Energy management method and system of intelligent micro-grid

Abstract

The invention discloses an energy management method and system for an intelligent micro-grid, and belongs to the technical field of power electronics, and the method comprises the steps: carrying out the multi-source data collection of a running micro-grid, and obtaining the operation state data of the micro-grid; on the basis of the operation state data, predicting the source load power of the micro-grid in a future time period to obtain a probability distribution characteristic value of the source load power; according to the probability distribution characteristic value and the operation state data, through a preset reward function, energy management optimization is carried out on interaction of a generator set, an energy storage system and a power grid of the micro-grid, and an energy scheduling decision is constructed; and sending an energy regulation and control instruction to the micro-grid according to the energy scheduling decision. Therefore, by implementing the method and the device, the problem that the reliability of energy scheduling and management is insufficient due to the fact that the influence of uncertain factors such as energy storage equipment and environment change on energy management in the operation process of the micro-grid is not considered in the prior art can be solved.

Description

Technical Field

[0001] This application belongs to the field of power electronics technology, specifically relating to an energy management method and system for a smart microgrid. Background Technology

[0002] With the increasing penetration of distributed power generation and energy storage technologies in the power grid, smart microgrids have become a core component of future power systems. However, traditional microgrid energy management systems rely excessively on accurate source-load power forecasts when dealing with microgrids with a high proportion of renewable energy integration. In the event of sudden weather changes, the forecast error increases, and dispatch commands will be unable to effectively address the power imbalance in the microgrid. This can lead to high electricity purchase costs or, in severe cases, system voltage and frequency instability, affecting power supply reliability.

[0003] Existing energy management methods often base scheduling on worst-case scenarios, leading to prolonged periods of low utilization of expensive resources such as energy storage and significantly sacrificing operational economics. Furthermore, current technologies focus solely on minimizing operating costs within the short-term window, neglecting the impact of uncertainties during microgrid operation. Simultaneously, these methods heavily rely on precisely defined mathematical models of objective functions and constraints, making them ill-suited to the complex nonlinear variations in equipment operating parameters and market conditions within microgrids. Consequently, they struggle to balance long-term optimality with real-time adaptability in energy scheduling decisions. Summary of the Invention

[0004] This application proposes an energy management method and system for smart microgrids, which can solve the problem that the existing technology does not take into account the impact of uncertain factors such as energy storage devices and environmental changes during the operation of microgrids on energy management, resulting in insufficient reliability of energy dispatch and management.

[0005] A first aspect of this application provides an energy management method for a smart microgrid, the method comprising:

[0006] Multi-source data acquisition is performed on the operating microgrid to obtain the microgrid's operating status data;

[0007] Based on the operating status data, the source-load power of the microgrid in future time periods is predicted to obtain the probability distribution characteristic value of the source-load power.

[0008] Based on the probability distribution characteristic values ​​and the operating status data, energy management optimization is performed on the interaction between the generator sets, energy storage systems and the power grid of the microgrid through a preset reward function, and energy dispatch decisions are constructed.

[0009] Based on energy dispatch decisions, energy regulation commands are issued to the microgrid.

[0010] The aforementioned scheme acquires multi-source data from the operating microgrid to obtain real-time operational data on various electrical devices, providing a precise physical data foundation for subsequent high-dimensional state space construction. Then, based on the acquired data, the source-load probability distribution for future periods is predicted, enabling the microgrid's energy management to proactively address system uncertainties and thus obtain energy management decisions more aligned with the microgrid's operational conditions. Based on the probability distribution characteristic values ​​and the operational state data, an energy dispatch decision mechanism is constructed to maintain stable microgrid operation. A reward function is also introduced during the decision generation process to assess the uncertainties of microgrid operation and the risks associated with the decision. The reward value provided by the reward function influences the agent's decision generation, thereby ensuring the reliability of grid dispatch.

[0011] In one possible implementation of the first aspect, multi-source data acquisition is performed on the operating microgrid to obtain the microgrid's operating status data, specifically as follows:

[0012] Data is collected from new energy generator sets, energy storage devices, loads, and grid interfaces in the microgrid at a preset frequency to obtain multi-source data. The multi-source data includes the real-time output of new energy generator sets, user-side load demand, the state of charge of energy storage devices, and real-time electricity price information of the main grid.

[0013] The multi-source data is fused to obtain the operating status data.

[0014] The above solution collects various operational data from various electrical devices in the microgrid, enabling energy management of the microgrid from aspects such as generator sets, energy storage devices and grid interfaces. Moreover, the collection of multi-source data ensures the comprehensiveness and accuracy of operational status data, achieving effective and refined energy management.

[0015] In one possible implementation of the first aspect, based on the operating status data, the source-load power of the microgrid in future time periods is predicted to obtain the probability distribution characteristic value of the source-load power, specifically as follows:

[0016] The operating status data is input into a preset deep learning prediction model. Based on the risk level coefficient, the model uses the current user-side load demand, renewable energy output, and real-time electricity price information to predict the load, renewable energy output, and energy storage charge status of the microgrid within a preset future time window, and outputs the probability distribution feature value.

