Microgrid control method, program product, and readable storage medium

Abstract

The application provides a micro-grid control method, a program product and a readable storage medium. The micro-grid control method comprises the following steps: collecting operation data of a micro-grid and each network type device of the micro-grid, and environment data of an environment in which the micro-grid is located; performing multi-source feature fusion processing on the collected operation data and environment data, and predicting energy output, load demand and power fluctuation amplitude of the micro-grid in a future preset time period based on the fused features; optimizing voltage parameters and frequency parameters of each network type device based on a prediction result and a current state of each network type device; and issuing the optimized voltage parameters and frequency parameters as parameter adjustment instructions to corresponding network type devices, so that each network type device operates based on the issued parameter adjustment instructions. The application has the advantage that the voltage parameters and frequency parameters of each network type device can be optimized, thereby laying a key foundation for stable and efficient operation of the micro-grid.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of micro-grid control, and in particular to a micro-grid control method, a program product and a readable storage medium. BACKGROUND

[0002] Grid-forming inverter is a key device for connecting distributed energy with power grid, which can actively adjust output voltage and frequency, has similar functions to traditional synchronous generator, and provides support for autonomous control of micro-grid. In the scene of photovoltaic storage and charging heat micro-grid, multiple grid-forming devices are operated in parallel, facing three core challenges: first, the voltage / frequency is easy to fluctuate, and it is difficult to maintain system stability; second, the power distribution of multiple devices is uneven, and the problem of circulating current and loss is prominent; third, the dynamic adaptation ability is weak, and it is unable to adapt to the complex changes of photovoltaic output and load demand.

[0003] In the prior art, although multi-frequency adaptive virtual impedance, sliding mode inertia enhancement and other technologies can improve parallel performance, RBF neural network algorithm can also adjust virtual inertia and damping, there are obvious defects: most of the control parameters are statically configured, which is difficult to adapt to the complex dynamic characteristics of photovoltaic storage and charging heat scene, and it is unable to realize real-time optimization of voltage / frequency parameters; control decisions are concentrated in local or single central controller, lack of global coordination, and it is difficult to solve the problem of uneven power distribution of multiple devices; AI algorithm application is limited to single device parameter adjustment, and it does not cover multi-source data and system global, which cannot meet the requirements of dynamic adaptation and global stability. SUMMARY

[0004] An object of the present application is to provide a micro-grid control method, a program product and a readable storage medium which at least partially solve the above problems.

[0005] A further object of the present application is to reduce voltage and frequency abnormalities from the source by optimizing the voltage parameters and frequency parameters of each grid-forming device, and to lay a key foundation for stable operation of the micro-grid.

[0006] Another further object of the present application is to realize real-time optimal configuration of voltage parameters and frequency parameters, coordinate power distribution among multiple devices, and reduce circulating current and loss.

[0007] In particular, according to a first aspect of the present application, the present application provides a micro-grid control method, comprising:

[0008] Collecting operation data of a micro-grid and each grid-forming device thereof, and environmental data of an environment in which the micro-grid is located;

[0009] Performing multi-source feature fusion processing on the collected operation data and environmental data, and predicting energy output, load demand and power fluctuation amplitude of the micro-grid in a future preset time period based on the fused features;

[0010] based on the prediction result and the current state of each of the network-forming devices, optimize voltage parameters and frequency parameters of each of the network-forming devices;

[0011] The optimized voltage parameters and frequency parameters are used as parameter adjustment instructions and are sent to the corresponding network-forming devices, so that each network-forming device operates based on the sent parameter adjustment instructions.

[0012] Optionally, the network-forming devices include photovoltaic inverters, energy storage converters, charging piles, and heat pumps, and each network-forming device is integrated with a network-forming virtual synchronous machine. The step of performing multi-source feature fusion processing on the collected operation data and environment data includes:

[0013] Based on the operation data of the photovoltaic inverter, a fluctuation sequence of photovoltaic output is extracted. Based on the operation data of the energy storage converter, a change rate of charging and discharging power is extracted. Based on the operation data of the charging pile, time sequence data of the number of accesses is extracted as time sequence features.

[0014] Based on the operation data of the energy storage converter, a state of charge is extracted. Based on the operation data of the heat pump, an operation mode is extracted. Based on the operation data of the microgrid, a bus voltage deviation and a frequency fluctuation amplitude are extracted as state features.

[0015] Based on the environment data, light intensity, environmental temperature, and power grid voltage stability level are extracted as environmental features.

[0016] The time sequence features, state features, and environmental features are integrated to construct a multi-dimensional feature matrix for prediction.

[0017] Optionally, the step of predicting the energy output, load demand, and power fluctuation amplitude of the microgrid in a future preset time period based on the fused features includes:

[0018] The intelligent prediction model is used to learn the time sequence dependency in the multi-dimensional feature matrix. The model inference outputs the prediction result of the energy output, load demand, and power fluctuation amplitude of the microgrid in the future preset time period.

[0019] Optionally, the step of optimizing the voltage parameters and frequency parameters of each of the network-forming devices based on the prediction result and the current state of each of the network-forming devices includes:

[0020] A deep reinforcement learning model is used for parameter optimization. The state space of the deep reinforcement learning model includes the prediction result output by the intelligent prediction model and the control parameters of the virtual synchronous machine integrated in each network-forming device.

[0021] The action space of the deep reinforcement learning model is defined as adjustment actions of virtual inertia, damping coefficient and virtual impedance parameters of each virtual synchronous machine;

[0022] A reward function of the deep reinforcement learning model is constructed, and the reward function is a weighted combination function of frequency deviation, power allocation error and device loss rate;

[0023] Through continuous interaction between the deep reinforcement learning model and the microgrid operation environment, the cumulative reward is maximized to generate a parameter configuration strategy, and optimization of voltage parameters and frequency parameters of each network-forming device is completed.

