Behavior portrait modeling method for typical adjustable equipment on residential user side

Abstract

The invention relates to the technical field of user behavior portraits, in particular to a behavior portrait modeling method for typical adjustable equipment on a residential user side. Comprising the following steps: based on a mechanism model analysis method, aiming at electrochemical energy storage equipment and an electric vehicle, constructing a physical constraint model with adjustable capability and adjustable cost; and establishing a dynamic mapping model of the real-time scheduling instruction and the scheduling cost. According to the method, the dynamic constraint model of the electrochemical energy storage charge capacity and the charging and discharging power is established, and the electric quantity-time window constraint of the electric vehicle is established, so that the adjustable power interval is corrected in real time, the scheduling instruction is ensured to accord with the actual operation state of the equipment, and the precision and timeliness of adjustable capability evaluation are improved; the problem that static capability estimation does not accord with reality is solved; based on the coupling relation between the time-of-use electricity price and the equipment operation power, dynamic and accurate measurement and calculation of the adjustable cost are realized; the dynamic capability and the real-time cost are fused through deep reinforcement learning, and the problem of extensive cooperation of multiple devices is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of user behavior profiling, and in particular to a method for modeling behavior profiles of typical adjustable devices on the residential user side. Background Art

[0002] Distributed, adjustable resources such as electrochemical energy storage devices and electric vehicles on the residential user side are becoming important elements for the flexible regulation of the power grid. These devices can respond to the grid's dispatching instructions, such as peak shaving and valley filling and frequency regulation, through dynamic charging and discharging. However, their large-scale access also places demands on refined user-side regulation technology. Existing technologies often rely on rule-based control or single-objective optimization for dispatching various types of adjustable devices. While these technologies can achieve basic command responses, they have significant shortcomings in systematically integrating the physical characteristics, economic costs, and real-time status of the devices, making it difficult to meet the dual regulation goals of "safe response and economic efficiency."

[0003] However, existing technologies have the following key defects in equipment characteristic modeling and coordinated scheduling: First, the assessment of adjustable capacity is rough: the existing technology uses the static empirical coefficient method to estimate the adjustment capacity, and does not integrate dynamic variables such as equipment operating status and environmental parameters. In particular, for adjustable equipment with strong time-varying, timing coupling and uncertainty characteristics such as electrochemical energy storage and electric vehicles, the static adjustable capacity is significantly inconsistent with the actual situation; Second, the measurement of adjustable cost is distorted: the existing technology relies on static experience and subjective estimation, and does not consider the dynamic coupling effect between the operating power change of adjustable equipment and time-of-use electricity prices. The adjustable cost is difficult to dynamically and accurately calculate; In view of this, we propose a behavioral portrait modeling method for typical adjustable equipment on the residential user side. Summary of the Invention

[0004] The purpose of the present invention is to provide a behavioral portrait modeling method for typical adjustable equipment on the residential user side to solve the problems of rough adjustable capacity evaluation and distorted adjustable cost estimation proposed in the above background technology.

[0005] To solve the above technical problems, the present invention provides a behavior profile modeling method for typical adjustable devices on the residential user side, comprising the following steps: S100, Mechanism Model Analysis: Based on the mechanism model analysis method, a physical constraint model of adjustable capacity and adjustable cost is constructed for electrochemical energy storage devices and electric vehicles; S200, Deep Reinforcement Learning Modeling: Based on the external characteristic modeling method of deep reinforcement learning, this method uses external observable states (such as the physical state of the equipment, the power grid, and the economic state) to establish a dynamic mapping model between real-time dispatch instructions and dispatch costs, and construct an adjustable capacity model and an adjustable cost model suitable for electrochemical energy storage devices and electric vehicles. S300 , generating a behavior profile: Based on the physical constraint model and dynamic mapping model of S100 - S200 , generating a behavior profile including an adjustable time period, an adjustable capacity threshold, and an adjustable cost.

[0006] As a further improvement of the present technical solution, the step of constructing a physical constraint model for the electrochemical energy storage device in S1 includes the following steps: S110.1: Modeling the charge and discharge power boundaries, establishing mutually exclusive constraints between the charge and discharge power and state variables: ; ; in, Indicates the discharge power of electrochemical energy storage equipment; Indicates the minimum value of discharge power; Indicates the maximum value of discharge power; A binary variable representing the discharge state (value 0 or 1), used to indicate whether the device is in the discharge state: Indicates the charging power of the electrochemical energy storage device; Indicates the minimum charging power of the electrochemical energy storage device; Indicates the maximum value of electric power; A binary variable representing the state of charge; + ; S110.2: Dynamic modeling of charge capacity, constructing the dynamic equation of charge capacity and charge and discharge power: ; in, express The state of charge of the energy storage device at the moment (range 0 1), which indicates the ratio of the remaining energy storage capacity to the rated capacity; is the time interval; represents the net power, and , when discharging >0 (SOC decreases), when charging <0 (SOC increased); Indicates the rated capacity of the energy storage device; express State of charge at the moment; express Net power at the moment; Indicates the minimum value of the state of charge; Maximum state of charge. Used to ensure that the SOC is within the physically allowed range (e.g. battery SOC is usually 0.2 SOC 0.8, avoid overcharge / overdischarge), and It indicates the minimum and maximum value of the battery state of charge.

