Dispatch and control methods and energy storage systems

Abstract

This application discloses a scheduling control method and an energy storage system. The method is applied to a smart distribution box and includes: collecting electrical characteristic data of multiple load circuits connected to the smart distribution box; using a machine learning model to predict the LRA (Low-Range Activated Load) inrush current and inrush probability of each load circuit; calculating dynamic reserved power based on the LRA inrush current and inrush probability; the dynamic reserved power is used to reserve power buffer for the load circuits to cope with the LRA inrush current when participating in VPP (Variable Power Controller) scheduling; reporting the dynamic reserved power to the VPP cloud platform; receiving and executing VPP scheduling instructions from the VPP cloud platform, the VPP scheduling instructions including VPP scheduling power; and sending VPP scheduling instructions to the energy storage device, causing the energy storage device to output or input power according to the VPP scheduling power. Thus, this application, by predicting load inrush current and dynamically reserving power through the smart distribution box, ensures the load inrush response capability when participating in VPP scheduling, solving the scheduling failure problem caused by the inrush current of motor-type loads.

Description

Technical Field

[0001] This application relates to the field of energy storage technology, and in particular to a scheduling control method and an energy storage system. Background Technology

[0002] With the development of new energy sources, residential energy storage is rapidly becoming widespread. While it offers functions such as self-consumption, peak shaving and valley filling, and emergency backup power, it is limited to individual households and cannot achieve unified scheduling and optimized allocation of green energy at the regional level. Therefore, Virtual Power Plant (VPP) technology has emerged. VPP technology is an energy management network that aggregates distributed energy sources such as residential energy storage, photovoltaics, and electric vehicles through advanced information technology and software systems. It has no physical power plant; its core lies in communication and aggregation. It can coordinate and optimize resources through a cloud-based control center, participate in the electricity market, and provide ancillary services such as peak shaving and frequency regulation to the grid to generate revenue. Currently, VPP is evolving towards large-scale and standardized development, aggregating more diverse resources, continuously improving its integration with new power systems, and its market size is constantly expanding.

[0003] In traditional VPP dispatching applications, the virtual power plant dispatching strategy mainly relies on basic electrical parameters such as the current capacity and charging / discharging power of the user-side energy storage battery for decision-making and control. When the virtual power plant issues a dispatching command and the energy storage battery responds to the dispatching demand by performing a discharge operation, if a motor-type load such as a refrigerator or air conditioner happens to be starting up, the instantaneous high power surge caused by its high locked rotor current (LRA) can easily exceed the real-time response capability of the home energy storage system, thereby causing the virtual power plant dispatching response to fail and unable to stably meet the grid dispatching requirements. Summary of the Invention

[0004] This application aims to address at least one of the technical problems existing in the prior art. Therefore, this application provides a scheduling control method and an energy storage system.

[0005] This application proposes a power dispatch control method for a home energy storage system, applied to a smart distribution box. The home energy storage system further includes an energy storage device electrically connected to the smart distribution box. The method includes: Collect electrical characteristic data of multiple load circuits connected to the intelligent distribution box; Based on the electrical characteristic data, a machine learning model is used to predict the LRA starting inrush current and starting probability of each load circuit; The dynamic reserved power is calculated based on the LRA initiation inrush current and the initiation probability; the dynamic reserved power is used to reserve power buffer for the load circuit to cope with the LRA initiation inrush current when participating in VPP scheduling; The dynamically reserved power is reported to the VPP cloud platform; Receive and execute VPP scheduling instructions from the VPP cloud platform, wherein the VPP scheduling instructions include VPP scheduling power, and the VPP scheduling power is determined based on grid scheduling requirements and the dynamically reserved power. Send the VPP scheduling command to the energy storage device so that the energy storage device outputs or inputs power according to the VPP scheduling power.

[0006] In some embodiments, the method further includes: Monitor the actual LRA startup inrush current during load startup; The machine learning model is corrected online based on the actual LRA starting inrush current.

[0007] In some implementations, the electrical characteristic data includes at least one of time-series power data, environmental parameters, and historical load operation data.

[0008] In some implementations, the machine learning model includes a dual-channel long short-term memory network, and the step of using the machine learning model to predict the starting inrush current and starting probability of each load circuit LRA based on the electrical characteristic data includes: Based on the time-series power data, environmental parameters, and historical load operation data, a feature vector that conforms to the input of the machine learning model is constructed. The feature vector is input into the dual-channel long short-term memory network. The first channel of the dual-channel long short-term memory network processes the timing power data, and the second channel of the dual-channel long short-term memory network processes the environmental parameters and the load operation history data, and outputs the LRA start-up inrush current and the start-up probability.

[0009] In some implementations, the calculation expression for the dynamic reserved power includes:

[0010] in, This is the load weighting coefficient. For safety reasons, The range is 1.1-1.2. The time decay coefficient, For the start probability, The starting inrush current for LRA.

[0011] This application also proposes a power dispatch control method for a home energy storage system, applied to a VPP cloud platform. The home energy storage system further includes an energy storage device electrically connected to a smart distribution box, and the VPP cloud platform is communicatively connected to the smart distribution box. The method includes: The system receives grid dispatching requests and the dynamic reserved power sent by the intelligent distribution box. The dynamic reserved power is used to reserve power buffer for the load circuit to cope with the LRA start-up inrush current when participating in VPP dispatching. VPP dispatch instructions are generated based on the power grid dispatch requirements and the dynamically reserved power. The VPP scheduling command is sent to the smart distribution box, which then sends the VPP scheduling command to the energy storage device so that the energy storage device can output or input power according to the VPP scheduling power.

[0012] In some implementations, generating VPP dispatch instructions based on the grid dispatch requirements and the dynamically reserved power includes: Determine whether the grid dispatch demand is less than or equal to the available capacity, wherein the available capacity is determined based on the inverter peak power and the dynamically reserved power. If the grid dispatch demand is less than or equal to the available capacity, generate an immediate execution instruction; or When the grid dispatch demand exceeds the available capacity, a negotiated execution instruction is generated.

[0013] In some embodiments, the method further includes: Obtain electricity service pricing; The VPP dispatch instruction is generated based on the electricity service price, the grid dispatch demand, and the dynamic reserved power.

[0014] This application also proposes a home energy storage system, comprising: The intelligent distribution box connects multiple load circuits and is configured to collect electrical characteristic data of the loads. Based on the electrical characteristic data, it uses a machine learning model to predict the LRA inrush current and inrush probability of each load circuit, and calculates dynamic reserved power based on the LRA inrush current and the inrush probability. The dynamic reserved power is used to reserve power buffer for the load circuit to cope with the LRA inrush current when participating in VPP dispatch. The VPP cloud platform is communicatively connected to the intelligent distribution box, configured to receive the dynamically reserved power, and generate and issue VPP dispatch commands to the intelligent distribution box based on grid dispatch requirements and the dynamically reserved power; and An energy storage device is communicatively connected to the intelligent distribution box and configured to execute the VPP scheduling command to output or input power according to the VPP scheduling power.

[0015] In some embodiments, the intelligent distribution box includes: Multiple detection loops are configured to collect the electrical characteristic data; An edge computing unit, connected to the detection loop, is configured to predict the LRA start-up inrush current and start-up probability of each load loop using a machine learning model based on the electrical characteristic data, and to calculate the dynamic reserved power based on the LRA start-up inrush current and the start-up probability.

[0016] In some implementations, the VPP cloud platform includes: The cluster scheduling engine is configured to receive grid scheduling requests and generate VPP scheduling instructions based on the grid scheduling requests and the dynamically reserved power. A dynamic reserved power database is configured to store the dynamic reserved power reported from the smart distribution box; The electricity price signal interface is configured to receive external electricity price information to assist in generating the VPP scheduling instructions.

