An energy production scheduling optimization decision system and method

Abstract

The application discloses a kind of based on energy production scheduling optimization decision system and method, belong to energy management and production scheduling technical field, obtain the operating parameter of various energy equipment, construct initial operating state set;Extract the edge response characteristic parameter under high-frequency variable load condition;Graph neural network model is constructed, the relative response distance of current state and edge state of equipment is calculated, and edge criticality judging function is constructed based on classification model;Relative response distance is mapped as scheduling priority index, and equipment ranking list is generated;According to real-time load and predicted fluctuation range, combined with priority selection response equipment combination, and risk correction is carried out to scheduling instruction;The robustness of scheduling scheme under disturbance condition is verified by simulation, if it does not meet the requirements, then feedback optimization.Edge characteristic parameters are updated after each scheduling cycle, and the application can realize dynamic perception, adaptive scheduling and risk avoidance under multi-source energy system, improve scheduling stability and intelligence.

Description

Technical Field

[0001] This invention relates to the field of energy management and production scheduling technology, specifically to an energy production scheduling optimization decision-making system and method. Background Technology

[0002] Against the backdrop of the rapid development of new energy sources such as wind power, photovoltaics, and biomass energy, the power grid system has placed higher demands on its ability to dispatch multi-source heterogeneous energy. Different types of energy have significant differences in terms of production stability, response speed, energy density, and cost constraints. Especially in multi-source hybrid energy supply scenarios, such as remote mountainous areas, island microgrids, and military outposts, traditional energy dispatching methods are difficult to balance the dynamic matching of energy supply and demand with the optimal coordination of multiple energy sources.

[0003] In existing technologies, most energy dispatch strategies are based on static prediction, which ignore the dynamic performance changes of production equipment at different operating stages. Especially when equipment is aging or operating inefficient, it is impossible to adjust the energy distribution ratio in real time, which leads to excessive energy consumption or frequent start-stop of some key equipment, increases system losses and affects the stability of energy supply.

[0004] For example, in remote border power supply points, under a multi-source energy structure with photovoltaic as the main source, wind power as a supplement, and diesel as a backup, the large fluctuations in wind and solar resources and the high error in load forecasting often lead to the frequent use of diesel generators for emergency power generation, which increases operating costs and is not conducive to carbon emission reduction.

[0005] In addition, existing optimization scheduling methods often match resources based on global load or average historical load, failing to identify the "edge response capability" of specific energy equipment in a specific time period, that is, the critical point at which equipment switches from inefficiency to high consumption under critical conditions, thus missing the optimal scheduling window and causing energy redundancy or waste. Summary of the Invention

[0006] The purpose of this invention is to provide an energy production scheduling optimization decision-making system and method to address the shortcomings of the prior art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a decision-making method based on energy production scheduling optimization, comprising:

[0008] S100: Obtain the operating parameter sequence of various energy equipment in the target energy supply area, including power output curve, response time, current efficiency, and historical load characteristics, and construct the initial operating state set E0;

[0009] S200. Based on the initial operating state set E0, extract the edge response characteristic parameters R of each energy device under high-frequency load conditions, including edge load threshold, cut-in power gradient, and stabilization time delay.

[0010] S300. Based on the edge response characteristic parameter R, calculate the relative response distance D between the current state of each device and the edge response state, and construct the edge critical judgment function F(x) to determine whether it is close to the critical operating point.

[0011] S400: Map the relative response distance D of all devices to a unified scheduling priority index L, and establish a sorting list Q for multi-source devices;

[0012] S500: Obtain the real-time load demand P(t) and the predicted load fluctuation range ΔP(t) at the current moment. Combine the sorting list Q and select the response equipment combination G according to priority so that it meets the load range demand of P(t) + ΔP(t).

[0013] S600: Based on the response status, edge response characteristic parameter R, and critical state evaluation value of the devices in the response device combination G, the power command of the devices within the scheduling cycle is corrected to generate the scheduling control set U.

[0014] S700: Simulate the generated scheduling control set U to evaluate its robustness and response delay within ΔP(t). If the dynamic robustness threshold is met, output U to the scheduling system; otherwise, return to S400 to adjust the Q list.

[0015] S800. After each scheduling cycle is completed, the actual response data of each device is compared with the predicted edge response feature parameter R, R is updated and fed back to step S200 to complete the closed-loop optimization.

[0016] Preferably, the calculation of the relative response distance D between the current state of each device and the edge response state includes:

[0017] Based on the edge response feature parameter set R, a device response relationship graph is constructed with energy devices as nodes and the response correlation between devices as edges. The current operating status parameters of each energy device and the corresponding edge response feature parameters are used as node feature inputs.

[0018] Graph neural network feature learning is performed on the device response relationship graph. The node features are mapped to a unified latent space through multi-layer graph convolution operations to obtain the latent space state vector corresponding to each energy device at the current time.

[0019] In the latent space, the distance between the current state vector of each energy device and its corresponding edge response state vector is calculated to obtain the relative response distance D, which represents the degree to which the current state of the device deviates from the edge response state.

[0020] Preferably, the steps for constructing the edge criticality determination function F(x) include:

[0021] Based on the relative response distance D and the edge response feature parameter set R of the device, the state feature vector of the energy device at the current moment is extracted, including power load level, regulation speed and stability delay;

[0022] The state feature vector is normalized and combined with historical edge response samples. A supervised learning method is used to train a nonlinear classification model to establish a mapping relationship between the equipment operating state and the critical state.

[0023] The nonlinear classification model is embedded as an edge critical determination function F(x), which is used to output whether the current device state is close to its corresponding edge operating region. The output of the function F(x) is a binary label or a probability value.

[0024] The device status is classified and labeled according to the output of F(x), and marked as "critically close" or "operating stably".

