A cloud-edge-end collaborative distributed optimization scheduling method for high proportion of renewable energy

Abstract

The application is suitable for power distribution network system dispatching technical field, and provides a cloud edge end collaborative distributed optimization dispatching method for high proportion renewable energy, which comprises the following steps: multiple time and space scale prediction is carried out through a cloud dispatching platform, and a robustness budget parameter is constructed, an optimization algorithm container image is actively pushed to the edge based on the prediction information, and an optimization strategy package containing target weight and boundary constraint is generated by fusing the robustness parameter and is issued; after each edge autonomous cluster receives the strategy and the image, the algorithm is loaded and a distributed robust optimization model aiming at maximizing immunity uncertainty is constructed based on local data to generate a local optimal dispatching instruction set and issue execution; while the end side device executes the instruction, it monitors the deviation and actively reports the exception to trigger a local optimization closed loop. The application can effectively improve the economy, safety and agility of high proportion renewable energy power distribution network operation.

Description

Technical Field

[0001] This invention relates to the field of power distribution network system dispatching technology, and in particular to a cloud-edge-device collaborative distributed optimization dispatching method for high proportions of renewable energy. Background Technology

[0002] With the widespread integration of high-proportion distributed renewable energy into the power grid, the uncertainty and control complexity of system operation have increased significantly. Current optimization scheduling methods mostly adopt a centralized architecture, the core of which relies on a cloud-based central controller for global data aggregation and unified optimization calculations. This system first collects monitoring data and forecast information from all distributed power sources and loads across the entire network, constructs and solves a large-scale deterministic or stochastic optimization model at the cloud center, calculates scheduling instructions for all controllable devices, and then distributes these instructions to each terminal for execution through the communication network. Edge computing nodes in this architecture primarily undertake data acquisition, forwarding, and instruction execution functions; they themselves lack the ability to make autonomous optimization decisions based on local information.

[0003] However, the real-time uploading and centralized processing of massive amounts of data across the entire network under the aforementioned architecture places enormous pressure on cloud computing power and communication bandwidth, resulting in delays in generating and distributing optimization decisions, making it difficult to meet the rapid response requirements for sudden events such as photovoltaic power drops. Furthermore, a single point of failure in the central server could directly paralyze the entire scheduling system. Secondly, deterministic optimization relies heavily on a single prediction scenario, and its scheduling scheme is highly susceptible to safety issues such as voltage exceeding limits and power backfeeding when prediction errors are large. While stochastic optimization and other methods that consider uncertainty can be used, their computational complexity is higher, and they typically rely on precise probability distribution assumptions, which are difficult to meet in practice. Therefore, existing technologies struggle to collaboratively ensure the economy, security, and response speed of system operation in complex and uncertain environments.

[0004] In view of this, a cloud-edge-device collaborative distributed optimization scheduling method for high proportion of renewable energy is proposed. Summary of the Invention

[0005] This invention provides a cloud-edge-device collaborative distributed optimization scheduling method for high proportions of renewable energy, which is used to solve the problem of difficulty in coordinating the economics, security and response speed of system operation in complex and uncertain environments.

[0006] This invention provides a cloud-edge-device collaborative distributed optimization scheduling method for high-proportion renewable energy, which is collaboratively executed by a cloud scheduling platform, at least one edge autonomous cluster, and multiple end-side execution terminals deployed in the distribution network, including: The cloud-based scheduling platform performs the following steps: Multi-temporal and spatial scale predictions are made based on the acquired data, and robust budget parameters for describing source load uncertainty are constructed based on the prediction error quantification results. The optimal pre-push decision is calculated based on the prediction information, and the corresponding optimized algorithm container image is proactively deployed to the target edge autonomous cluster based on the optimal pre-push decision; Based on the robust budget parameters and global optimization objectives, an optimization strategy package is generated and distributed to each edge autonomous cluster. The edge autonomous cluster performs the following steps: Receive the optimization strategy package and the deployed optimization algorithm container image issued by the cloud scheduling platform, and simultaneously obtain local data; Load the optimization algorithm container image to start the preset distributed robust optimization algorithm; Based on the objective function weights, global constraint boundaries, and robust budget recommendations in the optimization strategy package, and combined with the local data, the preset distributed robust optimization algorithm is used to construct and solve an optimization model aimed at maximizing the ability to resist uncertainty, thereby generating an optimal set of dispatch instructions suitable for the local power distribution network area. The optimal scheduling instruction set is issued to the end-side execution terminals within the jurisdiction; The terminal executes the following steps: Receive and execute the corresponding control instructions from the optimal scheduling instruction set; It monitors its own operating status and command execution deviations in real time. If a continuous deviation is detected that exceeds the threshold, it will proactively report the abnormal alarm information to the edge autonomous cluster to trigger a new round of local optimization.

