S21. Match the corresponding general mechanism equation according to the equipment type of the industrial load. The equipment type includes thermal inertial equipment and power equipment. The general mechanism equation corresponding to thermal inertial equipment is the energy conservation formula, and the general mechanism equation corresponding to power equipment is the first-order dynamic formula.

[0084] It is understandable that the equipment types for extracting industrial loads include thermal inertial equipment and power equipment, and simplified mechanistic equations (i.e., general mechanistic equations) are used according to different types. This embodiment uses physical equations to characterize the inherent constraints of industrial loads under power adjustment, ensuring that the identification results meet physical feasibility and are sustainable. and theoretical response limit Provide baseline values. For equipment categories with clear physical quantity relationships, simplified mechanistic equations are used. For example, for thermally inertial equipment (kilns, hot water tanks, etc.), energy conservation approximation is used:

[0085] ,

[0086] in, For heat capacity, For the rate of temperature change, For equipment temperature, For ambient temperature, The heat loss coefficient, The conversion efficiency from input power to heat load. This refers to the electrical power of the device.

[0087] For power equipment, a first-order dynamics approximation can be used:

[0088] ,

[0089] in, It is a time constant. The rate of change of power, To control the input or desired power.

[0090] Furthermore, thermal inertial equipment refers to industrial equipment with significant heat storage and heat buffering characteristics, while power equipment refers to industrial equipment whose core function is to convert electrical energy into mechanical energy.

[0091] S22, Based on the data base, the mechanism model parameters in each general mechanism equation are identified to obtain the mechanism model.

[0092] Understandably, data related to the general mechanism equation of industrial load is extracted from the data base, and the extracted data is substituted into the general mechanism equation to solve for unknown parameters (i.e., mechanism model parameters) through mathematical fitting.

[0093] Specifically, for the parameters of the mechanism model (such as...) , , , Online identification is performed using least squares or extended Kalman filtering.

[0094] Feature statistics are performed using the features extracted in step S1 for each device. In the window Above calculation:

[0095] ,

[0096] in, Average power, For the first The device at the time point The observed power, and , For the first The device at the time point The normalized power data.

[0097] ,

[0098] in, The standard deviation is denoted as .

[0099] ,

[0100] in, For volatility.

[0101] ,

[0102] in, Peak-to-valley difference, representing the difference between the maximum and minimum observed power within the window. Indicates the time window All moments within Power observations Take the maximum value. Indicates the time window All moments within Power observations Take the minimum value.

[0103] The response rate is measured using the sliding difference method. An empirical estimate is made, which represents the maximum instantaneous slope observed within the window. :

[0104] ,

[0105] in, For the first The equipment is at all times The actual observed power value.

[0106] The information obtained through the above formulas fills in the blind spots in the mechanistic model predictions and provides data support for the flexibility boundary.

[0107] S23 uses a mechanism model and data-driven method to dynamically identify the flexibility of each industrial load, obtain the actual adjustment capacity of each industrial load, and add a unique identifier and traceability information to the actual adjustment capacity of each industrial load to obtain an individual flexibility label.

[0108] Understandably, based on the mechanistic model in step S22 and combined with real-time process data in the data base, the theoretical boundary of the industrial load regulation capacity is obtained. Using recent operating data (e.g., minute-level data from the last 7 days) in the data base, theoretical parameters are corrected through a data-driven method to eliminate deviations between the model and reality, thereby obtaining the actual regulation capacity. The actual regulation capacity includes the maximum power, minimum power, maximum duration, minimum confidence value, and production or quality impact cost factors of the industrial load.

