A new energy micro-grid intelligent energy management control method and system

Abstract

This invention relates to an intelligent energy management and control method for new energy microgrids, belonging to the field of microgrid operation control. The application includes the following steps: constructing a multi-source heterogeneous sensor network to generate a standardized time-series dataset; embedding physical constraints of new energy output into a physical information neural network architecture to output multi-time-scale power predictions; implementing distributed model predictive control based on a consensus algorithm to generate multi-agent collaborative scheduling commands; quantifying lifetime loss through a differentiable energy storage attenuation empirical model and dynamically correcting power commands to achieve coordinated optimization of energy balance and lifetime protection; and conducting multi-physics field coupling pre-simulation based on a digital twin to generate robust control commands with pre-simulation verification. This method significantly improves predictive extrapolation and interpretability, extends the service life of energy storage, and enhances the system's operational robustness and economy under strong disturbances.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent management of microgrids, and in particular to an intelligent energy management and control method and system for new energy microgrids. Background Technology

[0002] With the large-scale integration of distributed photovoltaic, wind power, and energy storage, renewable energy microgrids have become a key form for improving grid flexibility and renewable energy absorption capacity. However, microgrid operation faces the dual uncertainties of strong fluctuations in renewable energy and randomness in load, placing extremely high demands on the prediction accuracy, optimization efficiency, and control robustness of energy management.

[0003] Existing technologies primarily rely on purely data-driven prediction models and centralized optimization architectures. While these methods can maintain operation under stable conditions, prediction accuracy drops sharply and interpretability is lacking during extreme weather changes or equipment malfunctions. Centralized architectures face communication bottlenecks and single-point-of-failure risks as node size increases, making them ill-suited for the plug-and-play requirements of massive distributed resources. More critically, existing solutions treat energy storage systems as ideal energy buffers, leading to frequent deep charging and discharging that results in actual service life far shorter than designed expectations and high life-cycle costs, thus increasing the maintenance costs of microgrid systems. Therefore, improvements are needed. Summary of the Invention

[0004] To improve the economic efficiency of microgrid systems, this application provides a smart energy management and control method and system for new energy microgrids.

[0005] Firstly, the above-mentioned inventive objective of this application is achieved through the following technical solution: A smart energy management and control method for a new energy microgrid, the method comprising the following steps: Data on the operation status of the power grid and environmental monitoring in the microgrid are collected through a multi-source heterogeneous sensor network. After spatiotemporal alignment and outlier removal, a standardized time-series operation dataset is generated. The time-series running dataset is input into the multi-time-scale power prediction model. The multi-time-scale power prediction model uses a physical information neural network architecture to embed the physical constraint equation of new energy output and outputs the power generation and load demand prediction results corresponding to different time scales. The prediction results are fed into the distributed model prediction control layer, which implements multi-agent collaborative optimization based on the consensus algorithm to solve for the operation scheduling instructions of each distributed power source. The operation scheduling command is sent to the virtual energy storage aggregation and lifetime awareness optimization module. The lifetime awareness optimization module quantifies the cumulative loss of cycle life caused by frequent charging and discharging through a differentiable energy storage decay empirical model to generate a dynamic correction power command. The dynamic correction power command is used to achieve the coordinated optimization of energy balance and lifetime protection. Based on the modified power command, hierarchical collaborative control is executed. When a grid disturbance event is detected, an elastic control mode is triggered. The control strategy is pre-simulated using a digital twin and multi-physics coupling is performed to generate a robust control command sequence with pre-simulation verification.

[0006] By adopting the above technical solutions, the deep integration of physical information constraints and distributed intelligent collaboration systematically solves the core technical bottlenecks of high-proportion renewable energy microgrids in terms of extreme condition adaptability, system scalability, energy storage full-cycle economics, and control strategy predictability. This method not only significantly improves the physical rationality and long-term generalization ability of power prediction, but also greatly reduces the dependence on high-quality labeled data. At the same time, through a decentralized multi-agent collaborative architecture, it realizes plug-and-play and fault-tolerant operation of massive distributed resources. The lifetime awareness optimization module effectively balances short-term energy efficiency and long-term equipment health, extending the service life of the energy storage system. The digital twin pre-simulation mechanism provides a means of pre-physical verification of control strategies, significantly enhancing the system's anti-disturbance capability and operational safety in highly uncertain environments. The overall solution forms an adaptive, self-optimizing, and self-verifying intelligent energy management closed loop, effectively reducing grid maintenance costs and improving the economic benefits of the grid system.

[0007] In a preferred embodiment, this application can be further configured such that the multi-timescale power prediction model construction process includes: A physical information neural network backbone is constructed, whose input layer receives the time-series running dataset, the hidden layer adopts a fully connected structure, and the nonlinear physical equation between photovoltaic power output and irradiance and the cubic relationship constraint between wind turbine output and wind speed are embedded in the loss function. The physical constraint equations are automatically differentiated, and the derivative residuals are added to the network loss as a penalty term to ensure that the prediction results satisfy the power ramp-up rate limit and the law of energy conservation. An adaptive time window mechanism is adopted, using a second-level sampling interval and shallow network structure for ultra-short-term predictions, and an hour-level sampling interval and deep temporal convolution structure for medium- and long-term predictions, to achieve decoupled output across multiple time scales. An uncertainty quantization branch is integrated into the model output layer, and a prediction interval is generated through Monte Carlo deactivation, providing confidence weights for subsequent robust optimization.

[0008] In a preferred example, this application can be further configured such that the step of feeding the prediction results into a distributed model prediction control layer, wherein the distributed model prediction control layer implements multi-agent cooperative optimization based on a consensus algorithm to obtain the operation scheduling instructions for each distributed power source, includes the following steps: Each distributed resource in the microgrid is abstracted into an independent intelligent agent. Each intelligent agent maintains a local cost function, which includes a power generation cost term, a deviation penalty term, and a coupling constraint term. A consistent communication topology is constructed, in which the agent only exchanges the power plan trajectory in the predicted time domain with neighboring nodes, and the local optimum is solved iteratively by the Lagrange multiplier method. An asynchronous update mechanism is introduced, and each agent dynamically adjusts the consensus iteration frequency according to its local computing load. After convergence, it generates active and reactive power adjustment instructions, energy storage power instructions and load scheduling instructions for each node. The generated instruction sequence is checked for feasibility. If it violates the physical constraints of the device, constraint tightening and re-optimization are initiated until a feasible solution set is obtained.

[0009] In a preferred embodiment, this application can be further configured as follows: The step of sending the operation scheduling command to the virtual energy storage aggregation and lifetime-aware optimization module, wherein the lifetime-aware optimization module quantifies the cumulative loss of cycle life due to frequent charging and discharging through a differentiable energy storage degradation empirical model to generate a dynamically corrected power command, includes the following steps: A differentiable empirical model for energy storage degradation is constructed. The empirical model for energy storage degradation takes the cumulative charge-discharge depth, the number of cycles, and the temperature as inputs, and outputs the capacity degradation rate and the health status assessment value. The energy storage charging and discharging power command is input into the attenuation model to calculate the lifetime loss gradient in the future prediction time domain. A lifetime protection regularization term is added to the optimization objective. The weight of the lifetime protection regularization term is dynamically adjusted according to the real-time electricity price and the energy storage replacement cost to achieve a multi-objective trade-off between economy and lifetime. The gradient descent method is used to solve for the corrected charge and discharge power command, so that the corrected command reduces life loss while satisfying power balance, and outputs health status warning information.