[0025] Preferably, generating the scheduling control set U includes the following steps:

[0026] Obtain the response status parameters, corresponding edge response characteristic parameters R, and evaluation results of the edge criticality determination function F(x) for each energy device in the response device combination G during the current scheduling cycle, and use the output of F(x) as the device critical state identifier value;

[0027] Based on the critical state identifier value of the equipment, the energy equipment in the response equipment combination G is divided into safe regulation equipment and restricted regulation equipment, and the corresponding power regulation upper limit value and power change rate constraint are determined respectively.

[0028] Based on the power regulation upper limit, power change rate constraint and edge response characteristic parameter R, the original power dispatch instructions of each energy device are corrected in amplitude and slope to generate dispatch control set U.

[0029] Perform a timing consistency check on the scheduling control set U within the scheduling cycle. If the check passes, output the scheduling control set U for execution; otherwise, update the device's response status parameters and re-correct the power command.

[0030] Preferably, the step of simulating the generated scheduling control set U includes:

[0031] The scheduling control set U is input into the multi-device joint scheduling model to construct the time-series simulation trajectory of the response behavior of each device within the scheduling cycle;

[0032] By introducing a load disturbance sequence ΔP(t) and using typical high, medium, and low disturbance samples within the predicted fluctuation range as input, the response results of the scheduling control set U under different load disturbance conditions are simulated.

[0033] Statistical analysis was performed on the response time, stability delay, and output deviation of each device in the simulation results to obtain the maximum response delay and minimum output stability margin of the scheduling control set U under different disturbance conditions.

[0034] The robustness of the scheduling control set U within the load disturbance range ΔP(t) is determined based on whether the maximum response delay is less than the set response time limit threshold T3 and whether the minimum output stability margin is greater than the safety margin threshold M1.

[0035] Preferably, the selection of the response device combination G in step S500 includes: accumulating output from high-priority devices based on the sorting list Q until P(t) + ΔP(t) is satisfied; eliminating devices in the "critical approach" state or with insufficient regulation capacity; prioritizing combinations with fewer devices, higher energy efficiency, or response distance far from the edge state; and ensuring that no device in the final combination is in an unstable operating zone.

[0036] Preferably, updating the edge response feature parameter R in step S800 includes: comparing the collected actual response data with the predicted edge response feature parameter set R used before the current scheduling period; comparing the deviation between the actual value and the predicted value, and calculating the correction factor; and dynamically adjusting each feature parameter using a sliding weighted update algorithm based on the deviation between the actual and the predicted values: let Ri be the current edge response feature parameter value, and Ria be the actual observed value in this period; let the weighted update coefficient γ∈[0,1], then the updated edge response feature parameter value is: .

[0037] The present invention also provides an energy production scheduling optimization decision-making system, comprising:

[0038] The data acquisition module obtains the operating parameter sequence of various energy equipment in the target energy supply area, including power output curves, response time, current efficiency, and historical load characteristics, and constructs the initial operating state set E0;

[0039] The edge response feature modeling module extracts edge response feature parameters R of each energy device under high-frequency load variation conditions based on the initial operating state set E0, including edge load threshold, cut-in power gradient, and stabilization time delay.

[0040] The graph neural network recognition module calculates the relative response distance D between the current state of each device and the edge response state based on the edge response feature parameter R, and constructs the edge critical determination function F(x) to determine whether it is close to the critical operating point.

[0041] The scheduling priority evaluation module maps the relative response distance D of all devices to a unified scheduling priority index L, and establishes a sorting list Q of multi-source devices;

[0042] The equipment combination selection module obtains the real-time load demand P(t) and the predicted load fluctuation range ΔP(t) at the current moment. Combined with the sorting list Q, it selects the response equipment combination G according to priority so that it meets the load range demand of P(t) + ΔP(t).

[0043] The power command correction and generation module, based on the response status of the devices in the response device combination G, the edge response characteristic parameter R and the critical state evaluation value of F(x), corrects the power command of the devices within the scheduling cycle and generates the scheduling control set U.

[0044] The robustness verification module simulates the generated scheduling control set U to evaluate its robustness and response delay within ΔP(t). If the dynamic robustness threshold is met, U is output to the scheduling system; otherwise, the module returns to the scheduling priority evaluation module to adjust the Q list.

[0045] The closed-loop optimization module compares the actual response data of each device with the predicted edge response feature parameter R after each scheduling cycle, updates R, and feeds it back to the edge response feature modeling module to complete the closed-loop optimization.

[0046] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0047] 1. The energy production scheduling optimization decision-making method provided by this invention can accurately identify the edge response characteristics of energy equipment under high-frequency load conditions. By constructing a relative response distance and edge criticality judgment function, it achieves early perception and dynamic assessment of the critical state of equipment operation, significantly improving the risk control capability of scheduling strategies. Compared with traditional scheduling methods based on static capacity or empirical thresholds, this invention has stronger adaptability and intelligent judgment capabilities, effectively preventing equipment from entering overload or unstable states, and ensuring the safe and stable operation of multi-source energy systems under complex operating conditions.

[0048] 2. This invention introduces graph neural networks for equipment state modeling and response distance calculation, and combines simulation verification and closed-loop optimization mechanisms to construct a data-driven, model-learnable, and self-iteratively optimizing intelligent scheduling system. This system not only achieves quantitative ranking of scheduling priorities and dynamic optimal selection of equipment combinations, but also ensures the stability of scheduling commands under load fluctuations through power command correction and simulation robustness evaluation, exhibiting excellent real-time performance, robustness, and energy efficiency adaptability. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0050] Figure 1 This is a flowchart of the method of the present invention.

[0051] Figure 2 This is a flowchart of the system modules of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Example 1, please refer to Figure 1 As shown in this embodiment, a decision-making method based on energy production scheduling optimization includes:

[0054] S100: Obtain the operating parameter sequence of various energy equipment in the target energy supply area, including power output curve, response time, current efficiency, and historical load characteristics, and construct the initial operating state set E0.