[0109] Furthermore, the mechanistic model and data-driven output are unified into a flexible vector that is quantifiable, comparable, and traceable. With flexibility label First, a flexibility vector is generated for each industrial load. :

[0110] ,

[0111] in, The maximum power of the equipment is indicated by its rated power. Safety limits related to mechanism or process Conservative values ​​are generally used. . This indicates the minimum power of the equipment, determined by the minimum allowable power of the process. With safety lower limit Decision, and . Indicates at a given power offset The maximum duration under the condition of (which can be taken as the upper or lower limit of power offset) is obtained by mechanism simulation or empirical calculation: .in, For the production impact function, This is an acceptable threshold. This represents the smaller reliable value between the upper bound of the mechanism and the empirical value. It is a cost factor affecting production or quality, representing the cost of output affected by deviation from the baseline per unit time, which can be provided by the process or finance department or estimated based on experience. The confidence level of the above quantities is obtained from residual statistics.

[0112] Flexibility label It can be represented as:

[0113] ,

[0114] in, The resource ID representing the flexibility label, Indicates the time stamp used to generate or update the flexibility label. Indicates the source of this flexibility metric, including all... Write back the TSDB from step S1, with the version or timestamp for traceability.

[0115] Furthermore, unique identifiers refer to the exclusive identification information of industrial loads, while traceability information refers to the information that records the entire process of tag generation, including the time stamp of flexibility tag generation or update, the source of generation, etc.

[0116] S3. Construct a directed load interaction network based on the individual flexibility labels, determine the constraint function based on the directed load interaction network, and obtain the group response capability based on the directed load interaction network and the constraint function.

[0117] Understandably, organizing a large number of industrial load units and their flexibility labels into a unified interactive network reveals the energy, information, and constraint coupling relationships between different nodes.

[0118] Among them, the directed load interaction network refers to a structured network constructed with industrial loads as nodes, the constraint function refers to a mathematical function extracted from the edge attributes of the directed load interaction network, and the group response capability refers to the total regulation capability that the group of industrial loads can provide in a coordinated manner under the constraint function of the directed load interaction network.

[0119] In some embodiments, step S3 (constructing a directed load interaction network based on the individual flexibility labels, determining a constraint function based on the directed load interaction network, and obtaining the group response capability based on the directed load interaction network and the constraint function) includes S31-S33:

[0120] S31 uses industrial loads as nodes and connects them to each other. It uses the actual adjustment capacity parameter in the individual flexibility label as the node attribute, obtains the quantitative constraint information carried by the directed edges of the connecting nodes as the edge attribute, and sets the directionality of the edges between nodes according to the direction of the coupling relationship to construct a directed load interaction network.

[0121] Understandably, industrial loads are used as nodes and interconnected. The actual adjustment capacity parameter in the individual flexibility label is used as a node attribute, and the quantified constraint information carried by the directed edges connecting the nodes is used as an edge attribute. Based on the coupling relationship information (physical coupling / process coupling) in the data base, directed edges are added between interacting nodes. The direction of the edge is consistent with the direction of the coupling relationship. All nodes, quantified attributes, and directed edges are integrated to form a complete directed load interaction network, which displays and describes the energy, information, and constraint coupling relationships between different nodes.

[0122] Among them, coupling relationship refers to the interactive association between industrial loads, and the orientation of edge refers to the directional attribute of the edge connecting nodes.

[0123] Constructing a directed graph Representing the relationships between different nodes:

[0124] ,

[0125] in, For a directed load interaction network, each node is a set of nodes. For an industrial load or plant, It is a set of directed edges. It is a directed edge.

[0126] The attributes of a node include the node's... upper limit of power offset Power offset lower limit Durability of load regulation Time constant , can be represented as:

[0127] ,

[0128] in, These are node attributes.

[0129] Edge properties include power coupling coefficient This is used to represent the direct impact of power changes on a node, and to control the coupling coefficient. , used to represent nodes With nodes The degree of dependency in communication or scheduling. The attributes of an edge can be represented as:

[0130] ,

[0131] in, This refers to the edge attribute.

[0132] S32. Based on the node and edge attributes of the directed load interaction network, determine the constraint function.