[0007] Furthermore, the step of performing multi-temporal-scale predictions based on the acquired data, and constructing robust budget parameters to describe source load uncertainties based on the prediction error quantification results, includes: Acquire historical meteorological data, historical power data, and real-time data streams, and extract time-series features, meteorological features, and date features to construct training and inference datasets; Prediction models with different time resolutions are used to perform short-term and ultra-short-term predictions on the constructed dataset. The short-term prediction is used to output the sequence of predicted values ​​for new energy output and load in the first future time period, and the ultra-short-term prediction is used to perform rolling corrections on the short-term prediction results and output the predicted power values ​​in the second future time period. The prediction error is calculated based on the ultra-short-term predicted value and the actual measured value, and the prediction error is modeled as a random distribution. Based on the random distribution characteristics of the prediction error, an uncertainty set is constructed with the ultra-short-term prediction value as the center and the robust budget parameter as the radius.

[0008] Furthermore, the step of calculating the optimal predictive decision based on the prediction information includes: Obtain the task trigger probability, network latency status, and real-time available storage capacity of each edge autonomous cluster from the prediction information; Based on the task trigger probability, network latency status, and real-time available storage capacity, the pre-push value for each target edge autonomous cluster is calculated for each optimized algorithm container image to be pre-pushed. With the goal of maximizing the overall pre-push value and the available storage capacity of each edge autonomous cluster as a constraint, a resource pre-push optimization model is constructed. Solve the resource pre-push optimization model to obtain the optimal pre-push decision that indicates whether each optimization algorithm container image should be pre-pushed to each target edge autonomous cluster.

[0009] Furthermore, the pre-push value of each optimized algorithm container image to be pre-pushed for each target edge autonomous cluster is calculated using the following dynamic value evaluation model: in: Indicates time mirror Pre-push to edge autonomous clusters The predictive value; This represents an edge autonomous cluster within a specific future time window, obtained based on predictive information. Image execution is required. The probability of the corresponding computational task; This indicates the transition from cloud-based scheduling platforms to edge autonomous clusters. Network latency; This is the delay sensitivity coefficient; Indicates mirror image Size; Represents an edge autonomous cluster In time Available storage capacity; Indicates to the edge autonomous cluster Fixed communication costs for pushing images; , , , These are configurable coefficients used to balance the weights of various factors.

[0010] Furthermore, the step of generating an optimization strategy package based on the robust budget parameters and the global optimization objective includes: Determine the weighting coefficients of the objective function used to balance system operating costs and operational risk immunity; Determine the global security constraint boundaries for the operation of the distribution network; Based on the robustness budget parameters and combined with global security margin analysis, the recommended robustness budget values ​​to be distributed to each edge autonomous cluster are determined. The objective function weight coefficients, global safety constraint boundaries, and robust budget recommendations are encapsulated into an optimization strategy package.

[0011] Furthermore, based on the robustness budget parameters and combined with global security margin analysis, the recommended robustness budget values ​​to be distributed to each edge autonomous cluster are determined, including: Obtain the source load prediction error distribution for each edge autonomous cluster in the corresponding region, and use it as an assessment of the local uncertainty level of the region; Based on global power flow analysis of the distribution network, assess the safety margin of critical lines and nodes; Based on the local uncertainty level and global security margin of each region, a suitable robust budget recommendation value is determined for each edge autonomous cluster through preset decision rules. Among them, a relatively conservative robust budget recommendation value is assigned to regions with high local uncertainty levels and tight global security margins.

[0012] Furthermore, the construction of the optimization model includes: A robust optimization model is established with the objective of maximizing the robustness budget parameter. The mathematical expression is as follows: in: The robustness budget parameter to be maximized represents the scheduling scheme's immunity to source load uncertainties; The vector of decision variables includes energy storage charging and discharging power and controllable load adjustment. The local system operating cost is calculated based on the weight coefficients of the objective function. This represents the baseline operating cost obtained based on deterministic forecast data; This is a preset percentage increase in operating costs; To account for the impact of uncertainty, the first Item execution constraints, For constraint sets; For the first Predicted values ​​for each source of uncertainty; , Decision variables The lower and upper limits of technology.

[0013] Furthermore, the operational constraints include power balance constraints, node voltage safety constraints, line transmission capacity constraints, and distributed resource operation constraints.

[0014] Furthermore, the built-in mathematical programming solver is invoked to solve the optimization model, and the optimal scheduling instruction set is generated based on the output optimal decision variables and the corresponding maximum achievable robustness level.

[0015] Furthermore, the system monitors its own operational status and instruction execution deviations in real time. If a continuous deviation exceeds a threshold, it proactively reports an anomaly alarm to its respective edge autonomous cluster to trigger a new round of local optimization, including: The actual operating electrical quantities of the local equipment are collected in real time and compared with the corresponding set values ​​in the received optimal scheduling instruction set to calculate the instruction execution deviation. Within a set rolling time window, continuously determine whether the instruction execution deviation exceeds a preset deviation threshold; If, within the rolling time window, the number of times the instruction execution deviation exceeds the deviation threshold reaches a preset number, it is determined that there is a continuous deviation, and an abnormal alarm message containing device identifier, deviation amount, timestamp, and abnormality type is generated. The abnormal alarm information is actively reported to the corresponding edge autonomous cluster, so that after receiving the abnormal alarm information, the edge autonomous cluster will re-execute the optimization calculation based on the latest local data to generate updated scheduling instructions.