[0024] Optionally, based on the prediction results and the current state of each network-forming device, the step of optimizing the voltage parameters and the frequency parameters of each network-forming device comprises:

[0025] Obtaining operation state parameters of each network-forming device, the operation state parameters comprising current output voltage, output frequency and control parameters of the virtual synchronous machine of the device;

[0026] Based on the prediction results of the energy output, the load demand and the power fluctuation amplitude of the microgrid in the future preset time period, the voltage parameters and the frequency parameters of each network-forming device are iteratively optimized by the deep reinforcement learning model;

[0027] Outputting expected voltage parameters and expected frequency parameters that meet the target.

[0028] Optionally, the microgrid control method further comprises:

[0029] Detecting the power fluctuation amplitude of the microgrid;

[0030] When the power fluctuation amplitude exceeds a preset threshold, triggering an emergency optimization mechanism, shortening the parameter optimization execution period of the deep reinforcement learning model, and reissuing the parameter adjustment instruction to each network-forming device after rapid re-optimization of the deep reinforcement learning model.

[0031] Optionally, the microgrid control method further comprises:

[0032] Periodically using historical data generated by the operation of the microgrid to perform offline incremental training on the deep reinforcement learning model, and dynamically updating the optimization strategy of the deep reinforcement learning model to adapt to long-term changes in the operation characteristics of the microgrid.

[0033] Optionally, the operation data and the environment data are collected for a first preset time period;

[0034] Inference of the intelligent prediction model is performed for a second preset time length, and parameter optimization of the deep reinforcement learning model is performed for a third preset time length longer than the second preset time length.

[0035] According to a second aspect of the present application, the present application provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of the microgrid control method of any one of the above.

[0036] According to a third aspect of the present application, the present application provides a machine-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the microgrid control method of any one of the above.

[0037] The microgrid control method of the present application avoids single perspective limitations to comprehensively perceive system states by collecting full-dimensional data of the microgrid as a whole, network-type devices and the environment; predicts future energy output, load demand and power fluctuations after multi-source feature fusion processing, changes passive response to active prediction, and reserves parameter adjustment time; and further optimizes voltage and frequency parameters based on the prediction results and the current state of the device, and issues the parameters as instructions for execution. In this way, the voltage and frequency parameters of each network-type device can be optimized to reduce voltage and frequency abnormalities from the source and improve system anti-interference capability.

[0038] Further, the microgrid control method of the present application significantly improves prediction accuracy and foresight by learning time sequence dependency in the multi-dimensional feature matrix through an intelligent prediction model and inferring prediction results. The intelligent model can deeply mine dynamic correlations among multi-dimensional features such as photovoltaic output fluctuations, load changes and environmental factors, and more accurately capture complex dynamic rules than traditional methods; the prediction results of energy output, load demand and power fluctuations based on this output provide a reliable foresight basis for subsequent parameter optimization, making voltage and frequency parameter adjustment more targeted and further enhancing the ability to predict and respond to intermittency and load randomness among network-type devices, laying a key foundation for stable and efficient operation of the microgrid.

[0039] Further, the micro-grid control method of the present application significantly improves the accuracy and synergy of global regulation of the micro-grid by optimizing parameters through a deep reinforcement learning model. The model incorporates the prediction results and the current control parameters of the virtual synchronous machine into the state space, enabling comprehensive consideration of the future and current states of the system; adjusts the core parameters such as virtual inertia and damping coefficient as the action space, which directly affects the voltage and frequency characteristics of the network-type equipment; through the fusion of the reward function of frequency deviation, power distribution error and equipment loss rate, a multi-objective optimization guide of stability, balance and economy is constructed. By continuously interacting with the running environment to maximize the cumulative reward, it can not only dynamically adapt to scenarios such as photovoltaic output fluctuation and load random change, realize real-time optimal configuration of voltage / frequency parameters, but also coordinate the power distribution among multiple devices, reduce circulating current and loss, ensure stable operation of the system, and improve energy utilization efficiency and equipment life.

[0040] The above and other objects, advantages and features of the present application will become more apparent from the following detailed description of specific embodiments thereof, taken in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0041] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are intended to illustrate preferred embodiments of the present application, and should not be considered limiting of the present application. Indeed, the drawings themselves can comprise embodiments of the present application which can be employed in any combination or arrangement. In the drawings:

[0042] Figure 1 is a schematic flow chart of a micro-grid control method according to an embodiment of the present application;

[0043] Figure 2 is a schematic flow chart of multi-source fusion processing according to an embodiment of the present application;

[0044] Figure 3 is a schematic diagram of a computer program product according to an embodiment of the present application;

[0045] Figure 4 is a schematic diagram of a machine-readable storage medium according to an embodiment of the present application;

[0046] Figure 5 is a schematic diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0047] Those skilled in the art should understand that the embodiments described below are only a part of the embodiments of the present application, not all embodiments of the present application, and are intended to explain the technical principles of the present application, not to limit the protection scope of the present application. Based on the embodiments provided by the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor shall fall within the protection scope of the present application.

[0048] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered listing of executable instructions for implementing logical functions, and can be specifically embodied in any computer-readable medium for use by an instruction execution system, apparatus or device, such as a computer-based system, a system including a processor, or other system that can fetch the instructions from the instruction execution system, apparatus or device and execute the instructions, or in conjunction with these instructions execution system, apparatus or device.

[0049] The embodiment provides a micro-grid control system, which comprises an edge AI control layer, a local execution layer and a perception layer.