[0133] Understandably, constraint functions are established based on the interactive network to ensure that physical boundary conditions and technological dependencies between nodes are considered during group scheduling. First, the power feasible region for each node is defined:

[0134] ,

[0135] in, , This indicates the flexibility boundary identified by the node in step S2.

[0136] Edge attributes refer to the quantified constraint information carried by directed edges in a directed load interaction network.

[0137] In addition to the flexibility boundaries of the nodes themselves, the interrelationships between devices and between devices and the system also need to be considered. Therefore, coupling constraint functions are defined. , used to represent nodes The power change acts on the node through the coupling parameters. This makes the node The power change is:

[0138] ,

[0139] Next, a group power balance constraint is defined to ensure that the aggregated adjustable power must match the scheduling target; otherwise, the flexibility provided is ineffective. Its formula is expressed as:

[0140] ,

[0141] in, This indicates the total amount of industrial load. Contribute to renewable energy This indicates the charging and discharging power of the energy storage system. This indicates the total load demand of the distribution network.

[0142] S33, calculate the adjustable power capacity of the group equipment based on the directed load interaction network and constraint function, and obtain the group response capability based on the adjustable power capacity.

[0143] It should be noted that adjustable power capacity includes adjustable power capacity up and adjustable power capacity down.

[0144] Specifically, based on interactive networks and constraint functions, the collaborative response behavior of the population load is analyzed to identify key nodes and emergent characteristics. In population response analysis, it is necessary to integrate the dispersed individual flexibility into a population-level metric for unified optimization. Therefore, a population flexibility aggregation metric is defined to describe the total adjustment space of the population over a given time period.

[0145] ,

[0146] in, The power capacity can be increased for the group. This indicates that the group can reduce its power capacity. For a single device at any time The amount of power adjustment is increased. For a single device at any time The amount of power adjustment.

[0147] In addition, group response capability refers to the comprehensive regulation capability that multiple industrial loads can provide through coordinated regulation under the constraints of a directed load interaction network.

[0148] Next, to measure the overall adjustment speed of the grid after receiving a regulation command, a grid response rate is defined to ensure the grid's rapid adjustment in response to frequency fluctuations or voltage over-limits:

[0149] ,

[0150] in, This represents the response time constant for a single node. This indicates the adjustment capacity of the node. This represents the weighted average response rate, reflecting the overall response capability of the population.

[0151] S4. Construct a multi-objective optimization model based on the group response capability, and generate a hierarchical response scheduling scheme based on the multi-objective optimization target model.

[0152] Understandably, based on the interactive analysis in step S3 and combined with market electricity prices, renewable energy output forecasts, and process constraints, a hierarchical multi-objective collaborative optimization model is established to generate long-term and short-term response scheduling schemes at the equipment level, plant level, and regional level.

[0153] Among them, the multi-objective optimization model refers to a mathematical model that includes multiple optimization objectives, and the hierarchical response scheduling scheme refers to a hierarchical scheduling strategy that matches different parameters of the group's response capability (response rate, duration, adjustment cost) with the task requirements according to the priority of the scheduling task, so as to form a hierarchical scheduling strategy with different priorities.

[0154] In some embodiments, step S3 (constructing a multi-objective optimization model based on the group response capability and generating a hierarchical response scheduling scheme according to the multi-objective optimization objective model) includes S41-S43:

[0155] S41 uses the group response capability as the constraint boundary and constructs a multi-objective optimization model through weighted summation.

[0156] It is understandable that the constraint boundary refers to the insurmountable system regulation limit in the group response capability, including the upper limit of the total adjustable power of the group, the upper limit of the total response rate, and the upper limit of the total duration.

[0157] In some embodiments, step S41 (constructing a multi-objective optimization model by weighted summation) includes S411:

[0158] S411, the multi-objective optimization model is obtained through the following formula:

[0159] ,

[0160] in, For operating costs, For emissions costs (or carbon targets). For output (or production impact penalty). for Target weights for Target weights for The target weight, The actual power of the node. For the charging and discharging power of the energy storage system, To optimize the objective, This indicates that among all feasible load regulation and energy storage dispatch, the search should be made to allow... The smallest set of scheduling schemes.