[0055] In the energy production scheduling optimization decision-making method described in this invention, step S100 aims to comprehensively acquire the operating status information of various energy equipment in the target energy supply area, and extract and store their parameters in a structured manner according to a unified standard, so as to provide a reliable data foundation for subsequent identification of edge response characteristics and scheduling priority ranking.

[0056] Within the target energy supply area, there are typically multiple energy types, such as photovoltaic power generation systems, wind power generation systems, biomass boilers, diesel generator sets, gas turbines, and energy storage systems. Different types of equipment exhibit significantly different dynamic response characteristics during actual operation. Therefore, this step requires establishing separate operating parameter acquisition models for each type of equipment to construct a unified initial operating state set E0. The "operating parameter sequence" mentioned in this step mainly includes, but is not limited to, the following four types of key data:

[0057] Power output curve P(t): Records the actual power output value of the equipment at different time points, usually collected at minute or second intervals, forming a time series curve. This data reflects the basic load response capability of the equipment and is a key basis for judging the load-bearing range.

[0058] Response time (Tr): This refers to the time required for a device to achieve stable output after receiving a scheduling command, describing its dynamic adjustment capability. Response time can be obtained by comparing the scheduling system logs with real-time operational data.

[0059] Current efficiency η(t): The energy efficiency ratio of the equipment at its current operation, such as the conversion efficiency of photovoltaic modules or the thermal efficiency of diesel generator sets. This parameter is used to determine whether the equipment is operating in a high-efficiency range.

[0060] Historical load characteristics H1: Typical load levels (such as maximum load, average load, load fluctuation amplitude, etc.) of the equipment in the past N scheduling cycles are used to evaluate the stability of equipment operation and its load adaptability.

[0061] The above parameters are collected in real time through the scheduling master station, the equipment controller (such as a PLC), the edge computing unit, or the remote monitoring system. For older equipment that does not have real-time communication capabilities, data access can be achieved by installing external data acquisition modules (such as smart meters, current transformers, wireless transmission terminals, etc.).

[0062] The collected multi-source heterogeneous data needs to be processed in a unified structure. The system uses a standardized interface to store various operating parameters into the status database using the device ID as the index, forming the following initial operating status set E0: E0={e1,e2,...,ei,...,en}; where ei=(Pi(t),Tri,ηi(t),Hli); where n is the number of devices, and ei represents the current operating status of the i-th device.

[0063] S200. Based on the initial operating state set E0, extract the edge response characteristic parameters R of each energy device under high-frequency load variation conditions, including edge load threshold, cut-in power gradient, and stabilization time delay.

[0064] After completing step S100 and obtaining the initial operating state set E0 of each energy device in the target power supply area, this step aims to further characterize the dynamic response characteristics of each energy device under high-frequency load conditions, extract the edge response feature parameter R that can reflect the critical operating state of the device, and provide a quantitative basis for subsequent scheduling priority assessment and critical state determination.

[0065] The high-frequency load variation condition refers to an operating scenario where load demand changes rapidly and continuously within adjacent scheduling cycles. This scenario is common in microgrids, islanded power supply systems, and emergency power supply systems. Under this condition, energy equipment often experiences frequent power ramp-ups and drops, and its operating state is more likely to approach or enter inefficient or even unstable regions. Therefore, it is necessary to focus on analyzing its edge response behavior.

[0066] Based on the initial operating state set E0, the power output curve P(t) of each energy device and the historical load characteristics H1 are first analyzed for the rate of change within a short time scale. By setting a predetermined time window Δt, the power change amplitude ΔP and the change frequency f are statistically analyzed to identify the time period in which the equipment is in a high-frequency load change state.

[0067] Within the identified high-frequency load variation section, the following edge response characteristic parameters are extracted for each energy device:

[0068] Edge load threshold Le: refers to the critical load value at which equipment transitions from a stable and efficient operating state to a state of significant efficiency decline or unstable response during continuous load variation. This threshold is obtained by analyzing the inflection point relationship between efficiency η(t) and load level.

[0069] Cut-in power gradient Gp: This refers to the slope of the power output of the equipment as it approaches the edge load threshold, reflecting the equipment's adjustment sensitivity within the critical range. A larger power gradient indicates a more aggressive response to dispatch commands, making it more prone to shocks or fluctuations.

[0070] Stabilization time delay Ts: refers to the time required for the equipment to reach stable output again from the start of power adjustment under high-frequency load conditions. It is used to measure the recovery capability of the equipment in the critical operating range.

[0071] The aforementioned edge response characteristic parameters can be obtained through statistical modeling of high-frequency sampled data, including but not limited to: slope fitting, piecewise regression analysis, and calculation of the mean and variance of a sliding window. Different parameter extraction models can be used for different types of energy equipment to adapt to their physical characteristics and control methods.

[0072] After parameter extraction, the edge response feature parameters corresponding to each energy device are combined to construct an edge response feature parameter set R, which is expressed as: R={r1,r2,...,ri,...,rn}, where ri=(Lei,Gpi,Tsi); where n is the number of energy devices, and ri represents the edge response feature description of the i-th energy device under high-frequency variable load conditions.

[0073] Through step S200, the present invention can explicitly quantify the critical operating characteristics that were originally implicit in the operating data, so that subsequent scheduling decisions no longer rely solely on static capacity or average efficiency, but are based on the actual response capability of the equipment under extreme or adverse operating conditions, thereby providing a reliable foundation for building the edge criticality determination model in step S300.

[0074] S300. Based on the edge response characteristic parameter R, calculate the relative response distance D between the current state of each device and the edge response state, and construct the edge critical determination function F(x) to determine whether it is close to the critical operating point.

[0075] First, a response relationship graph is constructed, treating each energy device in the energy supply system as a node. Edges in the graph represent response coupling relationships between any two devices, such as load coordination, time-adjacent adjustment complementarity, or historical state correlation. Coupling relationships can be determined by statistically analyzing the correlation coefficients of load changes for each device over multiple scheduling cycles, and setting a correlation threshold T1 (e.g., T1=0.7). When the Pearson correlation coefficient between two devices exceeds this threshold, an undirected edge is established in the graph.