[0017] The future time window is determined by a set future time period; the probability distribution characteristic values ​​include the expected value and the standard deviation.

[0018] The above scheme introduces a deep learning model for probabilistic prediction, which can more accurately characterize the uncertainty of key variables such as future load and renewable energy output.

[0019] In one possible implementation of the first aspect, the risk level coefficient is specifically:

[0020] Collect real-time meteorological data of the area where the microgrid is located, and construct a risk level coefficient based on the fluctuation of the real-time meteorological data;

[0021] The standard deviation of the probability distribution characteristic values ​​is dynamically adjusted using a risk level coefficient.

[0022] When the risk level coefficient exceeds the first threshold, the standard deviation is increased.

[0023] The above scheme takes into account the fluctuations in real-time meteorological data to construct a risk level coefficient. The uncertainty of the prediction is dynamically quantified through the risk level coefficient, which equips the intelligent agent with risk perception capabilities and obtains more accurate and reliable energy scheduling decisions.

[0024] In one possible implementation of the first aspect, based on the probability distribution characteristic values ​​and the operating state data, energy management optimization is performed on the interaction between the microgrid's generator sets, energy storage systems, and the grid through a preset reward function to construct energy dispatch decisions, specifically as follows:

[0025] The probability distribution feature values ​​and the operating state data are fused to obtain a high-dimensional state vector.

[0026] Based on the decision-making network of the intelligent agent, corresponding control actions are generated for the interaction of the generator set, energy storage system and power grid through high-dimensional state vectors, and the control actions are optimized through the reward function to construct energy scheduling decisions.

[0027] The above scheme first constructs a high-dimensional state vector in a high-dimensional state space through data fusion, and then generates corresponding control actions for various types of electrical equipment through an intelligent agent, realizing multi-objective and collaborative optimization scheduling of complex microgrid systems. By comprehensively considering multiple factors such as energy sources, loads, storage, and grid, and automatically learning optimization strategies, the scheme improves the intelligence level of energy management decisions and overall energy efficiency.

[0028] In one possible implementation of the first aspect, the control action is optimized through the reward function, specifically as follows:

[0029] The reward function is constructed based on the microgrid's operating costs, carbon emissions, load occupancy, and risk penalty term; wherein, the risk penalty term is obtained by setting the safe operating power value of the microgrid based on the high-dimensional state vector.

[0030] The reward function is used to set a safety boundary for the control action, and the control action is optimized based on the safety boundary.

[0031] The above scheme introduces a reward function to constrain energy management decisions, ensuring that the execution of decisions does not affect the safe operation of the microgrid and effectively preventing unsafe operations such as overcharging, over-discharging, and exceeding power limits. Furthermore, by constructing the reward function from multiple objectives, including operating costs, carbon emissions, load sufficiency, and risk penalties, it can achieve multiple objectives of energy management.

[0032] In one possible implementation of the first aspect, an energy regulation command is issued to the microgrid based on an energy dispatch decision, specifically as follows:

[0033] Based on energy dispatch decisions, energy regulation commands are sent to the microgrid;

[0034] According to the energy regulation command, the execution control unit configured in the microgrid adjusts the charging and discharging power of the energy storage device, the output power of the thermal power generator set, and the energy flow mode between the microgrid and the main grid.

[0035] The above scheme clarifies the conversion and execution path from optimization decisions to specific equipment actions, ensuring the feasibility and effectiveness of energy dispatch decisions. By executing the control unit to coordinate and precisely regulate each key component, optimized control of power balance within the microgrid and its interaction with the main grid is achieved.

[0036] One possible implementation of the first aspect also includes:

[0037] When the energy regulation command causes the microgrid's operating state to deviate from the preset desired state, a negative reward value will be generated and fed back to the agent.

[0038] The negative reward value is used to constrain the energy scheduling decision-making process.

[0039] The above scheme introduces a closed-loop negative feedback mechanism based on actual operating conditions. When the performance deviates from expectations, negative reward feedback can correct and constrain the agent's decision-making strategy in real time, enabling the system to learn online and continuously improve, thereby enhancing the adaptability and long-term optimization effect of the energy management method. Attached Figure Description

[0040] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0041] Figure 1 This is a schematic flowchart illustrating a smart microgrid energy management method according to an embodiment of this application;

[0042] Figure 2 This is a structural diagram of an energy management method for a smart microgrid provided in an embodiment of this application;

[0043] Figure 3 This is a structural diagram of a terminal device provided in an embodiment of this application. Detailed Implementation

[0044] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0045] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0046] First Embodiment

[0047] A microgrid is a small-scale autonomous power generation and distribution system that organically combines distributed generators, energy storage systems, controllable loads, and local monitoring and protection devices. It can operate in parallel with the main power grid or, in the event of grid failure or other needs, operate independently in islanded mode. Energy management is the core means for achieving safe, stable, economical, and efficient operation of a microgrid. Traditional energy management methods heavily rely on predicted source and load power, failing to consider the impact of weather changes, leading to significant errors in predictions under sudden weather events. Furthermore, the uncertainty of predictions is not taken into account during the prediction process, making it difficult to ensure the reliability of the forecasts.

