A micro-grid hierarchical rolling collaborative optimization scheduling method under high-altitude extreme weather environment

Abstract

The application relates to the field of power system automation, and specifically discloses a micro-grid hierarchical rolling collaborative optimization scheduling method under an extreme weather environment on a plateau. The method realizes accurate perception and prediction of weather mutations by constructing a multi-source weather monitoring network and an attention mechanism neural network system; realizes global optimization and local response collaboration by adopting a hierarchical scheduling architecture at the system level, the micro-grid level and the device level; establishes a multi-objective optimization function considering economic, reliable and resilient targets, and introduces robust constraints such as a plateau attenuation coefficient and temperature compensation; uses an improved multi-agent deep deterministic policy gradient algorithm for rolling solution, and triggers an emergency mechanism when a mutation is identified; and realizes model adaptive iteration based on operation data through federated learning. The application can significantly improve the accuracy of weather mutation identification and the real-time performance of scheduling, enhance the resilience and power supply guarantee rate of the micro-grid on the plateau, and ensure the safe operation of key loads under extreme environments.

Description

Technical Field

[0001] This application belongs to the field of microgrid optimization scheduling technology, specifically involving a hierarchical rolling collaborative optimization scheduling method for microgrids under extreme meteorological conditions in high-altitude areas. Background Technology

[0002] Plateau regions (altitude ≥ 2500 meters) are characterized by abundant sunshine but drastic radiation fluctuations, high wind speeds and variable directions, extremely low temperatures and large diurnal temperature ranges, and frequent extreme weather events (such as sudden blizzards and short-term heavy rainfall). At the same time, these regions are mostly far from the main power grid, and microgrids often operate in an isolated grid or weak grid-connected mode of "photovoltaic + wind power + energy storage + emergency power". They are the core carriers for ensuring the power supply for local residents' lives, industrial production (such as lithium carbonate mining), and critical loads (such as clinics and communication base stations).

[0003] Current microgrid dispatching methods are mostly designed for plains or conventional meteorological environments. When applied to extreme weather scenarios in plateau regions, they suffer from the following core drawbacks: 1. Poor weather adaptability: Traditional dispatching methods do not fully consider the sudden and nonlinear characteristics of extreme weather changes on the plateau. They rely solely on short-term weather forecasts or linear regression of single historical data, making it difficult to capture instantaneous fluctuations in photovoltaic power caused by cloud changes, sudden increases and decreases in wind power output caused by strong winds, and equipment failure risks caused by blizzards. This can easily lead to delayed dispatching decisions, resulting in power outages or equipment overloads.

[0004] 2. Insufficient dispatching accuracy: Meteorological stations are sparse in plateau areas, and meteorological forecasting errors are large. Traditional static dispatching methods cannot dynamically adjust dispatching strategies according to real-time meteorological changes, resulting in low renewable energy absorption rates, unreasonable charging and discharging of energy storage equipment, and even problems such as low-temperature degradation of energy storage and decreased equipment insulation, which affect the stability of microgrid operation.

[0005] 3. Poor hierarchical coordination: Existing hierarchical scheduling mostly adopts a "master-slave" architecture. Information exchange between the upper-level system-level scheduling and the lower-level microgrid-level scheduling is not timely. It does not take into account the differences in the adjustment capabilities of different equipment (photovoltaic, wind power, energy storage, emergency power supply) under extreme weather conditions in plateau areas. It cannot achieve multi-energy synergy and complementarity, and it is difficult to respond quickly in the event of sudden extreme weather changes to ensure power supply to critical loads.

[0006] 4. Insufficient resilience: In the face of high impact low probability (HILF) weather events in high-altitude areas, traditional dispatching methods rely on pre-set plans and lack dynamic adaptive capabilities. In the event of a sudden disaster, the system experiences long power outages and slow recovery, making it difficult to meet the high demand for power supply reliability in high-altitude areas.

[0007] Therefore, given the unique characteristics of extreme weather changes on plateaus, there is an urgent need for a microgrid scheduling method that can achieve real-time sensing, hierarchical coordination, and rolling optimization, in order to solve the problems of poor adaptability, low accuracy, and insufficient resilience in existing technologies, and ensure the safe and stable operation of microgrids under extreme weather conditions. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the prior art and provide a hierarchical rolling optimization scheduling method for microgrids under extreme weather conditions in plateau regions. By constructing a closed-loop architecture of "weather perception - hierarchical decision-making - rolling optimization", the invention enables dynamic adaptive scheduling of microgrids under extreme weather change scenarios, improves power supply reliability and renewable energy absorption rate, reduces operating costs, and enhances system resilience to quickly respond to various disturbances caused by weather changes.