[0076] For each device node, the current operating status parameters and edge response feature parameters are extracted as node feature inputs. Operating status parameters include, but are not limited to, current power output value, current efficiency, and response time. Edge response feature parameters include edge load threshold, cut-in power gradient, and settling time delay. All features are combined into a vector form and normalized to ensure stability during subsequent graph neural network training.

[0077] The constructed device response relationship graph and node features are input into the graph neural network model. The graph neural network adopts a multi-layer graph convolutional network (GCN) structure. Each layer achieves feature transfer between nodes by aggregating the features of neighboring nodes and updating them with weighted averages.

[0078] Specifically, the computation of each layer in a graph neural network can be described as follows: the feature vector of the current node is aggregated with the feature vectors of its neighboring nodes in a structured weighted manner. The weight parameters are obtained through backpropagation training, and the activation function is the ReLU function. After multi-layer graph convolution operations, each device node will obtain a low-dimensional dense vector representation, called the device's "latent space state vector" at the current time step. The dimension is generally set to 64 or 128 dimensions, adjusted according to task complexity and data volume. After obtaining the latent space state vectors of all devices, the latent space representation of the current state and the latent space representation of its historical edge response states are extracted and vector distance is measured. The relative response distance is calculated using Euclidean distance, which is the square root of the sum of the squared differences of each dimension between the current state vector and the edge state vector, expressed as the relative response distance D between the device's current state and the edge state. The smaller this value, the closer it is to the edge response interval.

[0079] It should be noted that the graph neural network training employs a supervised training method based on edge state labels. The training data consists of device response states from historical scheduling cycles, and the edge state labels are obtained by dividing the probability thresholds generated by the F(x) function. Cross-entropy loss is used as the loss function, and the Adam optimizer is employed for optimization. The learning rate ranges from 0.001 to 0.005, and the number of training epochs is limited to 300. An early stopping strategy is used to prevent overfitting. After training, the model accuracy must be above 90% before it can be used for actual scheduling decisions.

[0080] Based on the aforementioned relative response distance D and edge response feature parameter set, the current state feature vector of the device is extracted. This vector includes: the current power load level (i.e., the ratio of actual load to device capacity), the adjustment speed (i.e., the rate of change of output power per unit time), and the stabilization delay (i.e., the time from the start of power fluctuation to reaching stable output). All indicators are obtained through real-time monitoring and calculation of historical response behavior.

[0081] The aforementioned state feature vectors are standardized, and known edge response state samples from historical operational data are used as the training set to construct a nonlinear classification model using supervised learning methods. In this invention, Support Vector Machine (SVM), Gradient Boosting Decision Tree (GBP), or Multi-layer Perceptron (MLP) are preferably used as the model's basic structure. The model training process includes: fitting state features to edge labels (e.g., whether the model is in an edge response region) using samples from the training set; evaluating model performance using cross-validation; and optimizing the model structure and hyperparameters using accuracy and recall as metrics. After model training is complete, the model is saved as an edge critical decision function F(x).

[0082] During the scheduling decision-making process, the current state feature vector of each energy device is input into the decision function F(x), and the output is a probability value P∈[0,1], representing the probability that the device's current state is in the edge zone. A decision threshold T2 can be set (e.g., T2=0.8). When P≥T2, it is considered "critically close"; otherwise, it is considered "operating stably".

[0083] The classification results will be directly used for subsequent scheduling priority assessment: devices that are critically close to the state will be placed in a lower priority by the scheduling logic to avoid scheduling instructions from further increasing the device load or causing instability risks, thereby improving the robustness and scheduling security of the overall system.

[0084] S400: Map the relative response distance D of all devices to a unified scheduling priority index L, and establish a sorting list Q for multi-source devices.

[0085] In step S300, the relative response distance D of each energy device at the current moment has been obtained. This distance measures the degree of deviation between the current operating state of the device and its corresponding edge response state. Since the edge state distribution and D value range of different types of devices may differ, the relative response distance needs to be standardized first.

[0086] The standardization method employs a minimum-maximum normalization strategy: Let the response distance of the current device be Di, and let the minimum D value of all devices in the system be Dmin and the maximum D value be Dmax. Then, the standardized D value D'i can be expressed as: After standardization, the D' value of all devices is mapped to the interval [0,1], where a smaller value indicates that the device is closer to the edge response state, and a larger value indicates that the operation is safer.

[0087] Based on the standardized D' value, this invention introduces a scheduling priority index L to comprehensively evaluate the scheduling suitability of equipment at the current moment. The construction of the scheduling priority index Li considers the following three factors:

[0088] The relative response distance weighting factor α represents the degree of dominance of D' over L;

[0089] The equipment energy efficiency factor ηi represents the efficiency per unit of energy output; the higher the value, the higher the priority.

[0090] The load adaptability index βi is constructed based on the equipment's historical load fluctuation capability. Equipment with strong fluctuation adaptability should be given higher priority.

[0091] The specific calculation expression of the scheduling priority index Li is as follows: Where α ∈ [0,1] is the response distance sensitive weight set by the user; w1 and w2 are the weighting coefficients of the energy efficiency factor and the load adaptation factor, satisfying w1+w2=1.

[0092] Through the above calculations, each device will obtain a scheduling priority index Li in the range [0,1]. The higher the value of Li, the more suitable the device is to participate in the load response scheduling at the current moment, and the stronger the stability and efficiency guarantee.

[0093] Based on the calculation result of the scheduling priority index L, all devices are sorted in descending order according to the size of Li, generating a scheduling sorting list Q: Q={d1, d2, ..., dn}, where L(d1) ≥ L(d2) ≥ ... ≥ L(dn); where each element di in Q represents an energy device, and the sorting result is used as the basis for device selection and resource allocation priority in subsequent scheduling strategies.

