Multi-source distributed resource collaborative real-time control method, system, equipment and medium

Abstract

The invention relates to the technical field of power system real-time control, and discloses a multi-source distributed resource collaborative real-time control method, system and device and a medium. The method comprises the following steps: constructing a multi-target scheduling model fusing user comfort and power grid operation indexes; constructing an adaptive hybrid intelligent algorithm based on a non-dominated sorting genetic algorithm, a deep reinforcement learning algorithm and a particle swarm optimization algorithm; based on the real-time data, performing adaptive solution on the multi-target scheduling model according to an adaptive hybrid intelligent algorithm to obtain a minute-level optimization decision scheme; and issuing a control command according to the minute-level optimization decision scheme so as to realize deep cooperative control of the multi-source distributed resources. According to the method, the user comfort is used as a constraint to be fused into a power grid optimization target, and deep cooperative control of multi-source distributed resources, namely photovoltaic-energy storage-load, is realized through autonomous learning and optimization of an adaptive hybrid intelligent algorithm, so that the power grid peak clipping and voltage regulation capabilities are effectively enhanced while the user satisfaction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of real-time control of power systems, and in particular, to a multi-source distributed resource collaborative real-time control method, system, device, and medium. Background Art

[0002] In modern distribution networks, the high proportion of photovoltaic access and the use of electric air-conditioning loads have led to an increase in the peak-valley difference of electricity consumption and frequent voltage fluctuations. Traditional regulation technologies mainly include electricity consumption guidance based on time-of-use electricity prices, energy storage charge and discharge control with pre-set strategies, and simple direct control of air-conditioning loads. These methods have reduced the peak load to a certain extent, but have their own limitations:

[0003] Time-of-use electricity price regulation: By raising the electricity price during peak hours and lowering the electricity price during off-peak hours to guide users to shift electricity consumption peaks. This strategy is essentially a fixed rule, and users passively respond to electricity price signals. For example, the common time-of-use electricity price (TOU) has a lower rate at night and a higher rate during the evening peak. However, the real-time electricity price changes frequently, and ordinary users are difficult to adjust their electricity consumption behavior in a timely manner. Relying solely on the price mechanism of time-of-use electricity price regulation cannot accurately control loads such as air conditioners and cannot take into account the distribution network voltage index.

[0004] Energy storage fixed strategy control: Many energy storage systems adopt pre-set threshold or clock control strategies, such as charging when the photovoltaic output is excessive and discharging during the evening peak to cut the peak. Such fixed strategies are not optimized in combination with real-time working conditions and are often set according to the average situation, making it difficult to adapt to actual load and photovoltaic fluctuations. At the same time, most traditional strategies take economic benefits or peak shaving as a single goal. When pursuing peak shaving and valley filling, the problems of user comfort and local voltage over-limit may be ignored.

[0005] Direct control of air-conditioning loads: As the main adjustable load, the common regulation method for air-conditioning loads is to use simple on-off control or unified temperature increase during the peak period of the power grid. For example, in emergency demand response, some air conditioners are directly turned off for a period of time, which reduces the load but significantly reduces user comfort and easily causes user dissatisfaction. Or, requiring all users to raise the air-conditioning temperature above 28°C during the peak period, this one-size-fits-all control lacks pertinence and ignores the comfort preferences of different users. The result is often a decline in the user experience, which is not conducive to the continuous implementation of demand response in the long run.

[0006] In summary, the existing regulation technologies have the problem of acting independently: either focusing on grid-side indicators (such as peak shaving and voltage reduction), or focusing on user-side demands (comfort and power saving), lacking means to consider both aspects in an overall way. For example, the demand response of air-conditioning loads can improve the peak shaving ability of the power grid. However, if the user comfort constraint is not introduced, it may cause user complaints and it is difficult to prevent voltage over-limit in a timely manner. With the intelligent development of the power distribution and utilization system, there is an urgent need for a new method to achieve the collaborative optimization of user-side comfort and grid operation indicators, while ensuring the user experience and improving the power grid peak shaving and voltage regulation effects. Summary of the Invention

[0007] The main object of the present invention is to provide a multi-source distributed resource collaborative real-time control method, system, device and medium, aiming to solve at least one of the above technical problems.

[0008] To achieve the above object, the present invention provides a multi-source distributed resource collaborative real-time control method, including:

[0009] Constructing a multi-objective scheduling model that integrates user comfort and grid operation indicators;

[0010] Constructing an adaptive hybrid intelligent algorithm based on the non-dominated sorting genetic algorithm, deep reinforcement learning algorithm and particle swarm optimization algorithm;

[0011] Obtaining real-time data, and adaptively solving the multi-objective scheduling model based on the real-time data according to the adaptive hybrid intelligent algorithm to obtain a minute-level optimization decision plan;

[0012] Issuing control commands according to the minute-level optimization decision plan to achieve the deep collaborative control of multi-source distributed resources.

[0013] In some embodiments, the constructing of the multi-objective scheduling model that integrates user comfort and grid operation indicators includes:

[0014] Constructing an objective function according to the load peak-valley difference index, voltage deviation index, temperature deviation index and energy storage utilization rate index;

[0015] Setting multiple constraint conditions according to power balance constraints, photovoltaic output limits, air-conditioning comfort limits, energy storage SOC limits, voltage range limits and control interval limits;

[0016] Constructing a multi-objective scheduling model according to the objective function and multiple constraint conditions.

[0017] In some embodiments, the optimization objectives of the multi-objective scheduling model are to minimize the load peak-valley difference index, minimize the voltage deviation index, minimize the temperature deviation index and maximize the energy storage utilization rate.

[0018] In some embodiments, the obtaining of real-time data includes:

[0019] Collecting the observed data of the solar irradiance per minute, and obtaining the regional illumination data according to the observed data;

[0020] Obtaining the community load data;

[0021] Taking the regional illumination data and the community load data as the input real-time data.