[0017] The scheduling control method and home energy storage system described in this application collect load electrical characteristic data through a smart distribution box, use a machine learning model to predict the starting inrush current and starting probability of LRA (Low-Range Refrigerant Controller), and dynamically calculate the reserved power, which is then reported to the VPP (Virtual Power Plant) cloud platform. The VPP cloud platform generates scheduling instructions based on grid scheduling needs and the dynamically reserved power, and sends them to the smart distribution box, which then forwards them to the energy storage devices for execution. In this way, on the one hand, proactive prediction and power reservation for starting impacts of motor-type loads are achieved, greatly improving the stability and reliability of home energy storage systems participating in virtual power plant scheduling; on the other hand, during the scheduling execution process, the end-cloud collaboration between the VPP cloud platform and the smart distribution box can optimize the scheduling strategy in real time according to the actual load starting situation, avoiding scheduling response failures caused by load impacts, and continuously improving the prediction accuracy of the machine learning model through an online correction mechanism; in addition, it can also achieve efficient aggregation and optimized scheduling of distributed energy storage resources, ensuring reliable load operation while taking into account resource utilization and economic benefits, providing a complete solution for the large-scale participation of residential energy storage in grid interaction.

[0018] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0019] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a schematic diagram of a home energy storage system in some embodiments of this application; Figure 2 This application provides a schematic diagram of the power-time curve at the moment of air conditioner startup in certain embodiments; Figure 3A schematic diagram of the power-time curve of the refrigerator at the moment of startup in some embodiments of this application; Figure 4 A schematic diagram of the power-time curve of the water pump at the moment of startup in some embodiments of this application; Figure 5 This application includes a household total power load curve in certain embodiments; Figure 6 This is a schematic diagram of an intelligent distribution box in some embodiments of this application; Figure 7 This is a schematic diagram of the VPP cloud platform in some embodiments of this application; Figure 8 This is a flowchart illustrating a scheduling and control method for an intelligent distribution box in certain embodiments of this application. Figure 9 This is a schematic diagram comparing the total load and reserved power of certain implementation methods of this application with traditional scheduling strategies; Figure 10 This is a schematic diagram comparing the success rates of scheduling instruction execution and traditional scheduling strategy instruction execution in certain embodiments of this application; Figures 11-12 This is a flowchart illustrating the energy storage scheduling and control method according to certain embodiments of this application; Figure 13 This is a schematic diagram comparing the reserved power curves of certain embodiments of this application with those of a conventional fixed reserved strategy; Figures 14-15 This is a flowchart illustrating the energy storage scheduling method according to certain embodiments of this application; Figure 16 This is a schematic diagram comparing the VPP schedulable capacity of certain implementations of this application with traditional scheduling strategies; Figure 17 This is a flowchart illustrating the energy storage scheduling method according to certain embodiments of this application.

[0020] Explanation of icon numbers Home energy storage system 100, smart distribution box 10, detection circuit 11, edge computing unit 12, VPP cloud platform 20, cluster scheduling engine 21, dynamic reserved power database 22, electricity price signal interface 23, energy storage device 30. Detailed Implementation

[0021] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0022] With the rapid development of new energy technologies, home energy storage systems are rapidly gaining popularity worldwide. Home energy storage offers functions such as self-generation and consumption, peak shaving and valley filling, and emergency backup power, effectively improving the economy and reliability of household electricity use. However, because home energy storage is typically limited to independent operation within a single household, it is difficult to achieve unified scheduling and optimized allocation of green energy at the regional level, thus limiting its ability to support stable grid operation over a wider area.

[0023] Against this backdrop, Virtual Power Plant (VPP) technology has emerged and developed rapidly. VPP integrates widely distributed distributed energy resources such as residential energy storage, photovoltaic power generation, and electric vehicle charging stations through advanced information and communication technologies and software aggregation platforms, constructing a virtualized energy management network. This network has no physical power plant; its core lies in resource aggregation and collaborative control. Through a cloud-based control center, it coordinates and optimizes various resources in real time. It can participate in the electricity market by providing ancillary services such as peak shaving and frequency regulation to generate revenue, and it can also provide flexible support for new power systems. Currently, VPP technology is evolving towards large-scale and intelligent development, aggregating increasingly diverse resource types, continuously deepening its integration with the power system, and expanding its market size.

[0024] In some related technical solutions, virtual power plant dispatch strategies primarily rely on the basic electrical parameters of user-side energy storage batteries for decision-making, such as the battery's current capacity, state of charge, and maximum allowable charge / discharge power. When the virtual power plant issues a discharge dispatch command based on grid demand, the home energy storage system responds and executes according to the preset power value. However, this static parameter-based dispatch method has significant drawbacks in practical applications: if, during the discharge process, a high-power motor load such as a refrigerator or air conditioner happens to start up on the user side, the resulting high locked rotor current (LRA) will cause a momentary power surge. This surge power often far exceeds the real-time dynamic response capability of the home energy storage system, leading to inverter overload protection or output voltage drops, causing the dispatch response to fail, and the virtual power plant cannot stably meet the grid dispatch requirements.

[0025] In view of this, please refer to Figure 1This application proposes a home energy storage system 100, which includes a smart distribution box 10, a VPP cloud platform 20, and an energy storage device 30. The smart distribution box 10 connects to multiple load circuits and is configured to collect electrical characteristic data of the loads. Based on this data, it uses a machine learning model to predict the LRA inrush current and inrush probability of each load circuit, and calculates dynamic reserved power based on the LRA inrush current and inrush probability. This dynamic reserved power is used to buffer the load circuits against the LRA inrush current during VPP dispatch. The VPP cloud platform 20 is communicatively connected to the smart distribution box 10 and is configured to receive the dynamic reserved power. It also generates and sends VPP dispatch commands to the smart distribution box 10 based on grid dispatch requirements and the dynamic reserved power. The energy storage device 30 is communicatively connected to the smart distribution box 10 and is configured to execute VPP dispatch commands to output or input power according to the VPP dispatch power.

[0026] It is worth noting that the home energy storage system 100 refers to a distributed energy storage system deployed in a home setting, including a smart distribution box 10 and energy storage devices 30, used to realize home electricity management and participate in virtual power plant dispatch.

[0027] The intelligent distribution box 10 refers to a local intelligent power distribution device deployed on the user side, connecting multiple load circuits and energy storage device 30. The intelligent distribution box 10 contains high-precision sensors, data acquisition modules, edge computing units, and communication modules, and is responsible for collecting electrical characteristic data of each load circuit in real time, running machine learning models to predict load start-up behavior, calculating dynamic reserved power, and interacting with the VPP cloud platform 20.

[0028] VPP cloud platform 20 refers to a virtual power plant management platform deployed in the cloud, which communicates with a large number of smart distribution boxes 10 on the user side. VPP cloud platform 20 aggregates all connected household energy storage resources, receives information such as dynamic reserved power and energy storage device status reported by each smart distribution box 10 in real time, generates differentiated VPP dispatch instructions according to the grid dispatch requirements, and sends the instructions to each smart distribution box 10.

[0029] Energy storage device 30 refers to an execution device used to store and release electrical energy, typically including an energy storage battery, a bidirectional converter, and a battery management system. Energy storage device 30 is connected to the intelligent distribution box 10 via communication, and interacts with the VPP cloud platform 20 through the intelligent distribution box 10. It receives and executes VPP dispatch commands, outputs or inputs power according to the command requirements, and simultaneously feeds back its own operating status to the intelligent distribution box 10 in real time.

[0030] A load circuit refers to any circuit of electrical equipment connected to the output side of the intelligent distribution box 10, including but not limited to lighting circuits, socket circuits, air conditioning circuits, and refrigerator circuits. Each load circuit may contain one or more electrical devices, and their electrical characteristic data is collected in real time through the detection circuit within the intelligent distribution box 10. Electrical characteristic data refers to various parameter data reflecting the electrical characteristics of the load circuit, including but not limited to time-series power data, instantaneous voltage, instantaneous current, active power, current change rate, and harmonic content. This data is used to input machine learning models to predict the startup behavior of each load circuit.

[0031] Machine learning models are predictive models trained based on historical operating data, and can be implemented using algorithms such as neural networks, long short-term memory networks, and gradient boosting decision trees. The model takes electrical characteristic data as input and outputs the LRA starting inrush current and starting probability for each load circuit.