[0048] like Figure 1 As shown, to address the problem in existing technologies that fail to consider the impact of uncertainties such as energy storage devices and environmental changes during microgrid operation on energy management, resulting in insufficient reliability of energy dispatch and management, the first embodiment of this application provides a detailed flowchart of an energy management method for smart microgrids. The energy management method for smart microgrids in this embodiment includes steps S1 to S4, detailed below:

[0049] Step S1: Collect multi-source data from the operating microgrid to obtain the operating status data of the microgrid.

[0050] The energy management method provided in this application is mainly accomplished by three major units: a field data acquisition unit, a core computing service unit, and an execution control unit. The field data acquisition unit is used to collect real-time operational and environmental data from distributed power sources, energy storage devices, loads, and grid interfaces within the microgrid. The core computing service unit communicates with the field data acquisition unit to receive the real-time collected data and perform data prediction and energy dispatch decision generation. The execution control unit communicates with the core computing service unit to receive control commands corresponding to the energy dispatch decisions and control the power operation of each electrical device.

[0051] The on-site data acquisition unit consists of a high-precision smart meter, environmental sensors, and a remote terminal unit. The environmental sensors include an irradiance meter and an anemometer. The on-site data acquisition unit collects operational data from electrical equipment in the microgrid, such as new energy generator sets, energy storage devices, loads, and grid interfaces, at a frequency of seconds, enabling the acquisition of multi-source data.

[0052] The collected multi-source data mainly includes: real-time output power of photovoltaic / wind turbine units, user-side load demand, state of charge of energy storage devices, and real-time electricity price information of the main power grid.

[0053] Then, the multi-source data is fused to obtain the microgrid's operating status data.

[0054] The on-site data acquisition unit enables high-frequency sampling and multi-source fusion, providing a precise physical data foundation for the subsequent construction of a high-dimensional state space.

[0055] Step S2: Based on the operating status data, predict the source-load power of the microgrid in future time periods to obtain the probability distribution characteristic value of the source-load power.

[0056] The core computing service unit provided in this application embodiment employs an industrial-grade edge computing server equipped with a high-performance GPU. Deep learning prediction models and agents are also deployed within the core computing service unit. The agent is based on DRLAgent, which is an agent implementing reinforcement learning strategies through deep neural networks to make decisions in complex environments to maximize long-term rewards. Utilizing parallel computing capabilities, the core computing service unit runs complex deep neural networks and executes PPO reinforcement learning algorithms in real time. Compared to traditional servers, this unit can handle high-dimensional matrix operations, making it possible to apply complex probabilistic prediction and deep reinforcement learning to millisecond-level real-time control.

[0057] After receiving the operational status data uploaded by the field data acquisition unit, the core computing service unit inputs the data into a preset deep learning prediction model to predict the source-load power within a future time window.

[0058] Specifically, to address the issue that simply predicting source load power alone cannot avoid uncertainty risks, this application embodiment, based on the degree of risk, uses current user-side load demand, renewable energy output, and real-time electricity price information to predict the microgrid's load, renewable energy output, and energy storage state of charge within a preset future time window, obtaining corresponding probability distribution characteristic values. Specifically, user-side load demand, renewable energy output, and real-time electricity price information from the previous time period are extracted from the operational status data; and a corresponding future time window is set according to the future period to be predicted.

[0059] The probability distribution feature value is essentially the probability distribution of each predicted target in the future time period, including the expected value and standard deviation of the prediction. This probability distribution feature value can characterize the operating state of each electrical device in the future time period. The state vector of each electrical device corresponding to the probability distribution feature value is shown below:

[0060]

[0061] In the formula, S t For the state vector, SOC t The current state of charge of the energy storage. These are the current load demand, renewable energy output, and real-time electricity price, respectively. This represents the expected values ​​of load demand and renewable energy output within a future time period τ. The standard deviation within the future time period τ; The level of risk.

[0062] Furthermore, to respond to changes in the external environment, this embodiment of the application also collects real-time meteorological data of the microgrid's area, and then sets a corresponding risk level coefficient based on the fluctuation of the real-time meteorological data to dynamically quantify the uncertainty of the prediction. When the weather changes abruptly (i.e., when the fluctuation exceeds a set threshold), the risk level coefficient increases, and then the standard deviation of the probability distribution characteristic value is dynamically adjusted by the increased risk level coefficient, that is, the standard deviation is increased, so that the agent not only knows "how much electricity will be generated in the future", but also knows "how accurate the prediction is", thereby possessing risk perception capabilities.