[0007] As a further improvement of the present technical solution, in the adjustable cost modeling of the electrochemical energy storage device in S1, the opportunity cost calculation is implemented based on the time-of-use electricity price, and the calculation formula is as follows: ; in, represents the total opportunity cost of electrochemical energy storage devices; Indicates the total number of time periods within the time period; Indicates the current period index; Indicates the number of energy storage units; Indicates the number index of the energy storage unit; express Real-time electricity prices for the time period; Indicates the lowest electricity price during the entire time period; Indicates the highest electricity price during the entire time period; Indicates the Energy storage equipment in Discharge power during the time period; Indicates the Energy storage equipment in Charging power during the time period.

[0008] As a further improvement of the present technical solution, the physical constraint model constructed for the electric vehicle in S1 includes the following steps: S120.1: Charge and discharge power state constraints, establish constraints on electric vehicle charge and discharge power and state variables: ; ; in, Indicates the discharge power of the equipment; Indicates the minimum allowable power during discharge; Indicates the maximum allowable power during discharge; A binary variable representing the discharge state; Indicates the charging power of the device; Indicates the minimum allowable power during charging; Indicates the maximum allowable power during charging; A binary variable representing the state of charge; + ; S120.2: Charge capacity balance modeling, constructing the electric vehicle SOC dynamic equation: ; in, express State of charge at all times; represents the net power; Represents a time interval (e.g., hours, matching energy units); Indicates the rated capacity of the equipment; Used to limit In a safe range (such as battery 0.2-0.8, avoid overcharging / discharging).

[0009] As a further improvement of this technical solution, in the electric vehicle adjustable cost modeling in S1, the time-of-use cost calculation is based on the time-of-use electricity price, and the economic cost of charging and discharging is quantified by the following formula: ; in, represents the total time-sharing cost of electric vehicles; Indicates the total number of time periods within the time period; Indicates the current period index; Indicates the number of electric vehicles involved in the calculation (positive integer); Indicates the electric vehicle number index; express Real-time electricity price during time period; Indicates the lowest electricity price during the period; Indicates the highest electricity price during the period; Indicates the car Discharge power during the time period; Indicates the car Charging power during the time period.

[0010] As a further improvement of this technical solution, the deep reinforcement learning modeling in S2 includes the following steps: S210.1. Define observable state variables as the current period and real-time dispatching instructions for active power of power grid ,Right now: , where: Indicates that the system is The state vector at the moment; S210.1. Define the action variable as the real-time dispatch power of the adjustable unit within the residential user. ,Right now Indicates that the system is The action variable at the moment; among them, Satisfy grid dispatch instruction constraints: = .

[0011] As a further improvement of this technical solution, the reward and optimization of deep reinforcement learning in S2 includes the following steps: S220.1. Define real-time rewards The negative value of the total dispatch cost of energy storage and electric vehicles, the cumulative reward is the sum of real-time rewards within the scheduling period: ;in, It is a real-time reward, which is the negative value of the energy storage and electric vehicle dispatching cost. The lower the cost, the higher the reward, which drives optimization. Scheduling costs for electrochemical energy storage devices; Scheduling costs for electric vehicles; ;in, Cumulative rewards, which is the sum of all real-time rewards during the entire period, reflect the overall optimization goal (minimizing total cost); is the total number of time periods in the scheduling cycle; is the time period index; For real-time rewards; S220.2. Solve the optimal scheduling power by maximizing the cumulative reward: ; in, Indicates the real-time dispatching power of the adjustable units on the user side; represents the optimal dispatch power; Express Ask for The maximum parameter value is the optimal scheduling power.

[0012] As a further improvement of this technical solution, the preprocessing before deep reinforcement learning modeling in S2 includes the following steps: S230.1. Use the improved isolation forest algorithm to process raw power data (such as electricity price, power, SOC, etc.) to identify and remove outliers (noise); S231.1. Standardize the cleaned data. The calculation formula is: ; in, is the data mean (reflecting the trend of the data, such as the average level of historical electricity prices), is the standard deviation (reflecting the degree of data dispersion, such as the amplitude of power fluctuation).