[0010] In a preferred embodiment, this application can be further configured to: construct a differentiable empirical model for energy storage degradation, wherein the empirical model takes cumulative charge-discharge depth, cycle count, and temperature as inputs, and outputs capacity degradation rate and health status assessment value, wherein the construction of the differentiable empirical model for energy storage degradation includes the following steps: Collect historical operating data of energy storage devices, including charge / discharge depth sequences, terminal voltage curves, and ambient temperature records; Gaussian process regression was used to fit the capacity decay surface, and an empirical function with cycle depth, temperature and number of cycles as independent variables was constructed. By analytically differentiating the empirical function, the sensitivity gradient of the capacity decay rate to the charge and discharge power is obtained. By embedding a temperature acceleration factor into the model, the decay rate is increased according to the Arrhenius equation when the ambient temperature exceeds a threshold, thus achieving temperature-dependent lifetime sensing.

[0011] In a preferred embodiment, this application can be further configured such that, in the step of performing hierarchical cooperative control based on the modified power command and triggering a resilient control mode when a grid disturbance event is detected, the event-triggered resilient control mode includes: Real-time monitoring of power grid frequency, voltage amplitude, and power fluctuation rate; and construction of disturbance intensity assessment indicators. When the evaluation index exceeds the preset trigger threshold, the elastic control mode is activated, the economic optimization target is frozen, and power balance and power quality are prioritized. Construct a fast-response control law, and achieve rapid frequency and voltage support through the coordinated control of energy storage virtual inertia and distributed power source droop control; After the disturbance subsides, the system gradually switches back to the economic optimization mode, and a smooth transition function is used during the switching process to avoid secondary shocks.

[0012] In a preferred embodiment, this application can be further configured as follows: In the step of performing hierarchical collaborative control based on the modified power command, triggering a resilient control mode when a grid disturbance event is detected, and relying on a digital twin to perform multi-physics coupling pre-simulation of the control strategy to generate a robust control command sequence with pre-simulation verification, the digital twin pre-simulation process includes the following steps: Construct a multi-physics coupled digital twin of a microgrid, integrating electromagnetic transient models, thermodynamic models, and equipment aging models; The modified power command sequence is injected into the twin, and the power flow distribution, node voltage curves and equipment temperature field after execution are simulated in parallel. During the rehearsal process, random disturbances were injected to conduct stress tests and evaluate the robustness margin of the control strategy under extreme conditions. If the simulation results meet both safety and economic objectives, the instruction sequence is confirmed to be executable; otherwise, it is fed back to the distributed model predictive control layer for re-optimization, forming a decision-simulation-verification closed loop.

[0013] Secondly, the above-mentioned inventive objective of this application is achieved through the following technical solutions: A smart energy management and control device for a new energy microgrid, the device comprising: a standardized time-series operation dataset generation unit, used to collect grid operation status data and environmental monitoring data in the microgrid through a multi-source heterogeneous sensor network, and generate a standardized time-series operation dataset after spatiotemporal alignment and outlier removal; The power and load demand prediction unit is used to input the time-series operation dataset into the multi-time-scale power prediction model. The multi-time-scale power prediction model uses a physical information neural network architecture to embed the physical constraint equation of new energy output and outputs the power generation and load demand prediction results corresponding to different time scales. The operation scheduling instruction generation unit is used to feed the prediction results into the distributed model prediction control layer. The distributed model prediction control layer realizes multi-agent collaborative optimization based on the consensus algorithm and solves for the operation scheduling instructions of each distributed power source. The dynamic power correction command generation unit is used to send the operation scheduling command to the virtual energy storage aggregation and lifetime awareness optimization module. The lifetime awareness optimization module quantifies the cumulative loss of cycle life caused by frequent charging and discharging through a differentiable energy storage decay empirical model to generate dynamic power correction commands. The robust control command sequence generation unit is used to perform hierarchical collaborative control based on the modified power command. When a grid disturbance event is detected, it triggers the elastic control mode and relies on the digital twin to perform multi-physics coupling pre-simulation of the control strategy to generate a robust control command sequence with pre-simulation verification.

[0014] Thirdly, the above-mentioned objectives of this application are achieved through the following technical solutions: An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described intelligent energy management and control method for new energy microgrids.

[0015] Fourthly, the above-mentioned objectives of this application are achieved through the following technical solutions: A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described intelligent energy management and control method for new energy microgrids. Attached Figure Description

[0016] Figure 1 This is a flowchart of a smart energy management and control method for a new energy microgrid in one embodiment of this application; Figure 2 This is a schematic block diagram of a smart energy management and control device for a new energy microgrid in one embodiment of this application; Figure 3This is a schematic diagram of an electronic device according to an embodiment of this application.

[0017] Icon labels: 1. Standardized time-series operation dataset generation unit; 2. Power and load demand prediction unit; 3. Operation scheduling instruction generation unit; 4. Dynamically corrected power instruction generation unit; 5. Robust control instruction sequence generation unit. Detailed Implementation

[0018] The present application will be further described in detail below with reference to the accompanying drawings.

[0019] In one embodiment, such as Figure 1 As shown, this application discloses a smart energy management and control method for new energy microgrids, which specifically includes the following steps: S10: Collect power grid operation status data and environmental monitoring data in the microgrid through a multi-source heterogeneous sensor network, and generate a standardized time-series operation dataset after spatiotemporal alignment and outlier removal; In this embodiment, the power grid operation status data includes electrical parameters of distributed power sources, energy storage devices, flexible loads, and the power grid bus. Specifically, this step constructs a microgrid-wide data sensing system through a multi-source heterogeneous sensor network. In a typical island microgrid scenario, distributed photovoltaic inverters upload DC-side voltage and current and AC-side power factor of each array; energy storage converters report battery pack state of charge, terminal voltage, and internal temperature; flexible load smart meters transmit the adjustable capacity and response priority of equipment such as air conditioners and water pumps; and power grid bus measurement devices record three-phase voltage amplitude, frequency, and harmonic components. Simultaneously, environmental monitoring equipment collects irradiance, wind speed, temperature, and humidity data. The system performs spatiotemporal alignment on the collected heterogeneous data, for example, matching photovoltaic power data with environmental irradiance data using minute-level timestamps, eliminating abnormal jump values ​​caused by communication delays or sensor malfunctions, and finally generating a unified time-series operation dataset containing electrical parameters, environmental parameters, and equipment status, providing a standardized input basis for subsequent prediction and optimization.

[0020] S20: Input the time-series running dataset into the multi-time-scale power prediction model. The multi-time-scale power prediction model uses a physical information neural network architecture to embed the physical constraint equation of new energy output and outputs the power generation and load demand prediction results corresponding to different time scales. Specifically, this step achieves accurate power prediction across multiple time scales. The time-series dataset generated in S10 is input into the physical information neural network. This network embeds the nonlinear mapping relationship between photovoltaic power output and irradiance, the cubic ratio constraint between wind turbine output and wind speed, and its derivative residual penalty term in the hidden layer. For example, for the scenario of rapid cloud cover at midday, the network not only learns historical power decay patterns but also enforces the physical constraint of the maximum ramp rate when irradiance suddenly drops, avoiding ultra-fast power changes in the prediction results that violate equipment characteristics. The model adopts an adaptive time window mechanism: ultra-short-term prediction (future 15 minutes) uses second-level sampling data and a shallow network structure to meet real-time scheduling requirements; short-term prediction (future 4 hours) uses 15-minute average data and a temporal convolutional module; medium- and long-term prediction (future 72 hours) uses hourly data and introduces a meteorological model attention mechanism. Finally, the predicted curves of photovoltaic power generation, wind turbine output, and adjustable load demand for each time period are output, and the prediction confidence interval is given simultaneously, providing a quantitative basis for uncertainty in subsequent optimized scheduling.