[0032] Please see Figures 2-4 , Figures 2-4 The figures show the power-time curves for air conditioners, refrigerators, and water pumps at startup. As shown, motor-type loads generate a very short but extremely high instantaneous power surge at startup. The peak current corresponding to this surge power is the LRA startup surge current, which can be several to tens of times the rated current.

[0033] Please see Figure 5 , Figure 5 This diagram illustrates the total household load power curve, showcasing a real-world scenario of the combined effects of multiple impact loads. In a household electrical environment, the starting times of motor-driven loads such as air conditioners, refrigerators, and water pumps are random. When multiple impact loads start sequentially or even simultaneously within a similar timeframe, their instantaneous power surges superimpose, creating a composite impact power far exceeding that of a single load. This superposition effect places higher demands on the instantaneous overload capacity of the household energy storage system, further exacerbating the risk of dispatch response failure.

[0034] The aforementioned instantaneous power surge originates from the electrical characteristics of motor-type loads. At startup, before the rotor rotates, the surge current flowing through the stator windings reaches its maximum value, known as the LRA startup surge current. This current typically reaches 5-8 times the motor's rated current, and although its duration is short, the instantaneous power surge is extremely large. When a home energy storage system responds to a VPP dispatch command to perform a discharge operation, if such a surge load happens to start, its instantaneous power surge can easily exceed the system's real-time response capability, leading to an output voltage drop or overload protection triggering, thereby causing dispatch response failure. Therefore, accurately predicting the LRA startup surge current is crucial for ensuring 100% reliable and stable operation of home energy storage systems and achieving virtual power plant dispatch response.

[0035] The start-up probability refers to the likelihood that a motor-type load will transition from a standby or stopped state to a starting state at a given moment or within a given time period. The start-up probability is generally predicted based on the load's historical operating patterns and current electrical characteristics, and its value ranges from 0 to 1; a higher value indicates a higher probability of the load starting.

[0036] Dynamic reserved power refers to the power margin that a home energy storage system 100 proactively reserves to cope with instantaneous load surges when responding to grid dispatch demands. The dynamic aspect is that the reserved power is not a fixed value, but is calculated and adjusted in real time based on the predicted LRA initiation surge current and initiation probability from the predictive model, ensuring that the power margin matches the load surge risk. Compared to related technologies that reserve fixed power or make no reservations at all, dynamic reserved power maximizes resource utilization while ensuring dispatch reliability, avoiding capacity waste due to excessive reservations or dispatch failures caused by insufficient reservations.

[0037] VPP dispatch instructions refer to the control instructions generated and sent to the smart distribution box 10 by the VPP cloud platform 20 based on the grid dispatch requirements and the dynamic reserved power reported by each user. These instructions include parameters such as target power value, dispatch direction, and execution time period.

[0038] In one embodiment, a home energy storage system 100 is deployed in a residential household. A smart distribution box 10 connects to multiple load circuits, including air conditioning, refrigerator, and lighting circuits. The energy storage device 30 includes a lithium iron phosphate battery pack and a bidirectional converter, with a rated power of 6kW. The energy storage device 30 communicates with the smart distribution box 10 via a CAN bus. A VPP cloud platform 20 is deployed on a cloud server and communicates with the smart distribution box 10 via a 4G network.

[0039] During normal operation, the intelligent distribution box 10 collects electrical characteristic data of each load circuit in real time at a sampling frequency of 1kHz, including time-series power data, current, voltage, etc. The edge computing unit built into the intelligent distribution box 10 runs a pre-trained machine learning model to predict that the LRA starting inrush current of the air conditioning circuit is 46A and the starting probability is 35%, and the LRA starting inrush current of the refrigerator circuit is 28A and the starting probability is 20%. Based on the prediction results, the intelligent distribution box 10 calculates the current dynamic reserved power as 3.5kW and uploads it to the VPP cloud platform 20 through the communication module.

[0040] The VPP cloud platform 20 receives dynamic reserved power data from this user and a large number of other users, aggregating it to form a network-wide dispatchable resource pool. When the power grid dispatch center issues a peak-shaving demand, the VPP cloud platform 20, based on the dynamic reserved power of each user, generates a VPP dispatch instruction for that household: rated power of energy storage device 30 6kW - dynamic reserved power 3.5kW = discharge power 2.5kW, dispatch duration 30 minutes, and sends the VPP dispatch instruction to the smart distribution box 10.

[0041] After receiving the VPP dispatch command, the intelligent distribution box 10 forwards it to the energy storage device 30. The battery management system of the energy storage device 30 confirms that the current state of charge is sufficient, and the bidirectional converter begins to discharge at a power of 2.5kW, supplying electrical energy to the household load through the intelligent distribution box 10. During the discharge process, if the air conditioner actually starts, the energy storage device 30 uses the reserved 3.5kW power to buffer and support the instantaneous impact, avoiding dispatch interruption; if the air conditioner does not start at the end of the discharge cycle, the reserved power is not used, and the intelligent distribution box 10 recalculates and updates the dynamic reserved power in the next dispatch cycle.

[0042] Thus, in this embodiment, a cloud-edge-device collaborative home energy storage system 100 is constructed through a three-tier architecture consisting of a smart distribution box 10, a VPP cloud platform 20, and an energy storage device 30. The smart distribution box 10 enables local sensing of load characteristics and dynamic calculation of reserved power, ensuring real-time response to load surges; the VPP cloud platform 20 performs aggregated scheduling and global optimization of resources, enhancing the regulation capabilities of the virtual power plant; and the energy storage device 30 reliably executes commands to complete energy storage and release. This system balances local response timeliness with the globality of cloud scheduling, providing a complete solution for the large-scale participation of residential energy storage in virtual power plant services.

[0043] Please see Figure 6 In some embodiments, the intelligent distribution box 10 includes multiple detection loops 11 and an edge computing unit 12. The multiple detection loops 11 are configured to collect electrical characteristic data; the edge computing unit 12 is connected to the detection loops 11 and is configured to predict the LRA starting inrush current and starting probability of each load loop using a machine learning model based on the electrical characteristic data, and to calculate the dynamic reserved power based on the LRA starting inrush current and starting probability.

[0044] It is worth noting that the detection circuit 11 refers to the electrical parameter acquisition channel inside the intelligent distribution box 10 that corresponds one-to-one with each load circuit in the household. Each detection circuit 11 includes a current sensor, a voltage sensor, and a signal conditioning circuit, used to collect the electrical characteristic data of the corresponding load circuit in real time, and transmit the collected data to the edge computing unit 12 for processing. The number of detection circuits can be flexibly configured according to the actual number of load circuits in the household.

[0045] The edge computing unit 12 refers to the local computing module integrated within the intelligent distribution box 10. It includes a processor, memory, and communication interface, and is used to run machine learning models, process data collected by the detection loop, and calculate the dynamic reserved power. As the core computing component of the intelligent distribution box 10, the edge computing unit 12 undertakes local data processing and real-time control functions. It can independently complete load sensing, start-up prediction, and reserved power calculation without cloud intervention, ensuring the real-time response to load impacts.

[0046] In one embodiment, a smart distribution box 10 is deployed in a residential household and includes multiple detection loops 11 and an edge computing unit 12. The multiple detection loops 11 are connected one-to-one with each load loop in the household. In this embodiment, the household loads include four loops: air conditioner, refrigerator, lighting, and television. The distribution box 10 is equipped with four detection loops 11 accordingly. Each detection loop 11 has a built-in high-precision current transformer and voltage sampling circuit. It collects the current and voltage waveform data of the corresponding load in real time at a sampling frequency of 1kHz, and performs preprocessing such as filtering, amplification, and analog-to-digital conversion. The processed electrical characteristic data is then sent to the edge computing unit 12 through an internal bus.

[0047] The edge computing unit 12 is connected to all detection loops 11 via an internal bus. The edge computing unit 12 uses an ARM processor, with built-in memory and storage space, and pre-stores a trained prediction model and a dynamic reserved power calculation program.