[0013] As a further improvement of the present technical solution, the behavior profile generation in S300 includes the following steps: S310, combining the charge capacity constraint of electrochemical energy storage devices and generating the charging and discharging of electric vehicles Based on the charge and discharge power constraints of electrochemical energy storage devices (such as S110.1 、 、 and ) and the charge capacity range of electric vehicles (such as in S120.2 、 ), determine the user side in each time period Instructions to the power grid The limit of regulatory capacity ;Right now: ; ; in, Indicates time The minimum adjustable power; Indicates time Maximum adjustable power; 、 They represent the lower limit of discharge power and the upper limit of charge power of electrochemical energy storage devices respectively; 、 They represent the lower limit of discharge power and the upper limit of charging power of electric vehicles respectively; S320. Combine the dispatch cost calculations of electrochemical energy storage devices and electric vehicles to generate a time-of-use cost-capacity curve: Ability Dimension: Indicates the adjustable power range on the user side; Cost dimension: Indicates adjustment The real-time scheduling cost.

[0014] As a further improvement of the present technical solution, the behavior profile generation in S300 further includes the following steps: S330, real-time collection of the charge capacity of the electrochemical energy storage device and the remaining power of the electric vehicle, when the device status changes (such as When the fluctuation is ≥10% or new equipment is connected, the adjustable power range is recalculated. , ensuring that the portrait is consistent with the actual adjustment capabilities of the device; S340. Divide the generated cost-capability curve into three response intervals: Green zone: low-cost, high-feasibility adjustment range (such as energy storage charging during low electricity price periods); Yellow zone: medium cost, feasible adjustment range (such as electric vehicle charging during flat electricity prices); Red zone: high cost, limited regulation range (such as energy storage discharge during peak electricity price periods).

[0015] Different intervals are marked with different colors to provide an intuitive reference for scheduling decisions.

[0016] S350: Inputting the cost-capability curve generated in S320 as prior knowledge into the deep reinforcement learning model in S200, including the following steps: S350.1. Add profiling features to the state space (e.g., cost sensitivity level, adjustable priority); S350.2. Introduce a profile-guided term into the reward function (e.g., give extra rewards when selecting green interval adjustments) to optimize the economy and feasibility of the scheduling strategy.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention establishes a dynamic constraint model for electrochemical energy storage charge capacity and charge / discharge power, as well as an electric vehicle power-time window constraint, to adjust the adjustable power range in real time, ensuring that dispatch instructions are consistent with the actual operating status of the equipment. This improves the accuracy and timeliness of adjustable capacity assessment and resolves the problem of static capacity estimation not matching actual conditions. 2. Based on the coupling relationship between time-of-use electricity prices and equipment operating power, this invention constructs a three-dimensional cost model of "power-price-time". It quantifies the benefits of off-peak charging opportunities and the actual losses of peak discharge in real time, and achieves dynamic and accurate calculation of adjustable costs. This solves the problem of traditional static cost estimation being out of touch with actual scenarios, and provides data support for economic optimization scheduling. 3. This invention uses deep reinforcement learning to integrate dynamic capabilities and real-time costs to form a dual-objective scheduling strategy, giving priority to low-cost, highly feasible equipment. This minimizes the total user-side scheduling cost while meeting grid control requirements, improving equipment utilization efficiency and user benefits, and solving the problem of extensive multi-device collaboration. 4. This invention generates a two-dimensional "capacity-cost" behavior profile, transforming it into an intuitive three-dimensional decision-making interface. It uses color gradients or contour lines to identify cost levels, providing an intuitive economic reference for scheduling decisions. This improves the efficiency of control strategy execution and user experience, and solves the problem of unintuitive operation caused by insufficient dynamic behavior analysis. 5. The present invention designs a dynamic portrait update mechanism to monitor changes in device status in real time, automatically adjust the adjustable power range and cost model, ensure real-time synchronization of the portrait with the actual status of the device, improve the control system's ability to respond to changes in device status, solve the problem of lag in traditional strategies, and enhance the system's robustness and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the overall steps of the present invention; Figure 2 This is a diagram of the preprocessing steps before deep reinforcement learning modeling in the present invention. DETAILED DESCRIPTION