[0063] Step S3: Based on the probability distribution characteristic value and the operating status data, energy management optimization is performed on the interaction between the generator sets, energy storage system and grid of the microgrid through a preset reward function to construct energy dispatching decisions.

[0064] To enhance the ability to perceive prediction uncertainty, the probability distribution feature values ​​and the operational state data are fused to obtain a high-dimensional state vector. This high-dimensional state vector is then input into the agent to generate energy scheduling decisions.

[0065] The high-dimensional state vector is analyzed through a policy network of intelligent agents, which continuously interact with the environment during the analysis process. This generates corresponding optimal control actions for the interaction between the microgrid's generators, energy storage systems, and the power grid, thus constructing energy dispatch decisions. The expression for each control action is as follows:

[0066]

[0067] In the formula, A t For the control action at time t in the future, These are control commands exchanged between energy storage devices, generator sets, and the power grid to schedule output power. This is a wind / solar curtailment command.

[0068] In the process of constructing energy dispatch decisions, the agent also provides a reward function to continuously optimize the control actions according to the actual operating environment of the microgrid, so that the output energy dispatch decisions can better cope with the uncertainties brought about by environmental fluctuations.

[0069] For example, to achieve high reliability without manually setting hard constraints, the agent interacts with the simulated environment of the microgrid through a reward function, learning strategies for pre-charging and discharging energy storage devices or reserving backup capacity under different levels of prediction uncertainty to avoid potential power shortage risks and obtain corresponding control actions. Therefore, the role of the reward function is actually to train and correct the agent's behavior, optimizing the agent's decision-making capabilities.

[0070] Specifically, the reward function is mainly composed of a weighted average of operating cost, carbon emission, load occupancy level, and risk penalty term. The decision generation is adjusted by modifying these weighting factors. Load occupancy level characterizes the microgrid's power supply reliability and its ability to provide sufficient electricity to users. The calculation logic for the risk penalty term is as follows: Real-time monitoring of the deviation between the actual power value and the predicted expected value; if the deviation exceeds the confidence interval defined by the risk level, a penalty value that increases sharply with the deviation magnitude is calculated using a nonlinear function and added to the total reward function. When the risk level is low, the agent tends to execute an arbitrage strategy prioritizing economic efficiency; when the risk level is high, the agent tends to execute a conservative strategy prioritizing reliability, automatically increasing the reserve power of the energy storage system.

[0071] Optionally, in this embodiment of the application, the risk level is extracted from the high-dimensional state vector, and then the safe operating power value of the microgrid is set according to the risk level to obtain the confidence interval, and then the risk penalty term is obtained.

[0072] By optimizing energy dispatch decisions through reward functions, the energy dispatch decisions output by the embodiments of this application can not only adapt to environmental changes caused by the aging of microgrid equipment or seasonal changes, but also improve the economic benefits of energy management.

[0073] The expression for the reward function is as follows:

[0074]

[0075] In the formula, r t Let λ be the reward function. Cost , λ Env , λ Reliability , λ Risk These are the weighting factors for operating cost, carbon emissions, load factor, and risk penalty. For the operating cost component, λ Cost The larger the value, the lower the operating cost. For carbon emissions, λ Env The larger the value, the lower the carbon emissions; λ represents the degree of full load. Reliablity The larger the value, the more stable the microgrid system. This is a risk penalty item.

[0076] The expression for the risk penalty item is:

[0077]

[0078] In the formula, The actual power value used by the agent to make decisions. The predicted power expectation value is obtained through the probability distribution characteristic value; The safety boundary is set mainly by the standard deviation of the probability distribution characteristic value. For example, when k is 3, 3σ represents the corresponding confidence interval. This is a hard constraint violation flag used to verify whether the generated control actions violate the set safety boundaries.

[0079] When the agent is informed of significant uncertainty through the obtained high-dimensional state vector, the safety margin in the reward function widens. If the agent's output decision exceeds this safety margin, the risk penalty term increases sharply, leading to a decrease in the total reward. This forces the agent to learn a rule: when predictions are inaccurate (i.e., σ is large), more energy storage margin must be maintained, and full charging and discharging should be avoided to prevent high risk penalties. This mechanism transforms complex chance-constrained planning into a reward signal that the neural network can directly learn, enabling the output energy scheduling decisions to better adapt to changes in the microgrid environment.

[0080] Step S4: Based on the energy dispatch decision, issue an energy regulation command to the microgrid.

[0081] Finally, the output energy dispatch decision is transmitted to the execution control unit. The energy dispatch decision is sent to the microgrid in the form of energy regulation commands. Based on these commands, the execution control unit configured within the microgrid adjusts the charging and discharging power of the energy storage devices, the output power of the thermal power generators, and the energy flow pattern between the microgrid and the main grid. The execution control unit includes an energy storage converter, a distributed power controller, and a grid connection point circuit breaker and controller. The energy storage converter responds to the charging and discharging power commands to adjust the charging and discharging power of the energy storage devices and regulate energy throughput. The distributed power controller responds to the active power setting commands to adjust the output power of thermal power generators such as diesel generators or gas turbines. The grid connection point circuit breaker and controller respond to the grid connection power exchange commands to control the energy flow pattern between the microgrid and the main grid and the islanding mode switching.