[0022] In some embodiments, the adaptively solving the multi-objective scheduling model based on the real-time data according to the adaptive hybrid intelligent algorithm to obtain a minute-level optimization decision plan includes:

[0023] Processing the regional illumination data of the real-time data to obtain the minute-level output prediction of the photovoltaic system;

[0024] Taking the minute-level output prediction as the photovoltaic output limit constraint condition of the multi-objective scheduling model;

[0025] Performing global optimization on the multi-objective scheduling model according to the non-dominated sorting genetic algorithm of the adaptive hybrid intelligent algorithm to obtain a set of candidate solution sets;

[0026] Receiving the real-time data according to the deep reinforcement learning algorithm of the adaptive hybrid intelligent algorithm and outputting continuous control variables;

[0027] Iteratively updating the candidate solution set according to the particle swarm optimization algorithm of the adaptive hybrid intelligent algorithm to obtain the optimal solution;

[0028] Obtaining a minute-level optimization decision plan according to the optimal solution and the continuous control variables.

[0029] In some embodiments, the performing global optimization on the multi-objective scheduling model according to the non-dominated sorting genetic algorithm of the adaptive hybrid intelligent algorithm to obtain a set of candidate solution sets includes:

[0030] Solving the multi-objective scheduling model according to the non-dominated sorting genetic algorithm of the adaptive hybrid intelligent algorithm, and searching for the Pareto optimal solution set by means of population evolution;

[0031] Normalizing and weighting each optimization objective of the multi-objective scheduling model to solve a single objective to obtain an equilibrium solution;

[0032] Taking the Pareto optimal solution set or the equilibrium solution as a set of candidate solution sets.

[0033] In some embodiments, the method further includes:

[0034] During adaptive solution, if the time scale is hourly or the environment changes, the non-dominated sorting genetic algorithm of the adaptive hybrid intelligent algorithm is triggered for global optimization to update the overall optimization strategy set;

[0035] If the time scale is minute-level, the deep reinforcement learning algorithm of the adaptive hybrid intelligent algorithm is triggered to output continuous control variables;

[0036] The particle swarm optimization algorithm of the adaptive hybrid intelligent algorithm is embedded in the decision-making per minute, and the continuous control variables output by the deep reinforcement learning algorithm are fine-tuned, or the solution is improved between two calls of the non-dominated sorting genetic algorithm.

[0037] In addition, to achieve the above object, the present invention also proposes a multi-source distributed resource collaborative real-time control system, including:

[0038] A model construction module for constructing a multi-objective scheduling model that integrates user comfort and power grid operation indicators;

[0039] An algorithm combination module for constructing an adaptive hybrid intelligent algorithm based on the non-dominated sorting genetic algorithm, the deep reinforcement learning algorithm, and the particle swarm optimization algorithm;

[0040] A data acquisition and decision-making module for acquiring real-time data and adaptively solving the multi-objective scheduling model based on the real-time data according to the adaptive hybrid intelligent algorithm to obtain a minute-level optimization decision-making scheme;

[0041] A collaborative control module for issuing control commands according to the minute-level optimization decision-making scheme to achieve deep collaborative control of multi-source distributed resources.

[0042] In addition, to achieve the above object, the present invention also proposes an electronic device, the electronic device includes: a memory, a processor, and a multi-source distributed resource collaborative real-time control program stored on the memory and executable on the processor, and the multi-source distributed resource collaborative real-time control program is configured to implement the multi-source distributed resource collaborative real-time control method as described above.

[0043] In addition, to achieve the above object, the present invention also proposes a storage medium, the storage medium stores a multi-source distributed resource collaborative real-time control program, and the multi-source distributed resource collaborative real-time control program is used to cause the processor to execute to implement the multi-source distributed resource collaborative real-time control method as described above.

[0044] The present invention provides a multi-source distributed resource collaborative real-time control method, including: constructing a multi-objective scheduling model that integrates user comfort and power grid operation indicators; constructing an adaptive hybrid intelligent algorithm based on the non-dominated sorting genetic algorithm, the deep reinforcement learning algorithm, and the particle swarm optimization algorithm; obtaining real-time data, and adaptively solving the multi-objective scheduling model based on the real-time data according to the adaptive hybrid intelligent algorithm to obtain a minute-level optimization decision plan; issuing control commands according to the minute-level optimization decision plan to achieve deep collaborative control of multi-source distributed resources. In the present invention, user comfort is incorporated as a constraint into the power grid optimization objective, and through the autonomous learning and optimization of the adaptive hybrid intelligent algorithm, deep collaborative control of multi-source distributed resources, namely photovoltaic-storage-load, is achieved, thereby effectively enhancing the power grid's peak shaving and voltage regulation capabilities while improving user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic structural diagram of an electronic device for the hardware operating environment related to the solution of an embodiment of the present invention;

[0046] Figure 2 It is a schematic flowchart of an embodiment of the multi-source distributed resource collaborative real-time control method of the present invention;

[0047] Figure 3 It is a schematic flowchart of the AI collaborative real-time control solution related to the solution of an embodiment of the present invention;

[0048] Figure 4 It is a schematic diagram of the change of the voltage at the common connection point under two control solutions related to the solution of an embodiment of the present invention;

[0049] Figure 5 It is a schematic diagram of the comparison of key performance indicators of two control solutions related to the solution of an embodiment of the present invention;

[0050] Figure 6 It is a structural block diagram of an embodiment of the multi-source distributed resource collaborative real-time control system of the present invention.

[0051] The realization, functional features, and advantages of the objectives of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0053] It should be noted that all directional indications (such as up, down, left, right, front, back...) in the embodiments of the present invention are only used to explain the relative positional relationship, movement conditions, etc. between components in a specific posture (as shown in the attached drawings). If this specific posture changes, the directional indications will also change accordingly.

[0054] In addition, the descriptions involving "first", "second", etc. in the present invention are only for descriptive purposes, and should not be construed as indicating or implying their relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In addition, the technical solutions between various embodiments can be combined with each other, but it must be based on the fact that those of ordinary skill in the art can implement them. When the combination of technical solutions results in contradictions or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the protection scope required by the present invention. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0055] Refer to Figure 1 , Figure 1 which is a schematic structural diagram of an electronic device for the hardware operating environment involved in the solution of the embodiment of the present invention.