[0009] This invention provides a hierarchical rolling optimization scheduling method for microgrids under extreme weather conditions in high-altitude areas. The microgrid includes distributed power sources (photovoltaic modules, wind turbines), energy storage systems (low-temperature resistant lithium iron phosphate energy storage, supercapacitors), emergency power sources (diesel generators), critical loads (medical, communication), general loads, and a dispatch control center. The method includes the following steps: Step 1: Constructing a perception and prediction system for extreme weather changes on the plateau 1.1 Establish a multi-source meteorological monitoring network: Within the microgrid coverage area, meteorological monitoring nodes are deployed to collect meteorological parameters such as altitude, temperature, wind speed, irradiance, snowfall, and sandstorm intensity in real time, combined with satellite cloud image data. At the same time, equipment environmental parameters such as snow cover of photovoltaic modules and icing status of wind turbines are monitored to form a multi-dimensional meteorological-equipment status dataset. 1.2 Extreme Weather Change Identification: Based on multi-source monitoring data, an extreme weather change identification model is constructed using an attention mechanism combined with an LSTM neural network. Change thresholds are set (such as sudden increase in wind speed ≥8m / s, sudden decrease in irradiance ≥50%, sudden decrease in temperature ≥10℃). Extreme weather change events (strong wind, blizzard, sandstorm, instantaneous irradiance fluctuation) are identified in real time, and the change level (general, severe, extreme) is output. 1.3 Short-term rolling forecast: For the identified meteorological abrupt events, a short-term rolling forecast model is constructed by coupling the WRF meteorological model and the microgrid equipment simulation model. With a rolling step of 15 minutes, the model predicts the trend of meteorological parameter changes and the output fluctuation range of photovoltaic and wind power in the next 1-4 hours. At the same time, it predicts the state of charge (SOC) of the energy storage system and the equipment failure rate, providing data support for subsequent hierarchical scheduling and mitigating the cumulative effect of forecast errors.

[0010] Step 2: Construct a hierarchical scheduling architecture for the microgrid A three-layer, two-phase hierarchical architecture is adopted to achieve synergy between global optimization and local response, specifically including: 2.1 Upper layer: System-level collaborative optimization layer (dispatch and control center): With multiple objectives such as minimizing the overall operating cost of the microgrid, maximizing power supply reliability, and maximizing the renewable energy absorption rate, it coordinates the energy interaction, backup sharing, and power allocation of the tie line with the main grid (if any) of multiple lower-level microgrid units. At the same time, it embeds constraints such as equipment overload, voltage safety, and low-temperature operation of energy storage to generate a global dispatch benchmark scheme. 2.2 Middle Layer: Microgrid-level Autonomous Optimization Layer (each microgrid unit controller): Receives the global baseline scheme from the upper layer, combines the real-time meteorological monitoring data of this unit with the equipment operating status (such as photovoltaic snow melting status, wind turbine operating status), optimizes the output of distributed power sources and energy storage charging and discharging strategies with the goal of minimizing the operating cost of this unit and maximizing the load power supply guarantee rate, generates local scheduling schemes, and feeds back the adjustment capacity margin of this unit to the upper layer; 2.3 Lower layer: Equipment-level execution control layer (each equipment controller): Receives local scheduling plans from the middle layer, controls the operation of photovoltaic modules (snow melting, tilt adjustment), wind turbines (speed, yaw adjustment), energy storage systems (charging and discharging power, constant temperature control), and emergency power supplies (start and stop status) in real time, and collects equipment operation data in real time, feeding it back to the middle and upper layers to form an information closed loop; 2.4 Two-stage interaction mechanism: Under normal meteorological conditions, the upper-level global optimization is the main focus, and the middle-level local adjustment is the auxiliary focus; when extreme weather changes occur, the mechanism switches to prioritize the middle-level local response and supplement the upper-level global coordination to ensure the real-time nature and flexibility of scheduling decisions.

[0011] Step 3: Construct a hierarchical rolling optimization scheduling model 3.1 Objective Function Setting: A multi-objective optimization function is adopted, taking into account the three major objectives of economy, reliability, and resilience, as detailed below: (1) Economic objectives: Minimize the total operating cost of the microgrid, including the generation cost of distributed power sources, the charging and discharging loss cost of energy storage equipment, the start-up and shutdown cost of emergency power sources and fuel costs, equipment maintenance costs, and the cost of purchasing electricity from the main grid (if connected to the grid).

[0012] in It is the cost of photovoltaic power generation. It is the cost of wind power generation. Costs associated with energy storage charging and discharging losses. For emergency power supply operating costs, For equipment maintenance costs, Costs related to line and equipment wear and tear.

[0013] Reliability objectives: Maximize the power supply reliability of critical loads, ensure uninterrupted power supply to critical loads such as medical and communication systems during extreme weather changes, and reduce system downtime.

[0014] in For the normal power supply time of critical loads, Total scheduling time, target value .

[0015] Resilience Target: Maximize the Microgrid Resilience Index (MRI), which covers six dimensions including power availability, fault recovery time, and critical load availability, to enhance the system's ability to withstand extreme weather impacts and recover quickly.

[0016] in For power availability, The time required for fault recovery For critical load guarantee rate, Weights for each dimension, target value .