[0082] Furthermore, when the execution of energy regulation commands causes the microgrid's operating state to deviate from the preset expected state, the core computing service unit will generate a high negative reward value and feed it back to the agent. This negative reward value will constrain the energy scheduling decision-making process, thereby inhibiting the agent from taking aggressive scheduling behaviors that exceed the tolerance range of prediction uncertainty in future decisions.

[0083] Therefore, the core computing service unit periodically collects the deviation between the actual execution effect and the predicted data fed back by the execution control unit, stores these deviation data as new training samples in the experience replay pool, and uses these new samples to fine-tune the parameters of the deep learning prediction model and the policy network parameters of the agent online, so that the output energy scheduling decision can adapt to environmental changes caused by the aging of microgrid equipment or seasonal changes.

[0084] Implementing the embodiments of this application has the following beneficial effects:

[0085] This application embodiment acquires real-time operational data of the microgrid on various electrical devices through multi-source data collection, providing a precise physical data foundation for subsequent high-dimensional state space construction. Then, based on the acquired data, the source-load probability distribution for future periods is predicted, enabling the microgrid's energy management to proactively address system uncertainties and thus obtain energy management decisions more aligned with the microgrid's operational conditions. Based on the probability distribution characteristic values ​​and the operational state data, an energy dispatch decision mechanism is constructed to maintain stable microgrid operation. A reward function is also introduced during the decision generation process to assess the uncertainties of microgrid operation and the risks associated with the decision. The reward value provided by the reward function influences the agent's decision generation, thereby achieving reliable grid dispatch.

[0086] Second Embodiment

[0087] Furthermore, in order to implement the energy management system of the smart microgrid corresponding to the above method embodiments, and to achieve the corresponding functions and technical effects, Figure 2 A structural diagram of an energy management system for a smart microgrid is provided. For ease of explanation, only the parts relevant to this embodiment are shown. The energy management system for the smart microgrid provided in this embodiment includes:

[0088] The data acquisition module 201 is used to acquire multi-source data from the operating microgrid to obtain the operating status data of the microgrid.

[0089] The energy management method provided in this application is mainly accomplished by three major units: a field data acquisition unit, a core computing service unit, and an execution control unit. The field data acquisition unit is used to collect real-time operational and environmental data from distributed power sources, energy storage devices, loads, and grid interfaces within the microgrid. The core computing service unit communicates with the field data acquisition unit to receive the real-time collected data and perform data prediction and energy dispatch decision generation. The execution control unit communicates with the core computing service unit to receive control commands corresponding to the energy dispatch decisions and control the power operation of each electrical device.

[0090] The on-site data acquisition unit consists of a high-precision smart meter, environmental sensors, and a remote terminal unit. The environmental sensors include an irradiance meter and an anemometer. The on-site data acquisition unit collects operational data from electrical equipment in the microgrid, such as new energy generator sets, energy storage devices, loads, and grid interfaces, at a frequency of seconds, enabling the acquisition of multi-source data.

[0091] The collected multi-source data mainly includes: real-time output power of photovoltaic / wind turbine units, user-side load demand, state of charge of energy storage devices, and real-time electricity price information of the main power grid.

[0092] Then, the multi-source data is fused to obtain the microgrid's operating status data.

[0093] The on-site data acquisition unit enables high-frequency sampling and multi-source fusion, providing a precise physical data foundation for the subsequent construction of a high-dimensional state space.

[0094] The future source load prediction module 202 is used to predict the source load power of the microgrid in the future period based on the operating status data, and obtain the probability distribution characteristic value of the source load power.

[0095] In this embodiment of the application, the operating status data is input into a preset deep learning prediction model. Based on the risk level coefficient, the model uses the current user-side load demand, renewable energy output, and real-time electricity price information to predict the load, renewable energy output, and energy storage charge status of the microgrid within a preset future time window, and outputs the probability distribution feature value.

[0096] The future time window is determined by a set future time period; the probability distribution characteristic values ​​include the expected value and the standard deviation.

[0097] The decision generation module 203 is used to optimize energy management for the interaction between the generator sets, energy storage systems and the grid in the microgrid based on the probability distribution feature values ​​and the operating status data, and to construct energy dispatching decisions through a preset reward function.

[0098] In this embodiment of the application, the probability distribution feature values ​​and the running state data are fused to obtain a high-dimensional state vector;

[0099] Based on the decision-making network of the intelligent agent, corresponding control actions are generated for the interaction of the generator set, energy storage system and power grid through high-dimensional state vectors, and the control actions are optimized through the reward function to construct energy scheduling decisions.

[0100] The energy management module 204 is used to issue energy regulation commands to the microgrid based on energy dispatch decisions.