[0161] Operating costs can be specifically expressed as:

[0162] ,

[0163] in, For the operating costs of the equipment, For energy storage charging power, For energy storage discharge power, The operating cost of the equipment when charging energy storage. This refers to the operating cost of the equipment during energy storage and discharge.

[0164] Carbon emission costs or carbon targets can be expressed as:

[0165] ,

[0166] in, Indicates device Carbon intensity or carbon price converted into currency Indicates the carbon emission intensity of renewable energy. It indicates renewable energy output.

[0167] The purpose of output or production impact penalties is to monetize losses in quality or output that deviate from the baseline, which can be expressed as:

[0168] ,

[0169] in, Indicates baseline power. Indicates the first Taiwan's industrial load The production or quality at any given moment affects the cost factor.

[0170] The constraint set for the above objective function includes:

[0171] (1) Individual power boundary: Ensure that the equipment's safety limits are not exceeded.

[0172] ,

[0173] in, For the first Taiwan's industrial load at any time Minimum permissible operating power, For the first Taiwan's industrial load at any time The maximum permissible operating power.

[0174] (2) Response rate limit: Ensure the executability of the device after the command is issued.

[0175] ,

[0176] in, For the first Taiwan's industrial load at any time Operating power.

[0177] (3) Group power balance constraint: ensuring supply and demand balance

[0178] ,

[0179] (4) Energy storage dynamics and boundary conditions

[0180] ,

[0181] ,

[0182] in, For the first At any given moment, the state of charge of the energy storage system. For the first At any given moment, the state of charge of the energy storage system. Charging efficiency, , The minimum allowable charged energy of the energy storage system, The maximum allowable charged energy of the energy storage system. This represents the maximum charging power of the energy storage system. This represents the maximum discharge power of the energy storage system.

[0183] S42 integrates a multi-objective optimization model and uses robust optimization to handle renewable energy output and opportunity constraints to handle electricity prices, forming an optimization model with uncertainty handling.

[0184] Understandably, robust optimization is an existing optimization method for bounded uncertainty, while chance constraint is an existing optimization method for probabilistic uncertainty. An optimization model refers to a mathematical model that integrates multi-objective optimization, robust optimization, and chance constraint. Renewable energy output refers to the electrical power output by power generation equipment per unit time.

[0185] Specifically, to address the uncertainties in renewable energy output, load forecasting errors, and flexibility labels, robust optimization and chance constraints are employed to provide reliability assurance in scheduling decisions. The choice between these two methods is based on the system's tolerance for worst-case scenarios and computational capabilities.

[0186] Robust optimization is for uncertainties Setting an indeterminate set This ensures that the optimization objective satisfies the constraints even in the worst-case scenario. If the renewable energy forecast error... ,in, The bounded fluctuation range of renewable energy output deviation. This represents the prediction error for renewable energy.

[0187] The most conservative form of the group equilibrium constraint is written as:

[0188] ,

[0189] in, This indicates the projected output of renewable energy. This represents the predicted load power. This indicates the error in load forecasting.

[0190] The chance constraint is satisfied in a probabilistic sense, for example, "at least..." "The probability of satisfying supply and demand is higher," which is more economical but probabilistically allows for efficiency defaults. For example, constraining group equilibrium:

[0191] ,

[0192] in, , For renewable energy forecasting errors, , For load error, It is a probabilistic constraint to ensure that the constraint holds under high probability.

[0193] If we assume that the uncertain terms follow a Gaussian distribution and are independent, they can be transformed into deterministic constraints:

[0194] ,

[0195] in, The standard deviation of the uncertainties in supply and demand. For standard normal quantiles, The cumulative distribution function of the standard normal distribution inverse function, This represents the low probability corresponding to the confidence level.