[0016] As can be seen from the above technical solutions, the present invention has the following advantages: This invention utilizes a cloud-based scheduling platform to perform multi-temporal and spatiotemporal scale predictions and construct robust budget parameters. Based on the predicted information, it proactively pre-pushes optimization algorithm container images to the edge and integrates robust parameters to generate optimization strategy packages containing target weights and boundary constraints, which are then distributed. Each edge autonomous cluster receives the strategy and images, loads the algorithm, and constructs a distributed robust optimization model based on local data to maximize immunity to uncertainty. It generates a local optimal scheduling instruction set and distributes it for execution. While executing instructions, edge devices monitor deviations and proactively report anomalies to trigger local optimization loop closure. This invention firstly deploys critical computing capabilities to the network edge in advance through proactive pre-push of container images, effectively avoiding task cold start delays and significantly improving the system's response speed to load fluctuations. Secondly, by employing a distributed architecture, it decomposes global optimization into multiple locally autonomous optimization problems that can be processed in parallel, greatly reducing cloud computing and communication pressure and avoiding the risk of centralized single-point failures. Thirdly, the edge clusters perform robust optimization based on information gap decision theory, and their scheduling schemes can proactively resist power fluctuations under worst-case scenarios within a given cost tolerance, significantly enhancing the system's ability to operate safely in the face of uncertainty. This invention, through the deep integration of prediction-driven resource pre-provisioning, distributed autonomous decision-making, and robust optimization techniques, can effectively improve the economy, security, and agility of high-proportion renewable energy distribution network operation. Attached Figure Description

[0017] Figure 1 This is a system architecture diagram of a cloud-edge-device collaborative distributed optimization scheduling method for high-proportion renewable energy in this invention; Figure 2 This is a schematic diagram of the execution process of the cloud scheduling platform in this invention; Figure 3 This is a schematic diagram of the execution process of the edge autonomous cluster in this invention; Figure 4 This is a schematic diagram of the execution process of the terminal on the end side in this invention. Detailed Implementation

[0018] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] Example 1 The method of this invention is based on a system architecture comprising a cloud-based scheduling platform, at least one edge autonomous cluster, and multiple end-side execution terminals deployed in the power distribution network. Please refer to... Figure 1 The cloud-based scheduling platform, acting as the global information processing and strategy guidance center, is responsible for macro-level forecasting, resource pre-configuration, and strategy generation. The edge autonomous cluster, serving as the regional autonomy and decision-making center, receives cloud resources and strategies and performs distributed optimization calculations based on local information. The end-side execution terminal, as the sensing and execution unit, is responsible for high-fidelity execution of instructions and proactive status feedback. Through collaborative efforts, these three components jointly achieve rapid, robust, and autonomous operation of the power distribution network.

[0020] The method of the present invention will be described in detail below: Please see Figure 2 The cloud-based scheduling platform performs the following steps: 101. Perform multi-temporal-scale predictions based on the acquired data, and construct robust budget parameters to describe the uncertainty of source loads based on the prediction error quantification results; Multi-temporal and spatial scale prediction here refers to making predictions for variables such as photovoltaic power output and load demand at different time granularities and spatial ranges. The robustness budget parameter is a non-negative scalar, defining the radius of the range around the predicted value where the actual power is allowed to fluctuate. Its physical meaning represents the maximum severity of prediction errors that subsequent optimized scheduling schemes can tolerate.

[0021] 1. Acquire historical meteorological data, historical power data, and real-time data streams, and extract time-series features, meteorological features, and date features to construct training and inference datasets; 2. Prediction models with different time resolutions are used to perform short-term and ultra-short-term predictions on the constructed dataset. The short-term prediction is used to output the predicted value sequence of new energy output and load in the first future time period, while the ultra-short-term prediction is used to perform rolling correction on the short-term prediction results and output the predicted power value in the second future time period. 3. Calculate the prediction error based on the ultra-short-term predicted value and the actual measured value, and model the prediction error as a random distribution; 4. Based on the random distribution characteristics of prediction errors, an uncertainty set is constructed with the ultra-short-term prediction value as the center and the robust budget parameter as the radius.