[0056] As Figure 1 shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard). Optionally, the user interface 1003 may further include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless-fidelity (Wi-Fi) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM memory), or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0057] Those skilled in the art can understand that Figure 1 the structure shown in

[0058] As shown Figure 1 in FIG. 1, the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and a multi-source distributed resource collaborative real-time control program.

[0059] In Figure 1 the electronic device shown in FIG. 2, the network interface 1004 is mainly used for data communication with a network server; the user interface 1003 is mainly used for data interaction with a user; the processor 1001 and the memory 1005 in the electronic device of the present invention may be disposed in the electronic device, and the electronic device calls the multi-source distributed resource collaborative real-time control program stored in the memory 1005 through the processor 1001 and executes the multi-source distributed resource collaborative real-time control method provided by the embodiments of the present invention.

[0060] The present invention provides a multi-source distributed resource collaborative real-time control method, system, device, and medium.

[0061] An embodiment of the present invention provides a multi-source distributed resource collaborative real-time control method. Referring to Figure 2 FIG. 3 Figure 2 is a schematic flowchart of an embodiment of the multi-source distributed resource collaborative real-time control method of the present invention.

[0062] As shown Figure 2 in FIG. 4, the multi-source distributed resource collaborative real-time control method includes:

[0063] Step S100: Construct a multi-objective scheduling model that integrates user comfort and power grid operation indicators;

[0064] Step S200: Construct an adaptive hybrid intelligent algorithm based on a non-dominated sorting genetic algorithm, a deep reinforcement learning algorithm, and a particle swarm optimization algorithm;

[0065] Step S300: Obtain real-time data, and adaptively solve the multi-objective scheduling model based on the real-time data according to the adaptive hybrid intelligent algorithm to obtain a minute-level optimization decision-making scheme;

[0066] Step S400: Issue a control command according to the minute-level optimization decision-making scheme to achieve deep collaborative control of multi-source distributed resources.

[0067] It should be noted that the execution subject in this embodiment may be an electronic device, and the electronic device may be a computer device with data processing functions, or other devices that can implement the same or similar functions. This embodiment does not make any restrictions in this regard. In this embodiment, a computer device is used as an example for illustration.

[0068] It can be understood that this embodiment is described by taking the multi-source distributed resource collaborative real-time control based on artificial intelligence as an example. Aiming at the problem that the existing technology fails to effectively integrate user comfort and grid operation objectives, in this embodiment, a method of an integrated real-time control architecture including load forecasting, multi-objective optimization control, and adaptive decision-making is proposed by combining power system real-time control and artificial intelligence technology. The method described in this embodiment incorporates multi-source distributed resources such as photovoltaic power generation, energy storage units, and air-conditioning loads into unified scheduling, and realizes minute-level real-time optimization through an adaptive hybrid intelligent algorithm to improve the flexibility and stability of the distribution network. The following is an explanation in combination with specific steps.

[0069] In one embodiment, a multi-objective scheduling model that integrates user comfort and grid operation indicators is constructed, including: constructing an objective function according to the load peak-valley difference index, voltage deviation index, temperature deviation index, and energy storage utilization rate index; setting multiple constraint conditions according to power balance constraints, photovoltaic output limits, air-conditioning comfort limits, energy storage SOC limits, voltage range limits, and control interval limits; constructing a multi-objective scheduling model according to the objective function and multiple constraint conditions.

[0070] In one embodiment, the optimization objectives of the multi-objective scheduling model are to minimize the load peak-valley difference index, minimize the voltage deviation index, minimize the temperature deviation index, and maximize the energy storage utilization rate.

[0071] Specifically, a multi-objective scheduling model that integrates user comfort and grid operation indicators is constructed, incorporating peak shaving and valley filling, voltage stability, and the deviation of user room temperature comfort, etc. into a unified optimization framework to achieve coordination on both the supply and demand sides.

[0072] Exemplarily, the multi-objective scheduling model of this embodiment simultaneously considers key indicators on both the power grid and user sides, including four indicators. Peak-valley load difference indicator: It reflects the flatness of the distribution network load curve. One of the optimization objectives is to minimize the peak-valley load difference, that is, to reduce the difference between the maximum load and the minimum load, so as to relieve the power supply pressure during the peak period of the power grid and achieve peak shaving and valley filling. Voltage deviation indicator: It measures the degree to which the distribution network voltage level deviates from the nominal value. The second optimization objective is to minimize the voltage deviation, that is, to maintain the bus voltage at each moment as close as possible to the nominal voltage (for example, 1.0 p.u.) as much as possible, and reduce the voltage fluctuation and over-limit risk. User temperature control deviation indicator (temperature deviation indicator): It represents the influence degree of air-conditioning control on the indoor temperature of users. The third optimization objective is to minimize the user comfort deviation. By moderately regulating the operation of the air conditioner, the room temperature is made as close as possible to the user-set value. For example, taking the air-conditioning temperature control as an example, the temperature control deviation can be defined as the absolute value or the sum of squares of the difference between the actual room temperature and the user-set temperature. Energy storage utilization rate indicator: It represents the contribution degree of the energy storage device in peak shaving, valley filling and voltage regulation. The fourth optimization objective is to maximize the energy storage utilization rate, that is, to increase the charge and discharge times and depth of the energy storage device, and make full use of its capacity to support the operation of the power grid without significantly shortening its life.

[0073] It should be noted that the overall objective of the method regulation in this embodiment is to control the deviation of user comfort to the minimum and improve the utilization efficiency of assets such as energy storage while reducing the peak load and stabilizing the voltage. This is a multi-objective optimization problem, and there may be trade-offs between various objectives, which need to be considered comprehensively.