[0017] 3.2 Constraint Setting: Considering the characteristics of extreme weather on the plateau, the following constraints are set: (1) Power balance constraint: In each level of dispatch, the sum of distributed power output, energy storage charging and discharging power, emergency power output, and main grid power purchase (if any) is equal to the sum of load power consumption and equipment power loss.

[0018] in Photovoltaic plateau attenuation coefficient (when irradiance ≤ 200W / ㎡), =0.5-0.7; under normal irradiation Take 0.85). The plateau attenuation coefficient of the wind turbine (at wind speeds of 12-18 m / s). =0.6; at normal wind speed =0.8), For energy storage discharge power, For energy storage discharge power, For emergency power output, For critical load power, For normal load consolidation rate, Normal load interruption coefficient (during extreme sudden changes) =0.8), Power purchased / sold from the main grid (positive for purchased power, negative for sold power).

[0019] (2) Equipment operation constraints: The output of photovoltaic modules shall not exceed the maximum available output under extreme weather conditions (considering snow accumulation and irradiance fluctuations); the output of wind turbines shall not exceed the wind speed limit (during strong winds). (80% of rated output to avoid overload from strong winds); the state of charge (SOC) of the energy storage system should be maintained between 20% and 80%. In low-temperature environments (≤-10℃), the lower limit of SOC should be increased to 30% to avoid excessive discharge at low temperatures leading to capacity decay; the emergency power supply should not be started or stopped more than 3 times per day, and the start-up response time should be [not specified]. 30s.

[0020] (3) Meteorological adaptation constraints: In the event of extreme weather changes, priority should be given to ensuring the power supply of critical loads, limiting the power of non-critical loads, and at the same time, equipment protection strategies (such as photovoltaic snow melting and wind turbine shutdown) should be activated.

[0021] (4) Robust boundary constraints for strong uncertainty: By setting robust boundaries, the scheduling scheme can still operate stably within the range of uncertainty. Specifically, it includes robust boundary constraints for photovoltaic output, robust boundary constraints for wind turbine output, robust boundary constraints for load demand, and robust boundary constraints for power balance.

[0022] Robust Boundary Constraints for Photovoltaic Output: Taking into account uncertainties such as irradiance, robust boundary constraints for photovoltaic output are set: , in To contribute to the actual development of photovoltaics Contribute to photovoltaic forecasting, For robust margin of photovoltaic output, under normal weather conditions Generally, during mutation Severe / extreme mutations .

[0023] Wind turbine output robustness boundary constraints: Taking into account wind speed uncertainty, output robustness boundaries are set as follows: , in For the actual output of the wind turbine, To predict the output of wind turbines, For wind turbine output robustness margin, under normal weather conditions Generally, during mutation Severe / extreme mutations When strong winds suddenly change, the lower bound is set to 0.

[0024] Robust load boundary constraints: To address the large load fluctuations characteristic of high-altitude areas, robust load boundaries are set to prioritize and protect critical loads. , , in (Critical load robustness margin, small fluctuations) (Robust margin for ordinary loads, with large fluctuations) Ensures stable power supply to critical loads and adapts to fluctuations for ordinary loads.

[0025] Power balance robust constraint: Combining the above robust boundary, the power balance constraint is modified to ensure that power balance is still satisfied within the range of uncertainty. , By using robust boundary upper and lower limit constraints, power balance can still be achieved under extreme uncertainty scenarios, thus avoiding power outages.

[0026] Step 4: Layered rolling optimization solution and execution 4.1 Solution Algorithm Selection: The improved Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm is adopted, combined with the first-order optimality condition of KKT, to transform the mid-level local optimization problem into equivalent constraints, simplify the solution process, improve the solution efficiency, and ensure the real-time performance of the scheduling scheme. 4.2 Rolling optimization execution: With a rolling step size of 15 minutes in real time and an optimization time domain of 4 hours, after each rolling step is completed, the meteorological monitoring data, equipment operation data and short-term forecast results are updated, the hierarchical optimization model is resolved, a new scheduling scheme is generated, and the scheduling scheme of the previous cycle is replaced, so as to realize the rolling closed loop of "prediction-optimization-execution-update". 4.3 Extreme Weather Sudden Change Response: When an extreme weather sudden change event is detected, the emergency dispatch mechanism is triggered. The middle-level microgrid unit quickly adjusts the dispatch strategy, prioritizes the discharge of the energy storage system and the start-up and shutdown of the emergency power supply to ensure the power supply to the critical load. The upper-level dispatch center coordinates the energy complementarity of each microgrid unit, optimizes the allocation of backup resources, and shortens the fault recovery time. 4.4 Operational Status Feedback and Adjustment: The device-level execution layer provides real-time feedback on the operational status of the equipment. If equipment failure occurs (such as wind turbine icing and shutdown, or energy storage capacity decay), the middle and upper layers adjust the scheduling scheme in a timely manner to ensure the stable operation of the microgrid.