[0022] Specifically, the system aggregates historical meteorological data, such as irradiance and temperature, historical photovoltaic power output and load power data, as well as real-time sensor data streams from the edge. Through feature engineering methods, a feature set for model training and inference is extracted and constructed from the raw data. For example, the rolling average and standard deviation of historical power are calculated as time-series features describing its variation patterns; the gradient change of irradiance is calculated as a meteorological feature reflecting instantaneous weather changes; and date types, such as weekdays and holidays, are encoded as date features. A sequence-to-sequence long short-term memory network model based on an attention mechanism is used for short-term prediction. This model operates on an hourly cycle, outputting a sequence of predicted renewable energy output and load values ​​for the next 24 hours. Subsequently, an online learning support vector regression model is used for ultra-short-term prediction. This model operates on a minute cycle, taking short-term predicted values, recent actual power sequences, and real-time meteorological deviations as inputs, and performs rolling corrections on the power values ​​for the next 15 minutes, outputting a final predicted value with higher temporal resolution.

[0023] The system continuously compares ultra-short-term forecasts with actual measurements and calculates the forecast error. The prediction error at time t is modeled as a random variable, and based on historical data analysis, it is assumed to follow a normal distribution with a mean of zero. Its mathematical expression is: in: for Prediction error at time, for The power value actually measured at any given time. for Ultra-short-term power prediction values ​​at time t. The mean is 0 and the variance is The normal distribution for The standard deviation corresponding to the time-time prediction error.

[0024] Based on the aforementioned probability distribution model of prediction error, for each future scheduling time... Define a set of uncertainties This set contains all values ​​for a given robust budget parameter. The mathematical expression for the possible actual power value is as follows: in: for The actual power value that may occur at any given time. for Ultra-short-term power prediction values ​​at any given time.

[0025] 102. Calculate the optimal pre-deployment decision based on the prediction information, and actively deploy the corresponding optimization algorithm container image to the target edge autonomous cluster based on the optimal pre-deployment decision; This step proactively deploys potentially needed optimization algorithms to the edge in advance to reduce task startup latency and improve system response speed. Specifically, the pre-pushing of container images based on predictive information involves value assessment and optimization decisions, including the following sub-steps: 1. Obtain the task trigger probability, network latency status, and real-time available storage capacity of each edge autonomous cluster from the prediction information; 2. Based on the task trigger probability, network latency status, and real-time available storage capacity, calculate the pre-push value for each target edge autonomous cluster for each optimized algorithm container image to be pre-pushed; 3. With the goal of maximizing the overall pre-push value and the available storage capacity of each edge autonomous cluster as a constraint, a resource pre-push optimization model is constructed; 4. Solve the resource pre-push optimization model to obtain the optimal pre-push decision that indicates whether the container images of each optimization algorithm should be pre-pushed to each target edge autonomous cluster.

[0026] Specifically, the task trigger probability refers to the probability derived from the prediction result of step 101 (the photovoltaic output fluctuation rate exceeds the set threshold) that a specific edge cluster needs to execute a corresponding optimization calculation task within a future time window; network latency refers to the network communication latency from the cloud to each edge cluster; and the real-time available storage capacity of each edge autonomous cluster refers to the current available storage space of each edge cluster. Pre-push value is a quantitative indicator used to comprehensively measure the expected benefits and costs of pushing a specific image to a specific edge cluster. The calculation of this value comprehensively considers the urgency of task execution, network transmission efficiency, and storage resource pressure. The resource pre-push optimization model is a constrained 0-1 knapsack problem. Its goal is to select a set of images to push under the storage capacity constraints of each edge cluster, maximizing the total pre-push value of all pushed images. The mathematical expression of this model is: in: This indicates the total number of images to be pre-produced; This represents the total number of edge autonomous clusters; Indicates time mirror Pre-push to edge autonomous clusters The predictive value; For 0-1 decision variables, a value of 1 indicates mirroring. Push to edge cluster A value of 0 indicates that no push notifications will be sent. Indicates mirror image Size; Represents an edge autonomous cluster In time Available storage capacity; This is the proportion of storage space reserved for local operation of edge clusters. The above optimization model is solved using a dynamic programming algorithm to obtain the optimal set of 0-1 decision variables, i.e., the optimal pre-push decision. This decision clarifies which edge clusters each image should be pushed to.

[0027] In this embodiment, the pre-push value of each optimized algorithm container image to be pre-pushed for each target edge autonomous cluster is calculated using the following dynamic value evaluation model: in: Indicates time mirror Pre-push to edge autonomous clusters The predictive value; This represents an edge autonomous cluster within a specific future time window, obtained based on predictive information. Image execution is required. The probability of the corresponding computational task; This indicates the transition from cloud-based scheduling platforms to edge autonomous clusters. Network latency; This is the delay sensitivity coefficient; Indicates mirror image Size; Represents an edge autonomous cluster In time Available storage capacity; Indicates to the edge autonomous cluster Fixed communication costs for pushing images; , , , These are configurable coefficients used to balance the weights of various factors.

[0028] The model's first term represents the task triggering benefit, the second the network latency benefit, the third the storage pressure cost, and the fourth the fixed communication cost. After obtaining the optimal pre-push decision through solving, the cloud scheduling platform proactively pushes the corresponding optimized algorithm container image from the image repository to the local storage of each target edge autonomous cluster via the network, completing the pre-deployment of computing resources.