[0019] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention. Example 1

[0020] like Figure 1-Figure 2 As shown, this embodiment provides a behavior profile modeling method for typical adjustable devices on the residential user side, including: S100, Mechanism Model Analysis: Based on the mechanism model analysis method, a physical constraint model of adjustable capacity and adjustable cost is constructed for electrochemical energy storage devices and electric vehicles; In this step, constructing a physical constraint model for the electrochemical energy storage device in S1 includes the following steps: S110.1: Modeling the charge and discharge power boundaries, establishing mutually exclusive constraints between the charge and discharge power and state variables: ; ; in, Indicates the discharge power of electrochemical energy storage equipment; Indicates the minimum value of discharge power; Indicates the maximum value of discharge power; A binary variable representing the discharge state (value 0 or 1), used to indicate whether the device is in the discharge state: Indicates the charging power of the electrochemical energy storage device; Indicates the minimum charging power of the electrochemical energy storage device; Indicates the maximum value of electric power; A binary variable representing the state of charge; + ; S110.2: Dynamic modeling of charge capacity, constructing the dynamic equation of charge capacity and charge and discharge power: ; in, express The state of charge of the energy storage device at the moment (range 0 1), which indicates the ratio of the remaining energy storage capacity to the rated capacity; is the time interval; represents the net power, and , when discharging >0 (SOC decreases), when charging <0 (SOC increased); Indicates the rated capacity of the energy storage device; express State of charge at the moment; express Net power at the moment; Indicates the minimum value of the state of charge; Maximum state of charge. Used to ensure that the SOC is within the physically allowed range (e.g. battery SOC is usually 0.2 SOC 0.8, avoid overcharge / overdischarge), and It indicates the minimum and maximum value of the battery state of charge.

[0021] In this step, in the adjustable cost modeling of the electrochemical energy storage device in S1, the opportunity cost calculation is based on the time-of-use electricity price, and the calculation formula is as follows: ; in, represents the total opportunity cost of electrochemical energy storage devices; Indicates the total number of time periods within the time period; Indicates the current period index; Indicates the number of energy storage units; Indicates the number index of the energy storage unit; express Real-time electricity prices for the time period; Indicates the lowest electricity price during the entire time period; Indicates the highest electricity price during the entire time period; Indicates the Energy storage equipment in Discharge power during the time period; Indicates the Energy storage equipment in Charging power during the time period.

[0022] In this step, the physical constraint model constructed for the electric vehicle in S1 includes the following steps: S120.1: Charge and discharge power state constraints, establish constraints on electric vehicle charge and discharge power and state variables: ; ; in, Indicates the discharge power of the equipment; Indicates the minimum allowable power during discharge; Indicates the maximum allowable power during discharge; A binary variable representing the discharge state; Indicates the charging power of the device; Indicates the minimum allowable power during charging; Indicates the maximum allowable power during charging; A binary variable representing the state of charge; + ; S120.2: Charge capacity balance modeling, constructing the electric vehicle SOC dynamic equation: ; in, express State of charge at all times; represents the net power; Represents a time interval (e.g., hours, matching energy units); Indicates the rated capacity of the equipment; Used to limit In a safe range (such as battery 0.2-0.8, avoid overcharging / discharging).

[0023] In this step, in the electric vehicle adjustable cost modeling in S1, the time-of-use cost calculation is based on the time-of-use electricity price, and the economic cost of charging and discharging is quantified by the following formula: ; in, represents the total time-sharing cost of electric vehicles; Indicates the total number of time periods within the time period; Indicates the current period index; Indicates the number of electric vehicles involved in the calculation (positive integer); Indicates the electric vehicle number index; express Real-time electricity price during time period; Indicates the lowest electricity price during the period; Indicates the highest electricity price during the period; Indicates the car Discharge power during the time period; Indicates the car Charging power during the time period.

[0024] As a further illustration of this step, the physical constraint model of the electrochemical energy storage device in this embodiment ensures operational safety and accuracy through multi-dimensional rules, including: Power-SOC correlation: When When <0.3, the discharge power is limited to 80% of the rated power (the coefficient can be customized) to avoid deep discharge; State of Charge Evolution: A dynamic equation including the self-discharge rate (0.001 / h) accurately describes SOC changes and reflects the natural loss of the battery. Real-time cost parameters: electricity price Real-time acquisition, daily cycle electricity price extreme value ( 、 ) covers the entire day and supports real-time cost optimization.

[0025] S200, Deep Reinforcement Learning Modeling: Based on the external characteristic modeling method of deep reinforcement learning, this method utilizes external observable states to establish a dynamic mapping model between real-time dispatch instructions and dispatch costs, and construct an adjustable capacity model and an adjustable cost model suitable for electrochemical energy storage devices and electric vehicles. In this step, the deep reinforcement learning modeling in S2 includes the following steps: S210.1. Define observable state variables as the current period and real-time dispatching instructions for active power of power grid ,Right now: , where: Indicates that the system is The state vector at the moment; S210.1. Define the action variable as the real-time dispatch power of the adjustable unit within the residential user. ,Right now Indicates that the system is The action variable at the moment; among them, Satisfy grid dispatch instruction constraints: = This setting forces the power of user-side equipment to be consistent with the grid dispatch instructions, ensuring the active power balance of the grid (such as frequency regulation and load control). It is the core constraint of grid-user collaborative control and supports applications such as virtual power plants and distributed energy aggregation.