[0021] S30: The prediction results are fed into the distributed model prediction control layer. The distributed model prediction control layer realizes multi-agent collaborative optimization based on the consensus algorithm and solves the operation scheduling instructions of each distributed power source. In this embodiment of the application, the operation scheduling instructions include active and reactive power adjustment instructions, energy storage charging and discharging power instructions and load flexible scheduling instructions. Specifically, this step completes distributed collaborative optimization decision-making. Within the microgrid, each photovoltaic inverter, energy storage converter, and flexible load controller is abstracted as an independent intelligent agent. Each agent constructs a local cost function, encompassing multiple objectives such as minimizing generation costs, minimizing power deviation, and optimizing voltage levels. Each agent exchanges its power plan trajectory for the next hour only with its neighboring nodes through a low-bandwidth communication network, iteratively updating the Lagrange multipliers using a consensus algorithm. For example, when photovoltaic output is excessive, the photovoltaic agent reduces its own generation cost weight, the energy storage agent increases its charging revenue weight, and the load agent prioritizes adjustable capacity response. These three agents reach a power balance consensus through information exchange. After iterative convergence, each agent generates locally executable active and reactive power adjustment commands, initial values ​​for energy storage charging and discharging power, and a flexible load scheduling plan, achieving decentralized global collaborative optimization and avoiding the risk of single-point failure of the centralized controller.

[0022] S40: The operation scheduling command is sent to the virtual energy storage aggregation and lifetime awareness optimization module. The lifetime awareness optimization module quantifies the cumulative loss of cycle life caused by frequent charging and discharging through a differentiable energy storage decay empirical model to generate a dynamic correction power command. The dynamic correction power command is used to achieve coordinated optimization of energy balance and lifetime protection. Specifically, this step achieves energy storage lifetime sensing and power command optimization. After receiving the initial charge / discharge power value generated by S30, the energy storage converter calls a differentiable energy storage degradation empirical model. This model takes the cumulative charge / discharge depth sequence, battery pack temperature curve, and historical cycle count as input to quantitatively calculate the capacity degradation rate and health status evolution trend under the given power command. For example, if the initial command includes continuous deep charge / discharge cycles, the model predicts a significant increase in capacity degradation rate. In this case, the module dynamically increases the weight of the lifetime protection regularization term in the optimization objective, moderately sacrificing economic efficiency to reduce the charge / discharge depth. The corrected power command is solved using the gradient descent method, adjusting the originally planned frequent high-power charge / discharge to multiple shallow charge / discharge modes. At the same time, health status warning information is output to prompt maintenance personnel to intervene in advance, achieving coordinated optimization of energy balance and lifetime protection.

[0023] S50: Based on the modified power command, perform hierarchical collaborative control. When a grid disturbance event is detected, trigger the elastic control mode and rely on the digital twin to perform multi-physics coupling pre-simulation of the control strategy to generate a robust control command sequence with pre-simulation verification. Specifically, this step completes the active disturbance response and strategy pre-simulation verification. The system continuously monitors the grid frequency and voltage amplitude. When it detects that the frequency drop exceeds the threshold due to a sudden large load surge, it triggers the elastic control mode, immediately freezing the economic optimization target and prioritizing system stability. The energy storage converter switches to virtual inertia control mode, releasing high power to support the frequency in a short time; distributed photovoltaics switch from maximum power point tracking mode to power-limited operation to prevent overvoltage. At the same time, the digital twin receives the corrected control command sequence, deduces the grid power flow distribution after execution based on the electromagnetic transient model, and predicts battery temperature field changes in conjunction with the energy storage thermodynamic model. The pre-simulation shows that the strategy can quickly restore the frequency to the rated range and the equipment temperature does not exceed the limit. After confirming the robustness and feasibility of the command, it is issued for execution. After the disturbance subsides, the system gradually restores the economic optimization mode through a smooth transition function, avoiding the secondary impact of mode switching.

[0024] For steps S10-S50, this application systematically solves the core technical bottlenecks of high-proportion renewable energy microgrids in terms of extreme condition adaptability, system scalability, energy storage full-cycle economics, and control strategy predictability through the deep integration of physical information constraints and distributed intelligent collaboration. This method not only significantly improves the physical rationality and long-term generalization ability of power prediction, but also greatly reduces the dependence on high-quality labeled data. At the same time, through a decentralized multi-agent collaborative architecture, it realizes plug-and-play and fault-tolerant operation of massive distributed resources. The lifetime awareness optimization module effectively balances short-term energy efficiency and long-term equipment health, extending the service life of the energy storage system. The digital twin pre-simulation mechanism provides a means of pre-physical verification of control strategies, significantly enhancing the system's anti-disturbance capability and operational safety in highly uncertain environments. The overall solution forms an adaptive, self-optimizing, and self-verifying intelligent energy management closed loop, effectively reducing grid maintenance costs and improving the economic benefits of the grid.

[0025] Furthermore, by adopting the above technical solutions, the multi-source heterogeneous sensor network collects and generates standardized time-series operational datasets. The system can integrate multi-dimensional information from distributed power sources, energy storage, loads, and environmental parameters, effectively eliminating the problem of spatiotemporal asynchrony of heterogeneous data, providing a high-quality data foundation for subsequent decision-making, and improving the data completeness and consistency of energy management. The multi-timescale power prediction model using a physical information neural network architecture embeds the physical constraint equations of new energy output into the network loss function, ensuring that the prediction results naturally satisfy the power ramp-up rate limit and the law of energy conservation. This significantly enhances the extrapolation and interpretability of ultra-short-term, short-term, and medium-to-long-term predictions, effectively avoiding the prediction inaccuracies of pure data-driven models under extreme conditions. The distributed model prediction control layer based on a consensus algorithm decouples centralized optimization into collaborative decision-making by multiple agents. Each agent only needs to exchange local information with neighboring nodes to achieve global optimality, significantly reducing communication bandwidth requirements and dependence on the central node, improving the system's scalability and fault tolerance in communication-constrained environments, and avoiding single points of failure. Risks; The virtual energy storage aggregation and lifetime awareness optimization module quantifies the cumulative loss of cycle life due to frequent charging and discharging into an optimizable target through a differentiable empirical model of energy storage decay. It dynamically corrects power commands to balance the relationship between energy balance and lifetime protection, effectively extending the service life of the energy storage system, reducing the total life cycle cost, and resolving the inherent contradiction between economy and durability. The hierarchical collaborative control and event-triggered elastic control mode enable the system to focus on economic optimization during daily operation and automatically switch to a safety-first mode when grid disturbance events occur. It provides frequency and voltage support through the collaborative operation of energy storage virtual inertia and distributed power droop control. After the disturbance subsides, it smoothly transitions back to the economic mode, achieving a dynamic balance between operational economy and power supply robustness. Relying on a multi-physics coupled digital twin to pre-validate the control strategy, the system can deduce the impact of commands on grid power flow, node voltage, and equipment temperature before command execution and inject random disturbances to assess the safety margin under extreme conditions, forming a decision-making-pre-validation closed loop, which significantly improves the anti-disturbance capability and operational safety of the control strategy.