[0074] Specifically, to achieve the above overall objective of regulation, this embodiment constructs an optimization model (multi-objective scheduling model) including four objective functions and multiple constraint conditions. Among them, the objective function is expressed by the weighted sum or multi-objective evolutionary algorithm, and can be formally described as:

[0075] min F={F1(ΔP), F2(ΔU), F3(ΔT), -F4(U stor )}

[0076] Here, F1 corresponds to the peak-valley load difference ΔP indicator; F2 corresponds to the voltage deviation ΔU indicator; F3 corresponds to the temperature deviation ΔT indicator; F4 corresponds to the energy storage utilization rate U stor indicator (taking the negative sign means maximizing). Exemplarily, when solving, the non-dominated sorting genetic algorithm can be used to directly search for the Pareto solution set, or each objective can be normalized and weighted to solve a single objective to obtain an equilibrium solution.

[0077] Specifically, to ensure the feasibility of the optimization solution, the following constraint conditions are imposed in the optimization model:

[0078] Power balance constraint: At each moment, the power supply and demand inside the microgrid need to be balanced, that is, Pgrid +P PV +P dis = P L +P ch ; where P grid is the power taken from the main grid (positive value) or fed to the main grid (negative value); P PV is the PV output; P dis and P ch are the energy storage discharge / charge powers respectively; P L is the load power. This constraint (power balance constraint) ensures the conservation of energy in the system. Here, a microgrid refers to a small power generation and distribution system composed of distributed power sources, energy storage devices, energy conversion devices, relevant loads, and monitoring and protection devices, which can operate either in parallel with the external large power grid or independently.

[0079] PV output limit: Photovoltaic power generation is uncontrollable, and its upper output limit is restricted by instantaneous irradiance and inverter capacity. The constraint form is as follows: 0 ≤ P PV (t) ≤ P PV,max (t); where P PV,max (t) is determined by irradiance prediction. In this embodiment, the PV output can be regarded as a given parameter during optimization, without actively controlling the PV, but the energy storage and load can be adjusted to absorb or curtail the PV power.

[0080] Air conditioning comfort limit: The adjustment range of the air conditioning load is restricted by the acceptable comfort range of the user. For example, it can be set that when participating in the regulation, the room temperature is allowed to deviate from the user-set value within the upper and lower limits (such as ±2°C), corresponding to the constraint: Tset - ΔTmax ≤ Troom(t) ≤ Tset + ΔTmax; where Tset represents the target temperature value set by the user on the air conditioning control panel according to their own comfort requirements; Troom(t) represents the temperature value in the space where the air conditioner is located, and this value will change continuously during the operation of the air conditioner; ΔTmax represents the allowable deviation amplitude, that is, the user's acceptable degree of room temperature fluctuation; equivalently, the air conditioner power adjustment ratio can be restricted not to exceed a certain amplitude. In typical demand response, the room temperature is controlled within the range of 23 - 28°C (with 26°C as the comfort reference) and the fluctuation does not exceed 0.5°C.

[0081] Energy storage SOC limit: The charge and discharge of the energy storage battery are restricted by its state of charge (SOC) to avoid overcharging and over-discharging. The constraint is as follows: SOCmin ≤ SOC(t) ≤ SOC max; Exemplarily, SOCmin can be taken as 20% (discharge depth 80%), and SOC max can be taken as 100% or slightly lower. In addition, the energy storage charge and discharge power are also restricted by the rated power: -P rated ≤ P dis (t) ≤ P rated ; the negative sign indicates charging.

[0082] Voltage range limit: The voltage of each node in the distribution network needs to be maintained within the legal range (generally ±5% of the nominal value). For the local point of common coupling, the constraint is: 0.95 ≤ U(t) ≤ 1.05 p.u.; for a multi-node system, the voltage constraints of all critical nodes need to be satisfied.

[0083] Control interval limit: Due to the lag and frequency limitations of actual control execution, in this embodiment, the regulation period can be set to 1 minute, that is, the control instruction is updated every 1 minute. For continuously controlled air conditioners and energy storage, 1 minute is sufficient to meet the physical response without additional constraints; however, for some discrete-action devices, constraints can be added to avoid frequent switching. In addition, slope constraints can be set to limit the change amplitude of adjacent control instructions to prevent control oscillations.

[0084] It can be understood that, combining the above objective function and multiple constraint conditions, the formed optimization problem belongs to multi-objective non-linear constrained optimization. Since the objectives include continuous variables (such as air conditioner power / temperature, energy storage power, etc.), and there are conflicts between the objectives, in this embodiment, an adaptive hybrid intelligent algorithm is used for solution.

[0085] In one embodiment, an adaptive hybrid intelligent algorithm is constructed based on the non-dominated sorting genetic algorithm, the deep reinforcement learning algorithm, and the particle swarm optimization algorithm.

[0086] Specifically, in this embodiment, an adaptive hybrid intelligent algorithm is used for solution, and the algorithm framework of the adaptive hybrid intelligent algorithm is as follows:

[0087] 1. Global multi-objective search (non-dominated sorting genetic algorithm NSGA-II): First, use the non-dominated sorting genetic algorithm NSGA-II to perform global optimization on the multi-objective problem. As a multi-objective evolutionary algorithm, the non-dominated sorting genetic algorithm NSGA-II searches for the Pareto optimal solution set by means of population evolution. Through selection, crossover, and mutation operations, the NSGA-II algorithm explores solutions with different objective trade-offs in the global scope. For example, different control strategies that emphasize comfort or peak shaving, the NSGA-II algorithm can provide a set of diverse candidate solutions (candidate solution sets) for subsequent decision-making.

[0088] 2. Continuous control strategy optimization (DDPG): To meet the real-time continuous control requirements, the deep reinforcement learning algorithm DDPG is introduced. This deep reinforcement learning algorithm DDPG can train an agent to make optimal control decisions in a continuous action space. In this embodiment, the deep reinforcement learning algorithm DDPG agent receives the current power system state (such as observation data of load, photovoltaic, SOC, room temperature deviation, etc.) and outputs continuous control variables (such as energy storage charge and discharge power commands, air conditioner temperature setting adjustment amounts). The deep reinforcement learning algorithm DDPG can adopt an actor-critic structure and can continuously learn the environmental feedback online to optimize its policy network parameters. In particular, the deep reinforcement learning algorithm DDPG can output accurate control signals, such as how many kilowatts the energy storage should discharge in a certain minute and how many degrees the air conditioner temperature should be increased. This fine-grained control ability ensures the accuracy and rapid response of the adjustment of continuous variables.