[0026] In this step, the reward and optimization of deep reinforcement learning in S2 includes the following steps: S220.1. Define real-time rewards The negative value of the total dispatch cost of energy storage and electric vehicles, the cumulative reward is the sum of real-time rewards within the scheduling period: ;in, It is a real-time reward, which is the negative value of the energy storage and electric vehicle dispatching cost. The lower the cost, the higher the reward, which drives optimization. Scheduling costs for electrochemical energy storage devices; Scheduling costs for electric vehicles; ;in, Cumulative rewards, which is the sum of all real-time rewards during the entire period, reflect the overall optimization goal (minimizing total cost); is the total number of time periods in the scheduling cycle; is the time period index; For real-time rewards; S220.2. Solve the optimal scheduling power by maximizing the cumulative reward: ; in, Indicates the real-time dispatching power of the adjustable units on the user side; represents the optimal dispatch power; Express Ask for The maximum parameter value is the optimal scheduling power.

[0027] In this step, the preprocessing before deep reinforcement learning modeling in S2 includes the following steps: S230.1. Use the improved isolation forest algorithm to process raw power data (such as electricity price, power, SOC, etc.) to identify and remove outliers (noise); S231.1. Standardize the cleaned data. The calculation formula is: ; in, is the data mean (reflecting the trend of the data, such as the average level of historical electricity prices), is the standard deviation (reflecting the degree of data dispersion, such as the amplitude of power fluctuation).

[0028] It is understandable that after standardization, the data satisfies a mean of 0 and a standard deviation of 1, eliminating dimensional differences (such as the difference between the electricity price unit yuan / kWh and the power unit kW), making the input features of deep reinforcement learning models (such as the DQN algorithm used for energy storage scheduling) consistent, accelerating model training convergence, and improving the generalization ability for complex scenarios in the power system (such as peak-valley switching and power command mutations).

[0029] As a further explanation of this step, in order to break the limitation of the original model of "relying only on grid instructions", a four-dimensional state vector can be added in S210.1 , achieving real-time perception of time, grid demand, energy storage and electric vehicle charge status: Time and grid dimensions: through Adapt the time-sharing strategy and use Responding to the dynamic demands of the power grid, the energy storage dispatch is deeply coupled with the timing characteristics of the power grid; State of Charge dimension: and Respectively linking electrochemical energy storage to the physical constraints of electric vehicles (e.g., SOC safety intervals) to ensure that dispatch instructions do not exceed the energy storage capacity of the device (e.g., prohibiting forced discharge of low SOC energy storage to avoid device damage); This extension solves the problem of the original model being out of touch with the actual state of the equipment, enabling scheduling decisions to meet grid requirements while strictly adhering to the physical limitations of energy storage units. It also provides basic state perception capabilities for the coordinated optimization of multiple types of energy storage (such as vehicle-grid interaction and hybrid energy storage control).

[0030] As a further explanation of this step, in S210.1 of this embodiment, it can be Supplementary physical constraints: ; in, ; By integrating the power boundaries of energy storage (S110.1) and electric vehicles (S120.1), it is possible to: Discharge side: The lower limit is the minimum discharge power of the two (for example, if electric vehicles need to maintain their endurance, their lower limit is more stringent) to avoid excessive discharge of the equipment; Charging side: The upper limit is the maximum charging power of the two (the energy storage charging hardware limit may be lower) to avoid overcharging of the device.

[0031] This constraint breaks through the limitations of the original model, which "only matched grid commands," by incorporating the physical limitations of user-side devices into scheduling. This ensures safe power operation (no overcharging or over-discharging) and supports multi-device collaboration (such as vehicle-grid interaction and hybrid energy storage control). For example, in V2G scenarios, vehicle range is prioritized, while charging is limited by energy storage hardware, improving system reliability and energy efficiency.

[0032] Furthermore, in S220.1, the reward function is expanded to ,pass =0.1 for energy storage ( ) and electric vehicles ( )of Deviations from 0.5 are penalized to achieve: Economic optimization: Minimizing the operating costs of energy storage and electric vehicles (e.g., arbitrage between peak and valley electricity prices, and reducing redundant charging and discharging); Device health: Suppresses deep charge and discharge (such as extreme conditions of SOC < 0.2 or > 0.8) to extend battery life (for example, when a lithium-ion battery is cycled in the SOC range of 0.2–0.8, the capacity decay rate is reduced by approximately 30%. Specific data needs to be calibrated based on the battery type). This design solves the defect of the original reward of "focusing on cost and ignoring equipment", so that the strategy can automatically protect the health of energy storage and electric vehicle batteries while reducing electricity costs (such as giving priority to devices with SOC close to 0.5 to avoid excessive consumption of extreme SOC devices). It is suitable for long-term optimization control in scenarios such as vehicle-grid interaction and hybrid energy storage. In addition, in this embodiment, The value of needs to be dynamically adjusted according to the equipment life model and cost weight, and can be calibrated through offline simulation (such as Monte Carlo simulation based on the battery aging model).