[0050] The edge AI control layer is a core decision unit of the whole micro-grid control system, and an intelligent coordination controller is constructed based on an RK3588 chip. The RK3588 chip has the advantages of high-performance computing power and low power consumption, and can meet the requirements of AI model reasoning and real-time optimization. The edge AI control layer integrates four core function modules: a multi-source data acquisition module collects the equipment operation data and environmental data uploaded by the perception layer through a communication link, and provides original data support for subsequent processing. An AI model training and reasoning module is responsible for loading and running a long short-term memory (LSTM) prediction model and a deep reinforcement learning (DRL) optimization model, and simultaneously supports offline incremental training, for example, updating model parameters during the idle period at night; a global parameter optimization module takes system stability and efficiency optimization as the target, combines the prediction results and equipment states, and calculates the optimal voltage and frequency control parameters of each network type device. A communication interaction module realizes bidirectional data interaction with the local execution layer, downwardly issues parameter adjustment instructions, upwardly receives equipment execution state feedback, and ensures real-time and reliable instruction transmission.

[0051] The local execution layer is deployed in each networked device, including photovoltaic inverters, energy storage converters, charging piles, heat pumps, etc. The core of the local execution layer is to integrate a networked virtual synchronous machine control module in these devices. The networked virtual synchronous machine can simulate the voltage and frequency characteristics of traditional synchronous generators and solve the stability problem of new energy devices during grid connection. The networked virtual synchronous machine control module includes three key units: a local power calculation unit that calculates the current active and reactive power output of the device in real time and provides device operating state data for the edge AI control layer. A double closed-loop control unit that accurately tracks the parameter instructions issued by the edge AI control layer through the nested control logic of voltage and current closed loops to avoid voltage and current overshoot. A parameter execution unit that converts the optimized virtual synchronous machine control parameters (such as virtual inertia and damping coefficient) into device hardware executable operations to directly adjust the device output characteristics.

[0052] The perception layer is the data collection foundation of the system. By deploying various high-precision sensors and monitoring modules, it realizes real-time collection of full-dimensional data. Among them, the electrical parameter sensors include voltage sensors, current sensors, and power sensors, such as direct current power sensors with an accuracy of 0.5 level, which can collect bus voltage, device output current, and system total power, etc. The environmental monitoring module includes photovoltaic irradiance sensors and temperature sensors. The photovoltaic irradiance sensor has a measurement range of 0-2000W / m² and can collect environmental factors that affect microgrid output and load. The device state monitoring module collects real-time data such as photovoltaic output, energy storage state of charge, charging pile connection number, and heat pump operation mode (such as heating / cooling), etc. Finally, all collected data is uploaded to the edge AI control layer.

[0053] After receiving the perception layer data, the edge AI control layer does not directly use it for optimization, but first performs multi-source feature fusion processing. By extracting key features, eliminating data redundancy, and mining feature correlations, a multi-dimensional feature matrix that accurately reflects the system state is constructed. The multi-dimensional feature matrix includes three types of features: time series features that reflect the change law of data over time, including photovoltaic output fluctuation sequence within 15 minutes, energy storage charge and discharge power change rate, and charging pile connection number time series data.

[0054] Among them, the photovoltaic output fluctuation sequence within 15 minutes reflects the short-term stability of photovoltaic output, such as the output drop caused by cloud cover, the energy storage charging and discharging power change rate reflects the energy storage charging and discharging speed, which can avoid overcharging and overdischarging, and the charging pile access quantity time sequence data reflects the load change trend, such as the surge of charging demand during the peak of going to work and going home; the state characteristics reflect the current running state of the system and the equipment, including the energy storage state of charge, the heat pump operation mode, the bus voltage deviation, and the frequency fluctuation amplitude. The energy storage state of charge determines the energy storage charge and discharge capacity, which can avoid forced discharge damage to the battery when the energy storage state of charge is too low. The heat pump operation mode needs to be adjusted according to the power supply strategy due to the large difference in heat pump power consumption in different modes. The bus voltage deviation reflects the voltage stability of the system, and a large deviation will affect the normal operation of the equipment. The frequency fluctuation amplitude reflects the balance of power supply and demand of the system, and a fluctuation beyond the range may trigger protection shutdown; the environmental characteristics reflect the influence of the external environment on the system, including the light intensity, the environmental temperature, and the grid voltage stability level. The light intensity directly determines the upper limit of photovoltaic output, the environmental temperature affects the heat pump load and battery performance, and the grid voltage stability level can determine whether the external grid is reliable, providing a basis for off-grid / on-grid switching.

[0055] Based on the above multi-dimensional feature matrix, a long short-term memory (LSTM) prediction model is used. The LSTM prediction model is good at capturing long-term dependencies in time series data, such as the daily periodicity of photovoltaic output and the time periodicity of charging pile load, which can effectively avoid the lag of traditional prediction methods. Finally, the prediction model outputs three types of key prediction results within the next 3 minutes: photovoltaic output prediction value can predict the power generation capacity of photovoltaic energy in the future, providing a basis for energy storage charging and discharging plan; load demand prediction value can predict the total power demand of charging piles, heat pumps and other equipment in the future, avoiding power shortage; power fluctuation amplitude prediction value can predict the change range of system power supply and demand difference, and identify fluctuation risks in advance, such as power shortage caused by sudden increase of load.

[0056] The edge AI control layer takes frequency fluctuation minimization (ensuring system stability), power distribution error minimization (avoiding equipment overload), and device loss equalization (extending device life) as three optimization objectives, constructs a deep reinforcement learning (DRL) optimization model, and gradually finds the optimal virtual synchronous machine control parameters through the cycle of state perception, action decision, and reward feedback.