[0094] S500: Obtain the real-time load demand P(t) and the predicted load fluctuation range ΔP(t) at the current moment. Combined with the sorting list Q, select the response equipment combination G according to priority so that it meets the load range demand of P(t) + ΔP(t).

[0095] First, the actual energy load on the user side or in the target area at the current moment is collected as the real-time load demand value, denoted as P(t), in kilowatts or megawatts. This value is collected in real time by metering devices at minute or second intervals and serves as the baseline load input for this round of dispatching.

[0096] To address the fluctuating characteristics of actual load, this implementation method employs a weighted moving average method to predict the short-term load fluctuation range ΔP(t). A time window length of T (e.g., 5 minutes) is set, and the load change values ​​within the T-period before P(t) are assigned progressively decreasing weighting coefficients to calculate the maximum possible load change amplitude within the next Δt (e.g., 1 minute). The prediction model is as follows:

[0097] The predicted fluctuation range ΔP(t) is equal to the sum of the load change rates over the past T time points multiplied by weighting coefficients. For example, if the load increase over the past 5 minutes is +20 kW and the weight sequence is [0.4, 0.3, 0.15, 0.1, 0.05], then ΔP(t) is the maximum weighted increment, which is 20 multiplied by the corresponding cumulative weight value.

[0098] Using the above method, the target load range is obtained as follows: target load range = [P(t), P(t) + ΔP(t)]; where ΔP(t) is the dynamic response margin that the system must reserve to absorb sudden load increases.

[0099] Using the scheduling priority sorting list Q established in step S400 above, all energy devices are traversed in descending order of priority. For each device, its available output capacity is extracted, that is, the maximum regulated output allowed in the current state without exceeding the edge response threshold, denoted as Ai.

[0100] The available output Ai of each device is accumulated to form a candidate device set. When the total accumulated output of the candidate devices is greater than or equal to the upper limit of the target load range P(t) + ΔP(t), the traversal stops, and the initial selection of the candidate set is completed.

[0101] Based on the output value of the device edge critical determination function F(x) determined in steps S300 and S400, devices whose F(x) output is "critically close" are eliminated;

[0102] Exclude devices that are in a shutdown, maintenance, initialization, or off-grid state;

[0103] Constrain the adjustable output range of the equipment, and only retain equipment with an adjustment margin greater than a set threshold (e.g., 10% of the equipment's rated capacity).

[0104] After screening, a set of effective response devices with actual adjustment capabilities and safe operation was formed.

[0105] From the selected set, using a heuristic selection strategy or optimization algorithm (such as a greedy algorithm or integer linear programming), determine the final combination of response devices G based on one of the following objectives:

[0106] Minimum number of devices target: Prioritize combinations that can cover the target load range with the fewest devices;

[0107] Maximum system energy efficiency target: Prioritize average efficiency η i The highest equipment set;

[0108] Minimum average relative response distance target: Prioritize relative response distance D i The device furthest from the edge state.

[0109] Finally, the equipment combination G is obtained, which satisfies the following conditions: the total adjustable output of all equipment in G is ≥ P(t) + ΔP(t) and no equipment in G is in a critical operating state, thus possessing response robustness and adjustment reliability.

[0110] By following the steps above, we can ensure that the selected response equipment is capable of meeting current and future short-term load changes, as well as meeting the requirements for safe operation and energy efficiency, thus providing high-quality input data for the subsequent generation of power dispatch instruction sets.

[0111] S600: Based on the response status, edge response characteristic parameter R, and critical state evaluation value of the devices in the response device combination G, the power command of the devices within the scheduling cycle is corrected to generate the scheduling control set U.

[0112] First, obtain the response status parameters of each energy device in the response device combination within the current scheduling cycle, mainly including the actual output power, instantaneous efficiency, load change rate and settling time at the current moment.

[0113] Subsequently, the edge response feature parameter set corresponding to the device is extracted, specifically including three dimensions: edge load threshold, power cut-in gradient, and stabilization time delay. These parameters have been generated by the aforementioned edge response feature extraction steps.

[0114] The current response state parameters and edge response feature parameters are used as inputs and fed into the constructed edge criticality determination function F(x). This function F(x) is a determination function trained based on a nonlinear supervised classification model (such as support vector machine or multilayer perceptron), and its output is a probability value representing the risk probability that the current device is in an edge critical state.

[0115] Set a critical threshold T2, for example, T = 0.75. When the output value of F(x) is greater than or equal to T2, the device is marked as "in a critical state"; when the output value of F(x) is less than T2, it is marked as "in a safe state". This marking result serves as the critical state identifier value of the device.

[0116] Based on the above critical state indicator values, the response device combinations are divided into two categories:

[0117] The safety regulating equipment, whose current operating state is far from the edge zone, can achieve a large power change within the scheduling cycle;

[0118] The restricted adjustment device is currently in a state close to the edge response state, allowing only small adjustments or maintaining stable operation.

[0119] For each safety regulating device, a power regulation upper limit is set based on its remaining adjustable capacity and historical regulation speed, denoted as Pmax; at the same time, the maximum allowable power change rate is set, denoted as Rmax.

[0120] For restricted regulation equipment, in order to avoid entering the unstable region and limit its regulation capability, its power regulation upper limit is set to 5% to 10% of the current power output value, while the power change rate is limited to within 50% of the change rate of the previous scheduling cycle.

[0121] The original scheduling instruction Pt (original target power value) for each device is modified. The modification method includes two parts:

[0122] Amplitude correction: Limit the original instruction Pt to below the device's corresponding Pmax. If Pt exceeds this range, compress it to Pmax.

[0123] Slope correction: If the power in the previous cycle was The current instruction is Pt, and the rate of change is calculated as follows: If the rate of change is greater than Rmax, then adjust Pt so that the rate of change equals Rmax.