[0048] Edge computing unit 12 receives electrical characteristic data uploaded by each detection loop 11; then runs machine learning model to generate LRA starting inrush current and starting probability for each load loop; calculates dynamic reserved power based on prediction results; and uploads dynamic reserved power and related data to VPP cloud platform through communication module.

[0049] In this embodiment, the detection circuit 11 of the air conditioning circuit collects voltage and current data during the standby state of the air conditioner and uploads it to the edge computing unit 12. After running a machine learning model, the edge computing unit 12 outputs a start probability of 35% and an LRA start-up inrush current of 46A for the air conditioning circuit; and a start probability of 20% and an LRA start-up inrush current of 28A for the refrigerator circuit. Based on these prediction results, the edge computing unit 12 calculates the current dynamic reserved power as 3.5kW and then uploads the result to the VPP cloud platform 20.

[0050] Thus, in this embodiment of the application, by setting up multiple detection loops 11 that correspond one-to-one with the load loops, fine monitoring of the electrical characteristics of each load is achieved; by integrating the edge computing unit 12, the predictive model inference and dynamic reserved power calculation are completed locally, which not only ensures the real-time performance of data processing, but also reduces the cloud communication pressure and data transmission volume, thereby improving the system's response speed and reliability.

[0051] Please see Figure 7 In some implementations, the VPP cloud platform 20 includes a cluster scheduling engine 21, a dynamic reserved power database 22, and a price signal interface 23. The cluster scheduling engine 21 is configured to receive grid dispatch requests and generate VPP dispatch instructions based on these requests and the dynamic reserved power. The dynamic reserved power database 22 is configured to store the dynamic reserved power reported from the smart distribution box 10. The price signal interface 23 is configured to receive external price information to assist in generating VPP dispatch instructions.

[0052] It is worth noting that the cluster scheduling engine 21 is used to aggregate all connected household energy storage resources. Based on the grid scheduling requirements, the dynamic reserved power reported by each user, and electricity price information, it coordinates and generates differentiated VPP scheduling instructions and sends them to the smart distribution box 10. The cluster scheduling engine 21 is responsible for the optimal allocation and scheduling decisions of resources.

[0053] The dynamic reserved power database 22 refers to the database module in the VPP cloud platform 20 used to store the dynamic reserved power data reported by each smart distribution box 10. It records and updates the reserved power information of all users in real time, providing data support for the cluster scheduling engine 21.

[0054] The electricity price signal interface 23 refers to the communication module that connects the VPP cloud platform 20 with the external electricity market or grid dispatch center. It is used to receive electricity price information and ancillary service price signals in real time, providing an economic decision-making basis for the generation of dispatch instructions. The interface supports parsing multiple data formats and pushes the parsed electricity price information to the cluster dispatch engine 21.

[0055] In one embodiment, the VPP cloud platform 20 is deployed on a cloud server, and the VPP cloud platform 20 includes a cluster scheduling engine 21, a dynamic reserved power database 22, and an electricity price signal interface 23.

[0056] The dynamic reserved power database 22 receives and stores dynamic reserved power data reported by the smart distribution boxes 10 of households in a given area in real time. Each data record includes information such as user ID, reporting time, dynamic reserved power value, and corresponding load type. The database adopts a distributed storage architecture, supports high-concurrency writes and fast queries, and provides real-time data support for the cluster scheduling engine 21.

[0057] The electricity price signal interface 23 connects to the power trading platform via a dedicated communication link to receive various electricity service price information in real time, including peak-shaving prices, frequency regulation prices, and spot market prices. The interface supports parsing multiple data formats and pushes the parsed electricity price information to the cluster scheduling engine 21.

[0058] The cluster scheduling engine 21 periodically reads the current reserved power data of all users from the dynamic reserved power database 22, obtains real-time electricity price information from the electricity price signal interface 23, and simultaneously receives scheduling requests issued by the power grid dispatch center. Based on the available capacity of each user, i.e., the rated power of the energy storage device 30 minus the dynamic reserved power, and in conjunction with the electricity price signal, the cluster scheduling engine 21 prioritizes scheduling users with low costs and fast response times, generates differentiated VPP scheduling instructions, and issues them for execution.

[0059] In this example, the power grid dispatch center issues the following dispatch request: 50MW of discharge power needs to be increased within the next 30 minutes for peak shaving during the evening rush hour. The cluster dispatch engine 21 queries the dynamic reserved power database 22 to obtain the available capacity of all online users and filters out the set of users with available capacity greater than 0. Combining the current peak shaving price of 0.5 yuan / kWh obtained from the electricity price signal interface 23, the engine uses an optimization algorithm to prioritize dispatching users with low costs and fast response times, generates differentiated dispatch instructions, and sends them to the smart distribution boxes 10 of each user.

[0060] Thus, in this embodiment, the VPP cloud platform 20, through the coordinated operation of the cluster scheduling engine 21, the dynamic reserved power database 22, and the electricity price signal interface 23, achieves efficient aggregation and intelligent scheduling of massive residential energy storage resources. The dynamic reserved power database 22 gathers user-side load characteristic information in real time, providing a data foundation for scheduling decisions; the electricity price signal interface 23 introduces market price signals, enabling scheduling decisions to take into account economic efficiency; the cluster scheduling engine 21 coordinates the overall situation, generating differentiated instructions adapted to the actual capabilities of each user, thereby enhancing the overall regulation capability and market competitiveness of the virtual power plant.

[0061] Please see Figure 8 This application also proposes a power dispatch control method for a home energy storage system 100, applied to a smart distribution box 10. The home energy storage system includes an energy storage device 30 electrically connected to the smart distribution box 10. The method includes: 01. Collect electrical characteristic data of multiple load circuits connected to the intelligent distribution box 10; 02. Based on electrical characteristic data, machine learning models are used to predict the LRA starting inrush current and starting probability of each load circuit. 03. Calculate the dynamic reserve power based on the LRA inrush current and inrush probability; the dynamic reserve power is used to reserve power buffer for the load circuit to cope with the LRA inrush current when participating in VPP dispatch. 04. Report the dynamic reserved power to the VPP cloud platform 20; 05. Receive and execute VPP scheduling instructions from VPP cloud platform 20. VPP scheduling instructions include VPP scheduling power, which is determined based on grid scheduling requirements and dynamic reserved power. 06. Send VPP scheduling instructions to the energy storage device so that the energy storage device can output or input power according to the VPP scheduling power.

[0062] It is worth noting that electrical characteristic data refers to various parameter data reflecting the electrical state of the load circuit. Machine learning models are predictive models trained based on historical operating data, used to predict the LRA starting inrush current and starting probability of each load circuit. The LRA starting inrush current refers to the peak value of the locked rotor current generated at the moment of startup of motor-type loads.

[0063] Startup probability refers to the likelihood that a load circuit will transition from a standby state to a startup state at a given moment. Dynamic reserved power refers to the power buffer proactively reserved to cope with potential LRA startup impacts on the load circuit. VPP cloud platform 20 refers to a virtual power plant management platform deployed in the cloud, responsible for resource aggregation, scheduling decisions, and command issuance.

[0064] VPP dispatch instructions refer to downlink control instructions generated by the VPP cloud platform 20 based on grid dispatch requirements and dynamically reserved power reported by users. VPP dispatch power refers to the target power value specified in the VPP dispatch instruction.

[0065] In one embodiment, a home energy storage system 100 is deployed in a residential household. The home energy storage system includes a smart distribution box 10 and an energy storage device 30. The smart distribution box 10 is connected to multiple load circuits; in this embodiment, an air conditioning circuit is used as an example. The energy storage device 30 has a rated power of 6kW and is electrically connected to the smart distribution box 10.

[0066] First, the intelligent distribution box 10 monitors the voltage and current of the air conditioner load circuit in real time. When the air conditioner is in standby mode, the intelligent distribution box 10 collects electrical characteristic data at the current moment, including the effective value of voltage, the effective value of current, active power, reactive power, and the harmonic component characteristics in the current waveform.