[0057] The DRL model is specifically designed as follows: The state space defines the perception range of the DRL model, including two parts of core information, one is the three types of prediction results output by the LSTM model, which can know the future risk in advance, and the other is the virtual synchronous machine control parameters (virtual inertia J, damping coefficient D, and virtual impedance Z) of the current devices, which reflect the current control state of the devices to ensure that the model decision considers future changes and is consistent with the actual ability of the devices. The action space defines the adjustment operations that the DRL model can perform, i.e., the adjustment range and step size of the virtual synchronous machine control (VSG) parameters of each device, which realizes multi-objective optimization through weighted combination, and the formula is:

[0058] R = a x (1 - | Af | / f0) + b x (1 - AP / Pn) + g x (1 - å (P_loss,i / P_rated,i) / n).

[0059] Wherein, R is the reward function, a, b, g are weight coefficients, Af is the frequency deviation, AP is the power distribution error, P_loss,i is the loss of the ith device, f0 is the rated frequency of the microgrid, Pn is the rated power of the grid-forming device in the microgrid, P_rated,i is the rated power of the ith grid-forming device; n is the total number of grid-forming devices in the microgrid.

[0060] The DRL model continuously interacts with the microgrid operating environment, such as outputting parameter adjustment actions according to the current state in each round of interaction, and calculating rewards according to the actual operating effect, and constantly optimizing the decision strategy, and finally outputting the globally optimal virtual synchronous machine control parameter combination, such as matching a larger virtual impedance for the photovoltaic inverter to suppress circulating current, and matching a larger virtual inertia for the energy storage converter to stabilize the frequency. The system realizes closed-loop control of data acquisition, prediction, optimization, and execution through clear execution cycle and interaction logic of each level, and the specific process is as follows:

[0061] In the data collection and upload link, the perception layer collects all data every 10 ms, and high-frequency collection can timely capture sudden changes, such as the sudden access of a charging pile. The local execution layer synchronously calculates the current active / reactive power of the device, and uploads the data to the edge AI control layer through communication (Baud rate 500 kbps, ensuring low delay). In the prediction model running link, the edge AI control layer runs the LSTM model every 50 ms. Since the prediction needs to be based on multiple sets of collected data (such as 10 ms collection, 50 ms can accumulate 5 sets of data), this period can ensure prediction accuracy and quickly update the prediction result to provide the latest basis for optimization. In the optimization and instruction issuing link, the edge AI control layer executes the DRL model optimization every 200 ms. Optimization requires multiple rounds of iterative calculations, and the longer period than prediction can ensure accuracy. After generating the control parameter adjustment instructions of the virtual synchronous machine of each device, it is issued to the local execution layer. In the local execution and stable control link, after receiving the instructions, the local execution layer adjusts the output voltage through the multi-frequency adaptive virtual impedance model. This model matches different impedance values for different frequency harmonic interference, can suppress circulating current, and combines the sliding mode inertia enhancement mechanism to quickly respond to frequency changes. For example, when the load suddenly increases, causing the frequency to drop, the virtual inertia is quickly increased to slow down the frequency drop speed and ensure that the device output meets the instruction requirements. In the emergency optimization mechanism link, when the edge AI control layer detects that the power fluctuation amplitude exceeds the preset threshold (such as fluctuation exceeding 20% of the rated power), it immediately triggers the emergency optimization mode, shortens the DRL model optimization period from 200 ms to 50 ms, and quickly corrects the parameters through high-frequency optimization to avoid fluctuations and ensure that the system can quickly recover to stability in extreme scenarios (such as sudden drop in photovoltaic output, multiple charging piles simultaneously accessing).

[0062] In the light storage and charging heat scene, the system faces complex situations such as photovoltaic output fluctuation affected by light, random changes in the number of charging piles, and temperature adjustment of heat pump load. This scheme uses a long short-term memory (LSTM) prediction model to predict future energy output, load demand, and power fluctuation trends in advance, and combines a deep reinforcement learning model to optimize network parameters, so that network parameters can adapt to complex fluctuations in the scene in advance. Through actual application verification, this method can reduce the frequency fluctuation amplitude of the microgrid by more than 40%, and the power distribution error is controlled within 5%, effectively avoiding the problem of system instability caused by parameter adjustment lag.

[0063] The scheme adopts a hierarchical control architecture of an edge AI control layer and a local execution layer, wherein the edge AI control layer is based on a global perspective, comprehensively optimizes parameters based on all device states and environment data, and ensures the globality and optimality of the decision; and the local execution layer focuses on receiving and executing the optimized instructions in real time and quickly responding to the device adjustment demand. This architecture design not only avoids the communication delay problem caused by centralized data transmission in the traditional central control mode, but also solves the defect that the single local control only focuses on the self state and is easy to fall into local optimum. Finally, the overall loss of the device is reduced by 15%-20%, and the economic efficiency and energy efficiency level of the microgrid operation are improved.

[0064] In view of the extreme situations that may occur in the light storage and charging heat scene, such as sudden concentrated access of charging piles, photovoltaic output drop caused by cloud layer blocking, etc., the scheme comprehensively captures the system operation state and external environment change through a multi-source feature fusion technology, timely identifies the power fluctuation risk, and at the same time designs an emergency optimization mechanism, which immediately shortens the optimization period of the deep reinforcement learning model when detecting that the power fluctuation amplitude exceeds the preset threshold, quickly recalculates and issues the optimal parameters. In the above extreme scenarios, the time for the system to recover to stability can be shortened to within 0.3 seconds, which greatly improves the ability of the microgrid to respond to sudden conditions compared with the existing technology 1-second recovery time standard, and guarantees the continuous and stable operation of the system.

[0065] Based on the above microgrid control system, the application provides a microgrid control method, Figure 1 is a schematic flow chart of the microgrid control method according to an embodiment of the application, as Figure 1 shown, the microgrid control method at least includes the following steps S102 to S108.