[0124] After completing the above dual correction, a new power instruction is generated, which constitutes the corrected power instruction set for the current scheduling period, denoted as U, where U={u1,u2,...,un}, n is the number of devices, and ui is the correction instruction for the i-th device.

[0125] To ensure that the revised power command does not cause system-level fluctuations or load abrupt changes during the scheduling cycle, this implementation performs timing consistency checks on the instruction set U. The checks include:

[0126] Total regulation range verification: Ensure that the total regulation power of all equipment in combination G still meets the target load range;

[0127] Power surge detection: Ensures that the rate of power change of any device does not exceed the device's operational stability threshold between two consecutive scheduling cycles;

[0128] Edge state re-check: The corrected device state is input again into the edge critical determination function F(x) to ensure that no new device will enter the critical state.

[0129] If the verification passes, the scheduling control set U is directly output as the execution instruction for the current scheduling cycle;

[0130] If the verification fails, update the response status parameters of each device in the current cycle, and return to the aforementioned power correction step to regenerate U until the verification passes.

[0131] Through the above process, risk constraints and dynamic corrections can be made on the power regulation behavior of response equipment during the scheduling cycle, effectively reducing the risk of equipment entering the unstable operating zone and improving the stability and robustness of the entire scheduling process.

[0132] S700: Simulate the generated scheduling control set U to evaluate its robustness and response delay within ΔP(t). If the dynamic robustness threshold is met, output U to the scheduling system; otherwise, return to S400 to adjust the Q list.

[0133] First, a multi-device joint scheduling model is constructed based on the operating characteristic parameters, edge response feature parameters, and physical output model of each energy device in the response device combination. This model adopts a discrete-time stepping method, defining the time span within the scheduling period as T, and the total number of simulation steps as n.

[0134] Using the scheduling control set U as input, the time-varying output trajectory of each device is established. The output power of each device is described by the following variables:

[0135] Initial power value (P0); command setpoint (from U); response delay time (given by the settling time delay in the edge response characteristic parameters); maximum adjustable speed (given by the power change rate constraint). During model runtime, the power changes of each device at each moment are recorded, forming a set of device time-series response trajectories, which serves as the basis for subsequent disturbance simulations.

[0136] Based on the constructed simulation trajectory, a disturbance load sequence ΔP(t) is further introduced to represent the possible load changes within the prediction time range. In this embodiment, three typical disturbance levels within ΔP(t) are selected:

[0137] Highly perturbated sample: ΔP1, take the maximum value of ΔP(t);

[0138] The perturbation sample is ΔP2, which takes the expected value of ΔP(t).

[0139] Low disturbance sample: ΔP3, taking the minimum or negative value of ΔP(t) to represent a load decrease.

[0140] For each disturbance sample ΔPi, it is superimposed on the scheduling model as an external load change input to simulate the actual execution performance of the scheduling control set U under different disturbance levels. This disturbance input can be set in time as abrupt (step), gradual (linear), or random fluctuation.

[0141] It should be noted that the injection methods for the disturbance sample ΔPi include step disturbance (ΔPi is abruptly increased at the m-th second), linear ramp disturbance (gradually introduced with a constant slope within the range of 0 to Δt), and Poisson distribution disturbance (simulating random fluctuations). When selecting the disturbance mode, the preferred method is step disturbance in the microgrid scenario, linear disturbance in the grid load forecasting scenario, and Poisson disturbance in the energy storage fluctuation scenario.

[0142] Statistical analysis of the response behavior was performed on the simulation results of each disturbance, and the following three core performance indicators were extracted:

[0143] Maximum response delay Trmax: refers to the longest response time in seconds required for the total output change of the equipment combination G to first reach the new target load demand from the moment the disturbance input occurs;

[0144] Minimum output stability margin Mmin: refers to the minimum redundancy capacity of the total power after all devices have stabilized their output within the scheduling period, relative to the target value, in kilowatts;

[0145] Response error range E: Used to measure whether the output trajectory exhibits nonlinear fluctuations such as overshoot or undershoot after a disturbance.

[0146] After statistical analysis of the above indicators, a performance evaluation vector of the scheduling control set U under the disturbance range of ΔP(t) is established as the basis for robustness analysis.

[0147] Set dynamic robustness evaluation thresholds, including:

[0148] Response time threshold T3: set to 20 seconds, meaning Trmax must not exceed 20 seconds;

[0149] Safety margin threshold M1: set to 5% of the target load, that is, Mmin must not be less than this value.

[0150] The judgment criteria are as follows:

[0151] If Trmax is less than T3 and Mmin is greater than M1, the scheduling control set U is considered to have sufficient robustness to meet the response requirements under dynamic load disturbances.

[0152] Otherwise, if U is deemed to lack robustness, the device priority ranking result needs to be updated, and the process returns to step S400 to regenerate the ranking list Q, thereby reselecting the device combination and correcting the scheduling instructions.

[0153] Once the scheduling control set U passes the robustness check and is confirmed to be able to operate stably within the load prediction fluctuation range ΔP(t), U can be output to the execution stage and enter the actual scheduling control phase.

[0154] The above methods enable the stability verification and adaptability evaluation of scheduling control strategies under different load change scenarios, significantly enhancing the controllability, response timeliness, and operational safety boundary awareness of multi-source equipment joint scheduling.

[0155] S800. After each scheduling cycle is completed, the actual response data of each device is compared with the predicted edge response feature parameter R, R is updated and fed back to step S200 to complete the closed-loop optimization.

[0156] At the end of each scheduling cycle, the actual operating data of all energy devices participating in the scheduling are recorded to form a sequence of device response behaviors. The data mainly includes the following parameters:

[0157] Actual power variation curve: records the instantaneous output of the equipment from the issuance of the dispatch command to the stable output of power;

[0158] Actual response time: The time from when the command is issued to when the first output reaches the target power, in seconds;

[0159] Actual settling delay: The time it takes for the device's output power to re-enter the stable region after a disturbance;

[0160] Output deviation value: refers to the deviation between the actual stable output power of the equipment and the target value, and the unit is kilowatt.