[0029] 103. Generate optimization strategy packages based on robust budget parameters and global optimization objectives, and distribute them to each edge autonomous cluster; This step encapsulates the global optimization goals and security requirements, along with the system uncertainty understanding constructed in step 101, into a lightweight instruction framework and distributes it to each edge autonomous cluster. This grants the edge sufficient autonomous decision-making space while ensuring global coordination and security. This includes the following sub-steps, with sub-step 3 further refined into three sub-processes: 1. Determine the weighting coefficients of the objective function used to balance system operating costs and operational risk immunity; 2. Determine the global security constraint boundaries for the operation of the distribution network; 3. Based on robust budget parameters and combined with global security margin analysis, determine the recommended robust budget values ​​to be distributed to each edge autonomous cluster; a. Obtain the source load prediction error distribution for each edge autonomous cluster in the corresponding region, as an assessment of the local uncertainty level of the region; b. Assess the safety margin of critical lines and nodes based on global power flow analysis of the distribution network; c. Based on the local uncertainty level and global security margin of each region, determine an appropriate robust budget recommendation value for each edge autonomous cluster through preset decision rules. Among them, a relatively conservative robust budget recommendation value is assigned to regions with high local uncertainty levels and tight global security margins.

[0030] 4. Encapsulate the objective function weight coefficients, global safety constraint boundaries, and robust budget recommendations into an optimization strategy package.

[0031] Specifically, balancing system operating costs and operational risk immunity refers to the trade-off between reducing network losses and curtailment costs in distribution network dispatching and enhancing the system's ability to withstand safety risks such as voltage exceedances and power backflow. The objective function weighting coefficients quantify this trade-off; their values ​​are set by the cloud-based dispatching platform based on global operating modes and electricity pricing policies, and then distributed to edge clusters to guide the construction of their local optimization models' objective functions. Global security constraint boundaries refer to the technical limitations that must be met to ensure the safe operation of the entire distribution network. These mainly include the upper and lower limits of power exchange with the upstream main grid, transmission capacity limits of key trunk lines, and the allowable voltage fluctuation range of the entire network; these boundaries are uniformly calculated and determined by the cloud based on grid topology and equipment parameters.

[0032] Based on the global robust budget parameters constructed in step 101, and considering the specific circumstances of each region, the cloud assigns appropriate robust budget recommendation values. Specifically, the cloud analyzes historical and real-time prediction error data for each edge cluster's region, calculating indicators characterizing the severity of local photovoltaic and load fluctuations, serving as an assessment of the region's local uncertainty level. The cloud performs global power flow calculations, analyzing the differences between the load rate of each critical line and the voltage amplitude of each node and their safe upper or lower limits at the current operating point, thereby quantifying the safety margin of the entire network and its local areas. The cloud implements a decision rule that for a region, if its local uncertainty level assessment value is high (large fluctuations) and its global safety margin is tight (close to the voltage upper limit), a relatively small (more conservative) budget recommendation value is assigned. This means that the optimization scheme of this edge cluster must be able to cope with more severe worst-case scenarios to prioritize global safety. Conversely, for regions with low uncertainty and sufficient safety margin, a relatively large (more aggressive) budget recommendation value can be assigned, giving its optimization model greater flexibility to pursue economic efficiency. Finally, the cloud scheduling platform encapsulates the parameters determined in the above steps, such as the weight coefficients, global constraint boundaries, and robust budget recommendations tailored for each edge cluster, into a structured data packet, namely the optimization strategy packet, and distributes it to the corresponding autonomous edge clusters through the communication network.

[0033] Please see Figure 3The edge autonomous cluster performs the following steps: 201. Receive the optimization strategy package and the deployed optimization algorithm container image from the cloud scheduling platform, and simultaneously obtain local data; The local data here includes: the latest power forecast value obtained from the local ultra-short-term forecast module, and real-time operating status quantities such as voltage, current, and power collected from smart meters, distribution terminals, and various inverter controllers.

[0034] 202. Load the optimization algorithm container image to start the preset distributed robust optimization algorithm; The edge autonomous cluster utilizes its built-in container runtime environment to load and instantiate optimization algorithm container images pre-push from the cloud. This container image is a software package encapsulating a complete runtime environment and algorithm logic, pre-configured with a distributed robust optimization algorithm based on information gap decision theory. The loading process essentially involves rapidly deploying and starting the algorithm and its dependent libraries on edge computing nodes, forming an independent, executable optimization computation instance. The pre-configured distributed robust optimization algorithm defines a set of uncertainties centered on the predicted value and seeks a scheduling scheme that ensures the system satisfies operational constraints under all possible uncertainty scenarios within this set, while maximizing the range of uncertainties (robustness level) that the scheme can immunize against.