[0066] Step S102, collecting operation data of the microgrid and each network type device thereof, and environment data of the environment where the microgrid is located.

[0067] Step S104, performing multi-source feature fusion processing on the collected operation data and environment data, and predicting the energy output, load demand and power fluctuation amplitude of the microgrid in a future preset time period based on the fused features.

[0068] Step S106, optimizing the voltage parameters and frequency parameters of each network type device based on the prediction result and the current state of each network type device.

[0069] Step S108, issuing the optimized voltage parameters and frequency parameters as parameter adjustment instructions to the corresponding network type device, so that each network type device operates based on the issued parameter adjustment instructions.

[0070] The micro-grid control method of the application avoids single perspective limitation to comprehensively perceive system state by collecting full-dimensional data of the micro-grid as a whole, the network-forming equipment and the environment; predicts future energy output, load demand and power fluctuation after multi-source feature fusion processing, changes passive response to active prediction, and reserves parameter adjustment time; and then optimizes voltage and frequency parameters in combination with the prediction result and the current state of the equipment, and issues the parameters as instructions for execution. In this way, the voltage parameters and the frequency parameters of each network-forming equipment can be optimized, voltage and frequency abnormalities can be reduced from the source, and the system anti-interference ability can be improved.

[0071] In an optional embodiment, the network-forming equipment can include a photovoltaic inverter, an energy storage converter, a charging pile and a heat pump, and each network-forming equipment is integrated with a network-forming virtual synchronous machine.

[0072] Figure 2 is a schematic flow chart of multi-source fusion processing according to an embodiment of the application, as Figure 2 As shown in the figure, the multi-source feature fusion processing on the collected operation data and environmental data can include the following steps S202 to S208.

[0073] Step S202, extracting a photovoltaic output fluctuation sequence based on the operation data of the photovoltaic inverter, extracting a charging and discharging power change rate based on the operation data of the energy storage converter, and extracting time sequence data of the access quantity based on the operation data of the charging pile as time sequence features.

[0074] Step S204, extracting a state of charge based on the operation data of the energy storage converter, extracting an operation mode based on the operation data of the heat pump, and extracting a bus voltage deviation and a frequency fluctuation amplitude based on the operation data of the micro-grid as state features.

[0075] Step S206, extracting light intensity, environmental temperature and power grid voltage stability level based on the environmental data as environmental features.

[0076] Step S208, integrating the time sequence features, the state features and the environmental features to construct a multi-dimensional feature matrix for prediction.

[0077] It can be understood that for different network-forming equipment such as photovoltaic inverters and energy storage converters, features strongly related to their core characteristics (such as photovoltaic output fluctuation sequence reflecting photovoltaic intermittency and energy storage state of charge reflecting energy storage regulation capacity) are extracted, and environmental features (such as light intensity directly related to photovoltaic output) are also included, realizing targeted fusion of multi-dimensional data of equipment, system and environment, avoiding irrelevant data interference, and providing high-quality input for prediction.

[0078] The time sequence feature captures the dynamic change trend (such as the increase and decrease law of the number of charging piles with time), the state feature reflects the current running baseline (such as the bus voltage deviation reflects the instantaneous stability of the system), and the environment feature reveals the external influencing factors (such as the temperature influences the heat pump load), and the multi-dimensional feature matrix formed by the integration of the three provides a comprehensive feature basis for the intelligent prediction model, and improves the prediction accuracy.

[0079] By extracting different types of features in layers, the multi-dimensional feature matrix can not only reflect the common law of the light storage and charging heat scene (such as the periodicity of power fluctuation), but also reflect the individual differences of equipment (such as the different response characteristics of energy storage and heat pump). The prediction model trained based on the matrix can more accurately capture the complex dynamics of the scene (such as the chain reaction caused by sudden change of light), provide reliable forward basis for subsequent parameter optimization, and enhance the prediction and response ability of the system to scene fluctuations.

[0080] In an optional embodiment, when predicting the energy output, load demand and power fluctuation amplitude of the micro-grid in the future preset time period based on the fused features, the intelligent prediction model can be used to learn the time sequence dependence relationship in the multi-dimensional feature matrix, and the prediction result of the energy output, load demand and power fluctuation amplitude of the micro-grid in the future preset time period is output by the model inference.

[0081] The micro-grid control method of the present application learns the time sequence dependence relationship in the multi-dimensional feature matrix by the intelligent prediction model and infers the prediction result, which significantly improves the prediction accuracy and forward-looking. The intelligent model can deeply mine the dynamic correlation between the multi-dimensional features such as photovoltaic output fluctuation, load change and environmental factors, and more accurately capture the complex dynamic law than traditional methods; based on the output energy output, load demand and power fluctuation prediction result, a reliable forward-looking basis is provided for subsequent parameter optimization, the voltage and frequency parameter adjustment is more targeted, and the prediction and response ability to the intermittency of network type equipment and the randomness of load is further enhanced, which lays a key foundation for stable and efficient operation of micro-grid.

[0082] In an optional embodiment, when optimizing the voltage parameters and frequency parameters of each network-forming device based on the prediction results and the current state of each network-forming device, a deep reinforcement learning model can be used for parameter optimization. The state space of the deep reinforcement learning model includes the prediction results output by the intelligent prediction model and the control parameters of the virtual synchronous machines integrated by each network-forming device. The action space of the deep reinforcement learning model is defined as the adjustment actions of the virtual inertia, damping coefficient and virtual impedance parameters of each virtual synchronous machine. A reward function of the deep reinforcement learning model is constructed, which is a weighted combination function of the frequency deviation, power distribution error and device loss rate. Through the continuous interaction between the deep reinforcement learning model and the microgrid operating environment, the cumulative reward is maximized to generate a parameter configuration strategy, and the optimization of the voltage parameters and frequency parameters of each network-forming device is completed.