[0033] It should be added that the improved isolation forest algorithm in this embodiment can use time window sliding detection, with a time period of 15 minutes and a sliding window of 45 minutes (3 time periods). Clustering anomaly identification is performed on the data within the window, rather than single-point detection, to effectively filter out noise. For example, if an isolated anomaly occurs in the power of a certain period due to sensor fluctuations, the window detection will compare the previous and next time periods. If there is only a single-point anomaly, it will be determined as noise, avoiding misjudgment, improving the robustness of data cleaning, and ensuring the continuity and reliability of input data. At the same time, this embodiment can normalize the electricity price, power, and SOC according to the scheduling cycle, eliminate dimensional differences (such as the unification of the scale of the electricity price yuan / kWh and the SOC percentage), provide standardized input for model training, and accelerate convergence (such as avoiding numerical dominance problems in multi-parameter optimization).

[0034] S300 , generating a behavior profile: Based on the physical constraint model and dynamic mapping model of S100 - S200 , generating a behavior profile including an adjustable time period, an adjustable capacity threshold, and an adjustable cost.

[0035] In this step, the behavior profile generation in S300 includes the following steps: S310, combining the charge capacity constraint of electrochemical energy storage devices and generating the charging and discharging of electric vehicles Based on the charge and discharge power constraints of electrochemical energy storage devices (such as S110.1 、 、 and ) and the charge capacity range of electric vehicles (such as in S120.2 、 ), determine the user side in each time period Instructions to the power grid The limit of regulatory capacity ;Right now: ; ; in, Indicates time The minimum adjustable power; Indicates time Maximum adjustable power; 、 They represent the lower limit of discharge power and the upper limit of charge power of electrochemical energy storage devices respectively; 、 They represent the lower limit of discharge power and the upper limit of charging power of electric vehicles respectively; It is understandable that ,This design can avoid the logical contradiction of simultaneous charging and discharging, and ensure that the capacity boundary calculation is consistent with physical reality.

[0036] S320. Combine the dispatch cost calculations of electrochemical energy storage devices and electric vehicles to generate a time-of-use cost-capacity curve: Ability Dimension: Indicates the adjustable power range on the user side; Cost dimension: Indicates adjustment The real-time scheduling cost.

[0037] It is worth mentioning that ,in, is the total number of energy storage devices; is the total number of electric vehicles; They correspond to the time-sharing costs of a single energy storage unit and electric vehicle respectively (such as the opportunity cost formula in S1); they solve the problem that the cost dimension does not clearly define the logic of multi-device aggregation.

[0038] In this step, the behavior profile generation in S300 further includes the following steps: S330, real-time collection of the charge capacity of the electrochemical energy storage device and the remaining power of the electric vehicle, when the device status changes (such as When the fluctuation is ≥10% or new equipment is connected, the adjustable power range is recalculated. , ensuring that the portrait is consistent with the actual adjustment capabilities of the device; S340. Divide the generated cost-capability curve into three response intervals: Green zone: low-cost, high-feasibility adjustment range (such as energy storage charging during low electricity price periods); Yellow zone: medium cost, feasible adjustment range (such as electric vehicle charging during flat electricity prices); Red zone: high cost, limited regulation range (such as energy storage discharge during peak electricity price periods).

[0039] Different intervals are marked with different colors to provide an intuitive reference for scheduling decisions.

[0040] S350: Inputting the cost-capability curve generated in S320 as prior knowledge into the deep reinforcement learning model in S200, including the following steps: S350.1. Add profiling features to the state space (e.g., cost sensitivity level, adjustable priority); S350.2. Introduce a profile-guided term into the reward function (e.g., give extra rewards when selecting green interval adjustments) to optimize the economy and feasibility of the scheduling strategy.

[0041] As a further illustration of this step, in S33, a configurable SOC fluctuation threshold may be introduced. (Default 10%), when the SOC change of energy storage or electric vehicle ≥ this threshold, the model parameter update is automatically triggered; This design has the following advantages: Flexible adaptation to heterogeneous devices: Targeting different types of devices, such as lithium batteries (sensitive, low threshold), lead-acid batteries (tolerant, high threshold), as well as scenarios such as vehicle-grid interaction (EV low threshold) and energy storage power stations (high threshold), the system dynamically adjusts thresholds to ensure a balance between state response and computing efficiency. For example, EV =0.8 can update the strategy after each significant discharge to avoid battery life exhaustion; =0.15Reduce invalid updates caused by daily fluctuations and improve operational efficiency; Dynamic optimization strategy: When SOC fluctuates significantly (such as deep charging and discharging, rapid charging and discharging), the model quickly updates parameters, such as adjusting discharge limits and optimizing charging and discharging priorities. This solves the "update lag" problem of the original model, enhances the perception and response capabilities of the actual status of the equipment, effectively reduces the damage to equipment life caused by extreme SOC conditions, and improves system reliability and energy efficiency in multi-device collaborative scenarios (such as hybrid energy storage and vehicle-grid interaction).