[0196] Robust optimization and the use of opportunity constraints: Robust optimization is used for critical safety constraints (such as equipment safety and chemical process thresholds); opportunity constraints are used for economically viable constraints such as supply and demand balance to achieve better economic efficiency. A two-layer / segmented strategy can also be used: opportunity constraints are used for the main problem, and robust guarantees are used for rescue actions triggered after constraint failure (such as ESS discharge).

[0197] S43, solve the optimization model in layers to generate a layered response scheduling scheme.

[0198] It is understandable that hierarchical solution refers to a solution method that, based on the dual logic of scheduling task priority and device response characteristics, breaks down an optimization model containing uncertainty into multiple levels for independent solution. Each level corresponds to a specific set of tasks and devices, and the results from each level are finally integrated. Adjustment power allocation refers to the specific adjustment power quota allocated to each device through hierarchical solution. Startup time refers to the specific moment when the device begins to execute the adjustment command, and duration refers to the length of time the device maintains the adjusted state.

[0199] A hierarchical strategy is adopted to optimize the coordination of massive industrial power distribution network flexibility resources. The optimization is divided into multiple levels, which can achieve parallel computing while retaining local private information. The hierarchical structure mainly includes three layers: the first is the device layer, which is the actuator end, where the actual execution is performed. Instructions and report local status and Secondly, the plant / station level is responsible for the equipment under its jurisdiction. The first layer is the aggregation layer, which addresses local sub-problems (cost minimization or customer preferences) and reports the aggregation flexibility available to external parties. The second layer is the dispatch layer, which is responsible for coordination at the network-wide or regional level, issuing price signals or target loads, and managing systems and constraints.

[0200] The basic idea of ​​hierarchical collaborative solution is to write the centralized problem as a block optimization with consistency constraints. Let each plant... Control of its equipment collection Introducing local decision variables (e.g., local) Vectors and globally consistent variables (For example, the entire network supply vector). The overall objective function is:

[0201] ,

[0202] Must meet (Consistent), among which Indicates the first The local objective function of each plant. Refers to globally consistent variables Corresponding to the plant / station The local components.

[0203] This is a distributed optimization algorithm that uses the Alternating Direction Multiplier Method (ADMM) to decompose the global problem into parallel solvable local subproblems, and achieves convergence by alternately updating local, global, and dual variables. The decomposable local subproblems of the overall objective function are as follows:

[0204] ,

[0205] in, For the factory station After the first After the second local optimization, the determined local decision variables are: Indicates the iteration count index. This represents the penalty factor, used to control the weight of the consistency term and affect the convergence speed and numerical stability. Indicates the first Globally consistent variables in the next iteration Indicates the first In the nth iteration The dual variables of each plant. This represents the set of local decision variables.

[0206] The update of a globally consistent variable is represented as:

[0207] ,

[0208] in, Indicates the first Globally consistent variables in the next iteration This indicates the total number of plants and stations.

[0209] The update of the dual variable is represented as:

[0210] ,

[0211] in, Indicates the first In the nth iteration The dual variables of each plant.

[0212] Alternatively, a price signaling method can be used. The distribution network publishes time-varying prices, and then each substation addresses its localized issues.

[0213] ,

[0214] Then the distribution network checks global constraints and updates them. Until convergence, among which, It refers to the Lagrange multipliers.

[0215] The above solution is implemented using rolling time domain (MPC), which means that the time domain optimization is solved in each real-time period and only the first step of control is performed, and then the forecast is updated and repeated in the next time step. The hierarchical ADMM and price signal are iterated several times in each MPC step to achieve acceptable consistency.

[0216] S5. Apply the scheduling scheme to the actual scenario, collect the actual response results and compare them with the prediction benchmark to obtain the comparison deviation, and correct the individual flexibility label and the group response capability based on the comparison deviation.