[0083] The prediction results output by the intelligent prediction model and the current control parameters of the virtual synchronous machine are included in the state space, which enables the model to know the future energy output, load demand and power fluctuation trend in advance (such as predicting the sudden drop of photovoltaic output in 10 minutes), and also enables the model to master the current running basis of the device (such as the current value of the virtual inertia, the current setting of the damping coefficient), thereby avoiding the short-sightedness caused by relying only on real-time state during optimization.

[0084] The action space is defined as the adjustment actions of the virtual inertia, damping coefficient and virtual impedance parameters, because these three parameters directly determine the voltage regulation and frequency regulation characteristics of the virtual synchronous machine: the virtual inertia affects the frequency change speed (the larger the inertia, the slower the frequency drop), the damping coefficient suppresses frequency oscillation (the larger the damping, the faster the frequency recovery), and the virtual impedance balances power distribution (impedance matching can reduce circulating current between devices). By accurately adjusting these three parameters, the optimization effect of the output voltage and frequency of the device can be directly converted, and the model action is strongly associated with the control target.

[0085] The reward function is designed as a weighted combination of the frequency deviation, power distribution error and device loss rate, which essentially clarifies the optimization priority of the model: when the frequency deviation weight is high, the system frequency stability is prioritized (to avoid triggering protection shutdown); when the power distribution error weight is high, the balance of the load of each device is emphasized (to prevent overloading of a single device); and when the device loss rate weight is high, the device life can be extended (to reduce operation and maintenance costs). The model continuously interacts with the microgrid environment (after each round of parameter adjustment, the reward is calculated based on the actual running frequency, power and loss data), constantly learns the parameter combination that can maximize the cumulative reward, and finally generates a globally optimal configuration strategy that takes into account multiple objectives, which not only solves the limitations of single-objective optimization, but also makes the parameter adjustment fit the actual operation demand of the microgrid.

[0086] The micro-grid control method of the application can significantly improve the accuracy and synergy of global regulation of the micro-grid by optimizing parameters of a deep reinforcement learning model. The model incorporates the prediction results and the current control parameters of the virtual synchronous machine into the state space, realizes comprehensive consideration of the future and current states of the system, takes the adjustment of core parameters such as virtual inertia and damping coefficient as the action space, directly acts on the voltage and frequency characteristics of the network-type equipment, and constructs a multi-objective optimization guide for stability, balance and economy by fusing the reward functions of frequency deviation, power distribution error and equipment loss rate.

[0087] By continuously interacting with the operating environment to maximize the cumulative reward, the real-time optimal configuration of the voltage / frequency parameters can be realized to dynamically adapt to scenarios such as photovoltaic output fluctuation and load random change, and the power distribution among multiple devices can be coordinated to reduce circulating current and loss, thereby ensuring stable operation of the system while improving energy utilization efficiency and equipment life.

[0088] In some embodiments, when optimizing the voltage parameters and frequency parameters of each network-type equipment based on the prediction results and the current state of each network-type equipment, the operating state parameters of each network-type equipment can be obtained, including the current output voltage, output frequency and control parameters of the virtual synchronous machine. Then, based on the prediction results of the energy output, load demand and power fluctuation amplitude of the micro-grid in a future preset time period, the voltage parameters and frequency parameters of each network-type equipment are iteratively optimized by a deep reinforcement learning model, and finally the desired voltage parameters and desired frequency parameters that meet the target are output.

[0089] Obtaining the current output voltage, frequency and virtual synchronous machine control parameters of the equipment can accurately grasp the real-time operating boundaries of the equipment (such as the current charge capacity limit of the energy storage and the output capacity of the inverter), avoid execution failure or hardware damage caused by optimizing parameters exceeding the physical threshold of the equipment, and make the optimization strategy more practical. Combined with the prediction results of future energy output, load and fluctuation, potential power imbalance risks can be avoided in advance; superimposed with the current state parameters of the equipment, both short-term emergency adjustment and long-term stability demand can be considered, and compared with optimization relying solely on historical data or real-time data, the system stability and equipment operating efficiency can be better balanced.

[0090] More importantly, the deep reinforcement learning model can explore the globally optimal parameter combination under multi-objective constraints such as frequency stability, power balance and minimum loss through multiple iterations, avoid local optimal solutions, and finally output the voltage parameters and frequency parameters as the desired voltage parameters and desired frequency parameters, which can more accurately match the operating requirements of the micro-grid and effectively reduce voltage deviation, frequency fluctuation and equipment loss.

[0091] In some embodiments, the power fluctuation amplitude of the micro-grid can be detected, and when the power fluctuation amplitude exceeds a preset threshold, an emergency optimization mechanism is triggered to shorten the parameter optimization execution period of the deep reinforcement learning model, and the parameter adjustment instruction is reissued to each network device after the deep reinforcement learning model is quickly re-optimized.

[0092] When the power fluctuation exceeds the threshold (such as sudden access of charging piles or sudden drop of photovoltaic output), the reinforcement learning optimization period is shortened (such as from 200 ms to 50 ms), a new optimal parameter instruction can be quickly generated to avoid voltage / frequency instability caused by optimization lag in the traditional fixed period, and the fluctuation is controlled in a small range to prevent triggering of device protection shutdown or system collapse.

[0093] In addition, the emergency optimization mechanism responds to the sudden scene, does not need to rely on slow adjustment of the regular period, and can balance power supply and demand in a short time through parameter re-optimization (such as increasing the discharge power of energy storage and adjusting virtual inertia to suppress frequency drop), so that the system quickly recovers to stability in extreme working conditions. Compared with the scheme without the emergency mechanism, the stable recovery time is greatly shortened (such as from 1 second to 0.3 seconds).