[0089] 3. Fast local search (PSO): To accelerate the solution convergence and further finely adjust the solution, the particle swarm optimization (PSO) algorithm is integrated in this embodiment. The particle swarm optimization algorithm PSO has the advantage of fast convergence speed and can perform local optimization in the solution space. Specifically, in this embodiment, the candidate solutions obtained by the non-dominated sorting genetic algorithm NSGA-II can be used as the initial particle swarm, and the particle swarm optimization algorithm PSO is used to iteratively update within the solution neighborhood to quickly approach the optimal solution. The particle swarm optimization algorithm PSO realizes the local exploitation of the current solution through the information sharing of particle individuals and the group. The particle swarm optimization algorithm PSO is complementary to the global exploration of the non-dominated sorting genetic algorithm NSGA-II, and the combination of the two improves the quality of the solution and the algorithm efficiency.

[0090] In one embodiment, real-time data is obtained, including: collecting the observation data of the solar irradiance every minute, and obtaining the regional lighting data according to the observation data; obtaining the community load data; and using the regional lighting data and the community load data as the input real-time data.

[0091] Specifically, data preparation: The regional lighting data and the residential community load data can be used as the input. The regional lighting data is sourced from the local solar irradiance observation every minute, and after processing, the minute-level output prediction of the photovoltaic system is obtained (for example, the noon peak is about 0.8 times the photovoltaic installation capacity, and it is reduced to 0.3 - 0.5 times on cloudy days). The residential community load data includes the electricity consumption differences between weekdays and weekends: the load is lower during the day on weekdays (some households go to work and leave home, only the basic load operates), and a peak appears around 19:00 in the evening; on weekends, due to households being at home, the use of air conditioners and electrical appliances increases, and there are peaks at noon and in the evening.

[0092] In one embodiment, based on the real-time data, the multi-objective scheduling model is adaptively solved according to the adaptive hybrid intelligent algorithm to obtain a minute-level optimization decision plan, including: processing the regional light data of the real-time data to obtain a minute-level output prediction of the photovoltaic system; using the minute-level output prediction as the photovoltaic output limit constraint condition of the multi-objective scheduling model; globally optimizing the multi-objective scheduling model according to the non-dominated sorting genetic algorithm of the adaptive hybrid intelligent algorithm to obtain a set of candidate solution sets; receiving the real-time data according to the deep reinforcement learning algorithm of the adaptive hybrid intelligent algorithm and outputting continuous control variables; iteratively updating the candidate solution set according to the particle swarm optimization algorithm of the adaptive hybrid intelligent algorithm to obtain an optimal solution; and obtaining a minute-level optimization decision plan according to the optimal solution and the continuous control variables.

[0093] In one embodiment, globally optimizing the multi-objective scheduling model according to the non-dominated sorting genetic algorithm of the adaptive hybrid intelligent algorithm to obtain a set of candidate solution sets, including: solving the multi-objective scheduling model according to the non-dominated sorting genetic algorithm of the adaptive hybrid intelligent algorithm, and searching for the Pareto optimal solution set by means of population evolution; normalizing and weighting each optimization objective of the multi-objective scheduling model to solve a single objective to obtain an equilibrium solution; and using the Pareto optimal solution set or the equilibrium solution as a set of candidate solution sets.

[0094] Specifically, as Figure 3 shown, the adaptive solution process: First, use the non-dominated sorting genetic algorithm NSGA-II to globally optimize the multi-objective problem, search for the Pareto optimal solution set by means of population evolution, and explore solutions with different objective trade-offs in the global scope through selection, crossover, and mutation operations, such as different control strategies that focus on comfort or peak shaving, to obtain a set of diverse candidate solutions (candidate solution sets) for subsequent decision-making reference; the deep reinforcement learning algorithm DDPG agent receives the current system state (such as real-time data observed for load, photovoltaic, SOC, room temperature deviation, etc.) and outputs continuous control variables (such as energy storage charge and discharge power commands, air conditioner temperature setting adjustment amounts); use the candidate solutions obtained by the non-dominated sorting genetic algorithm NSGA-II as the initial particle swarm, and use the particle swarm optimization PSO algorithm to iteratively update within the solution neighborhood to quickly approach the optimal solution.

[0095] In one embodiment, the method further includes: during adaptive solution, if the time scale is hourly or the environment changes, triggering the non-dominated sorting genetic algorithm of the adaptive hybrid intelligent algorithm for global optimization to update the overall optimization strategy set; if the time scale is minute-level, triggering the deep reinforcement learning algorithm of the adaptive hybrid intelligent algorithm to output continuous control variables; embedding the particle swarm optimization algorithm of the adaptive hybrid intelligent algorithm into the decision-making per minute, and fine-tuning the continuous control variables output by the deep reinforcement learning algorithm, or improving the solution between two calls of the non-dominated sorting genetic algorithm.

[0096] Specifically, as Figure 3 shown, the adaptive solution process: the above three algorithms (non-dominated sorting genetic algorithm, deep reinforcement learning algorithm, and particle swarm optimization algorithm) do not run in isolation. In this embodiment, an algorithm cooperation mechanism is designed: when the time scale is large (for example, every few hours) or the environment changes drastically, triggering the non-dominated sorting genetic algorithm NSGA-II for global optimization to update the overall optimization strategy set; on a short time scale (per minute), the deep reinforcement learning algorithm DDPG quickly outputs control according to the latest state; the particle swarm optimization algorithm PSO can be embedded in the decision-making per minute to fine-tune the actions (continuous control variables) given by the deep reinforcement learning algorithm DDPG, or continuously improve the solution (candidate solution set) between two calls of the non-dominated sorting genetic algorithm NSGA-II. The above three algorithm modules can achieve seamless switching by sharing the evaluation function and constraint information. For example, when the deviation between the prediction and the actual value increases and it is detected that the current strategy may not be optimal, it automatically switches to be re-optimized by the non-dominated sorting genetic algorithm NSGA-II; in the general state, the existing strategy is followed and only the deep reinforcement learning algorithm DDPG is used to adjust the micro-control. This adaptive mechanism in this embodiment ensures the adaptability of the algorithm to different scenarios.