[0035] 203. Based on the objective function weights, global constraint boundaries, and robust budget recommendations in the optimization strategy package, and combined with local data, a pre-defined distributed robust optimization algorithm is used to construct and solve an optimization model aimed at maximizing the ability to resist uncertainty, thereby generating an optimal set of dispatch instructions suitable for the local power distribution network area. 1. Establish a robust optimization model with the objective of maximizing the robustness budget parameter. The mathematical expression is as follows: in: The robustness budget parameter to be maximized represents the scheduling scheme's immunity to source load uncertainties; The vector of decision variables includes energy storage charging and discharging power and controllable load adjustment. The local system operating cost is calculated based on the weight coefficients of the objective function. This represents the baseline operating cost obtained based on deterministic forecast data; This is a preset percentage increase in operating costs; To account for the impact of uncertainty, the first Item execution constraints, For constraint sets; For the first Predicted values ​​for each source of uncertainty; , Decision variables The lower and upper limits of technology.

[0036] 2. Operational constraints include power balance constraints, node voltage safety constraints, line transmission capacity constraints, and distributed resource operation constraints.

[0037] Specifically, power balance constraints require that at each node and during each scheduling period, the sum of power injected by all power sources (including the main grid, distributed photovoltaic, and energy storage discharge) equals the sum of power consumed by all loads (including base load, controllable load, and energy storage charging). Node voltage safety constraints require that the voltage amplitude of all nodes be maintained within the range allowed by technical standards. Line transmission capacity constraints require that the apparent power or active power flowing through each line does not exceed its thermal stability limit or stable transmission limit. Distributed resource operation constraints are physical limitations on specific controllable devices. For example, for energy storage devices, these include upper and lower limits for charging and discharging power, as well as dynamic changes and upper and lower limits for state of charge; for adjustable loads, these include constraints on power regulation range and regulation rate.

[0038] 3. Call the built-in mathematical programming solver to solve the optimization model, and generate the optimal scheduling instruction set based on the output optimal decision variables and the corresponding maximum realizable robustness level.

[0039] Worst-case scenario construction and model transformation: For different types of operational constraints, the actual values ​​of the uncertainty sources are replaced with their worst-case values ​​under a given robust budget recommendation. For constraints that must be strictly satisfied, the worst-case scenario is the minimum actual photovoltaic output; for constraints to prevent exceeding limits, the worst-case scenario is the maximum actual photovoltaic output. Substituting these worst-case values ​​into the uncertainty constraints of the original model transforms them into deterministic constraints that are only related to the decision variables.

[0040] Solving deterministic optimization problems: The edge cluster invokes its built-in or container-encapsulated mathematical programming solver to numerically solve the transformed model, which aims to maximize performance and includes all deterministic constraints. The solver outputs the optimal solution, including the optimal decision variables (the sequence of optimal power setpoints for each device) and the maximum robustness achievable under this scheme. Based on the optimal decision variables, the system automatically generates a structured optimal scheduling instruction set containing device identifiers, timestamps, and power commands.

[0041] 204. Issue the optimal scheduling instruction set to the end-side execution terminals within the jurisdiction; After generating the local optimal scheduling instruction set, the edge autonomous cluster distributes it to each of its managed end-side execution terminals via standard industrial communication protocols. The distribution process ensures the accuracy, completeness, and timeliness of the instructions. For example, the charging and discharging power instructions for the energy storage system are distributed to the energy storage converter controller, and the setpoints for adjustable loads are distributed to the corresponding load controllers.

[0042] Please see Figure 4 The terminal executes the following steps: 301. Receive and execute the corresponding control instructions from the optimal scheduling instruction set; End-side execution terminals, such as energy storage converters, photovoltaic inverters, and controllable load controllers, receive a set of structured optimal scheduling instructions from their respective edge autonomous clusters via a communication network. The terminal devices parse this instruction set, extracting control instructions that match their own device identifier. These instructions typically include setpoints or adjustments for active and reactive power within a specific time period. Subsequently, the terminal device's built-in control logic or programmable logic controller converts these setpoints into specific device drive signals and controls the operation of power electronic switches or actuators, thereby precisely adjusting its output or absorbed power to ensure that its actual operating state tracks and approximates the requirements of the scheduling instructions.

[0043] 302. Monitor its own operating status and command execution deviation in real time. If a continuous deviation is detected that exceeds the threshold, it will proactively report the abnormal alarm information to the edge autonomous cluster to trigger a new round of local optimization.

[0044] 1. Real-time acquisition of local actual operating electrical quantities, comparison with corresponding set values ​​in the received optimal scheduling instruction set, and calculation of instruction execution deviation; 2. Within a set rolling time window, continuously determine whether the instruction execution deviation exceeds a preset deviation threshold; 3. If the number of times the instruction execution deviation exceeds the deviation threshold within the rolling time window reaches a preset number, it is determined that there is a continuous deviation, and an abnormal alarm message containing device identifier, deviation amount, timestamp and abnormality type is generated. 4. Proactively report abnormal alarm information to the corresponding edge autonomous cluster, so that after receiving the abnormal alarm information, the edge autonomous cluster will re-execute optimization calculation based on the latest local data to generate updated scheduling instructions.