[0101] Finally, the output energy dispatch decision is transmitted to the execution control unit. The energy dispatch decision is sent to the microgrid in the form of energy regulation commands. Based on these commands, the execution control unit configured within the microgrid adjusts the charging and discharging power of the energy storage devices, the output power of the thermal power generators, and the energy flow pattern between the microgrid and the main grid. The execution control unit includes an energy storage converter, a distributed power controller, and a grid connection point circuit breaker and controller. The energy storage converter responds to the charging and discharging power commands to adjust the charging and discharging power of the energy storage devices and regulate energy throughput. The distributed power controller responds to the active power setting commands to adjust the output power of thermal power generators such as diesel generators or gas turbines. The grid connection point circuit breaker and controller respond to the grid connection power exchange commands to control the energy flow pattern between the microgrid and the main grid and the islanding mode switching.

[0102] Furthermore, when the execution of energy regulation commands causes the microgrid's operating state to deviate from the preset expected state, the core computing service unit will generate a high negative reward value and feed it back to the agent. This negative reward value will constrain the energy scheduling decision-making process, thereby inhibiting the agent from taking aggressive scheduling behaviors that exceed the tolerance range of prediction uncertainty in future decisions.

[0103] Therefore, the core computing service unit periodically collects the deviation between the actual execution effect and the predicted data fed back by the execution control unit, stores these deviation data as new training samples in the experience replay pool, and uses these new samples to fine-tune the parameters of the deep learning prediction model and the policy network parameters of the agent online, so that the output energy scheduling decision can adapt to environmental changes caused by the aging of microgrid equipment or seasonal changes.

[0104] In some embodiments, the future source load prediction module 202 specifically comprises:

[0105] The core computing service unit provided in this application embodiment employs an industrial-grade edge computing server equipped with a high-performance GPU. Deep learning prediction models and agents are also deployed within the core computing service unit. The agent is based on DRLAgent, which is an agent implementing reinforcement learning strategies through deep neural networks to make decisions in complex environments to maximize long-term rewards. Utilizing parallel computing capabilities, the core computing service unit runs complex deep neural networks and executes PPO reinforcement learning algorithms in real time. Compared to traditional servers, this unit can handle high-dimensional matrix operations, making it possible to apply complex probabilistic prediction and deep reinforcement learning to millisecond-level real-time control.

[0106] After receiving the operational status data uploaded by the field data acquisition unit, the core computing service unit inputs the data into a preset deep learning prediction model to predict the source-load power within a future time window.

[0107] Specifically, to address the issue that simply predicting source load power alone cannot avoid uncertainty risks, this application embodiment, based on the degree of risk, uses current user-side load demand, renewable energy output, and real-time electricity price information to predict the microgrid's load, renewable energy output, and energy storage state of charge within a preset future time window, obtaining corresponding probability distribution characteristic values. Specifically, user-side load demand, renewable energy output, and real-time electricity price information from the previous time period are extracted from the operational status data; and a corresponding future time window is set according to the future period to be predicted.

[0108] The probability distribution feature value is essentially the probability distribution of each predicted target in the future time period, including the expected value and standard deviation of the prediction. This probability distribution feature value can characterize the operating state of each electrical device in the future time period. The state vector of each electrical device corresponding to the probability distribution feature value is shown below:

[0109]

[0110] In the formula, S t For the state vector, SOC tThe current state of charge of the energy storage. These are the current load demand, renewable energy output, and real-time electricity price, respectively. This represents the expected values ​​of load demand and renewable energy output within a future time period τ. The standard deviation within the future time period τ; The level of risk.

[0111] Furthermore, to respond to changes in the external environment, this embodiment of the application also collects real-time meteorological data of the microgrid's area, and then sets a corresponding risk level coefficient based on the fluctuation of the real-time meteorological data to dynamically quantify the uncertainty of the prediction. When the weather changes abruptly (i.e., when the fluctuation exceeds a set threshold), the risk level coefficient increases, and then the standard deviation of the probability distribution characteristic value is dynamically adjusted by the increased risk level coefficient, that is, the standard deviation is increased, so that the agent not only knows "how much electricity will be generated in the future", but also knows "how accurate the prediction is", thereby possessing risk perception capabilities.

[0112] In some embodiments, the decision generation module 203 specifically comprises:

[0113] To enhance the ability to perceive prediction uncertainty, the probability distribution feature values ​​and the operational state data are fused to obtain a high-dimensional state vector. This high-dimensional state vector is then input into the agent to generate energy scheduling decisions.

[0114] The high-dimensional state vector is analyzed through a policy network of intelligent agents, which continuously interact with the environment during the analysis process. This generates corresponding optimal control actions for the interaction between the microgrid's generators, energy storage systems, and the power grid, thus constructing energy dispatch decisions. The expression for each control action is as follows:

[0115]

[0116] In the formula, A t For the control action at time t in the future, These are control commands exchanged between energy storage devices, generator sets, and the power grid to schedule output power. This is a wind / solar curtailment command.