[0042] As a further illustration of this step, Furthermore, in S340, the cost function and historical cost statistical parameters is the historical average cost, is the cost standard deviation), and the operating status is divided into three levels: Green zone: when , the system is in an efficient and low-cost state, and can continue to optimize economic strategies (such as maximizing peak-valley arbitrage) while maintaining equipment health (such as SOC balance).

[0043] Yellow zone: When , the cost is within the normal fluctuation range, and it is necessary to monitor and fine-tune the strategy (such as adjusting the device power allocation weight to prevent the cost from drifting into the red range).

[0044] Red zone: when , the system is in a high-cost and low-efficiency state, and emergency optimization needs to be triggered immediately (such as switching equipment operation modes, resetting model parameters, and quickly reducing costs).

[0045] It should be noted that in the S350, by dividing cost sensitivity into three levels: green (level 3), yellow (level 2), and red (level 1), and integrating them into the state vector s'(t), and designing a reward enhancement item (an additional +5 reward in the green range), a closed-loop mechanism of "state perception-reward feedback" is constructed, namely: Refined state perception: The model can accurately identify current cost priorities (such as high-value operations in the green zone), optimize equipment selection (prioritizing energy storage / electric vehicles in the green zone), and improve resource utilization efficiency. Strengthening the reward mechanism: giving higher rewards to low-cost behaviors (green zone), guiding strategy learning (more frequent selection of green zone actions during training), and enhancing the long-term economic efficiency of the system; For example, in the case of electric vehicle scheduling, the model prioritizes idle vehicles with high SOC (green zone, level 3) on weekends for V2G discharge (to earn high returns) due to increased incentives, while vehicles with low SOC (yellow zone, level 2) on weekdays before commuting are encouraged to reduce discharge (to avoid impacting range and maintain normal incentives). This mechanism, through dynamic quantification and incentive guidance, achieves a cost-effective balance between multiple devices (energy storage, electric vehicles), improves optimization accuracy and response speed in scenarios such as vehicle-grid interaction and hybrid energy storage, and provides strong support for efficient system operation.

[0046] Those skilled in the art will appreciate that the process of implementing all or part of the steps of the above embodiments may be accomplished by hardware, or by instructing related hardware through a program, which may be stored in a computer-readable storage medium.

[0047] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A behavioral profile modeling method for typical adjustable devices on the residential user side, characterized in that: The steps include: S100, Mechanism Model Analysis: Based on the mechanism model analysis method, a physical constraint model of adjustable capacity and adjustable cost is constructed for electrochemical energy storage devices and electric vehicles; S200, Deep Reinforcement Learning Modeling: Based on the external characteristic modeling method of deep reinforcement learning, this method utilizes external observable states to establish a dynamic mapping model between real-time dispatch instructions and dispatch costs, and construct an adjustable capacity model and an adjustable cost model suitable for electrochemical energy storage devices and electric vehicles. S300 , generating a behavior profile: Based on the physical constraint model and dynamic mapping model of S100 - S200 , generating a behavior profile including an adjustable time period, an adjustable capacity threshold, and an adjustable cost.

2. The behavioral portrait modeling method of typical adjustable equipment on the residential user side according to claim 1 is characterized in that: Constructing a physical constraint model for the electrochemical energy storage device in S1 includes the following steps: S110.1: Modeling the charge and discharge power boundaries, establishing mutually exclusive constraints between the charge and discharge power and state variables: ； ； in, Indicates the discharge power of electrochemical energy storage equipment; Indicates the minimum value of discharge power; Indicates the maximum value of discharge power; A binary variable representing the discharge state, used to indicate whether the device is in the discharge state: Indicates the charging power of the electrochemical energy storage device; Indicates the minimum charging power of the electrochemical energy storage device; Indicates the maximum value of electric power; A binary variable representing the state of charge; S110.2: Dynamic modeling of charge capacity, constructing the dynamic equation of charge capacity and charge and discharge power: ； in, express The state of charge of the energy storage device at all times; is the time interval; represents the net power; Indicates the rated capacity of the energy storage device; express State of charge at the moment; express Net power at the moment; Indicates the minimum value of the state of charge; Maximum state of charge.

3. The behavioral profile modeling method for typical adjustable devices on the residential user side according to claim 2 is characterized by: In the adjustable cost modeling of the electrochemical energy storage device in S1, the opportunity cost calculation is based on the time-of-use electricity price, and the calculation formula is as follows: ； in, represents the total opportunity cost of electrochemical energy storage devices; Indicates the total number of time periods within the time period; Indicates the current period index; Indicates the number of energy storage units; Indicates the number index of the energy storage unit; express Real-time electricity prices for the time period; Indicates the lowest electricity price during the entire time period; Indicates the highest electricity price during the entire time period; Indicates the Energy storage equipment in Discharge power during the time period; Indicates the Energy storage equipment in Charging power during the time period.