[0097] Exemplarily, the control period and implementation: This embodiment adopts a rolling optimization control method, and the basic control period is 1 minute. That is to say, every 1 minute, real-time data is re-obtained and an optimization decision (minute-level optimization decision plan) is executed once. This enables the power control system to respond to the rapid fluctuations of photovoltaic and load (such as sudden changes in light intensity and sudden increase in load) quasi-real-time. Compared with the traditional method of scheduling according to a 15-minute or longer cycle, the minute-level control in this embodiment is more refined and can make full use of the dynamic regulation capabilities of energy storage and flexible loads.

[0098] Exemplarily, in terms of the control period, it can be set according to actual needs. For example, in one embodiment, it is simulated to run continuously for 14 days (about 20,160 scheduling periods), and the effectiveness under different weather and different electricity consumption patterns is verified through a long-term verification algorithm. In actual applications, it can run continuously for 7×24 hours, repeating every day, and continuously correcting and optimizing the strategy. Due to the adoption of adaptive rolling optimization, this embodiment supports dynamic response capabilities: when abnormal situations beyond prediction occur (such as equipment failures and grid disturbances), the control output can be immediately adjusted in the next 1-minute cycle to ensure the safe and stable operation of the power system. This fast closed-loop control in this embodiment is particularly important for ensuring the stability of the distribution network under the access of high-penetration distributed power sources. In addition, the minute-level frequent optimization (minute-level optimization decision-making scheme) poses high requirements on the algorithm performance. This embodiment greatly improves the calculation and solution speed by introducing a hybrid of reinforcement learning and meta-heuristic algorithms, ensuring that the decision-making calculation is completed within 1 minute and the control command is issued, guaranteeing the real-time requirements from an engineering perspective.

[0099] In one example, to verify the effectiveness of the method described in this embodiment, a typical scenario of a residential community in a certain area in summer is selected for simulation testing. The community is connected with distributed photovoltaics and centralized energy storage, and the air-conditioning load of residents is relatively large. The simulation scenario includes continuous operation for 14 days, covering various situations such as sunny days, cloudy days, weekdays, and weekends. The specific simulation test steps are as follows:

[0100] 1. Data preparation: Use typical irradiance data of a certain area and load data of the residential community as inputs. The irradiance data is sourced from local minute-by-minute solar irradiance observations, and after processing, the minute-level power output prediction of the photovoltaic system is obtained (for example, the noon peak is about 0.8 times the photovoltaic installation capacity, and it drops to 0.3 - 0.5 times on cloudy days). The load data of the residential community includes the electricity consumption differences between weekdays and weekends: the load is lower during the day on weekdays (some households go to work and leave home, only basic loads are operating), and a peak appears around 19:00 in the evening; on weekends, due to households being at home, the use of air conditioners and electrical appliances increases, and there are peaks at noon and in the evening. To reflect randomness, a certain amount of random disturbance is superimposed on the typical daily load curve to simulate the fluctuations of actual user behavior.

[0101] 2. Parameter setting: The energy storage battery capacity is set to 300 kWh, the initial SOC is 50%, and the upper limit of charge and discharge power is 100 kW; the adjustable total capacity of the air-conditioning group accounts for about 50% of the peak load of the community. The air-conditioning comfort limit is set with a room temperature allowable deviation of ±2°C, corresponding to a maximum downward adjustment of 50% and an upward adjustment (pre-cooling) of 50% for the air-conditioning load. The nominal voltage of the common connection point of the distribution network is 1.0 p.u., and the allowable range is 0.95 - 1.05 p.u. The control period Δt = 1 minute, and the total duration is 14 days. The above parameters all conform to the actual community scale and equipment specifications.

[0102] 3. Comparison of control schemes: Simulate the operation of the community for 14 days under the traditional control scheme and the scheme of this embodiment respectively:

[0103] Traditional control scheme: Adopt a typical time-sharing + fixed strategy for control. Specifically: Energy storage is fixed for charging at noon when there is surplus PV power, discharges to shave the peak from 19:00 to 21:00 during the evening peak, and does not operate during the remaining periods; The air conditioner does not participate in active regulation and only operates at the temperature set by the user (assumed to be a constant temperature of 24°C). There is no additional voltage control means, and voltage over-limit is only slowly adjusted by the conventional transformer tap during the day (assuming a fixed tap for the transformer in the simulation). This traditional control scheme reflects the regulation ability under the existing technical level.

[0104] Scheme of this embodiment: Operate the AI collaborative control system (implement the multi-source distributed resource collaborative real-time control method), sense the state and optimize every minute. Both energy storage and air conditioner participate in regulation. When the PV output is high and the voltage rises, the controller can increase the charging power of the energy storage in advance or lower the air conditioner temperature (increase the load) to absorb the surplus PV power; When the load is high and the voltage drops, the energy storage discharge is increased in time and the air conditioner temperature is raised (reduce the load) to support the voltage. All actions are kept within the range permitted by user comfort and equipment capacity. The scheme of this embodiment realizes the coordinated control of user comfort and grid indicators.