[0027] Step 5: Scheduling effect evaluation and adaptive optimization 5.1 Calculation of evaluation indicators: After each rolling cycle, calculate the microgrid operating cost, critical load power supply guarantee rate, renewable energy absorption rate, system resilience index (MRI), average outage time, and other evaluation indicators. 5.2 Adaptive Adjustment: Compare the evaluation indicators with the preset thresholds. If the preset requirements are not met, adjust the objective function weights of the hierarchical optimization model (such as increasing the reliability and resilience weights during extreme weather), the constraint thresholds, and the prediction model parameters, optimize the solution algorithm, and improve the scheduling adaptability. 5.3 Model Iteration and Update: Based on long-term operating data, federated learning method is used to update the extreme weather change identification model, short-term rolling prediction model and hierarchical optimization model, improve the adaptability of the model to the extreme weather characteristics of different areas of the plateau, and realize the continuous optimization of the scheduling strategy.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Strong adaptability to extreme weather: Through a multi-source meteorological monitoring network and abrupt change identification model, it can capture extreme meteorological abrupt events on the plateau in real time. Combined with short-term rolling forecasts, it can predict meteorological changes and equipment operating status in advance, solve the problem of response lag in traditional scheduling methods, and adapt to extreme scenarios such as strong winds, blizzards, and radiation fluctuations on the plateau. 2. High scheduling accuracy: It adopts a "three-layer two-stage" hierarchical architecture to achieve coordination between the upper-layer global optimization and the lower-layer local response. Combined with the rolling optimization mechanism, the scheduling plan is updated every 15 minutes, which effectively alleviates the accumulation of prediction errors, optimizes the energy storage charging and discharging strategy, improves the renewable energy consumption rate, and reduces equipment wear. 3. Significantly enhanced system resilience: By incorporating resilience indicators into a multi-objective optimization function and combining it with an emergency dispatch mechanism, backup resources can be quickly activated in the event of extreme weather changes to ensure power supply to critical loads, shorten power outage time, and enable the microgrid to maintain more than 90% power supply reliability in extreme weather events that occur once every 50 years. 4. Balancing economy and reliability: By optimizing the output of distributed power sources, the charging and discharging of energy storage, and the start-up and shutdown strategies of emergency power sources, the operating cost of the microgrid is minimized while ensuring power supply reliability. Compared with traditional dispatching methods, the operating cost is reduced by 8%-12%, and the power supply guarantee rate for critical loads is increased to over 98%. 5. Wide adaptability: It can be applied to plateau microgrids (island or weak grid connection mode) with different altitudes and load types. It can be implemented through software algorithm optimization without large-scale modification of existing equipment. It has strong engineering practicality and replicability and can be promoted and applied to extreme environment areas such as the Qinghai-Tibet Plateau and Liangshan high-altitude mountainous areas. Attached Figure Description

[0029] Figure 1 This is an overall flowchart of the microgrid hierarchical rolling optimization scheduling method of the present invention; Figure 2 This is a schematic diagram of the "three-layer, two-stage" hierarchical scheduling architecture of the present invention; Figure 3 This is a schematic diagram of the structure of the plateau extreme weather change identification and short-term rolling prediction model of the present invention. Detailed Implementation

[0030] Example 1: This invention provides a hierarchical rolling collaborative optimization scheduling method for microgrids under extreme weather conditions in high-altitude areas. The method includes the following specific steps: The first step is to construct a plateau extreme weather change perception and prediction system. This system consists of a multi-source meteorological data acquisition module, an environmental parameter mapping module, an attention mechanism recognition layer, and a long short-term memory prediction module. The multi-source meteorological data acquisition module is physically connected to a sensor array in the plateau microgrid area to acquire real-time total horizontal radiation, direct radiation, diffuse radiation, instantaneous wind speed, ambient temperature, and atmospheric pressure. The environmental parameter mapping module inputs the collected meteorological parameters into a preset equipment physical simulation model and outputs predicted values ​​for photovoltaic module backsheet temperature, wind turbine hub torque, and energy storage battery cell temperature rise rate. The attention mechanism recognition layer performs correlation weighting calculations between real-time meteorological parameters and historical normal fluctuation sequences to extract meteorological feature vectors. These feature vectors are input to the input of the long short-term memory neural network, and the recognition layer outputs level identifiers for general, severe, and extreme changes. The short-term rolling prediction module uses a preset time step as the sampling period and couples the hidden layer state of the long short-term memory neural network with the WRF meteorological numerical forecast results to output the distributed power output envelope, equipment random failure probability distribution, and energy storage charge state trajectory for future periods.

[0031] The second step is to construct a hierarchical dispatch architecture for the microgrid. This architecture adopts a three-layer logical topology consisting of a system-level coordination layer, a microgrid-level autonomous layer, and a device-level execution layer. The system-level coordination layer is connected to the microgrid-level autonomous layer via a communication network, and the microgrid-level autonomous layer is connected to the device-level execution layer via a fieldbus. The two-stage interaction mechanism includes a global optimization stage under normal operating conditions and a local response stage under extreme operating conditions. Under normal operating conditions, the system-level coordination layer acts as the logical master node and issues power allocation baseline commands downward. After receiving the baseline commands, the microgrid-level autonomous layer performs deviation correction based on local real-time monitoring data and feeds back the regulation capacity margin to the system-level coordination layer. Under extreme operating conditions, logical control is switched to the microgrid-level autonomous layer. Based on the mutation level flag output by the sensing system, the microgrid-level autonomous layer directly calls the preset emergency response action library and issues control commands to the device-level execution layer, while simultaneously transmitting state change information back to the system-level coordination layer.