[0161] The above data all come from equipment-level metering acquisition devices and scheduling instruction execution records, and are synchronized to the scheduling management platform for model updates.

[0162] The collected actual response data is compared one-to-one with the predicted edge response feature parameter set R used before the current scheduling cycle. Typical feature parameters included in R are:

[0163] Edge load threshold (predicted value): The critical load point at which equipment enters the region of rapid efficiency decline;

[0164] Power cut-in gradient (predicted value): the device's power response speed per unit time;

[0165] Stabilization time delay (predicted value): The predicted time required for the disturbance to reach stability.

[0166] Using the three parameters mentioned above as dimensions, the deviations between the actual and predicted values ​​are compared, and a correction factor is calculated. For example, if the actual stable delay of the equipment is more than 10% greater than the predicted delay, the delay parameter will be corrected upwards.

[0167] Based on the deviation between the actual and predicted values, a sliding weighted update algorithm is used to dynamically adjust each feature parameter. The update method is as follows:

[0168] Let Ri be the current edge response feature parameter value, and Ria be the actual observed value in this period;

[0169] Let the weighted update coefficient γ∈[0,1], then the updated Ri is: γ can be adaptively set according to the device type and stability, with a typical value range of 0.1 to 0.3, and is used to balance the weight relationship between historical trends and new observation data. The larger the γ, the higher the model's confidence in the current observation, which is suitable for devices with large fluctuations in response status.

[0170] The updated edge response parameter set Rnew replaces the original R as the edge feature input for the next scheduling cycle, ensuring that the edge state recognition model for each scheduling cycle is based on the latest actual dynamics. The updated edge response feature parameters Rnew are automatically fed back to step S200, i.e., the edge response feature extraction step, and serve as the initial feature input for subsequent scheduling cycles.

[0171] In the next round of scheduling optimization, all R-based computational steps (such as the calculation of edge response distance D, the output of critical decision function F(x), power command correction, and robustness simulation) are executed based on Rnew, realizing adaptive correction of scheduling parameters and forming a data-driven closed-loop optimization scheduling control system.

[0172] This closed-loop update mechanism can continuously calibrate the ability to judge the actual operating boundary of the equipment, improve the real-time performance, accuracy and stability of the scheduling strategy, and avoid misjudgment and scheduling deviation caused by long-term use of static parameters.

[0173] Example 2, please refer to Figure 2 As shown in this embodiment, an energy production scheduling optimization decision-making system includes:

[0174] The data acquisition module obtains the operating parameter sequence of various energy equipment in the target energy supply area, including power output curves, response time, current efficiency, and historical load characteristics, and constructs the initial operating state set E0;

[0175] The edge response feature modeling module extracts edge response feature parameters R of each energy device under high-frequency load variation conditions based on the initial operating state set E0, including edge load threshold, cut-in power gradient, and stabilization time delay.

[0176] The graph neural network recognition module calculates the relative response distance D between the current state of each device and the edge response state based on the edge response feature parameter R, and constructs the edge critical determination function F(x) to determine whether it is close to the critical operating point.

[0177] The scheduling priority evaluation module maps the relative response distance D of all devices to a unified scheduling priority index L, and establishes a sorting list Q of multi-source devices;

[0178] The equipment combination selection module obtains the real-time load demand P(t) and the predicted load fluctuation range ΔP(t) at the current moment. Combined with the sorting list Q, it selects the response equipment combination G according to priority so that it meets the load range demand of P(t) + ΔP(t).

[0179] The power command correction and generation module, based on the response status of the devices in the response device combination G, the edge response characteristic parameter R and the critical state evaluation value of F(x), corrects the power command of the devices within the scheduling cycle and generates the scheduling control set U.

[0180] The robustness verification module simulates the generated scheduling control set U to evaluate its robustness and response delay within ΔP(t). If the dynamic robustness threshold is met, U is output to the scheduling system; otherwise, the module returns to the scheduling priority evaluation module to adjust the Q list.

[0181] The closed-loop optimization module compares the actual response data of each device with the predicted edge response feature parameter R after each scheduling cycle, updates R, and feeds it back to the edge response feature modeling module to complete the closed-loop optimization.

[0182] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A decision-making method based on energy production scheduling optimization, characterized in that: include: S100: Obtain the operating parameter sequence of each energy device in the target energy supply area, including power output curve, response time, current efficiency, and historical load characteristics, and construct the initial operating state set E0; S200. Based on the initial operating state set E0, extract the edge response characteristic parameter set R of each energy device under high frequency load conditions, including edge load threshold, cut-in power gradient, and stabilization time delay. S300. Based on the edge response characteristic parameter set R, calculate the relative response distance D between the current state of each energy device and the edge response state, and construct the edge critical judgment function F(x) to determine whether it is close to the critical operating point. The calculation of the relative response distance D between the current state of each energy device and its edge response state includes: constructing a device response relationship graph based on the edge response feature parameter set R, with energy devices as nodes and response correlations between devices as edges, and using the current operating state parameters and corresponding edge response feature parameters of each energy device as node feature inputs; performing graph neural network feature learning on the device response relationship graph, mapping the node features to a unified latent space through multi-layer graph convolution operations to obtain the latent space state vector corresponding to each energy device at the current time; and calculating the distance between the latent space state vector corresponding to each energy device at the current time and its corresponding edge response state vector in the latent space to obtain the relative response distance D representing the degree to which the current state of each energy device deviates from the edge response state. The steps for constructing the edge criticality determination function F(x) include: extracting the state feature vectors of each energy device at the current moment based on the relative response distance D and the edge response feature parameter set R of the device, including power load level, adjustment speed and stabilization delay; normalizing the state feature vectors and training a nonlinear classification model using a supervised learning method in combination with historical edge response samples to establish a mapping relationship between the device operating state and the critical state; embedding the nonlinear classification model into the edge criticality determination function F(x), which outputs whether the current device operating state is close to its corresponding edge operating region, and the output of the function F(x) is a binary label or probability value; classifying and labeling the device operating state according to the output of F(x), marking it as critically close or operating stably; S400: Map the relative response distance D of all energy devices to a unified scheduling priority index L, and establish a sorting list Q for multi-source devices; S500: Obtain the real-time load demand P(t) and the predicted load fluctuation range ΔP(t) at the current moment. Combine the sorting list Q and select the response equipment combination G according to priority so that the response equipment combination G meets the load range demand of P(t) + ΔP(t). S600: Based on the response status of the devices in the response device combination G, the critical state evaluation value of the edge response characteristic parameter set R and F(x), the power command of the devices within the scheduling cycle is corrected to generate the scheduling control set U. S700: Simulate the generated scheduling control set U to evaluate its robustness and response delay within ΔP(t). If the dynamic robustness threshold is met, output U to the scheduling system; otherwise, return to S400 to adjust the sorting list Q. S800. After each scheduling cycle is completed, the actual response data of each energy device is compared with the predicted edge response feature parameter set R, the edge response feature parameter set R is updated and fed back to step S200 to complete the closed-loop optimization.