[0094] In some embodiments, the historical data generated by the operation of the micro-grid can be used to perform offline incremental training on the deep reinforcement learning model, and the optimization strategy of the deep reinforcement learning model can be dynamically updated to adapt to the long-term changes of the operation characteristics of the micro-grid.

[0095] It can be understood that in the long-term operation of the micro-grid, device performance degradation (such as energy storage capacity reduction), load characteristic change (such as charging pile user behavior change), and environmental influence trend adjustment (such as seasonal light change) may occur. Through offline incremental training based on historical data, the model can dynamically learn these long-term change rules, update the optimization strategy, avoid control precision degradation caused by the adaptability of the initial model, and ensure that the parameter optimization is still consistent with the actual scene in the long-term operation.

[0096] In this way, without interrupting the normal operation of the micro-grid, the historical data is trained based on the idle period at night, which does not affect the real-time control of the system, and the model can continuously absorb new operation experience (such as parameter adjustment cases in extreme weather), gradually improve the adaptability to complex scenes, and continuously optimize the frequency stability, power distribution, and device loss control effect compared with the fixed model.

[0097] The offline incremental training can be based on the existing model parameter update, without the need to build a new model from zero, reducing the algorithm power and time investment; at the same time, the historical data can be accumulated and reused for a long time, and as the data volume increases, the model optimization strategy will be more accurate, which is beneficial to improving the economy and reliability of the long-term operation of the micro-grid.

[0098] In some embodiments, the operation data and the environment data are collected for a first preset time length, the inference of the intelligent prediction model is performed for a second preset time length, and the parameter optimization of the deep reinforcement learning model is performed for a third preset time length longer than the second preset time length.

[0099] For example, the first preset time length is 10 ms, the second preset time length is 50 ms, and the third preset time length is 200 ms.

[0100] The 10 ms high-frequency collection of the operation and environment data can capture rapid fluctuations such as photovoltaic output sudden change and charging pile sudden access in real time, avoid state misjudgment caused by data lag, provide a high-fidelity data source for subsequent processing, and lay a foundation for sensing accuracy.

[0101] The 50 ms execution of the inference of the intelligent prediction model can construct complete time sequence features through 5 sets of 10 ms collected data, guarantee prediction accuracy, quickly update future output and load trend, provide near-real-time basis for parameter optimization, and balance real-time performance and prediction reliability.

[0102] The 200 ms execution of the parameter optimization of the deep reinforcement learning can reserve sufficient computing power to complete multiple iterations, ensure output of globally optimal parameters, avoid local optimum or system shock of short-period optimization, match device response speed, and guarantee stable parameter execution.

[0103] It should be understood that in some embodiments, each part can be realized by hardware, software, firmware or a combination thereof. In the above implementation, a plurality of steps or methods can be realized by software or firmware stored in a memory and executed by a suitable instruction execution system.

[0104] The embodiment also provides a computer program product 10, a machine readable storage medium 20 and a computer device 30. Figure 3 is a schematic diagram of a computer program product according to an embodiment of the present application, Figure 4 is a schematic diagram of a machine readable storage medium according to an embodiment of the present application, Figure 5 is a schematic diagram of a computer device according to an embodiment of the present application. The computer program product 10 comprises a computer program 11, which, when executed by the processor 32, realizes the steps of any of the microgrid control methods described above. The machine readable storage medium 20 has the computer program 11 stored thereon, and the computer program 11, when executed by the processor 32, realizes the steps of the microgrid control method of any of the embodiments described above. The computer device 30 can comprise a memory 31, a processor 32 and a computer program 11 stored on the memory 31 and running on the processor 32.

[0105] Computer program 11, which can also be referred to or described as a program, software, a software application, an app, a program of instructions, or simply an application program, is stored on any apparatus-readable medium, which is a tangible computer-readable storage medium, such as a floppy disk, a DVD, a CD, a Blu-ray Disc, a memory stick, a memory card, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read only memory (EPROM), electronically programmable read only memory (EPROM), electronically erasable programmable read only memory (EEPROM), a magnetic or optical card, a flash memory, a volatile memory, a non-volatile memory, a hard disk drive, or any other suitable apparatus-readable medium, including a combination of any of the above, wherein the memory medium is tangibly coupled to the computer to enable the computer to read information from and write information to the memory medium. When the computer program 11 is implemented as a computer program 11, the computer program 11 can be stored on or transmitted across one or more computer-readable media, such as any tangible computer-readable storage medium or media, or a combination thereof, and can be executed by one or more processors, such as one or more application processors, digital signal processors, microprocessors, or any other suitable processing device, or a combination thereof.

[0106] For the description of the present embodiment, the computer program product 10 is a product of manufacture that includes the computer program 11.

[0107] For the description of the present embodiment, the machine-readable storage medium 20 is a tangible device that is capable of storing and / or carrying the computer program 11, which can be any device that can contain, store, communicate, propagate or transport the program 11 for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the machine-readable storage medium 20 include the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch cards, and any suitable combination of the above, or any other device(s) that is normally used for carrying or storing computer programs 11 for instructions to be executed by an instruction execution system, apparatus, or device, and that can be read by an electronic circuit such as microprocessor(s) or a processor(s) of a computer.

[0108] The computer device 30 can be, for example, a server, a desktop computer, a notebook computer, a tablet computer, or a smart phone. In some examples, the computer device 30 can be a cloud computing node. The computer device 30 can be described in the general context of computer system-executable instructions, such as program modules, being executed by a computer system. Generally, program modules can include routines, programs, objects, components, logic, data structures, and so on that perform particular tasks or implement particular abstract data types. The computer device 30 can be practiced in distributed cloud computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed cloud computing environment, program modules can be located in both local and remote computer system storage media including memory storage devices.