[0026] For step S20, the multi-timescale power prediction model construction process includes: S21: Construct the backbone of the physical information neural network. Its input layer receives the time-series running dataset, the hidden layer adopts a fully connected structure, and the nonlinear physical equation between photovoltaic power output and irradiance and the cubic relationship constraint between wind turbine output and wind speed are embedded in the loss function. Specifically, the time-series dataset generated in S10 is input into the network input layer, and the hidden layers adopt a multi-layer fully connected structure. In the loss function design, for photovoltaic arrays, a nonlinear physical equation between photovoltaic output and irradiance is embedded, meaning the power output is limited by the maximum irradiance conversion efficiency and exhibits nonlinear saturation characteristics. In this embodiment, for wind turbines, a cubic relationship constraint between output and wind speed is embedded to ensure that the predicted power strictly follows aerodynamic laws. For example, when the wind speed is at its rated value, the predicted wind turbine power must satisfy the physical law that it is proportional to the cube of the wind speed. If the network output deviates from this constraint, the residual of the physical equation will be directly included in the total loss, forcing the network to correct its weight parameters, thereby ensuring that the prediction result always conforms to the physical characteristics of the equipment.

[0027] S22: Automatically differentiate the physical constraint equations and add the derivative residuals as a penalty term to the network loss so that the prediction results satisfy the power ramp-up rate limit and the law of energy conservation. Specifically, this step performs automatic differentiation on the embedded physical constraint equations. The system calculates the derivatives of the photovoltaic power-irradiance equation and the wind turbine power-wind speed equation with respect to the input variables, forming derivative residual terms. For example, when the rate of change of predicted power between adjacent sampling times exceeds the maximum allowable ramp rate of the equipment, the derivative residual of the ramp rate constraint will increase significantly. This residual is added as a penalty term to the network loss function, forcing the network to reduce the power change amplitude during that period. At the same time, the derivative constraint of the law of conservation of energy ensures that the total output of all distributed power sources maintains a dynamic balance with load demand and network losses, avoiding the physical paradox of power mismatch in the prediction results. Through the automatic differentiation mechanism, the network not only fits historical data during backpropagation but also actively follows physical laws, improving the rationality and security of the prediction results.

[0028] S23: An adaptive time window mechanism is adopted, using a second-level sampling interval and shallow network structure for ultra-short-term predictions, and an hour-level sampling interval and deep temporal convolution structure for medium- and long-term predictions, to achieve decoupled output across multiple time scales. Specifically, this step employs an adaptive time window mechanism to achieve decoupled output across multiple time scales. For ultra-short-term forecasts (the next 15 minutes), the network uses a second-level sampling interval and a three-layer shallow fully connected structure to quickly capture the dynamic characteristics of instantaneous cloud occlusion or load abrupt changes, meeting the low-latency requirements of real-time scheduling. For short-term forecasts (the next 4 hours), a 15-minute average sampling interval is used, and a temporal convolutional layer is introduced to extract the changing trends of meteorological conditions and load cycles. For medium- and long-term forecasts (the next 72 hours), an hourly sampling interval is used, and a deep temporal convolutional network is constructed to fuse numerical weather prediction model information, capturing long-cycle patterns such as sunrise and sunset, and seasonal variations. Networks at different time scales share the bottom-level feature extraction layer, but the top-level output structure is independent to avoid mutual interference and achieve parallel output of multi-time-scale forecast results.

[0029] S24: Integrate an uncertainty quantization branch into the model output layer, generate prediction intervals through Monte Carlo deactivation, and provide confidence weights for subsequent robust optimization; Specifically, this step integrates an uncertainty quantification branch into the model output layer. During the inference phase, the network performs random deactivation on the hidden layer neurons, generating a set of prediction results through multiple forward propagations. For example, for photovoltaic power prediction at the same time, ten deactivation runs may produce slightly different prediction values. The system calculates the mean of this set as the point prediction result and takes the upper and lower quantiles to form a prediction interval. The width of this prediction interval dynamically reflects the model's confidence in the prediction result at that time: during periods of drastic weather changes, the interval automatically widens to cover high uncertainty; during sunny and stable periods, the interval narrows to improve prediction accuracy. This uncertainty quantification result is passed to the subsequent optimization layer in the form of confidence weights, enabling scheduling decisions to proactively increase reserve capacity during periods of high uncertainty, thereby improving overall robustness.

[0030] For steps S21-S24, the multi-timescale power prediction model fundamentally solves the shortcomings of traditional prediction methods, such as poor extrapolation under extreme conditions and results that violate physical laws, through the deep fusion of physical information and deep learning. The embedding of physical constraint equations and the automatic differential penalty mechanism enable the network to achieve a balance between data fitting and physical consistency, significantly improving the rationality and safety of prediction results and preventing equipment overload or fluctuation exceeding limits due to prediction inaccuracies. The adaptive time window mechanism, through differentiated network structures and sampling strategies, achieves decoupled optimization of ultra-short-term, short-term, and medium-to-long-term predictions, meeting both the low-latency requirements of real-time scheduling and the macroscopic perspective of medium-to-long-term scheduling, thus improving the flexibility and comprehensiveness of the prediction system. The uncertainty quantification branch provides confidence assessment capabilities for the prediction results, enabling subsequent optimization decisions to dynamically perceive risk levels and proactively increase safety margins during periods of high uncertainty, effectively avoiding scheduling risks caused by the lack of confidence in traditional point prediction models. Overall, this model improves the prediction accuracy, physical rationality, and decision robustness of microgrids under strong uncertainty environments, laying a reliable forward-looking foundation for intelligent energy management.

[0031] In step S30: feeding the prediction results into the distributed model prediction control layer, which uses a consensus algorithm to achieve multi-agent collaborative optimization and obtain the operation scheduling instructions for each distributed power source, the steps include the following: S31: Abstract each distributed resource in the microgrid into an independent intelligent agent. Each intelligent agent maintains a local cost function, which includes a generation cost term, a deviation penalty term, and a coupling constraint term. S32: Construct a consistent communication topology, where the agent only exchanges power plan trajectories in the predicted time domain with neighboring nodes, and iteratively solves for local optimal solutions using the Lagrange multiplier method; S33: Introducing an asynchronous update mechanism, each agent dynamically adjusts the consensus iteration frequency according to its local computing load, and after convergence, generates active and reactive power adjustment instructions, energy storage power instructions and load scheduling instructions for each node. S34: Perform feasibility verification on the generated instruction sequence. If it violates the physical constraints of the device, start constraint compaction re-optimization until a feasible solution set is obtained.

[0032] In this embodiment of the application, for steps S31-S34: in an industrial park microgrid that includes distributed photovoltaic, energy storage systems and flexible loads, the distributed model predictive control layer realizes decentralized collaborative optimization scheduling.

[0033] First, step S31 abstracts each photovoltaic inverter, each energy storage converter, and each flexible load controller as an independent intelligent agent. For example, the local cost function maintained by a 500kWp photovoltaic array intelligent agent includes generation cost terms (such as curtailment penalty), power deviation penalty terms (deviation from predicted values), and voltage coupling constraint terms (related to the voltage of nearby busbars); the cost function of a 2MWh energy storage system intelligent agent includes charging and discharging loss costs, state-of-charge deviation penalties, and power balance coupling terms. Each intelligent agent only needs to know its local device parameters and neighbor node information, without needing to know the entire network topology.

[0034] Next, step S32 constructs a consistent communication topology. Agents employ a sparse connection approach, with each node exchanging power plan trajectories for the next hour only with its 2-3 neighboring agents. Based on the received neighbor trajectories, each agent iteratively updates its local decision using the Lagrange multiplier method: the photovoltaic agent dynamically adjusts its own curtailment based on the charging plans of its neighboring energy storage agents, while the energy storage agents optimize charging and discharging timing based on the load adjustment capabilities of its neighbors. During the iteration process, the Lagrange multipliers gradually converge, and the power plan trajectories of all agents become consistent, ultimately achieving a global power balance consensus and generating preliminary scheduling instructions for each node.