[0032] The third step is to construct a hierarchical rolling optimization scheduling model. This model consists of a multi-objective optimization function, a set of physical operational constraints, and robust boundary constraints with strong uncertainty. The multi-objective optimization function includes economic, reliability, and resilience objectives. The economic objective is formed by the linear summation of the following components: photovoltaic power generation operating cost, wind power generation operating cost, energy storage system charge-discharge cycle loss, emergency diesel generator start-up and fuel cost, equipment preventive maintenance cost, accelerated depreciation loss of equipment in high-altitude environments, and the cost of purchasing electricity from the main power grid.

[0033] The reliability objective is achieved by calculating the load shortage rate, which is the ratio of the amount of critical load unmet, the amount of general load interruption, to the total load demand. The resilience objective consists of multi-dimensional resilience indicators, including power availability, automatic fault recovery time, critical load continuity, frequency deviation after system disturbances, voltage fluctuation rate, and energy reserve adequacy. Each dimension is assigned a corresponding weight coefficient, and the microgrid resilience index is calculated through weighted summation.

[0034] The physical operational constraints include power balance constraints, distributed generation output constraints, energy storage system state constraints, and emergency power response constraints. The left side of the power balance constraint equation is the algebraic sum of photovoltaic (PV) output, wind power output, energy storage discharge power, emergency power output, and purchased power, where PV and wind power outputs are multiplied by the previously calculated plateau load power attenuation coefficient. The right side of the equation is the algebraic sum of energy storage charging power, load demand at each level, and system transmission losses. Load demand at each level is obtained by multiplying the total load by the corresponding load interruption coefficient. Energy storage system state constraints include upper and lower limits of state of charge, charging and discharging power change rate constraints, and continuous charging and discharging time constraints.

[0035] Robust boundary constraints for strong uncertainties are implemented by introducing an output robustness margin parameter. The predicted output, output robustness margin, and actual output satisfy an algebraic inequality relationship, meaning the actual output lies within a closed interval defined by the predicted output and the robustness margin. A mapping relationship is established between the output robustness margin and the meteorological abrupt change level output by the sensing system; as the abrupt change level increases, the robustness margin parameter increases synchronously. For extreme strong wind abrupt change conditions, the model sets the lower bound logical value of the turbine output to zero.

[0036] The fourth step involves solving the scheduling model using an improved deep reinforcement learning algorithm. The algorithm employs a multi-agent deep deterministic policy gradient architecture, with each controlled device corresponding to one agent. Each agent contains a policy network and an evaluation network. The policy network outputs the optimal scheduling action based on the current environment state, and the evaluation network scores the action and feeds it back to the policy network. The algorithm introduces the first-order optimality condition of KKT, transforming the objective function of autonomous optimization at the mid-level microgrid level into an equivalent algebraic constraint at the lower-level device execution layer, which is then embedded into the deep reinforcement learning experience replay pool.

[0037] The rolling optimization execution mechanism updates over a preset time step. At the start of each cycle, the algorithm obtains the latest meteorological parameter forecasts and equipment operating status, restarts the deep reinforcement learning algorithm to solve for and generate a sequence of scheduling plans for the future prediction time domain. The system only issues and executes the first time-domain instruction in this sequence; subsequent instructions are overwritten by the newly generated plan in the next cycle. When a sudden change response logic is triggered, the system interrupts the current rolling plan, and the logic jumps to the emergency response module, prioritizing the closure of the output circuits of the energy storage system and emergency power supply, and disconnecting the power distribution branches of non-critical loads.

[0038] The fifth step involves adaptive iterative optimization based on scheduling performance evaluation metrics. These metrics include preset cost deviation thresholds, resilience index lower limits, and frequency safety stability domains. At the end of each scheduling cycle, the system substitutes actual operational data into the evaluation metrics for verification. If a metric exceeds a preset threshold, the system initiates objective function weight adjustment logic, reconstructing the model by increasing the weight coefficients of the unmet criteria. Simultaneously, a cross-regional scheduling model collaborative update link is constructed using a federated learning protocol. Each microgrid node trains its model parameters locally, encrypting only the gradient updates after training before uploading them to the cloud parameter server. The cloud parameter server weights and aggregates the gradients from each node, generating globally optimal scheduling model parameters and distributing them to each microgrid node, achieving model parameter iteration without relying on raw privacy data.

[0039] In this invention, the high-altitude load power attenuation coefficient is calculated using an empirical formula based on altitude, air density, ambient temperature, and equipment rated parameters. The photovoltaic attenuation coefficient is negatively correlated with altitude, while the wind turbine attenuation coefficient is positively correlated with air density. The load interruption coefficient is preset according to the load level, with a zero interruption coefficient corresponding to a level one load.