[0067] Next, the intelligent distribution box 10 inputs the collected electrical characteristic data into a pre-trained machine learning model. This model, trained using the household's historical operating data, can predict the load's startup behavior based on the input electrical characteristic data. After inference calculation, the model outputs that the air conditioner has a 35% probability of starting at the current moment and predicts the LRA starting inrush current to be 46A.

[0068] Based on the LRA starting inrush current of 46A and the starting probability of 35% generated in the above steps, the intelligent distribution box 10 calculates the dynamic reserve power. The dynamic reserve power characterizes the power buffer that needs to be reserved to cope with the instantaneous power surge that may occur when the air conditioner starts. The calculation shows that the current required reserve power is approximately 3.5kW.

[0069] Subsequently, the intelligent distribution box 10 reports the calculated dynamic reserved power to the VPP cloud platform 20.

[0070] The VPP cloud platform 20 determines the power supply based on the grid dispatch demand of 4kW and the user's reported dynamic reserved power of 3.5kW: the sum of the dispatch demand of 4kW and the reserved power of 3.5kW is 7.5kW, which exceeds the rated power of the energy storage device 30 by 6kW. Therefore, the VPP cloud platform 20 adjusts the actual VPP dispatch power to the rated power minus the dynamic reserved power, i.e., 6kW minus 3.5kW equals 2.5kW, and generates a VPP dispatch command containing 2.5kW of discharge power, which is then sent to the smart distribution box 10.

[0071] After receiving the instruction, the intelligent distribution box 10 sends a VPP dispatch instruction containing 2.5kW discharge power to the energy storage device 30, which then begins discharging at 2.5kW. During the discharge process, if the air conditioner actually starts, the energy storage device 30 uses its reserved 3.5kW power to buffer and support instantaneous impacts, preventing output voltage drops or overload protection. If the air conditioner does not start at the end of the discharge cycle and the reserved power is not used, the intelligent distribution box 10 will repeat the above steps in the next dispatch cycle to update the dynamic reserved power value.

[0072] The technical advantages of this application can be further illustrated through comparative experiments with traditional scheduling strategies. Please refer to [link / reference]. Figure 9 and Figure 10 ,in Figure 9 This is a schematic diagram comparing the total load and reserved power of the implementation method of this application with that of the traditional scheduling strategy. Figure 10 This is a diagram showing the comparison of the success rates of scheduling instruction execution.

[0073] like Figure 9 As shown, at the moment of startup of motor-type loads, the traditional fixed reservation strategy suffers from insufficient available capacity, leading to a drop in output power and scheduling interruption. This application adopts a dynamic reservation strategy, which matches the available capacity with the load impact risk in real time, maintaining stable output during impacts. Based on this, as Figure 10 Experimental data shows that in household power consumption scenarios involving the starting impact of motor-type loads, traditional scheduling strategies fail due to overload impact, with an execution success rate of approximately 86.7%. This application improves the execution success rate of scheduling commands to 99.4% through a dynamic reserved power mechanism, significantly enhancing the reliability of household energy storage systems participating in virtual power plant scheduling.

[0074] Thus, in this embodiment, by locally collecting load data, predicting the LRA starting inrush current and starting probability, and dynamically calculating the reserved power by the intelligent distribution box 10, the home energy storage system 100 is able to cope with instantaneous power surges when responding to VPP dispatch, effectively solving the problem of dispatch response failure caused by the starting of motor-type loads. Simultaneously, the dynamic reservation strategy achieves real-time matching of power buffer and load risk, avoiding capacity waste caused by traditional fixed reservation methods and improving resource utilization; moreover, it eliminates the need for additional hardware compensation units, reducing system cost and complexity.

[0075] Please see Figure 11 In some implementations, the method further includes: 001, Monitor the actual LRA starting inrush current during load startup; 002. The machine learning model is corrected online based on the actual LRA starting inrush current.

[0076] It is worth noting that the actual LRA starting inrush current refers to the true value of the locked rotor current measured at the moment of actual load startup. When the intelligent distribution box 10 detects a load startup event through the current change rate, it records the current waveform at the moment of startup using high-frequency sampling, and extracts the maximum peak value from the current waveform, which is the actual LRA starting inrush current. This value serves as the true label of the model prediction result, used to evaluate the prediction error and guide model optimization.

[0077] Online correction refers to the process of adjusting the parameters of a machine learning model in real time based on newly acquired actual operating data. By comparing the actual LRA starting impact current with the model's predicted value, the prediction error is calculated, and incremental learning or backpropagation algorithms are used to fine-tune the model weights, enabling the model to adapt to dynamic changes in load behavior due to factors such as seasons and usage habits, thereby continuously improving prediction accuracy.

[0078] In one embodiment, the intelligent distribution box 10 continuously monitors the current changes in the load circuit during the execution of scheduling instructions, and triggers the model update process when a motor-type load start-up event is detected.

[0079] The intelligent distribution box 10 sets load start detection conditions: when the current change rate di / dt exceeds the preset threshold for three consecutive sampling points, and the current amplitude rises to several times the rated current, it is determined to be a motor load start event.

[0080] In this embodiment, the intelligent distribution box 10 detects that the air conditioner load circuit current rapidly increases from 0.4A to 43A within 3ms, determining that the air conditioner has actually started. The intelligent distribution box 10 immediately records the current waveform at the moment of startup, extracts the maximum peak value from the current waveform, and obtains the LRA startup inrush current as 43A, denoted as . .

[0081] The intelligent distribution box 10 uses this startup event as a new training sample to update the machine learning model online. First, training samples are constructed. The intelligent distribution box 10 reads the electrical feature data input to the machine learning model before the current start-up time from the historical database as sample features; the actual LRA start-up inrush current value of 43A is used as the sample label, and the two together form a complete training sample.

[0082] Next, the loss function value is calculated. The sample features are input into the current machine learning model to obtain the LRA starting impact current value output by the machine learning model. In this embodiment, the model's predicted value before the air conditioner starts is 46A, denoted as... The intelligent distribution box 10 uses a preset mean square error loss function to calculate the difference between the predicted and actual values. Specifically:

[0083] The loss function value is 9, indicating the degree of error in this prediction.

[0084] Then, the model is updated. The intelligent distribution box 10 uses the backpropagation algorithm to calculate the gradient of the loss function with respect to each model parameter layer by layer, starting from the output layer, to obtain the contribution direction of each weight parameter to the prediction error. The model weights are fine-tuned and updated according to the preset learning rate of 0.001, so that the model can more accurately output the LRA starting impact current value under similar working conditions in subsequent predictions.

[0085] Through continuous online corrections, the model has become increasingly accurate in understanding the load behavior characteristics of specific households, the prediction error has gradually decreased, and the calculation of dynamic reserved power has become more reasonable.

[0086] Thus, in this embodiment of the application, by monitoring the actual LRA starting inrush current during load startup and making online corrections to the machine learning model, the model can be continuously optimized based on actual operating data, gradually improving the prediction accuracy of LRA starting inrush current and startup probability, thereby providing a more reliable data foundation for the calculation of dynamic reserved power.

[0087] In some implementations, electrical characteristic data includes at least one of timing power data, environmental parameters, and historical load operation data.

[0088] It is worth noting that time-series power data refers to the power change data of the load over a continuous time series, including curves or sampling point sequences showing the changes of active power and reactive power over time. Time-series power data can reflect the operating mode, periodic characteristics, and power fluctuation patterns of the load, such as the standby power consumption fluctuation of air conditioners and the periodic start-stop characteristics of refrigerators, providing a time-series characteristic basis for predicting the starting inrush current and starting probability of LRAs.

[0089] Environmental parameters refer to various parameter data related to the operating environment of the load, including but not limited to ambient temperature, ambient humidity, light intensity, and time information. Environmental parameters affect the operating status and startup patterns of the load. For example, the startup probability of an air conditioner is closely related to the ambient temperature; the higher the ambient temperature, the greater the likelihood of the air conditioner starting, and the higher the corresponding startup probability.