[0105] 4. Simulation results: Record the performance indicators under the two schemes respectively, including the distribution network voltage, the net load power of the community, the SOC of the energy storage, and the deviation of the air conditioner room temperature, etc. Focus on analyzing the voltage curve and the load curve. Figure 4 Shows the change of the voltage at the point of common coupling under the two control schemes during the 13th - 14th days: As Figure 4 shown, the comparison of the voltage curves under the traditional control scheme ( Figure 4 the yellow solid line shown) and the AI control scheme of this embodiment ( Figure 4 the orange solid line shown), Figure 4 the red dashed line shown represents the voltage upper and lower limits of 0.95 and 1.05 p.u.; Figure 4 It shows that under the traditional control scheme, the voltage rises slightly during the day when the PV output is high, and drops significantly during the evening peak, once approaching the lower limit of 0.95. After adopting the scheme of this embodiment, the voltage curve is smoother: During the day, by increasing the pre-cooling of the air conditioner load and the charging of the energy storage, the voltage rise is alleviated; At night, by reducing part of the air conditioner load and the discharge of the energy storage, the voltage drop amplitude is reduced. Especially during the evening peak on the 14th day ( Figure 4At around 2500 minutes, the voltage of the traditional solution drops to approximately 0.96 p.u. (sometimes instantaneously below 0.95 p.u., resulting in an out-of-limit situation), while under the AI control solution of this embodiment, the voltage remains at around 0.98 p.u. without reaching the lower limit. This indicates that the solution of this embodiment can effectively suppress voltage out-of-limit and improve voltage stability. At the same time, the solution of this embodiment significantly reduces the peak load. For example Figure 5 As shown in the simulation on the evening of the 13th day of the community, the maximum net load of the community reached approximately 200 kW under traditional control, while the peak value dropped to around 160 kW under the AI control solution of this embodiment, with a peak shaving amplitude of approximately 20%. The load during the valley period (such as late at night and noon) increased slightly because the AI control solution of this embodiment increased the load during the low valley period by using energy storage charging and precooling. Overall, the solution of this embodiment makes the load curve smoother and significantly reduces the peak-to-valley difference.

[0106] Summarize and compare the simulation results of the two solutions to quantitatively evaluate the improvement of the indicators, such as Figure 5 As shown: The solution of this embodiment is superior to the traditional control solution in all indicators. Among them, with the solution of this embodiment, the maximum load is reduced by approximately 20% (from 200 kW to 160 kW), proving that the multi-source collaborative control effectively reduces the peak load; the average voltage deviation drops from 1.9% to 1.4%, indicating that the voltage level is closer to the nominal value and the power quality is improved; the energy storage utilization rate is increased by 50%, meaning that the energy storage device is used more fully in cyclic operation and plays a greater role in peak shaving and voltage regulation (in the traditional control solution, the energy storage only works during limited periods, while in the AI control solution of this embodiment, it charges and discharges frequently according to needs, improving the utilization rate); the deviation of user temperature control only increases by 0.5 °C, and the average room temperature is still close to the user setting, and the deviation remains within ±2 °C, which can be considered that the user comfort is hardly affected; most importantly, the number of voltage out-of-limit times drops from 2 to 0, and this embodiment achieves that the voltage is completely maintained within the permitted range throughout the cycle. These data fully illustrate that while significantly improving the grid operation indicators, this embodiment has little impact on user comfort, achieving the effect of taking both into account.

[0107] It should be noted that according to the above simulation test verification, the multi-source distributed resource collaborative real-time control method proposed in this embodiment realizes AI collaborative real-time control and can greatly improve the comprehensive performance of the distribution network operation: it can not only effectively cut peaks and fill valleys and alleviate voltage fluctuations, but also ensure the temperature comfort of users, reflecting a double improvement in being friendly to the power grid and users.

[0108] This embodiment provides a multi-source distributed resource collaborative real-time control method, including: constructing a multi-objective scheduling model that integrates user comfort and power grid operation indicators; constructing an adaptive hybrid intelligent algorithm based on the non-dominated sorting genetic algorithm, the deep reinforcement learning algorithm, and the particle swarm optimization algorithm; obtaining real-time data, and adaptively solving the multi-objective scheduling model based on the real-time data according to the adaptive hybrid intelligent algorithm to obtain a minute-level optimization decision scheme; and issuing control commands according to the minute-level optimization decision scheme to achieve in-depth collaborative control of multi-source distributed resources. In this embodiment, user comfort is incorporated as a constraint into the power grid optimization objective, and through the autonomous learning and optimization of the adaptive hybrid intelligent algorithm, in-depth collaborative control of multi-source distributed resources, namely photovoltaic-storage-load, is achieved, thereby enhancing user satisfaction while effectively enhancing the power grid's peak shaving and voltage regulation capabilities, and realizing intelligent real-time regulation of the distribution network.

[0109] In addition, an embodiment of the present invention also proposes a storage medium, on which a multi-source distributed resource collaborative real-time control program is stored. When the multi-source distributed resource collaborative real-time control program is executed by a processor, the steps of the multi-source distributed resource collaborative real-time control method described above are implemented.

[0110] Refer to Figure 6 , Figure 6 which is a structural block diagram of an embodiment of the multi-source distributed resource collaborative real-time control system of the present invention.

[0111] As Figure 6 shown, the multi-source distributed resource collaborative real-time control system includes:

[0112] A model construction module 10, configured to construct a multi-objective scheduling model that integrates user comfort and power grid operation indicators;

[0113] An algorithm combination module 20, configured to construct an adaptive hybrid intelligent algorithm based on the non-dominated sorting genetic algorithm, the deep reinforcement learning algorithm, and the particle swarm optimization algorithm;

[0114] A data acquisition and decision module 30, configured to obtain real-time data, and adaptively solve the multi-objective scheduling model based on the real-time data according to the adaptive hybrid intelligent algorithm to obtain a minute-level optimization decision scheme;

[0115] A collaborative control module 40, configured to issue control commands according to the minute-level optimization decision scheme to achieve in-depth collaborative control of multi-source distributed resources.

[0116] This embodiment provides a multi-source distributed resource collaborative real-time control system, which incorporates user comfort as a constraint into the power grid optimization goal. Through the autonomous learning and optimization of an adaptive hybrid intelligent algorithm, it realizes the in-depth collaborative control of multi-source distributed resources, namely photovoltaic - energy storage - load. Thus, while improving user satisfaction, it effectively enhances the power grid's peak shaving and voltage regulation capabilities, achieving intelligent real-time regulation of the distribution network.