[0035] Step S33 introduces an asynchronous update mechanism to adapt to heterogeneous computing capabilities. When the photovoltaic agent requires high-frequency calculations due to rapid changes in irradiance, its iteration cycle is shortened to 5 seconds; while the load agent's iteration cycle is extended to 30 seconds due to its gradual adjustment behavior. Each agent dynamically adjusts the consensus iteration frequency according to its own computing load, avoiding latency caused by synchronous waiting. After several rounds of asynchronous iteration, all agents converge to a stable solution, generating executable active and reactive power adjustment commands, energy storage power commands, and load flexible scheduling commands, significantly reducing scheduling cycle and communication overhead.

[0036] Finally, step S34 verifies the feasibility of the generated instruction sequence. The system detects that a charging power instruction generated by an energy storage agent exceeds its maximum allowable charging rate, violating the device's physical constraints. Therefore, it initiates constraint-tightening re-optimization. The energy storage agent re-introduces the power limit as a hard constraint into the local optimization problem, and neighboring agents adjust their power plans accordingly to compensate for the shortfall. After local iteration, a feasible solution set satisfying all device physical constraints is obtained, ensuring the generated scheduling instructions are practically executable.

[0037] Specifically, for steps S31-S34, the distributed model prediction control layer, through multi-agent abstraction and consistent collaboration, fundamentally overcomes the communication bottlenecks and single-point-of-failure limitations of traditional centralized optimization architectures. Each agent only needs local communication to achieve global optimum, significantly reducing communication bandwidth requirements and dependence on the central node, and improving the system's scalability and plug-and-play capability in environments with limited communication conditions. The asynchronous update mechanism enables heterogeneous devices to autonomously adjust their computation rhythm based on local dynamics, avoiding the inefficiency of centralized synchronous waiting and significantly improving the real-time performance and flexibility of optimization response. The distributed iterative solution of the Lagrange multiplier method, while ensuring global power balance, fully respects the autonomy of each agent, giving the system good fault tolerance; the failure of individual nodes does not affect the overall optimization process. The feasibility verification and constraint-based compaction re-optimization mechanism, as the final safeguard, ensures that all scheduling instructions strictly meet the physical limits of the equipment, avoiding the risk of execution failure or equipment damage due to constraint violations. This control layer enables a paradigm shift in microgrids from centralized management and control to distributed intelligent collaboration, providing innovative technical support for the efficient aggregation and autonomous operation of massive distributed resources.

[0038] In step S40: the operation scheduling command is sent to the virtual energy storage aggregation and lifetime awareness optimization module. The lifetime awareness optimization module quantifies the cumulative loss of cycle life due to frequent charging and discharging through a differentiable energy storage degradation empirical model to generate a dynamically corrected power command. This step includes the following steps: S41: Construct a differentiable empirical model for energy storage degradation, wherein the empirical model for energy storage degradation takes the cumulative charge-discharge depth, the number of cycles and the temperature as inputs, and outputs the capacity degradation rate and the health status assessment value. S42: Input the energy storage charging and discharging power command into the attenuation model to calculate the lifetime loss gradient in the future predicted time domain; S43: Add a lifetime protection regularization term to the optimization objective. The weight of the lifetime protection regularization term is dynamically adjusted according to the real-time electricity price and the energy storage replacement cost to achieve a multi-objective trade-off between economy and lifetime. S44: The gradient descent method is used to solve the corrected charge and discharge power command, so that the corrected command reduces life loss while satisfying power balance, and outputs health status warning information.

[0039] In this embodiment of the application, taking a microgrid in an industrial park operating under high summer temperatures as an example, the 2MWh energy storage system receives the initial charge and discharge power command generated by the S30 link, and plans to deeply charge during the midday solar power generation and deeply discharge during the evening load peak to achieve electricity price arbitrage.

[0040] S41: The empirical model for the degraded capacity of the energy storage system constructed in this step takes the current cumulative charge-discharge depth, the number of cycles achieved, and the real-time temperature of the battery compartment as inputs. The model calculates and outputs the capacity degradation rate and health status assessment value, and finds that deep charge-discharge cycles under high temperature conditions will accelerate capacity degradation.

[0041] S42: The step inputs the power command into the attenuation model and calculates the lifetime degradation gradient in the predicted time domain for the next 24 hours along the time axis. The results show that continuous deep charge and discharge operations in the current command sequence will lead to a significant increase in the lifetime degradation gradient, and the capacity degradation trend exceeds expectations.

[0042] S43: The step then dynamically adjusts the weight of the lifetime protection regularization term in the optimization objective. Although the current real-time electricity price is high and the economic benefits are considerable, the system automatically increases the lifetime protection weight based on the overall replacement cost assessment of the energy storage device, so that the optimization objective shifts from simply pursuing economic benefits to a multi-objective trade-off between economy and lifetime protection.

[0043] S44: Step S44 employs the gradient descent method to iteratively solve for the corrected charge and discharge power commands, optimizing the originally planned deep charge and discharge mode into a multiple shallow charge and discharge mode. This effectively reduces the depth of a single cycle, thereby significantly reducing lifespan loss while meeting the power balance requirements of the campus. Simultaneously, the module outputs health status warning information, prompting maintenance personnel to pay attention to the battery health degradation trend under high temperatures and recommending the activation of auxiliary cooling measures.

[0044] The virtual energy storage aggregation and lifetime-aware optimization module, through the gradient quantization capability of the differentiable decay model, achieves for the first time an online dynamic trade-off between energy storage economics and lifetime loss. Compared to traditional lifetime management methods that rely on fixed thresholds or post-event statistics, this module can quantify the cumulative impact on cycle life during the power command generation stage. By guiding command correction direction through gradient information, it fundamentally avoids irreversible capacity decay caused by deep charging and discharging. The dynamic weight adjustment mechanism enables the system to autonomously decide on protection strength based on real-time electricity prices and equipment costs. It moderately tolerates lifetime loss to obtain excess profits when electricity prices are extremely high, and strictly protects lifetime when electricity prices are stable, achieving optimal control of the entire life cycle cost. Health status early warning information provides a basis for preventive maintenance, shifting operation and maintenance from post-fault handling to pre-condition early warning, significantly reducing the risk of unplanned downtime. This module effectively resolves the inherent contradiction between the economics and durability of energy storage systems, significantly extends the service life of energy storage devices, and reduces the total life cycle investment cost of microgrids.

[0045] In S41: Construct a differentiable empirical model for energy storage degradation. The empirical model takes the cumulative charge-discharge depth, cycle count, and temperature as inputs, and outputs the capacity degradation rate and health status assessment value. The construction of the differentiable empirical model for energy storage degradation includes the following steps: S411: Collect historical operating data of energy storage devices, including charge / discharge depth sequences, terminal voltage curves, and ambient temperature records; S412: Gaussian process regression is used to fit the capacity decay surface, and an empirical function with cycle depth, temperature and number of cycles as independent variables is constructed. S413: Analytically differentiate the empirical function to obtain the sensitivity gradient of the capacity decay rate to the charge and discharge power; S414: Embed a temperature acceleration factor in the model. When the ambient temperature exceeds the threshold, the decay rate is increased according to the Arrhenius equation to achieve temperature-dependent lifetime perception.