[0090] Historical load operation data refers to the record of the load's operating status over a past period, including historical start and stop times, runtime, power curves, and start intervals. This data is used to train machine learning models, enabling them to learn load behavior patterns, such as the electricity consumption habits of specific households during specific time periods, thereby improving the accuracy of LRA (Local Load Regulator) inrush current and start probability predictions.

[0091] Thus, in this embodiment of the application, by introducing at least one of time-series power data, environmental parameters, and historical load operation data as electrical characteristic data, the machine learning model can capture the behavior patterns of the load from multiple dimensions, further improving the prediction accuracy of LRA start-up inrush current and start-up probability, providing a more reliable data foundation for the calculation of dynamic reserved power, thereby better ensuring the reliability of home energy storage system participating in VPP scheduling.

[0092] Please see Figure 12 In some implementations, the machine learning model includes a dual-channel long short-term memory network, and step 02 includes: 021. Construct feature vectors that conform to the input of the machine learning model based on time-series power data, environmental parameters, and historical load operation data; 022. The feature vector is input into the dual-channel long short-term memory network. The first channel of the dual-channel long short-term memory network processes the timing power data, and the second channel of the dual-channel long short-term memory network processes the environmental parameters and load operation history data, and outputs the LRA start-up inrush current and start-up probability.

[0093] It's worth noting that a dual-channel long short-term memory network refers to a deep learning network architecture with two parallel processing channels, used to simultaneously process input data of different types or sources. Each channel independently extracts deep features from its own input data, and then a feature fusion layer integrates the features extracted from the two channels, ultimately outputting a prediction result based on the fused combined features.

[0094] In this application, the first channel of the dual-channel Long Short-Term Memory (LSTM) network is dedicated to processing time-series power data, extracting temporal features of load power changes over time; the second channel is used to process environmental parameters and historical load operation data, extracting non-temporal features affecting load startup behavior. The feature vectors extracted by the two channels are fused in the deep layers of the network, ultimately outputting the prediction result. A feature vector is a numerical vector representation constructed from preprocessed time-series power data, environmental parameters, and historical load operation data, meeting the input requirements of a machine learning model, and containing the statistical features of the time-series power data sequence, environmental parameter values, and historical load operation data.

[0095] In one embodiment, the intelligent distribution box 10 is equipped with a trained dual-channel long short-term memory network model to predict the LRA starting inrush current and starting probability of each load circuit.

[0096] First, the intelligent distribution box 10 acquires the electrical characteristic data of the air conditioning load through a detection circuit, specifically including: time-series power data. The ambient temperature is Grid voltage Power factor Load start history interval Number of launches within the day .

[0097] Next, the intelligent distribution box 10 preprocesses the collected data to construct a feature vector X. The feature vector X consists of the electrical feature data collected above.

[0098] Then, the constructed feature vector X is input into a dual-channel long short-term memory network. The first channel receives the time-series power data sequence, extracts the time-series features of the air conditioner power changing over time through multi-layer LSTM units, and outputs the time-series feature vector. The second channel receives the ambient temperature. Grid voltage Power factor and load start history interval Number of times started within a day Non-time series data, such as the current time, are processed through a fully connected layer to extract non-time series features and output a non-time series feature vector. The feature vectors extracted from the two channels are concatenated in the fusion layer to form a comprehensive feature vector. After being mapped by the fully connected layer, the output layer simultaneously outputs the start-up probability and the predicted peak value of the locked rotor current.

[0099] Thus, in this embodiment, by employing a dual-channel long short-term memory network to process time-series power data and non-time-series data respectively, the machine learning model can fully mine the feature information of different types of data, avoiding the feature confusion problem when a single channel processes multi-source data, and further improving the prediction accuracy of LRA starting impact current and starting probability.

[0100] In some implementations, the calculation expression for dynamic reserved power includes:

[0101] in, This is the load weighting coefficient. For safety reasons, The range is 1.1-1.2. The time decay coefficient, For the start probability, The starting inrush current for LRA.

[0102] It is worth noting that the load weighting coefficient refers to the weight value set according to the impact of different types of loads on the intelligent distribution box 10. Different load types correspond to different load weighting coefficients. For example, the impact of air conditioner loads at startup is relatively large, so a higher weighting coefficient can be set; the impact of refrigerator loads at startup is relatively small, so a medium weighting coefficient can be set. The load weighting coefficient can be obtained through experimental calibration or historical data statistics and can be preset during system configuration.

[0103] The safety factor is a redundancy factor introduced to ensure the reliability and safety of the intelligent distribution box 10. Its functions are as follows: first, to compensate for the inherent prediction errors of the prediction model, ensuring sufficient power margin to cope with actual shocks even if the predicted values ​​deviate to some extent; second, to cope with fluctuations in operating conditions such as grid voltage fluctuations and changes in load characteristics; and third, to reserve space for the aging and performance degradation of the intelligent distribution box 10 itself. The specific value of the safety factor can be dynamically adjusted according to the operating status of the intelligent distribution box 10: a lower limit of 1.1 is used when the intelligent distribution box 10 is in good health and the ambient temperature is suitable; an upper limit of 1.2 is used when the intelligent distribution box 10's health deteriorates, the ambient temperature is too low or too high, or the grid stability is poor, to provide a greater safety margin.

[0104] Safety margin refers to the additional power reserve reserved to ensure the stable operation of the intelligent distribution box 10, used to cope with unforeseen factors such as instantaneous power fluctuations, communication delays, command execution deviations, and prediction errors. The safety margin can be a fixed value or dynamically adjusted according to the system's operating status, for example, increasing the margin when battery health declines. In available capacity calculations, the safety margin can be used as a deduction item to ensure that the intelligent distribution box 10 can still maintain safe operation under extreme conditions.

[0105] The time decay coefficient is a decay factor that reflects the timeliness of the prediction result. As the time interval between the prediction time and the actual start time of the load increases, the coefficient decreases accordingly, which is used to reflect the characteristic that the prediction confidence decreases over time.

[0106] After obtaining the LRA starting impact current and starting probability obtained from the aforementioned implementation method, the two can be substituted into the reserved power calculation formula of this implementation method to obtain the dynamic reserved power value that adapts to the current load impact risk.

[0107] In one embodiment, the weighting factor for the air conditioning load Take 1.2; the current intelligent distribution box 10 is operating well, and the safety factor is 1.2. Use 1.15; the interval between the prediction time and the scheduling time is 5 minutes, and the time decay coefficient is... Set to 0.1; predict the start probability output by the model. The starting inrush current of LRA is 0.35. The value is 46A; the interval between the predicted time t0 and the current time t is 5 minutes.

[0108] Substituting the values ​​from the above embodiments into the formula, the dynamic reserve power can be calculated to be 2.9 kW, meaning that a power margin of 2.9 kW is currently required to cope with the instantaneous power surge that may occur when the air conditioner starts.

[0109] Please see Figure 13 , Figure 13 This diagram illustrates a comparison of the reserved power curves of the implementation method described in this application and the traditional fixed reserved power strategy, corresponding to three typical load types: air conditioners, refrigerators, and water pumps. The dashed line represents the traditional fixed reserved power, while the solid line represents the reserved power dynamically calculated by this application based on predicted start-up probability and LRA start-up inrush current. As shown, traditional fixed reserved power leads to wasted capacity during low-risk periods and may be insufficient during high-risk periods; the dynamic reserved power of this application matches the actual impact risk of the load in real time, maximizing resource utilization while ensuring safety.

[0110] Thus, in this embodiment of the application, by introducing a calculation expression for dynamic reserved power, the load weight coefficient, safety coefficient and time decay coefficient in the expression are used to perform refined calculation of dynamic reserved power, so that the reserved power can more accurately match different load characteristics, operating conditions and prediction timeliness requirements. While ensuring the reliability of dispatching, the resource waste caused by excessive reservation is avoided, and the dispatching economy and adaptability of the intelligent distribution box 10 are further improved.