[0109] The computer device 30 can include a processor 32 adapted to execute instructions stored in a memory 31, which in operation provide the instructions with temporary storage. The processor 32 can be a single core processor, a multi-core processor, a computing cluster, or any number of other configurations. The memory 31 can include random access memory (RAM), read only memory, flash memory, or any other suitable memory system.

[0110] The computer device 30 can also include a network adapter / interface and an input / output (I / O) interface. The I / O interface allows for input and output of data with external devices that can be connected to the computer device. The network adapter / interface can provide for communication between the computer device and a network, typically illustrated as a communication network.

[0111] To this end, it will be understood that, although the present application has been described herein in terms of its several embodiments, it is not intended to be limited to the embodiments described herein, but rather is intended to be broadly within the scope of the claims so as to embrace all such modifications, equivalents, and alternatives. Accordingly, many modifications can be made by those skilled in the art to, and equated to, the above-described embodiments without departing from the scope of the present application.

Claims

1. A microgrid control method, characterized by, The method comprises the following steps: Collecting operation data of a micro-grid and each network-forming device thereof, and environmental data of an environment in which the micro-grid is located; Performing multi-source feature fusion processing on the collected operation data and environmental data, and predicting energy output, load demand, and power fluctuation amplitude of the micro-grid in a future preset time period based on the fused features; Optimizing voltage parameters and frequency parameters of each network-forming device based on the prediction result and the current state of each network-forming device; Distributing the optimized voltage parameters and frequency parameters as parameter adjustment instructions to the corresponding network-forming devices, so that each network-forming device operates based on the distributed parameter adjustment instructions; The network-forming devices include photovoltaic inverters, energy storage converters, charging piles, and heat pumps, and each network-forming device is integrated with a network-forming virtual synchronous machine, and the step of performing multi-source feature fusion processing on the collected operation data and environmental data comprises: Extracting a fluctuation sequence of photovoltaic output based on the operation data of the photovoltaic inverter, extracting a change rate of charging and discharging power based on the operation data of the energy storage converter, and extracting time sequence data of the number of accesses based on the operation data of the charging pile as time sequence features; Extracting a state of charge based on the operation data of the energy storage converter, extracting an operation mode based on the operation data of the heat pump, and extracting bus voltage deviation and frequency fluctuation amplitude based on the operation data of the micro-grid as state features; Extracting light intensity, environmental temperature, and power grid voltage stability level based on the environmental data as environmental features; Integrating the time sequence features, state features, and environmental features to construct a multi-dimensional feature matrix for prediction.

2. The micro-grid control method of claim 1, wherein the step of predicting energy output, load demand, and power fluctuation amplitude of the micro-grid in a future preset time period based on the fused features comprises: Learning time sequence dependency in the multi-dimensional feature matrix using an intelligent prediction model, and outputting prediction results of energy output, load demand, and power fluctuation amplitude of the micro-grid in a future preset time period through model reasoning.

3. The micro-grid control method of claim 2, wherein the step of optimizing voltage parameters and frequency parameters of each network-forming device based on the prediction result and the current state of each network-forming device comprises: Using a deep reinforcement learning model to optimize parameters, wherein a state space of the deep reinforcement learning model contains prediction results output by the intelligent prediction model, and control parameters of virtual synchronous machines integrated in each network-forming device; Defining an action space of the deep reinforcement learning model as adjustment actions of virtual inertia, damping coefficient, and virtual impedance parameters of each virtual synchronous machine; Constructing a reward function of the deep reinforcement learning model, wherein the reward function is a weighted combination function of frequency deviation, power distribution error, and device loss rate. ​ ​ By continuous interaction of the deep reinforcement learning model with the microgrid operating environment, the cumulative reward is maximized to generate a parameter configuration strategy, and optimization of the voltage parameters and frequency parameters of each network-forming device is completed.

4. The microgrid control method of claim 3, wherein, Based on the prediction results and the current state of each network-forming device, the step of optimizing the voltage parameters and frequency parameters of each network-forming device comprises: Obtaining the operating state parameters of each network-forming device, including the current output voltage, output frequency, and control parameters of the virtual synchronous machine; Based on the prediction results of the energy output, load demand, and power fluctuation amplitude of the microgrid within the future preset time period, a deep reinforcement learning model is used to iteratively optimize the voltage parameters and frequency parameters of each network-forming device; Output the desired voltage parameters and frequency parameters that meet the target.

5. The microgrid control method of claim 3, wherein Further comprising: Detecting the power fluctuation amplitude of the microgrid; When the power fluctuation amplitude exceeds a preset threshold, triggering an emergency optimization mechanism, shortening the parameter optimization execution period of the deep reinforcement learning model, and re-issuing the parameter adjustment instructions to each network-forming device after rapid re-optimization of the deep reinforcement learning model.

6. The microgrid control method of claim 3, wherein Further comprising: Periodically using historical data generated by the operation of the microgrid to perform offline incremental training on the deep reinforcement learning model, dynamically updating the optimization strategy of the deep reinforcement learning model to adapt to long-term changes in the operating characteristics of the microgrid.

7. The microgrid control method of claim 3, wherein, The operating data and environmental data are collected for a first preset time period; The inference of the intelligent prediction model is performed for a second preset time period, and the parameter optimization of the deep reinforcement learning model is performed for a third preset time period longer than the second preset time period.

8. A computer program product, characterised in that, A computer program that, when executed by a processor, implements the steps of the microgrid control method of any one of claims 1-7.

9. A machine-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, implements the steps of the microgrid control method of any one of claims 1-7.

Images