4. The behavioral profile modeling method for typical adjustable devices on the residential user side according to claim 3 is characterized in that: The physical constraint model constructed for the electric vehicle in S1 includes the following steps: S120.1: Charge and discharge power state constraints, establish constraints on electric vehicle charge and discharge power and state variables: ； ； in, Indicates the discharge power of the equipment; Indicates the minimum allowable power during discharge; Indicates the maximum allowable power during discharge; A binary variable representing the discharge state; Indicates the charging power of the device; Indicates the minimum allowable power during charging; Indicates the maximum allowable power during charging; A binary variable representing the state of charge; S120.2: Charge capacity balance modeling, constructing the electric vehicle SOC dynamic equation: ； in, express State of charge at all times; represents the net power; Indicates a time interval; Indicates the rated capacity of the equipment; Used to limit In the safe zone, and It indicates the minimum and maximum value of the battery state of charge.

5. The behavioral profile modeling method for typical adjustable devices on the residential user side according to claim 4 is characterized by: In the electric vehicle adjustable cost modeling in S1, the time-of-use cost calculation is based on the time-of-use electricity price, and the economic cost of charging and discharging is quantified by the following formula: ； in, represents the total time-sharing cost of electric vehicles; Indicates the total number of time periods within the time period; Indicates the current period index; Indicates the number of electric vehicles involved in the calculation; Indicates the electric vehicle number index; express Real-time electricity price during time period; Indicates the lowest electricity price during the period; Indicates the highest electricity price during the period; Indicates the car Discharge power during the time period; Indicates the car Charging power during the time period.

6. The behavioral profile modeling method for typical adjustable devices on the residential user side according to claim 1 is characterized in that: The deep reinforcement learning modeling in S2 includes the following steps: S210.

1. Define observable state variables as the current period and real-time dispatching instructions for active power of power grid ; S210.

1. Define the action variable as the real-time dispatch power of the adjustable unit within the residential user. ;in, Satisfy grid dispatch instruction constraints: = .

7. The behavioral profile modeling method for typical adjustable devices on the residential user side according to claim 6 is characterized by: The reward and optimization of deep reinforcement learning in S2 includes the following steps: S220.

1. Define real-time rewards The negative value of the total dispatch cost of energy storage and electric vehicles, the cumulative reward is the sum of real-time rewards within the scheduling period: ;in, For real-time rewards; Scheduling costs for electrochemical energy storage devices; Scheduling costs for electric vehicles; ;in, Cumulative rewards, which is the sum of all real-time rewards during the entire period, reflect the overall optimization goal; is the total number of time periods in the scheduling cycle; is the time period index; For real-time rewards; S220.

2. Solve the optimal scheduling power by maximizing the cumulative reward: ; in, Indicates the real-time dispatching power of the adjustable units on the user side; represents the optimal dispatch power; Express Ask for The maximum parameter value is the optimal scheduling power.

8. The behavioral profile modeling method for typical adjustable devices on the residential user side according to claim 7 is characterized by: The preprocessing before deep reinforcement learning modeling in S2 includes the following steps: S230.

1. Use the improved isolation forest algorithm to process the raw power data and identify and remove outliers; S231.

1. Perform standardization on the cleaned data.

9. The behavioral profile modeling method for typical adjustable devices on the residential user side according to claim 1 is characterized in that: The behavior profile generation in S300 includes the following steps: S310, combining the charge capacity constraint of the electrochemical energy storage device and generating the charge and discharge of the electric vehicle: Based on the charge and discharge power constraints of electrochemical energy storage devices and the charge capacity range of electric vehicles, the user side is determined in each period. Instructions to the power grid The limit of regulatory capacity ; S320. Combine the dispatch cost calculations of electrochemical energy storage devices and electric vehicles to generate a time-of-use cost-capacity curve: Ability Dimension: Indicates the adjustable power range on the user side; Cost dimension: Indicates adjustment The real-time scheduling cost.

10. The behavioral profile modeling method for typical adjustable devices on the residential user side according to claim 9 is characterized in that: The behavior profile generation in S300 further includes the following steps: S330, collecting the charge capacity of the electrochemical energy storage device and the remaining power of the electric vehicle in real time, and recalculating the adjustable power range when the device state changes; S340. Divide the generated cost-capability curve into three response intervals: S350: Inputting the cost-capability curve generated in S320 as prior knowledge into the deep reinforcement learning model in S200, including the following steps: S350.

1. Adding portrait features to the state space; S350.

2. Introduce a profile-guided term into the reward function to optimize the economy and feasibility of the scheduling strategy.

Images