[0117] In the process of constructing energy dispatch decisions, the agent also provides a reward function to continuously optimize the control actions according to the actual operating environment of the microgrid, so that the output energy dispatch decisions can better cope with the uncertainties brought about by environmental fluctuations.

[0118] For example, to achieve high reliability without manually setting hard constraints, the agent interacts with the simulated environment of the microgrid through a reward function, learning strategies for pre-charging and discharging energy storage devices or reserving backup capacity under different levels of prediction uncertainty to avoid potential power shortage risks and obtain corresponding control actions. Therefore, the role of the reward function is actually to train and correct the agent's behavior, optimizing the agent's decision-making capabilities.

[0119] Specifically, the reward function is mainly composed of a weighted average of operating cost, carbon emission, load occupancy level, and risk penalty term. The decision generation is adjusted by modifying these weighting factors. Load occupancy level characterizes the microgrid's power supply reliability and its ability to provide sufficient electricity to users. The calculation logic for the risk penalty term is as follows: Real-time monitoring of the deviation between the actual power value and the predicted expected value; if the deviation exceeds the confidence interval defined by the risk level, a penalty value that increases sharply with the deviation magnitude is calculated using a nonlinear function and added to the total reward function. When the risk level is low, the agent tends to execute an arbitrage strategy prioritizing economic efficiency; when the risk level is high, the agent tends to execute a conservative strategy prioritizing reliability, automatically increasing the reserve power of the energy storage system.

[0120] Optionally, in this embodiment of the application, the risk level is extracted from the high-dimensional state vector, and then the safe operating power value of the microgrid is set according to the risk level to obtain the confidence interval, and then the risk penalty term is obtained.

[0121] By optimizing energy dispatch decisions through reward functions, the energy dispatch decisions output by the embodiments of this application can not only adapt to environmental changes caused by the aging of microgrid equipment or seasonal changes, but also improve the economic benefits of energy management.

[0122] The expression for the reward function is as follows:

[0123]

[0124] In the formula, r t Let λ be the reward function. Cos t, λ Env , λ Reliability , λ Risk These are the weighting factors for operating cost, carbon emissions, load factor, and risk penalty. For the operating cost component, λ Cost The larger the value, the lower the operating cost. For carbon emissions, λ Env The larger the value, the lower the carbon emissions; λ represents the degree of full load. ReliablityThe larger the value, the more stable the microgrid system. This is a risk penalty item.

[0125] The expression for the risk penalty item is:

[0126]

[0127] In the formula, The actual power value used by the agent to make decisions. The predicted power expectation value is obtained through the probability distribution characteristic value; The safety boundary is set mainly by the standard deviation of the probability distribution characteristic value. For example, when k is 3, 3σ represents the corresponding confidence interval. This is a hard constraint violation flag used to verify whether the generated control actions violate the set safety boundaries.

[0128] When the agent is informed of significant uncertainty through the obtained high-dimensional state vector, the safety margin in the reward function widens. If the agent's output decision exceeds this safety margin, the risk penalty term increases sharply, leading to a decrease in the total reward. This forces the agent to learn a rule: when predictions are inaccurate (i.e., σ is large), more energy storage margin must be maintained, and full charging and discharging should be avoided to prevent high risk penalties. This mechanism transforms complex chance-constrained planning into a reward signal that the neural network can directly learn, enabling the output energy scheduling decisions to better adapt to changes in the microgrid environment.

[0129] Implementing the embodiments of this application has the following beneficial effects:

[0130] This application embodiment acquires real-time operational data of the microgrid on various electrical devices through multi-source data collection, providing a precise physical data foundation for subsequent high-dimensional state space construction. Then, based on the acquired data, the source-load probability distribution for future periods is predicted, enabling the microgrid's energy management to proactively address system uncertainties and thus obtain energy management decisions more aligned with the microgrid's operational conditions. Based on the probability distribution characteristic values ​​and the operational state data, an energy dispatch decision mechanism is constructed to maintain stable microgrid operation. A reward function is also introduced during the decision generation process to assess the uncertainties of microgrid operation and the risks associated with the decision. The reward value provided by the reward function influences the agent's decision generation, thereby achieving reliable grid dispatch.

[0131] Furthermore, Figure 3 This is a structural diagram of a terminal device provided in one embodiment of this application. Figure 3 As shown, the terminal device 3 of this embodiment includes: at least one processor 30 (in... Figure 3The processor 30 (only one is shown in the image) and a memory 31 and a computer program 32 stored in the memory 31 and executable on the at least one processor, wherein the processor 30 executes the computer program 32 to implement the steps of an energy management method for a smart microgrid as described in any one of the embodiments of this application.

[0132] The terminal device 3 may be a computing device such as a desktop computer, a cloud server, or a laptop computer, and the computing device may include, but is not limited to, a processor 30 and a memory 31. Figure 3 This is merely an example of terminal device 3 and does not constitute a limitation on terminal device 3. It may include more or fewer components than those shown in the figure.