[0040] In the sensing system, the sampling frequency of the multi-source meteorological data acquisition module matches the control frequency of the microgrid-level autonomous layer. The attention mechanism recognition layer adopts a multi-head attention structure to extract meteorological abrupt change features from both the temporal and spatial dimensions. The long short-term memory prediction module introduces a residual connection structure to nonlinearly fuse the long-term trend term of the WRF meteorological model with the short-term fluctuation term of the sensor data.

[0041] The communication network in the hierarchical scheduling architecture adopts an industrial Ethernet network with priority scheduling capabilities. Power allocation baseline commands issued by the system-level coordination layer carry priority tags, with local response commands under extreme conditions having the highest execution priority. The device-level execution layer includes photovoltaic inverter controllers, wind turbine pitch controllers, energy storage converter controllers, and load switch actuators.

[0042] In the multi-objective optimization model, the equipment loss component is calculated using the rainflow counting method to determine the stress cycles during equipment operation, and the equivalent cost is derived by combining this with the fatigue loss curve. The weighting coefficients of the toughness index are pre-set using the analytic hierarchy process (AHP) combined with a knowledge base of plateau operation experts, and are adjusted based on federated learning feedback during the adaptive iteration process.

[0043] The output robustness margin in the robust boundary constraints for strong uncertainties adopts an asymmetric construction. Under conditions of severe fluctuations in sunlight, the upper deviation margin of photovoltaic output is set to be smaller than the lower deviation margin. Under low-temperature conditions, the upper limit of the discharge power of the energy storage system is dynamically derated based on the real-time electrolyte temperature prediction value.

[0044] The improved multi-agent interaction logic in deep reinforcement learning algorithms adopts a centralized training and distributed execution model. During training, the evaluation network acquires the global state and action information of all agents; during execution, each agent's policy network outputs control actions based solely on local observations. The introduction of KKT conditions ensures that the solution to the multi-layered optimization problem lies within the globally feasible region.

[0045] The emergency response module in the rolling optimization execution mechanism is equipped with a hardware interlock mechanism. When the sensing system detects extreme changes and communication network anomalies, the device-level execution layer directly enters a safe operation mode based on locally preset logical thresholds.

[0046] The federated learning protocol in adaptive iterative optimization employs homomorphic encryption to process gradient updates. Without knowing the specific output, load, and energy storage status of each microgrid, the cloud parameter server uses an aggregation algorithm to evolve the global scheduling model.

[0047] This invention, through the aforementioned specific technical means, transforms the microgrid scheduling problem under extreme weather conditions in high-altitude regions into a closed-loop logical system with real-time perception, hierarchical collaboration, rolling optimization, and adaptive evolution. The modules are connected through clearly defined data interfaces and logical links, realizing a complete process from environmental perception to equipment control.

[0048] Example 2: As Figures 1-3 As shown, this invention proposes a hierarchical rolling collaborative optimization scheduling method for microgrids under extreme weather conditions in high-altitude regions. The invention will be further described in detail below with reference to specific embodiments.

[0049] In this implementation scenario, the microgrid is applied to a plateau village at an altitude of 3,500 meters. The microgrid operates in an isolated grid mode and is equipped with: 200kW bifacial snow-resistant photovoltaic modules (with an automatic snow melting system), 150kW low-wind-speed permanent magnet wind turbine (starting wind speed 2.5m / s), 100kW / 200kWh low-temperature resistant lithium iron phosphate energy storage system (equipped with a constant temperature chamber), 50kW biodiesel emergency generator, critical loads (10kW health clinic, 5kW communication base station), ordinary loads (50kW residential electricity), and multi-source meteorological monitoring nodes (3 nodes deployed, covering the entire microgrid area).

[0050] Step 1: Constructing a perception and prediction system for extreme weather changes on the plateau 1.1 Deploy meteorological monitoring nodes to collect real-time data on air temperature (-25℃ to 15℃), wind speed (0-25m / s), and irradiance (0-1200W / m²). 2 The system collects data on parameters such as snowfall, sandstorm intensity, etc., and also monitors the snow cover thickness of photovoltaic modules and the icing of wind turbine blades. The data collection frequency is 1 minute / time. 1.2 Set extreme weather change thresholds: wind speed increase ≥8m / s, irradiance decrease ≥50%, temperature drop ≥10℃, and snowfall ≥5mm / h. Use attention mechanism combined with LSTM neural network to identify extreme weather change events in real time, such as sudden strong wind (wind speed 18m / s) and blizzard (snowfall 8mm / h). 1.3 Construct a short-term rolling forecast model with a rolling step of 15 minutes to predict meteorological parameters for the next 4 hours, such as predicting strong winds lasting for 2 hours and irradiance dropping to 300W / m². 2 The following forecasts also predict that photovoltaic output will drop below 50kW, wind power output will rise to 120kW, and energy storage SOC will drop to 35%, providing data support for tiered dispatching.