[0046] Specifically, when constructing the empirical model for the conductable degradation of the 2MWh energy storage system, the system first executes step S411 to continuously collect historical operating data of the energy storage device, including the depth sequence of each charge-discharge cycle over the past year, the corresponding terminal voltage change curves, and the ambient temperature record of the battery compartment. Data shows that the subsequent capacity degradation of deep charge-discharge cycles performed during the high-temperature period in summer (ambient temperature exceeding 35 degrees Celsius) is significantly higher than that in spring and autumn.

[0047] Step S412 uses Gaussian process regression to fit the historical data. The model uses cycle depth, temperature, and cumulative cycle count as independent variables, and capacity decay rate as the dependent variable, constructing a three-dimensional empirical function surface. The regression revealed that when the cycle depth exceeds 60% of the rated capacity and the temperature is above 35 degrees Celsius, the capacity decay rate surface shows a steep upward region, indicating that this operating condition combination has a strong accelerating effect on lifespan loss.

[0048] Step S413 involves analytically differentiating the empirical function to obtain the sensitivity gradient of the capacity decay rate to charge / discharge power. The results show that, under high-temperature conditions, the sensitivity gradient of the capacity decay rate increases non-linearly for every unit increase in charge / discharge power, especially in the deep charge / discharge range, providing a clear direction for gradient descent in subsequent optimizations.

[0049] Step S414 embeds a temperature acceleration factor into the model. An ambient temperature threshold parameter is set; when the monitored value exceeds this threshold, the model dynamically increases the capacity degradation rate output according to the Arrhenius equation. That is, for every certain increase in temperature, the degradation rate increases exponentially, thereby achieving accurate perception and quantification of accelerated lifespan degradation under high-temperature conditions. The final differentiable degradation empirical model not only outputs the current health status assessment value but also provides real-time gradient feedback on lifespan loss to power commands, supporting dynamic optimization decisions in subsequent steps.

[0050] In step S50: Executing hierarchical cooperative control based on the modified power command, and triggering a resilient control mode when a grid disturbance event is detected, the event-triggered resilient control mode includes: S51: Real-time monitoring of grid frequency, voltage amplitude, and power fluctuation rate to construct disturbance intensity assessment indicators; S52: When the evaluation index exceeds the preset trigger threshold, activate the elastic control mode, freeze the economic optimization target, and prioritize power balance and power quality. S53: Construct a fast-response control law, and achieve rapid support for frequency and voltage through the coordinated control of energy storage virtual inertia and distributed power source droop control; S54: After the disturbance subsides, gradually switch back to the economic optimization mode. The switching process uses a smooth transition function to avoid secondary shocks.

[0051] Specifically, during the stable operation of the industrial park microgrid, the system continuously executes step S51 to monitor the grid frequency, voltage amplitude, and power fluctuation rate in real time, and constructs a comprehensive disturbance intensity assessment index. One afternoon, due to the sudden start-up of high-power equipment on the production line, the load demand surged instantly, causing the grid frequency to drop, the voltage amplitude to decrease synchronously, and the power fluctuation rate to rise sharply. The disturbance intensity assessment index rapidly increased and exceeded the preset trigger threshold.

[0052] Step S52 responds immediately, activating the flexible control mode, freezing the original economic optimization objectives, suspending considerations of power generation costs and electricity price arbitrage, and shifting the control priority to ensuring power balance and power quality. At this point, the system no longer pursues operational economy, but instead concentrates resources on dealing with disturbances.

[0053] Step S53 establishes a fast-response control law, switching the energy storage converter to virtual inertia control mode to simulate the rotational inertia characteristics of a synchronous generator. This allows for rapid release of stored energy to support the grid frequency during the initial stages of a disturbance. Simultaneously, the distributed photovoltaic inverter initiates droop control, automatically adjusting active power output based on frequency deviation to provide voltage support in conjunction with the energy storage system. Both systems coordinate their response pace rapidly through consistent communication, effectively suppressing further frequency and voltage drops.

[0054] As production load gradually stabilizes and the disturbance intensity assessment index falls below the threshold, step S54 initiates a gradual switchback to the economic optimization mode. The switchover process employs a smooth transition function, gradually reducing the virtual inertia control weight and slowly restoring the proportion of the economic optimization objective. This ensures a smooth transition of the control strategy within minutes, avoiding power surges and secondary disturbances during mode switching, and ultimately restoring the microgrid to an economically efficient normal operating state.

[0055] The event-triggered resilient control mode, through real-time monitoring and dynamic threshold assessment, achieves rapid perception and classification response to grid disturbances, transforming traditional passive post-fault handling into proactive defense. When a disturbance occurs, the mode switching mechanism decisively freezes economic targets and prioritizes safety and stability, avoiding conflicts between economic optimization and safety and stability objectives, and significantly improving the microgrid's operational resilience under extreme conditions. The collaborative design of energy storage virtual inertia control and distributed power source droop control fully utilizes the complementary characteristics of multiple resource types, achieving rapid frequency and voltage support, and effectively suppressing the spread and duration of disturbance impacts. The progressive smooth transition function solves the problem of secondary impacts during mode switching, enabling the system to seamlessly recover to the economically optimized operating state after the disturbance subsides, ensuring a balance between long-term economic efficiency and safety. This mode provides the microgrid with a dual-modal control architecture with environmental adaptability, significantly enhancing the power supply quality and equipment safety level in scenarios with high uncertainty and high penetration of new energy access.

[0056] In step S50: Based on the modified power command, hierarchical cooperative control is executed. When a grid disturbance event is detected, a flexible control mode is triggered. The control strategy is pre-simulated using a digital twin through multi-physics coupling, generating a robust control command sequence with pre-simulation verification. The digital twin pre-simulation process includes the following steps: S54: Construct a multi-physics coupled digital twin of a microgrid, integrating electromagnetic transient models, thermodynamic models, and equipment aging models; S55: Inject the modified power command sequence into the twin and simulate the power flow distribution, node voltage curves and equipment temperature field after execution in parallel. S56: Inject random disturbances during the rehearsal process to conduct stress tests and evaluate the robustness margin of the control strategy under extreme conditions. S57: If the simulation results meet the dual objectives of safety and economy, the instruction sequence is confirmed to be executable; otherwise, it is fed back to the distributed model predictive control layer for re-optimization, forming a decision-simulation-verification closed loop.

[0057] Specifically, before implementing hierarchical collaborative control in the industrial park microgrid, the system first completes step S54 to construct a multi-physics coupled digital twin of the microgrid. This twin integrates an electromagnetic transient model to characterize power flow and voltage dynamics, a thermodynamic model to simulate the temperature field of the energy storage battery compartment and the heat dissipation state of the photovoltaic inverter, and an equipment aging model to track the health degradation trend of key components, forming a virtual mirror that is synchronized with the actual physical system in real time.

[0058] In step S55, the modified energy storage charging and discharging power command sequence and distributed power dispatching commands from step S40 are injected into the twin. The twin performs parallel simulations of the operating status for the next 4 hours: the electromagnetic transient model calculates the power flow distribution of each branch, showing that the peak load rate of the main transformer reaches 80% of its rated capacity; the node voltage curve shows that the minimum voltage of the terminal bus remains within acceptable limits; the thermodynamic model simulates the temperature field of the energy storage battery compartment, predicting that continuous high-power charging and discharging will cause the maximum temperature to rise by five degrees Celsius. The multiphysics simulation results comprehensively reveal the global state of the system after the commands are executed.

[0059] Step S56 involves proactively injecting random disturbances during the pre-test to conduct stress testing. Random noise, consistent with actual fluctuation characteristics, is superimposed onto the photovoltaic output trajectory to simulate a scenario of rapidly changing cloud cover. Sudden load switching events are injected at load nodes to test the control strategy's resilience. Twin simulations show that after the added disturbances, the maximum frequency deviation remains within the allowable range, voltage fluctuations do not trigger protection actions, and the energy storage temperature rise remains within the thermal management tolerance, indicating that the control strategy has sufficient robustness margin.