[0045] Specifically, the terminal equipment uses its built-in sensors or measurement units to collect local actual operating electrical quantities in real time at a high sampling rate, such as actual output active power. reactive power Or the actual voltage and current values. Simultaneously, read the corresponding setpoints from the received scheduling instructions, such as the power setting. Instruction execution deviation Through comparative calculations, for example, the deviation calculation formula for active power is as follows: in: express Active power command execution deviation at any given time.

[0046] The terminal device maintains a rolling time window of a preset length, and within this window, it processes each calculated deviation value. The system will determine whether the deviation exceeds the preset threshold for this type of device or instruction. The judgment condition is Within this rolling time window, the counter increments each time a deviation exceeding the threshold event is detected. If the cumulative number of deviation exceeding the threshold reaches or exceeds a preset trigger count within this window, the terminal device determines that it has a persistent deviation. If the determination is successful, the terminal generates a structured anomaly alarm message. This message includes at least: the unique identifier of the terminal device that generated the alarm, the currently calculated deviation amount, the timestamp of the event, and the anomaly type determined based on the deviation characteristics. Upon receiving this alarm message, the edge cluster treats it as an important local state update event. Based on this new information and the latest local measurement and prediction data, the edge cluster will restart the local optimization calculation process starting from step 202, quickly solve for a new robust optimization scheduling scheme that considers the current abnormal operating conditions, and generate and issue an updated scheduling instruction set.

[0047] It is understood that those skilled in the art can combine various implementation methods in the above embodiments under the guidance of the above examples to obtain technical solutions with multiple implementation methods.

[0048] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A cloud-edge-device collaborative distributed optimization scheduling method for high-proportion renewable energy, characterized in that, The system is executed collaboratively by a cloud-based scheduling platform, at least one edge autonomous cluster, and multiple end-side execution terminals deployed in the power distribution network, including: The cloud-based scheduling platform performs the following steps: Multi-temporal and spatial scale predictions are made based on the acquired data, and robust budget parameters for describing source load uncertainty are constructed based on the prediction error quantification results. The optimal pre-push decision is calculated based on the prediction information, and the corresponding optimized algorithm container image is proactively deployed to the target edge autonomous cluster based on the optimal pre-push decision; Based on the robust budget parameters and global optimization objectives, an optimization strategy package is generated and distributed to each edge autonomous cluster. The edge autonomous cluster performs the following steps: Receive the optimization strategy package and the deployed optimization algorithm container image issued by the cloud scheduling platform, and simultaneously obtain local data; Load the optimization algorithm container image to start the preset distributed robust optimization algorithm; Based on the objective function weights, global constraint boundaries, and robust budget recommendations in the optimization strategy package, and combined with the local data, the preset distributed robust optimization algorithm is used to construct and solve an optimization model aimed at maximizing the ability to resist uncertainty, thereby generating an optimal set of dispatch instructions suitable for the local power distribution network area. The optimal scheduling instruction set is issued to the end-side execution terminals within the jurisdiction; The terminal executes the following steps: Receive and execute the corresponding control instructions from the optimal scheduling instruction set; It monitors its own operating status and command execution deviations in real time. If a continuous deviation is detected that exceeds the threshold, it will proactively report the abnormal alarm information to the edge autonomous cluster to trigger a new round of local optimization.

2. The cloud-edge-device collaborative distributed optimization scheduling method for high-proportion renewable energy as described in claim 1, characterized in that, The process of making multi-temporal-scale predictions based on the acquired data, and constructing robust budget parameters to describe source load uncertainties based on the prediction error quantification results, includes: Acquire historical meteorological data, historical power data, and real-time data streams, and extract time-series features, meteorological features, and date features to construct training and inference datasets; Prediction models with different time resolutions are used to perform short-term and ultra-short-term predictions on the constructed dataset. The short-term prediction is used to output the sequence of predicted values ​​for new energy output and load in the first future time period, and the ultra-short-term prediction is used to perform rolling corrections on the short-term prediction results and output the predicted power values ​​in the second future time period. The prediction error is calculated based on the ultra-short-term predicted value and the actual measured value, and the prediction error is modeled as a random distribution. Based on the random distribution characteristics of the prediction error, an uncertainty set is constructed with the ultra-short-term prediction value as the center and the robust budget parameter as the radius.

3. The cloud-edge-device collaborative distributed optimization scheduling method for high-proportion renewable energy as described in claim 1, characterized in that, The step of calculating the optimal predictive decision based on the predicted information includes: Obtain the task trigger probability, network latency status, and real-time available storage capacity of each edge autonomous cluster from the prediction information; Based on the task trigger probability, network latency status, and real-time available storage capacity, the pre-push value for each target edge autonomous cluster is calculated for each optimized algorithm container image to be pre-pushed. With the goal of maximizing the overall pre-push value and the available storage capacity of each edge autonomous cluster as a constraint, a resource pre-push optimization model is constructed. Solve the resource pre-push optimization model to obtain the optimal pre-push decision that indicates whether each optimization algorithm container image should be pre-pushed to each target edge autonomous cluster.