[0051] Step 2: Construct a hierarchical scheduling architecture for the microgrid 2.1 Upper-level system-level collaborative optimization layer: Set multi-objective weights (economic efficiency 40%, reliability 35%, resilience 25%) to coordinate the overall operation of the microgrid and generate a baseline dispatch plan: 50kW photovoltaic output, 100kW wind power output (limiting strong wind overload), 30kW energy storage discharge, and emergency power backup to ensure full power supply for critical loads and 80% power supply for ordinary loads; 2.2 Mid-level microgrid autonomous optimization layer: Receives the upper-level baseline scheme, combines the real-time monitoring of photovoltaic snow accumulation (coverage thickness 3mm), starts the photovoltaic snow melting system (consuming 5kW of power), adjusts the photovoltaic output to 45kW, and optimizes the energy storage discharge power to 35kW to ensure power balance, and feeds back the adjustment capacity margin (remaining energy storage discharge power 40kW) to the upper layer; 2.3 Lower-level equipment-level execution control layer: Receives the mid-level scheduling plan, controls the start of the photovoltaic module snow melting system, adjusts the wind turbine speed to a safe range (avoiding strong wind overload), discharges the energy storage system at 35kW power, keeps the emergency power supply in standby mode, collects equipment operation data in real time, and feeds it back to the mid-level and upper-level layers.

[0052] Step 3: Construct a hierarchical rolling optimization scheduling model 3.1 Objective function: Minimize total operating cost (generation cost + energy storage loss + maintenance cost + emergency power backup cost), maximize critical load power supply guarantee rate (100%), and maximize system resilience index (MRI≥0.85); 3.2 Constraints: Power balance (45kW PV + 100kW wind power + 35kW energy storage = 10kW critical load + 40kW general load + 5kW snow melting loss); Energy storage SOC ≥ 20% (currently 35%, meeting the constraint); Wind turbine output ≤ 120kW (currently 100%, meeting the constraint); The PV snow melting system is only activated when the snow thickness is ≥ 2mm.

[0053] Step 4: Layered rolling optimization solution and execution 4.1 An improved MADDPG algorithm is adopted, which simplifies the solution by combining KKT conditions. The solution time is ≤10 seconds, which meets the real-time scheduling requirements. 4.2 With a rolling step of 15 minutes, after 15 minutes, update the meteorological data (strong winds continue, wind speed 17m / s) and equipment data (photovoltaic snow accumulation thickness 2mm, snow melting system continues to operate), resolve the optimization model, and adjust the scheduling scheme: photovoltaic output 48kW, wind power output 105kW, energy storage discharge 27kW, to ensure power balance; 4.3 Extreme Weather Sudden Change Response: When extreme blizzard (snowfall of 10mm / h) is detected, emergency dispatch is triggered. The mid-level adjustment plan is as follows: the emergency power supply (output of 20kW) is activated, the energy storage discharge power is increased to 40kW, the power supply to ordinary loads is suspended, and priority is given to ensuring the power supply to health clinics and communication base stations; the upper level confirms the dispatch plan simultaneously to ensure system stability. 4.4 Equipment Status Feedback: Slight icing was detected on the fan blades and reported to the middle layer. The middle layer adjusted the fan output to 80kW to prevent the icing from worsening, and the upper layer updated the global plan simultaneously.

[0054] Step 5: Scheduling effect evaluation and adaptive optimization 5.1 Calculation and evaluation indicators: operating costs are reduced by 10% compared with traditional dispatching, critical load power supply guarantee rate is 100%, renewable energy consumption rate is 88%, system resilience index is 0.87, and average outage time is 0. 5.2 Adaptive Adjustment: If the evaluation indicators meet the preset requirements, maintain the current objective function weights and constraints; 5.3 Model Iteration: Based on one month of running data, the extreme weather change identification model was updated to improve the identification accuracy of blizzards and strong winds, optimize the short-term prediction model, and reduce prediction errors.

Claims

1. A hierarchical rolling collaborative optimization scheduling method for microgrids under extreme weather conditions in high-altitude areas, characterized in that, Includes the following steps: Step 1: Construct a plateau extreme weather change perception and prediction system. By collecting multi-source meteorological parameters and equipment environmental parameters, identify the level of extreme weather change in real time and make short-term rolling predictions. Step 2: Construct a hierarchical dispatch architecture for the microgrid, adopting a three-layer architecture of "system-microgrid-equipment", and realizing global optimization under normal operating conditions and local rapid response under extreme operating conditions through a two-stage interaction mechanism; Step 3: Construct a hierarchical rolling optimization scheduling model with multiple objectives of economy, reliability and resilience, and set operational constraints that take into account the attenuation effect of the plateau environment and robust boundary constraints with strong uncertainty. Step 4: Solve the model using an improved multi-agent deep deterministic strategy gradient algorithm, and perform rolling optimization and emergency response to extreme weather changes with a preset step size; Step 5: Based on the scheduling effect evaluation index, dynamically adjust the model weights and prediction parameters, and iterate the model through federated learning based on long-term running data.