[0060] Step S57 evaluates the pre-simulation results, confirming that it meets both safety and economic objectives: power quality indicators are qualified, equipment temperature and load rate do not exceed limits, and economic cost is close to the theoretical optimum. The system determines that the instruction sequence has strong robustness and high feasibility, and confirms its execution. If the pre-simulation results show that the voltage of a certain node exceeds the limit or the energy storage temperature rise exceeds the standard, the feedback mechanism is triggered, returning the evaluation results to the distributed model predictive control layer of step S30, automatically tightening voltage constraints or reducing the power command strength before re-optimization, forming a decision-making-pre-simulation-verification closed loop.

[0061] The digital twin pre-simulation process constructs a multi-physics coupled mirror in virtual space, enabling the advance simulation and verification of control strategy execution effects. This transforms traditional passive control, reliant on post-event response, into proactive, predictive decision-making. The integration of multi-physics models allows the pre-simulation to consider not only electrical steady-state conditions but also equipment thermal states and aging trends, avoiding latent fault risks caused by neglecting thermal effects or lifespan accumulation. The stress testing mechanism with random disturbance injection can assess the robustness of the strategy under extreme operating conditions in a harmless environment, quantifying safety margins and ensuring the strategy remains effective under real disturbance impacts. The closed-loop feedback design deeply integrates optimization and verification; strategies that fail verification are automatically re-optimized, ensuring strong robustness and high reliability of the final issued commands. This process significantly improves the operational safety and economy of microgrids under conditions of high uncertainty and high renewable energy penetration, providing a scientific virtual testbed for high-risk control decisions.

[0062] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0063] In one embodiment, a smart energy management and control device for a new energy microgrid is provided, which corresponds one-to-one with the smart energy management and control method for a new energy microgrid in the above embodiments. For example... Figure 2 As shown, the intelligent energy management and control device for the new energy microgrid includes a standardized time-series operation dataset generation unit 1, which is used to collect grid operation status data and environmental monitoring data in the microgrid through a multi-source heterogeneous sensor network, and generate a standardized time-series operation dataset after spatiotemporal alignment and outlier removal. The power and load demand prediction unit 2 is used to input the time-series operation dataset into the multi-time-scale power prediction model. The multi-time-scale power prediction model uses a physical information neural network architecture to embed the physical constraint equation of new energy output and outputs the power generation and load demand prediction results corresponding to different time scales. The operation scheduling instruction generation unit 3 is used to feed the prediction results into the distributed model prediction control layer. The distributed model prediction control layer realizes multi-agent collaborative optimization based on the consensus algorithm and solves the operation scheduling instructions of each distributed power source. The dynamic power correction command generation unit 4 is used to send the operation scheduling command to the virtual energy storage aggregation and lifetime awareness optimization module. The lifetime awareness optimization module quantifies the cumulative loss of cycle life caused by frequent charging and discharging through a differentiable energy storage decay empirical model to generate dynamic power correction commands. The robust control command sequence generation unit 5 is used to perform hierarchical collaborative control based on the modified power command. When a grid disturbance event is detected, it triggers the elastic control mode and relies on the digital twin to perform multi-physics coupling pre-simulation of the control strategy to generate a robust control command sequence with pre-simulation verification.

[0064] Specific limitations regarding the intelligent energy management and control device for new energy microgrids can be found in the limitations of the intelligent energy management and control method for new energy microgrids mentioned above, and will not be repeated here. Each module in the aforementioned intelligent energy management and control device for new energy microgrids can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the electronic device, or stored in the memory of the electronic device as software, so that the processor can call and execute the corresponding operations of each module.

[0065] In one embodiment, an electronic device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3As shown, this electronic device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The database stores the database. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a smart energy management and control method for a new energy microgrid.

[0066] In one embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: The grid operation status data and environmental monitoring data in the microgrid are collected by a multi-source heterogeneous sensor network. After spatiotemporal alignment and outlier removal, a standardized time-series operation dataset is generated. The grid operation status data includes electrical parameters of distributed power sources, energy storage devices, flexible loads and grid buses. The time-series running dataset is input into the multi-time-scale power prediction model. The multi-time-scale power prediction model uses a physical information neural network architecture to embed the physical constraint equation of new energy output and outputs the power generation and load demand prediction results corresponding to different time scales. The prediction results are fed into the distributed model prediction control layer, which implements multi-agent collaborative optimization based on the consensus algorithm to solve for the operation scheduling instructions of each distributed power source. The operation scheduling command is sent to the virtual energy storage aggregation and lifetime awareness optimization module. The lifetime awareness optimization module quantifies the cumulative loss of cycle life caused by frequent charging and discharging through a differentiable energy storage decay empirical model to generate a dynamic correction power command. The dynamic correction power command is used to achieve the coordinated optimization of energy balance and lifetime protection. Based on the modified power command, hierarchical collaborative control is executed. When a grid disturbance event is detected, an elastic control mode is triggered. The control strategy is pre-simulated using a digital twin and multi-physics coupling is performed to generate a robust control command sequence with pre-simulation verification.

[0067] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: The grid operation status data and environmental monitoring data in the microgrid are collected by a multi-source heterogeneous sensor network. After spatiotemporal alignment and outlier removal, a standardized time-series operation dataset is generated. The grid operation status data includes electrical parameters of distributed power sources, energy storage devices, flexible loads and grid buses. The time-series running dataset is input into the multi-time-scale power prediction model. The multi-time-scale power prediction model uses a physical information neural network architecture to embed the physical constraint equation of new energy output and outputs the power generation and load demand prediction results corresponding to different time scales. The prediction results are fed into the distributed model prediction control layer, which implements multi-agent collaborative optimization based on the consensus algorithm to solve for the operation scheduling instructions of each distributed power source. The operation scheduling command is sent to the virtual energy storage aggregation and lifetime awareness optimization module. The lifetime awareness optimization module quantifies the cumulative loss of cycle life caused by frequent charging and discharging through a differentiable energy storage decay empirical model to generate a dynamic correction power command. The dynamic correction power command is used to achieve the coordinated optimization of energy balance and lifetime protection. Based on the modified power command, hierarchical collaborative control is executed. When a grid disturbance event is detected, an elastic control mode is triggered. The control strategy is pre-simulated using a digital twin and multi-physics coupling is performed to generate a robust control command sequence with pre-simulation verification.

[0068] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0069] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0070] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A smart energy management and control method for a new energy microgrid, characterized in that, The method includes the following steps: collecting power grid operation status data and environmental monitoring data in the microgrid through a multi-source heterogeneous sensor network, and generating a standardized time-series operation dataset after spatiotemporal alignment and outlier removal; The time-series running dataset is input into the multi-time-scale power prediction model. The multi-time-scale power prediction model uses a physical information neural network architecture to embed the physical constraint equation of new energy output and outputs the power generation and load demand prediction results corresponding to different time scales. The prediction results are fed into the distributed model prediction control layer, which implements multi-agent collaborative optimization based on the consensus algorithm to solve for the operation scheduling instructions of each distributed power source. The operation scheduling command is sent to the virtual energy storage aggregation and lifetime awareness optimization module. The lifetime awareness optimization module quantifies the cumulative loss of cycle life caused by frequent charging and discharging through a differentiable energy storage decay empirical model to generate a dynamic correction power command. The dynamic correction power command is used to achieve the coordinated optimization of energy balance and lifetime protection. Based on the modified power command, hierarchical collaborative control is executed. When a grid disturbance event is detected, an elastic control mode is triggered. The control strategy is pre-simulated using a digital twin and multi-physics coupling is performed to generate a robust control command sequence with pre-simulation verification.