[0217] Understandably, validating the effectiveness of the hierarchical collaborative optimization strategy in real-world scenarios involves comparing actual response results with the prediction benchmark to verify power balance and response performance. Deviation feedback is then used for flexibility label correction and model parameter updates, forming a closed loop of "modeling—analysis—decision—validation—optimization" to ensure continuous improvement and stable operation of the method.

[0218] The actual response result refers to the actual operating data of the equipment (i.e., industrial load) captured by the data acquisition and processing system after the scheduling plan is implemented. The prediction benchmark refers to the preset data used for comparison with the actual response result. The comparison deviation refers to the quantitative difference between the actual response result and the prediction benchmark.

[0219] In some embodiments, step S5 (applying the scheduling scheme to a real-world scenario, collecting actual response results and comparing them with a prediction benchmark to obtain a comparison deviation, and correcting individual flexibility labels and group response capabilities based on the comparison deviation) includes S51-S52:

[0220] S51, Based on the pilot system or simulation platform, the scheduling scheme is applied to the actual load or virtual load cluster to obtain the actual response results.

[0221] It is understandable that, based on pilot systems or simulation platforms, the hierarchical scheduling strategy obtained from S4 will be implemented. (in Indicates hierarchy At any moment The optimal scheduling instructions are applied to the actual load or virtual load cluster to verify their effectiveness under real operating conditions.

[0222] Among them, the actual load cluster refers to a collaborative operation set composed of multiple industrial devices in the actual operation scenario of industrial power distribution network, while the virtual load cluster refers to a set of digital models constructed in the simulation platform that are equivalent to the physical characteristics of the actual load cluster.

[0223] S52 compares the actual response results with the prediction benchmark of the multi-objective optimization model to obtain the comparison bias, and corrects the individual flexibility label and the group response capability based on the comparison bias.

[0224] It is understandable that the actual execution results will be... With predicted load power The comparison was conducted to evaluate whether the load response performance met expectations and to verify the group power balance constraints. Whether it is true or not. Among them, It is the reference power obtained through S3 load interaction analysis and S4 collaborative optimization decision-making, and .in, This is the total global reference power. It means The global power up-adjustment deviation at any given moment. It means The global down-adjustment power deviation at any given time.

[0225] The comparison bias found during the verification Returning to S2 flexibility identification and S3 interaction analysis, this is used to correct the flexibility label and the group's responsiveness, thereby achieving iterative improvement and adaptive optimization of the method.

[0226] See Figure 2 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. The electronic device 40 includes: a processor 41, a memory 42, and a computer program; wherein...

[0227] The memory 42 is used to store the computer program, and the memory may also be flash memory. The computer program is, for example, an application program or functional module that implements the above method.

[0228] The processor 41 is configured to execute the computer program stored in the memory to implement the various steps performed by the device in the above method. For details, please refer to the relevant descriptions in the preceding method embodiments.

[0229] Alternatively, the memory 42 can be either standalone or integrated with the processor 41.

[0230] When the memory 42 is a device independent of the processor 41, the device may further include:

[0231] Bus 43 is used to connect the memory 42 and the processor 41.