2. The hierarchical rolling collaborative optimization scheduling method for microgrids under extreme meteorological conditions in high-altitude areas according to claim 1, characterized in that, In step 1, the meteorological mutation identification model is constructed based on the attention mechanism combined with the LSTM neural network. It compares real-time monitoring data with the preset mutation threshold and outputs mutation levels including general, severe and extreme. The short-term rolling prediction uses the preset time period as the step size and couples the WRF meteorological model and the equipment simulation model to predict the power fluctuation range, equipment failure rate and energy storage state of charge in the next 1-4 hours.

3. The hierarchical rolling collaborative optimization scheduling method for microgrids under extreme meteorological conditions in high-altitude areas according to claim 1, characterized in that, The interaction mechanism of the three-tier architecture in step 2 is as follows: The upper layer is a system-level collaborative optimization layer, which allocates power benchmarks based on global cost, reliability, and absorption rate. The middle layer is a microgrid-level autonomous optimization layer that combines real-time monitoring data to optimize distributed power generation and energy storage strategies, and provides feedback on the margin of regulation capability. The lower layer is the device-level execution control layer, which realizes the underlying control of each device and transmits operational data back in real time; The two-stage interaction mechanism prioritizes upper-level optimization under normal conditions, but switches to mid-level local response priority during extreme weather changes.

4. The hierarchical rolling collaborative optimization scheduling method for microgrids under extreme meteorological conditions in high-altitude areas according to claim 1, characterized in that, The economic objective of the multi-objective optimization function mentioned in step 3 is to minimize the total operating cost of the microgrid, as shown in the following formula: in, It is the cost of photovoltaic power generation. It is the cost of wind power generation. Costs associated with energy storage charging and discharging losses. For emergency power supply operating costs, For equipment maintenance costs, Costs related to line and equipment wear and tear.

5. The hierarchical rolling collaborative optimization scheduling method for microgrids under extreme meteorological conditions in high-altitude areas according to claim 1, characterized in that, The resilience objective of the multi-objective optimization function is to maximize the microgrid resilience index MRI, as shown in the following formula: in, For power availability, The time required for fault recovery For critical load guarantee rate, Weights for each dimension.

6. The hierarchical rolling collaborative optimization scheduling method for microgrids under extreme meteorological conditions in high-altitude areas according to claim 1, characterized in that, The operational constraints mentioned in step 3 include power balance constraints, the expression of which is: in, The photovoltaic plateau attenuation coefficient, The plateau attenuation coefficient of the wind turbine. For energy storage discharge power, For energy storage discharge power, For emergency power output, For critical load power, For normal load consolidation rate, This is the normal load interruption factor. The amount of electricity purchased / sold from the main grid.

7. The hierarchical rolling collaborative optimization scheduling method for microgrids under extreme meteorological conditions in high-altitude areas according to claim 6, characterized in that, The robust boundary constraints with strong uncertainty mentioned in step 3 are achieved by setting robust boundaries, including robust boundary constraints for photovoltaic power output, robust boundary constraints for wind turbine power output, robust boundary constraints for load demand, and robust boundary constraints for power balance. Robust boundary constraints for photovoltaic power output are given by the following formula: ， in, To contribute to the actual development of photovoltaics Contribute to photovoltaic forecasting, Provide robust margin for photovoltaic power generation; The robust boundary constraints for wind turbine output are given by the following formula: , in, For the actual output of the wind turbine, To predict the output of wind turbines, Provide robust margin for wind turbine output; The robust boundary constraints for load demand are defined by the following formula: ， ， in, It is the critical load robustness margin. It is the robustness margin under normal load; Power balance robust constraint, the formula is as follows: ， By using robust boundary upper and lower limits, power balance can still be achieved under extreme uncertainty scenarios, thus avoiding power outages.

8. The hierarchical rolling collaborative optimization scheduling method for microgrids under extreme meteorological conditions in high-altitude areas according to claim 1, characterized in that, The solution algorithm described in step 4 adopts the improved multi-agent deep deterministic policy gradient (MADDPG) algorithm, and introduces the first-order optimality condition of KKT to transform the mid-level optimization problem into an equivalent constraint of the lower-level execution layer, thereby realizing multi-level collaborative solution.

9. The hierarchical rolling collaborative optimization scheduling method for microgrids under extreme meteorological conditions in high-altitude areas according to claim 1, characterized in that, The rolling optimization execution described in step 4 includes: updating meteorological and equipment status data every 15 minutes, resolving the scheduling scheme for the next 4 hours and executing only the first time-domain instruction; when a sudden change response is triggered, prioritizing the activation of the energy storage system and emergency power supply, and limiting non-critical loads.

10. The hierarchical rolling collaborative optimization scheduling method for microgrids under extreme meteorological conditions in high-altitude areas according to claim 1, characterized in that, The adaptive iterative optimization described in step 5 compares the running evaluation index with the preset threshold. If the index is not met, the objective function weight and constraint threshold are adjusted. The federated learning protocol is used to collaboratively update the global prediction and scheduling model parameters without disclosing the privacy data of each micronet.

Images