2. The energy production scheduling optimization decision-making method according to claim 1, characterized in that: Generating the scheduling control set U includes the following steps: Obtain the response status parameters of each energy device in the response device combination G during the current scheduling cycle, the corresponding edge response characteristic parameter set R, and the evaluation value of the edge critical decision function F(x), and use the output of F(x) as the device critical state identifier value; Based on the critical state identifier value of the equipment, the energy equipment in the response equipment combination G is divided into safe regulating equipment and restricted regulating equipment, and the corresponding power regulation upper limit value and power change rate constraint are determined respectively. Based on the power regulation upper limit, power change rate constraint, and edge response characteristic parameter set R, the original power dispatch instructions of each energy device are corrected in amplitude and slope to generate a dispatch control set U. Perform a timing consistency check on the scheduling control set U within the scheduling cycle. If the check passes, output the scheduling control set U for execution; otherwise, update the device's response status parameters and re-correct the power command.

3. The energy production scheduling optimization decision-making method according to claim 1, characterized in that: The steps for simulating the generated scheduling control set U include: The scheduling control set U is input into the multi-device joint scheduling model to construct the time-series simulation trajectory of the response behavior of each device within the scheduling cycle; Introducing the predicted load fluctuation range ΔP(t), and using high, medium, and low disturbance samples within the predicted fluctuation range as inputs, the response results of the dispatch control set U under different load disturbance conditions are simulated; Statistical analysis was performed on the response time, stability delay, and output deviation of each device in the simulation results to obtain the maximum response delay and minimum output stability margin of the scheduling control set U under different disturbance conditions. The robustness of the scheduling control set U within the predicted load fluctuation range ΔP(t) is determined based on whether the maximum response delay is less than the set response time limit threshold T3 and whether the minimum output stability margin is greater than the safety margin threshold M1.

4. The energy production scheduling optimization decision-making method according to claim 1, characterized in that: Step S500 involves selecting the response device combination G, which includes: accumulating output from high-priority devices based on the sorting list Q until P(t) + ΔP(t) is satisfied; and eliminating devices that are in a critical near-critical state or have insufficient regulation capacity.

5. The energy production scheduling optimization decision-making method according to claim 1, characterized in that: Step S800, updating the edge response feature parameter set R, includes: comparing the collected actual response data with the predicted edge response feature parameter set R used before the current scheduling cycle; comparing the deviation between the actual and predicted values ​​and calculating the correction factor; and dynamically adjusting each edge response feature parameter using a sliding weighted update algorithm based on the deviation between the actual and predicted values: Let Ri be the current edge response feature parameter value, and Ria be the actual observed value in this cycle; let the weighted update coefficient γ∈[0,1], then the updated edge response feature parameter value is: .

6. An energy production scheduling optimization decision-making system, used to implement the energy production scheduling optimization decision-making method according to any one of claims 1-5, characterized in that: include: Data acquisition module: acquires the operating parameter sequence of each energy device in the target energy supply area, including power output curve, response time, current efficiency, historical load characteristics, and constructs the initial operating state set E0; Edge response feature modeling module: Based on the initial operating state set E0, extract the edge response feature parameter set R of each energy device under high frequency load conditions, including edge load threshold, cut-in power gradient, and stabilization time delay; The graph neural network recognition module calculates the relative response distance D between the current state of each energy device and the edge response state based on the edge response feature parameter set R, and constructs the edge critical determination function F(x) to determine whether it is close to the critical operating point. Scheduling priority evaluation module: Maps the relative response distance D of all energy devices to a unified scheduling priority index L, and establishes a sorting list Q for multi-source devices; Equipment combination selection module: Obtain the real-time load demand P(t) and the predicted load fluctuation range ΔP(t) at the current moment, and combine it with the sorting list Q to select the response equipment combination G according to priority, so that the response equipment combination G meets the load range demand of P(t) + ΔP(t); Power command correction and generation module: Based on the response status of the devices in the response device combination G, the edge response characteristic parameter set R and the critical state evaluation value of F(x), the power command of the devices within the scheduling cycle is corrected to generate the scheduling control set U; Robustness verification module: Simulates the generated scheduling control set U to evaluate the robustness and response delay of the scheduling control set U within ΔP(t). If the dynamic robustness threshold is met, output U to the scheduling system; otherwise, return to the scheduling priority evaluation module to adjust the sorting list Q. Closed-loop optimization module: After each scheduling cycle, the actual response data of each energy device is compared with the predicted edge response feature parameter set R, the edge response feature parameter set R is updated and fed back to the edge response feature modeling module to complete the closed-loop optimization.

Images