[0232] 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 or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for coordinating industrial resources in massive power distribution networks based on load interaction and response analysis, characterized in that, include: The system collects multi-source data and performs standardization processing on the multi-source data to obtain the data base corresponding to the multi-source data. Based on the aforementioned data foundation, the mechanistic model corresponding to each industrial load is determined. Individual flexibility labels are then obtained based on the mechanistic model and data-driven approach, including: The general mechanism equations are matched according to the equipment type of the industrial load. The equipment types include thermal inertial equipment and power equipment. The general mechanism equation for thermal inertial equipment is the energy conservation formula, and the general mechanism equation for power equipment is the first-order dynamic formula. Based on the data base, the mechanism model parameters in each general mechanism equation are identified to obtain the mechanism model; By using mechanistic models and data-driven methods, the flexibility of each industrial load is dynamically identified to obtain the actual adjustment capacity of each industrial load. A unique identifier and traceability information are added to the actual adjustment capacity of each industrial load to obtain an individual flexibility label. A directed load interaction network is constructed based on the individual flexibility labels. A constraint function is determined based on the directed load interaction network. The group response capability is obtained based on the directed load interaction network and the constraint function, including: Using industrial loads as nodes and connecting them to each other, the actual adjustment capacity parameter in the individual flexibility label is used as the node attribute, and the quantitative constraint information carried by the directed edges connecting the nodes is obtained as the edge attribute. The edge attribute includes the power coupling coefficient, which represents the degree of direct impact of the power change of one node on another node, and the control coupling coefficient, which represents the degree of dependence between nodes in communication or scheduling. The directionality of the edges between nodes is set according to the direction of the coupling relationship, and a directed load interaction network is constructed. Based on the node and edge attributes of the directed load interaction network, the constraint function is determined. The adjustable power capacity of the group equipment is calculated based on the directed load interaction network and constraint function, and the group response capability is obtained based on the adjustable power capacity. A multi-objective optimization model is constructed based on the group response capability, and a hierarchical response scheduling scheme is generated based on the multi-objective optimization model, including: A multi-objective optimization model is constructed by using a weighted summation method, with the group response capability as the constraint boundary. By integrating a multi-objective optimization model and employing robust optimization to handle renewable energy output and opportunity constraints to handle electricity prices, an optimization model with uncertainty handling is formed. The optimization model is solved hierarchically to generate a hierarchical response scheduling scheme. The scheduling scheme is applied to a real-world scenario, and the actual response results are collected and compared with the prediction benchmark to obtain the comparison deviation. Based on the comparison deviation, the individual flexibility label and the group response capability are corrected.

2. The method according to claim 1, characterized in that, The step of collecting multi-source data according to the data acquisition and processing system, and standardizing the multi-source data to obtain the data base corresponding to the multi-source data includes: The time base of multi-source data is unified to obtain a unified multi-source dataset. The multi-source data includes the operation data of the power distribution network, the operation data and process data of industrial load, the environmental data of the external environment, and the market data of the external market. The data from the multi-source dataset is processed to obtain a standard dataset; The data in the standard dataset is normalized to obtain a data base. The data in the data base is then uniformly stored and interfaced, and updated in real time.

3. The method according to claim 2, characterized in that, The process of unifying the time base of multi-source data to obtain a unified multi-source dataset includes: A unified time step is set, and low-frequency data with a frequency lower than the unified time step and high-frequency data with a frequency higher than the unified time step are collected. The low-frequency data in the multi-source data is interpolated, and the high-frequency data in the multi-source data is downsampled to obtain data with a unified time series. The multi-source dataset is determined based on the data with a unified time series.

4. The method according to claim 2, characterized in that, The process of detecting and processing data from multiple source datasets to obtain a standard dataset includes: Retrieve 3 The principle is to perform anomaly detection on the data in the multi-source dataset, obtain abnormal data, correct or remove the abnormal data, and supplement the removed abnormal data to obtain a standard dataset.

5. The method according to claim 1, characterized in that, The construction of the multi-objective optimization model through weighted summation includes: The multi-objective optimization model is obtained through the following formula. ； in, For operating costs, For emission costs, The purpose of output or production impact penalties is to monetize the loss of quality or output that deviates from the baseline. for The target weight, for The target weight, for The target weight, The actual power of the node. For the charging and discharging power of the energy storage system, To optimize the objective.

6. The method according to claim 1, characterized in that, The process of applying the scheduling scheme to a real-world scenario, collecting actual response results and comparing them with a predicted baseline to obtain the comparison deviation, and then correcting individual flexibility labels and group response capabilities based on the comparison deviation includes: Based on the pilot system or simulation platform, the scheduling scheme is applied to the actual load or virtual load cluster to obtain the actual response results; The actual response results are compared with the prediction benchmark of the multi-objective optimization model to obtain the comparison bias. Based on the comparison bias, the individual flexibility label and the group response capability are corrected.

Images