2. The intelligent energy management and control method for a new energy microgrid according to claim 1, characterized in that, The process for constructing the multi-timescale power prediction model includes: A physical information neural network backbone is constructed, whose input layer receives the time-series running dataset, the hidden layer adopts a fully connected structure, and the nonlinear physical equation between photovoltaic power output and irradiance and the cubic relationship constraint between wind turbine output and wind speed are embedded in the loss function. The physical constraint equations are automatically differentiated, and the derivative residuals are added to the network loss as a penalty term to ensure that the prediction results satisfy the power ramp-up rate limit and the law of energy conservation. An adaptive time window mechanism is adopted, using a second-level sampling interval and shallow network structure for ultra-short-term predictions, and an hour-level sampling interval and deep temporal convolution structure for medium- and long-term predictions, to achieve decoupled output across multiple time scales. An uncertainty quantization branch is integrated into the model output layer, and a prediction interval is generated through Monte Carlo deactivation, providing confidence weights for subsequent robust optimization.

3. The intelligent energy management and control method for a new energy microgrid according to claim 1, characterized in that, The step of feeding the prediction results into the distributed model prediction control layer, which uses a consensus algorithm to achieve multi-agent collaborative optimization and obtain the operation scheduling instructions for each distributed power source, includes the following steps: Each distributed resource in the microgrid is abstracted into an independent intelligent agent. Each intelligent agent maintains a local cost function, which includes a power generation cost term, a deviation penalty term, and a coupling constraint term. A consistent communication topology is constructed, in which the agent only exchanges the power plan trajectory in the predicted time domain with neighboring nodes, and the local optimum is solved iteratively by the Lagrange multiplier method. An asynchronous update mechanism is introduced, and each agent dynamically adjusts the consensus iteration frequency according to its local computing load. After convergence, it generates active and reactive power adjustment instructions, energy storage power instructions and load scheduling instructions for each node. The generated instruction sequence is checked for feasibility. If it violates the physical constraints of the device, constraint tightening and re-optimization are initiated until a feasible solution set is obtained.

4. The intelligent energy management and control method for a new energy microgrid according to claim 1, characterized in that, The step of sending the operation scheduling command to the virtual energy storage aggregation and lifetime awareness optimization module, in which the lifetime awareness optimization module quantifies the cumulative loss of cycle life due to frequent charging and discharging through a differentiable energy storage degradation empirical model to generate a dynamically corrected power command, includes the following steps: A differentiable empirical model for energy storage degradation is constructed. The empirical model for energy storage degradation takes the cumulative charge-discharge depth, the number of cycles, and the temperature as inputs, and outputs the capacity degradation rate and the health status assessment value. The energy storage charging and discharging power command is input into the attenuation model to calculate the lifetime loss gradient in the future prediction time domain. A lifetime protection regularization term is added to the optimization objective. The weight of the lifetime protection regularization term is dynamically adjusted according to the real-time electricity price and the energy storage replacement cost to achieve a multi-objective trade-off between economy and lifetime. The gradient descent method is used to solve for the corrected charge and discharge power command, so that the corrected command reduces life loss while satisfying power balance, and outputs health status warning information.

5. The intelligent energy management and control method for a new energy microgrid according to claim 4, characterized in that, The construction of a differentiable empirical model for energy storage degradation, which takes cumulative charge-discharge depth, cycle count, and temperature as inputs, and outputs capacity degradation rate and health status assessment value, includes the following steps: Collect historical operating data of energy storage devices, including charge / discharge depth sequences, terminal voltage curves, and ambient temperature records; Gaussian process regression was used to fit the capacity decay surface, and an empirical function with cycle depth, temperature and number of cycles as independent variables was constructed. By analytically differentiating the empirical function, the sensitivity gradient of the capacity decay rate to the charge and discharge power is obtained. By embedding a temperature acceleration factor into the model, the decay rate is increased according to the Arrhenius equation when the ambient temperature exceeds a threshold, thus achieving temperature-dependent lifetime sensing.

6. The intelligent energy management and control method for a new energy microgrid according to claim 1, characterized in that, In the step of performing hierarchical coordinated control based on the modified power command and triggering a resilient control mode when a grid disturbance event is detected, the event-triggered resilient control mode includes: Real-time monitoring of power grid frequency, voltage amplitude, and power fluctuation rate; and construction of disturbance intensity assessment indicators. When the evaluation index exceeds the preset trigger threshold, the elastic control mode is activated, the economic optimization target is frozen, and power balance and power quality are prioritized. Construct a fast-response control law, and achieve rapid frequency and voltage support through the coordinated control of energy storage virtual inertia and distributed power source droop control; After the disturbance subsides, the system gradually switches back to the economic optimization mode, and a smooth transition function is used during the switching process to avoid secondary shocks.

7. The intelligent energy management and control method for a new energy microgrid according to claim 6, characterized in that, In the step of executing hierarchical cooperative control based on the modified power command, triggering a resilient control mode when a grid disturbance event is detected, and performing multi-physics coupling pre-simulation of the control strategy using a digital twin to generate a robust control command sequence with pre-simulation verification, the digital twin pre-simulation process includes the following steps: Construct a multi-physics coupled digital twin of a microgrid, integrating electromagnetic transient models, thermodynamic models, and equipment aging models; The modified power command sequence is injected into the twin, and the power flow distribution, node voltage curves and equipment temperature field after execution are simulated in parallel. During the rehearsal process, random disturbances were injected to conduct stress tests and evaluate the robustness margin of the control strategy under extreme conditions. If the simulation results meet both safety and economic objectives, the instruction sequence is confirmed to be executable; otherwise, it is fed back to the distributed model predictive control layer for re-optimization, forming a decision-simulation-verification closed loop.

8. A smart energy management and control device for a new energy microgrid, applied to the smart energy management and control method for a new energy microgrid as described in any one of claims 1 to 7, characterized in that, The device includes: The standardized time-series running dataset generation unit (1) is used to collect power grid operation status data and environmental monitoring data in microgrids through multi-source heterogeneous sensor networks, and generate a standardized time-series running dataset after spatiotemporal alignment and outlier removal. The power and load demand prediction unit (2) is used to input the time series operation dataset into the multi-time scale power prediction model. The multi-time scale power prediction model uses a physical information neural network architecture to embed the physical constraint equation of new energy output and outputs the power generation and load demand prediction results corresponding to different time scales. The operation scheduling instruction generation unit (3) is used to feed the prediction results into the distributed model prediction control layer. The distributed model prediction control layer realizes multi-agent collaborative optimization based on the consensus algorithm and solves the operation scheduling instructions of each distributed power source. The dynamic power correction command generation unit (4) is used to send the operation scheduling command to the virtual energy storage aggregation and lifetime perception optimization module. The lifetime perception optimization module quantifies the cumulative loss of cycle life caused by frequent charging and discharging through a differentiable energy storage decay empirical model to generate a dynamic power correction command. The robust control command sequence generation unit (5) is used to perform hierarchical collaborative control based on the modified power command. When a grid disturbance event is detected, it triggers the elastic control mode and relies on the digital twin to perform multi-physics coupling pre-drills of the control strategy to generate a robust control command sequence with pre-drill verification.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent energy management and control method for a new energy microgrid as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent energy management and control method for a new energy microgrid as described in any one of claims 1 to 7.

Images