[0111] Please refer to this document. Figure 14 This application also proposes a power dispatch control method for a home energy storage system 100, applied to a VPP cloud platform 20. The home energy storage system includes an energy storage device 30 electrically connected to a smart distribution box 10. The method includes: 001, receive grid dispatch requirements and dynamic reserved power sent by smart distribution box 10. The dynamic reserved power is used to reserve power buffer for the load circuit to cope with the LRA start-up inrush current when participating in VPP dispatch. 002, Generate VPP dispatch instructions based on grid dispatch requirements and dynamically reserved power; 003, send VPP dispatch instructions to the smart distribution box 10, and the smart distribution box sends the VPP dispatch instructions to the energy storage device 30 so that the energy storage device can output or input power according to the VPP dispatch power.

[0112] It is worth noting that grid dispatch demand refers to the instruction information issued by the grid operator or power dispatch center that requires virtual power plants to provide specific power services.

[0113] In one specific example, the VPP cloud platform 20 is deployed on a cloud server and communicates with the smart distribution boxes 10 of multiple homes.

[0114] The VPP cloud platform 20 receives a power grid dispatch request from the power grid dispatch center: a 50MW increase in discharge power is required within the next 30 minutes. Simultaneously, the VPP cloud platform 20 receives a report from a household smart distribution box 10 regarding a dynamic reserved power of 4.25kW.

[0115] Based on the dynamic reserved power of 4.25kW reported by the smart distribution box 10 and the rated power of its energy storage device 30 of 6kW, the VPP cloud platform 20 calculates the available capacity to be 1.75kW. Based on this, the VPP cloud platform 20 generates a VPP dispatch instruction for the household: a discharge power of 1.5kW and a dispatch duration of 30 minutes, and sends the instruction to the household's smart distribution box 10.

[0116] Thus, in this embodiment of the application, the VPP cloud platform 20 receives the dynamic reserved power reported by the smart distribution box 10, and generates VPP dispatch instructions in a coordinated manner based on the grid dispatch requirements and issues them for execution, thereby realizing the efficient aggregation and optimized dispatch of massive household energy storage resources.

[0117] Please refer to this document. Figure 15 In some implementations, step 002 includes: 0021. Determine whether the grid dispatch demand is less than or equal to the available capacity. The available capacity is determined based on the inverter peak power and the dynamic reserved power. 0022, When the grid dispatch demand is less than or equal to the available capacity, generate an immediate execution instruction; or 0023, when the grid dispatch demand exceeds the available capacity, generate a negotiated execution instruction.

[0118] It is worth noting that available capacity refers to the maximum power value that the home energy storage system 100 can use to respond to grid dispatch under the current state. It is determined based on the inverter's peak power and dynamic reserved power, specifically the difference between the inverter's peak power and the dynamic reserved power. Available capacity characterizes the actual upper limit of power that the energy storage device 30 can safely use to respond to grid dispatch while ensuring load impact response capability and additional safety redundancy.

[0119] Inverter peak power refers to the maximum power value that the inverter in the energy storage device 30 can continuously output. It is an inherent parameter of the energy storage device 30 and is determined by the inverter's hardware specifications.

[0120] An immediate execution instruction refers to a VPP dispatch instruction type generated by the VPP cloud platform 20 when the grid dispatch demand does not exceed the available capacity, requiring the energy storage device 30 to execute immediately according to the dispatch demand power. This instruction includes parameters such as VPP dispatch power, dispatch direction, and execution time period, and is immediately executed after being forwarded to the energy storage device 30 through the smart distribution box 10.

[0121] Negotiated execution instructions are dispatch commands generated by the local controller when grid dispatch demand exceeds the available capacity of the home energy storage system. These instructions are used to negotiate with the virtual power plant platform. Negotiated execution instructions include derated execution instructions (containing adjusted executable power values), delayed execution instructions (containing suggested delay time information), and rejection instructions (containing feedback information indicating inability to execute). The local controller uploads the negotiated execution instructions to the virtual power plant platform, which then confirms or adjusts them based on the actual grid demand.

[0122] Negotiated execution instructions refer to the VPP dispatch instructions generated by the VPP cloud platform 20 when the grid dispatch demand exceeds the available capacity. These include derated execution instructions and delayed execution instructions. The derated execution instructions include the adjusted VPP dispatch power; the delayed execution instructions include the suggested delay time information. After the instructions are sent to the smart distribution box 10, the smart distribution box 10 forwards them to the energy storage device 30 for execution.

[0123] In the following two embodiments, after receiving the power grid dispatch request, the VPP cloud platform 20 performs hierarchical dispatch decisions based on the dynamic reserved power reported by the smart distribution box 10.

[0124] First, the VPP cloud platform 20 calculates the available capacity based on the inverter's peak power and the dynamic reserved power reported by the smart distribution box 10. In this embodiment, the inverter peak power of the energy storage device 30 is 6kW, and the current dynamic reserved power reported by the smart distribution box 10 is 2.9kW, so the available capacity is 6kW - 2.9kW = 3.1kW.

[0125] The VPP cloud platform 20 received a grid dispatch demand of 2.5kW for discharge. The platform compared the dispatch demand of 2.5kW with the available capacity of 3.1kW and determined that the dispatch demand was less than the available capacity.

[0126] Since the dispatch demand of 2.5kW is less than the available capacity of 3.1kW, the VPP cloud platform 20 determines that there is still sufficient remaining capacity to meet the dispatch demand while ensuring the ability to cope with load surges. Therefore, the VPP cloud platform 20 generates an immediate execution instruction, encapsulates the dispatch demand power of 2.5kW as the VPP dispatch power into a VPP dispatch instruction, and sends it to the intelligent wiring box 10.

[0127] In another embodiment, the VPP cloud platform 20 receives a grid dispatch demand of 4kW discharge, which is greater than the available capacity of 3.1kW. The platform determines that the dispatch demand is greater than the available capacity.

[0128] Since the scheduling demand of 4kW exceeds the available capacity of 3.1kW, the VPP cloud platform 20 determines that it cannot fully meet the scheduling demand while ensuring the ability to cope with load surges. Therefore, the platform generates a negotiated execution instruction.

[0129] Specifically, the platform calculates the currently available capacity of 3.1kW as the executable power and generates a derating execution instruction. This instruction includes the original scheduling demand information of 4kW and the actual VPP scheduling power information of 3.1kW, with the reason being reserved power occupation. The VPP cloud platform 20 sends the generated VPP scheduling instruction to the smart distribution box 10, which then forwards it to the energy storage device 30 for execution.

[0130] In another implementation, if the scheduling demand is time-elastic, the VPP cloud platform 20 can predict future changes in available capacity based on the time decay characteristics of the dynamically reserved power. For example, if it is expected that the dynamically reserved power will decrease to 2.0kW in 30 minutes, at which point the available capacity will increase to 4.0kW, the platform generates a delayed execution instruction, including information suggesting a 30-minute delay, and issues it as a VPP scheduling instruction.

[0131] Please see Figure 16 , Figure 16 This diagram illustrates a comparison of the dispatchable capacity of a virtual power plant platform using the implementation method of this application and traditional scheduling strategies. As shown, the traditional fixed reservation strategy results in significant fluctuations in dispatchable capacity and low utilization. This application, through a dynamic power reservation mechanism, releases unused reserved capacity in real time while ensuring the ability to cope with load surges, thus maintaining a high level of dispatchable capacity with smooth fluctuations. The capacity release gain area visually demonstrates the additional dispatchable capacity released by this application compared to the traditional strategy, effectively improving the resource aggregation efficiency of the virtual power plant.

[0132] Thus, in this embodiment, the VPP cloud platform 20 introduces a hierarchical scheduling mechanism, adopting either immediate execution or negotiated execution strategies based on the matching degree between scheduling needs and available capacity. This ensures the ability to cope with load shocks while avoiding overload or failure caused by scheduling needs exceeding system capacity, thereby improving the reliability and adaptability of home energy storage systems participating in virtual power plant scheduling.

[0133] Please see Figure 17 In some embodiments, the power dispatch control method further includes: 0003, Obtain electricity service pricing; 0004. Generate VPP dispatch instructions based on electricity service prices, grid dispatch requirements, and dynamic reserved power.