4. The cloud-edge-device collaborative distributed optimization scheduling method for high-proportion renewable energy as described in claim 1, characterized in that, The pre-push value of each optimized algorithm container image to be pre-pushed for each target edge autonomous cluster is calculated using the following dynamic value evaluation model: in: Indicates time mirror Pre-push to edge autonomous clusters The predictive value; This represents an edge autonomous cluster within a specific future time window, obtained based on predictive information. Image execution is required. The probability of the corresponding computational task; This indicates the transition from cloud-based scheduling platforms to edge autonomous clusters. Network latency; This is the delay sensitivity coefficient; Indicates mirror image Size; Represents an edge autonomous cluster In time Available storage capacity; Indicates to the edge autonomous cluster Fixed communication costs for pushing images; , , , These are configurable coefficients used to balance the weights of various factors.

5. The cloud-edge-device collaborative distributed optimization scheduling method for high-proportion renewable energy as described in claim 1, characterized in that, The generation of the optimization strategy package based on the robust budget parameters and the global optimization objective includes: Determine the weighting coefficients of the objective function used to balance system operating costs and operational risk immunity; Determine the global security constraint boundaries for the operation of the distribution network; Based on the robustness budget parameters and combined with global security margin analysis, the recommended robustness budget values ​​to be distributed to each edge autonomous cluster are determined. The objective function weight coefficients, global safety constraint boundaries, and robust budget recommendations are encapsulated into an optimization strategy package.

6. The cloud-edge-device collaborative distributed optimization scheduling method for high-proportion renewable energy as described in claim 1, characterized in that, Based on the robustness budget parameters and combined with global security margin analysis, the recommended robustness budget values ​​to be distributed to each edge autonomous cluster are determined, including: Obtain the source load prediction error distribution for each edge autonomous cluster in the corresponding region, and use it as an assessment of the local uncertainty level of the region; Based on global power flow analysis of the distribution network, assess the safety margin of critical lines and nodes; Based on the local uncertainty level and global security margin of each region, a suitable robust budget recommendation value is determined for each edge autonomous cluster through preset decision rules. Among them, a relatively conservative robust budget recommendation value is assigned to regions with high local uncertainty levels and tight global security margins.

7. The cloud-edge-device collaborative distributed optimization scheduling method for high-proportion renewable energy as described in claim 1, characterized in that, The construction of the optimization model includes: A robust optimization model is established with the objective of maximizing the robustness budget parameter. The mathematical expression is as follows: in: The robustness budget parameter to be maximized represents the scheduling scheme's immunity to source load uncertainties; The vector of decision variables includes energy storage charging and discharging power and controllable load adjustment. The local system operating cost is calculated based on the weight coefficients of the objective function. This represents the baseline operating cost obtained based on deterministic forecast data; This is a preset percentage increase in operating costs; To account for the impact of uncertainty, the first Item execution constraints, For constraint sets; For the first Predicted values ​​for each source of uncertainty; , Decision variables The lower and upper limits of technology.

8. The cloud-edge-device collaborative distributed optimization scheduling method for high-proportion renewable energy as described in claim 7, characterized in that, The operational constraints include power balance constraints, node voltage safety constraints, line transmission capacity constraints, and distributed resource operation constraints.

9. The cloud-edge-device collaborative distributed optimization scheduling method for high-proportion renewable energy as described in claim 7, characterized in that, The built-in mathematical programming solver is invoked to solve the optimization model, and the optimal scheduling instruction set is generated based on the output optimal decision variables and the corresponding maximum achievable robustness level.

10. The cloud-edge-device collaborative distributed optimization scheduling method for high-proportion renewable energy as described in claim 1, characterized in that, The system monitors its own operational status and instruction execution deviations in real time. If a continuous deviation exceeds a threshold, it proactively reports an anomaly alarm to its respective edge autonomous cluster to trigger a new round of local optimization, including: The actual operating electrical quantities of the local equipment are collected in real time and compared with the corresponding set values ​​in the received optimal scheduling instruction set to calculate the instruction execution deviation. Within a set rolling time window, continuously determine whether the instruction execution deviation exceeds a preset deviation threshold; If, within the rolling time window, the number of times the instruction execution deviation exceeds the deviation threshold reaches a preset number, it is determined that there is a continuous deviation, and an abnormal alarm message containing device identifier, deviation amount, timestamp, and abnormality type is generated. The abnormal alarm information is actively reported to the corresponding edge autonomous cluster, so that after receiving the abnormal alarm information, the edge autonomous cluster will re-execute the optimization calculation based on the latest local data to generate updated scheduling instructions.

Images