[0133] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the scope of protection of this application. In particular, it should be noted that any modifications, equivalent substitutions, or improvements made by those skilled in the art within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. An energy management method for a smart microgrid, characterized in that, include: Multi-source data acquisition is performed on the operating microgrid to obtain the microgrid's operating status data; Based on the operating status data, the source-load power of the microgrid in future time periods is predicted to obtain the probability distribution characteristic value of the source-load power. Based on the probability distribution characteristic values ​​and the operating status data, energy management optimization is performed on the interaction between the generator sets, energy storage systems and the power grid of the microgrid through a preset reward function, and energy dispatch decisions are constructed. Based on energy dispatch decisions, energy regulation commands are issued to the microgrid.

2. The energy management method for a smart microgrid according to claim 1, characterized in that, The process of acquiring multi-source data from the operating microgrid to obtain its operating status data specifically involves: Data is collected from new energy generator sets, energy storage devices, loads, and grid interfaces in the microgrid at a preset frequency to obtain multi-source data. The multi-source data includes the real-time output of new energy generator sets, user-side load demand, the state of charge of energy storage devices, and real-time electricity price information of the main grid. The multi-source data is fused to obtain the operating status data.

3. The energy management method for a smart microgrid according to claim 1, characterized in that, Based on the operating status data, the source-load power of the microgrid is predicted for future periods to obtain the probability distribution characteristic value of the source-load power, specifically as follows: The operating status data is input into a preset deep learning prediction model. Based on the risk level coefficient, the model uses the current user-side load demand, renewable energy output, and real-time electricity price information to predict the load, renewable energy output, and energy storage charge status of the microgrid within a preset future time window, and outputs the probability distribution feature value. The future time window is determined by a set future time period; the probability distribution characteristic values ​​include the expected value and the standard deviation.

4. The energy management method for a smart microgrid according to claim 3, characterized in that, The risk level coefficient is specifically as follows: Collect real-time meteorological data of the area where the microgrid is located, and construct a risk level coefficient based on the fluctuation of the real-time meteorological data; The standard deviation of the probability distribution characteristic values ​​is dynamically adjusted using a risk level coefficient. When the risk level coefficient exceeds the first threshold, the standard deviation is increased.

5. The energy management method for a smart microgrid according to claim 1, characterized in that, The step involves optimizing energy management based on the probability distribution characteristic values ​​and the operating status data, using a preset reward function, and constructing energy dispatch decisions for the interaction between the microgrid's generator sets, energy storage systems, and the grid. Specifically: The probability distribution feature values ​​and the operating state data are fused to obtain a high-dimensional state vector. Based on the decision-making network of the intelligent agent, corresponding control actions are generated for the interaction of the generator set, energy storage system and power grid through high-dimensional state vectors, and the control actions are optimized through the reward function to construct energy scheduling decisions.

6. The energy management method for a smart microgrid according to claim 5, characterized in that, The optimization of the control action through the reward function specifically includes: The reward function is constructed based on the microgrid's operating costs, carbon emissions, load occupancy, and risk penalty term; wherein, the risk penalty term is obtained by setting the safe operating power value of the microgrid based on the high-dimensional state vector. The reward function is used to set a safety boundary for the control action, and the control action is optimized based on the safety boundary.

7. The energy management method for a smart microgrid according to claim 1, characterized in that, The step of issuing energy regulation commands to the microgrid based on energy dispatch decisions specifically includes: Based on energy dispatch decisions, energy regulation commands are sent to the microgrid; According to the energy regulation command, the execution control unit configured in the microgrid adjusts the charging and discharging power of the energy storage device, the output power of the thermal power generator set, and the energy flow mode between the microgrid and the main grid.

8. The energy management method for a smart microgrid according to any one of claims 1 to 7, characterized in that, Also includes: When the energy regulation command causes the microgrid's operating state to deviate from the preset desired state, a negative reward value will be generated and fed back to the agent. The negative reward value is used to constrain the energy scheduling decision-making process.

9. An energy management system for a smart microgrid, characterized in that, include: Data acquisition module, future source load prediction module, decision generation module, and energy management module; The data acquisition module is used to acquire multi-source data from the operating microgrid to obtain the operating status data of the microgrid. The future source load prediction module is used to predict the source load power of the microgrid in the future time period based on the operating status data, and obtain the probability distribution characteristic value of the source load power. The decision generation module is used to optimize energy management for the interaction between the generator sets, energy storage systems and the grid in the microgrid based on the probability distribution feature values ​​and the operating status data, through a preset reward function, and to construct energy dispatch decisions. The energy management module is used to issue energy regulation commands to the microgrid based on energy dispatch decisions.

10. A terminal device, characterized in that, It includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the steps of the energy management method for a smart microgrid according to any one of claims 1 to 8.

Images