[0134] It is worth noting that the electricity service price refers to the price or rate paid by the VPP cloud platform or grid operator for different types of electricity ancillary services, including but not limited to peak-shaving prices, frequency regulation prices, and reserve prices. The electricity service price can fluctuate in real time according to market supply and demand, and can also be priced differently based on factors such as service type, response time, and response reliability. In this application, the electricity service price serves as an economic decision factor in the generation of VPP dispatch instructions, used to optimize the dispatch revenue of the residential energy storage system 100.

[0135] In one embodiment, the VPP cloud platform interacts with the power trading center via a communication module to obtain real-time electricity service price information. The price information is distributed in structured data format, including fields such as service type, valid price period, and unit price.

[0136] The VPP cloud platform obtained the following pricing information: the price for real-time frequency regulation service is 0.8 yuan / kWh, and the effective period is within 15 minutes from the current time; the price for real-time peak shaving service is 0.3 yuan / kWh, and the effective period is within 30 minutes from the current time.

[0137] The VPP cloud platform received a grid dispatch request: within the next 30 minutes, the home energy storage system is required to discharge at a power of 3kW for real-time peak shaving services. Simultaneously, the platform received a report from smart distribution box 10 indicating a dynamic reserved power of 2.9kW. Combined with the rated power of the energy storage device (6kW), the calculated available capacity is 3.1kW, which is insufficient to meet the 3kW dispatch request.

[0138] The VPP cloud platform 20, after checking electricity service prices, found that the price of real-time frequency regulation service was 0.8 yuan / kWh, significantly higher than the 0.3 yuan / kWh price of real-time peak regulation. After comprehensive decision-making, the platform generated a negotiated execution instruction: issuing a VPP dispatch instruction to the smart distribution box 10, suggesting either executing peak regulation service at the available capacity of 2.8kW, or suggesting converting this dispatch to frequency regulation service to utilize available capacity in the frequency regulation market for higher returns.

[0139] In another example, the virtual power plant platform issues a dispatch request for 2kW discharge, and the available capacity is sufficient. The platform finds that the current peak-shaving price is 0.4 yuan / kWh, which meets the preset execution threshold. Therefore, it generates an immediate execution instruction to fully respond to the dispatch request for 2kW and issues the VPP dispatch instruction to the smart distribution box 10.

[0140] Thus, in this embodiment of the application, the VPP cloud platform introduces electricity service prices as an economic decision factor for generating VPP dispatch instructions, enabling home energy storage systems to maximize revenue by optimizing dispatch behavior based on market price signals while meeting the needs of load shock response. This provides a dispatch decision scheme that balances technical feasibility and economic optimization for home energy storage to participate in the virtual power plant market.

[0141] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A power dispatching and control method for a home energy storage system, applied to a smart distribution box, wherein the home energy storage system further includes an energy storage device electrically connected to the smart distribution box, characterized in that, The method includes: Collect electrical characteristic data of multiple load circuits connected to the intelligent distribution box; Based on the electrical characteristic data, a machine learning model is used to predict the LRA starting inrush current and starting probability of each load circuit; The dynamic reserved power is calculated based on the LRA initiation inrush current and the initiation probability; the dynamic reserved power is used to reserve power buffer for the load circuit to cope with the LRA initiation inrush current when participating in VPP scheduling; The dynamically reserved power is reported to the VPP cloud platform; Receive and execute VPP scheduling instructions from the VPP cloud platform, wherein the VPP scheduling instructions include VPP scheduling power, and the VPP scheduling power is determined based on grid scheduling requirements and the dynamically reserved power. Send the VPP scheduling command to the energy storage device so that the energy storage device outputs or inputs power according to the VPP scheduling power.

2. The power dispatch control method according to claim 1, characterized in that, The method further includes: Monitor the actual LRA startup inrush current during load startup; The machine learning model is corrected online based on the actual LRA starting inrush current.

3. The power dispatch control method according to claim 1, characterized in that, The electrical characteristic data includes at least one of time-series power data, environmental parameters, and historical load operation data.

4. The power dispatch control method according to claim 3, characterized in that, The machine learning model includes a dual-channel long short-term memory network. The step of using the machine learning model to predict the starting inrush current and starting probability of each load circuit (LRA) based on the electrical characteristic data includes: Based on the time-series power data, environmental parameters, and historical load operation data, a feature vector that conforms to the input of the machine learning model is constructed. The feature vector is input into the dual-channel long short-term memory network. The first channel of the dual-channel long short-term memory network processes the timing power data, and the second channel of the dual-channel long short-term memory network processes the environmental parameters and the load operation history data, and outputs the LRA start-up inrush current and the start-up probability.

5. The power dispatch control method according to claim 1, characterized in that, The calculation expression for the dynamically reserved power includes: in, This is the load weighting coefficient. For safety reasons, The range is 1.1-1.

2. The time decay coefficient, For the start probability, The starting inrush current for LRA.

6. A power dispatching and control method for a home energy storage system, applied to a VPP cloud platform, wherein the home energy storage system further includes an energy storage device electrically connected to a smart distribution box, characterized in that... The VPP cloud platform is communicatively connected to the intelligent distribution box, and the method includes: The system receives grid dispatching requests and the dynamic reserved power sent by the intelligent distribution box. The dynamic reserved power is used to reserve power buffer for the load circuit to cope with the LRA start-up inrush current when participating in VPP dispatching. VPP dispatch instructions are generated based on the power grid dispatch requirements and the dynamically reserved power. The VPP scheduling command is sent to the smart distribution box, which then sends the VPP scheduling command to the energy storage device so that the energy storage device can output or input power according to the VPP scheduling power.

7. The power dispatch control method according to claim 6, characterized in that, The step of generating VPP dispatch instructions based on the power grid dispatch requirements and the dynamically reserved power includes: Determine whether the grid dispatch demand is less than or equal to the available capacity, wherein the available capacity is determined based on the inverter peak power and the dynamically reserved power. If the grid dispatch demand is less than or equal to the available capacity, generate an immediate execution instruction; or When the grid dispatch demand exceeds the available capacity, a negotiated execution instruction is generated.

8. The power dispatch control method according to claim 6, characterized in that, The method further includes: Obtain electricity service pricing; The VPP dispatch instruction is generated based on the electricity service price, the grid dispatch demand, and the dynamic reserved power.

9. A home energy storage system, characterized in that, include: The intelligent distribution box connects multiple load circuits and is configured to collect electrical characteristic data of the loads. Based on the electrical characteristic data, it uses a machine learning model to predict the LRA inrush current and inrush probability of each load circuit, and calculates dynamic reserved power based on the LRA inrush current and the inrush probability. The dynamic reserved power is used to reserve power buffer for the load circuit to cope with the LRA inrush current when participating in VPP dispatch. The VPP cloud platform is communicatively connected to the smart distribution box and is configured to receive the dynamic reserved power and generate and issue VPP dispatch instructions to the smart distribution box according to the grid dispatch requirements and the dynamic reserved power. and An energy storage device is communicatively connected to the intelligent distribution box and configured to execute the VPP scheduling command to output or input power according to the VPP scheduling power.

10. The home energy storage system according to claim 9, characterized in that, The intelligent power distribution box includes: Multiple detection loops are configured to collect the electrical characteristic data; An edge computing unit, connected to the detection loop, is configured to predict the LRA start-up inrush current and start-up probability of each load loop using a machine learning model based on the electrical characteristic data, and to calculate the dynamic reserved power based on the LRA start-up inrush current and the start-up probability.

11. The home energy storage system according to claim 9, characterized in that, The VPP cloud platform includes: The cluster scheduling engine is configured to receive grid scheduling requests and generate VPP scheduling instructions based on the grid scheduling requests and the dynamically reserved power. A dynamic reserved power database is configured to store the dynamic reserved power reported from the smart distribution box; The electricity price signal interface is configured to receive external electricity price information to assist in generating the VPP scheduling instructions.

Images