[0117] It should be noted that for the technical details not described in detail in this embodiment of the multi-source distributed resource collaborative real-time control system, reference can be made to the multi-source distributed resource collaborative real-time control method provided in any embodiment of the present invention as described above, and details will not be repeated here.

[0118] It should be understood that the above is only for illustration and does not constitute any limitation to the technical solution of the present invention. In specific applications, those skilled in the art can set according to needs, and the present invention does not limit this.

[0119] It should be noted that the above-described work process is only illustrative and does not limit the protection scope of the present invention. In actual applications, those skilled in the art can select some or all of them according to actual needs to achieve the purpose of the solution of this embodiment, and there is no limitation here.

[0120] In addition, it should be noted that in this article, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or system. Without more limitations, an element defined by the statement "including a..." does not exclude the existence of another identical element in the process, method, article or system including that element.

[0121] The serial numbers of the above embodiments of the present invention are only for description and do not represent the superiority or inferiority of the embodiments.

[0122] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-described example methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as a read-only memory (ROM) / RAM, magnetic disk, optical disk), and includes several instructions for causing a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in various embodiments of the present invention.

[0123] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. A multi-source distributed resource collaborative real-time control method, characterized in that: include: Construct a multi-objective dispatch model integrating user comfort and power grid operation indicators; Construct an adaptive hybrid intelligent algorithm based on non-dominated sorting genetic algorithm, deep reinforcement learning algorithm and particle swarm optimization algorithm; Acquire real-time data, and adaptively solve the multi-objective scheduling model according to the adaptive hybrid intelligent algorithm based on the real-time data to obtain a minute-level optimization decision-making plan; Control commands are issued according to the minute-level optimization decision-making scheme to achieve deep collaborative control of multi-source distributed resources.

2. The method according to claim 1, characterized in that The multi-objective dispatching model integrating user comfort and power grid operation indicators is constructed, including: Construct an objective function based on load peak-to-valley difference index, voltage deviation index, temperature deviation index and energy storage utilization index; Set multiple constraints based on power balance constraints, photovoltaic output limits, air conditioning comfort limits, energy storage SOC limits, voltage range limits, and control interval limits; A multi-objective scheduling model is constructed according to the objective function and multiple constraints.

3. The method according to claim 2, characterized in that The optimization objectives of the multi-objective scheduling model are to minimize the load peak-to-valley difference index, minimize the voltage deviation index, minimize the temperature deviation index and maximize the energy storage utilization rate.

4. The method according to claim 1, characterized in that The obtaining of real-time data comprises: Collecting minute-by-minute solar irradiance observation data, and obtaining regional illumination data based on the observation data; Obtain cell load data; The regional illumination data and the cell load data are used as input real-time data.

5. The method according to claim 1, characterized in that The method of adaptively solving the multi-objective scheduling model based on the real-time data according to the adaptive hybrid intelligent algorithm to obtain a minute-level optimization decision-making solution includes: Processing the regional illumination data of the real-time data to obtain a minute-level output forecast of the photovoltaic system; Using the minute-level output forecast as a photovoltaic output limit constraint of the multi-objective scheduling model; Performing global optimization on the multi-objective scheduling model according to the non-dominated sorting genetic algorithm of the adaptive hybrid intelligent algorithm to obtain a set of candidate solutions; Receive the real-time data according to the deep reinforcement learning algorithm of the adaptive hybrid intelligent algorithm and output continuous control variables; Iteratively updating the candidate solution set according to the particle swarm optimization algorithm of the adaptive hybrid intelligent algorithm to obtain an optimal solution; A minute-level optimization decision plan is obtained based on the optimal solution and the continuous control variables.

6. The method according to claim 5, characterized in that The non-dominated sorting genetic algorithm according to the adaptive hybrid intelligent algorithm performs global optimization on the multi-objective scheduling model to obtain a set of candidate solutions, including: Solving the multi-objective scheduling model according to the non-dominated sorting genetic algorithm of the adaptive hybrid intelligent algorithm, and obtaining the Pareto optimal solution set by searching in a population evolution manner; Normalizing and weighting each optimization objective of the multi-objective scheduling model to solve a single objective to obtain a balanced solution; The Pareto optimal solution set or the equilibrium solution is used as a set of candidate solution sets.

7. The method according to any one of claims 1 to 6, characterized in that The method further comprises: During adaptive solution, if the time scale is hourly or the environment changes, the non-dominated sorting genetic algorithm of the adaptive hybrid intelligent algorithm is triggered to perform global optimization and update the overall optimization strategy set; If the time scale is at the minute level, the deep reinforcement learning algorithm of the adaptive hybrid intelligent algorithm is triggered to output continuous control variables; The particle swarm optimization algorithm of the adaptive hybrid intelligent algorithm is embedded in the minute-by-minute decision making and fine-tunes the continuous control variables output by the deep reinforcement learning algorithm or improves the solution between two calls to the non-dominated sorting genetic algorithm.

8. A multi-source distributed resource collaborative real-time control system, characterized in that: include: Model building module, used to build a multi-objective dispatch model integrating user comfort and power grid operation indicators; Algorithm combination module, used to build adaptive hybrid intelligent algorithms based on non-dominated sorting genetic algorithm, deep reinforcement learning algorithm and particle swarm optimization algorithm; A data acquisition and decision-making module is used to acquire real-time data, and based on the real-time data, adaptively solve the multi-objective scheduling model according to the adaptive hybrid intelligent algorithm to obtain a minute-level optimization decision-making plan; The collaborative control module is used to issue control commands according to the minute-level optimization decision-making scheme to achieve deep collaborative control of multi-source distributed resources.

9. An electronic device, characterized in that: The electronic device includes: a memory, a processor, and a multi-source distributed resource collaborative real-time control program stored in the memory and executable on the processor, wherein the multi-source distributed resource collaborative real-time control program is configured to implement the multi-source distributed resource collaborative real-time control method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium stores a multi-source distributed resource collaborative real-time control program, and the multi-source distributed resource collaborative real-time control program is used to enable the processor to implement the multi-source distributed resource collaborative real-time control method as described in any one of claims 1 to 7 when executed.

Images