Source-load-storage multi-twinborn collaborative interaction method

Abstract

The invention relates to a source-load-storage multi-twinborn collaborative interaction method. The method comprises the following steps: acquiring operation parameters of source, load and storage equipment to construct a source-load-storage multi-twin body; performing multi-scale time sequence analysis on the twinborn body, and extracting supply and demand fluctuation characteristics including fluctuation frequency, fluctuation amplitude and fluctuation energy ratio; based on the supply and demand fluctuation characteristics, optimizing output distribution of the source-side power generation equipment and the energy storage device by adopting a learning model, and generating a source-load-storage cooperative scheduling scheme; based on the scheduling scheme, an improved particle swarm optimization algorithm is used for solving an energy flow parameter between the load-side electric load and the energy storage device, and a source-load-storage collaborative optimization instruction is generated; fusing the real-time operation parameters and the collaborative optimization instruction, and performing equipment health state prediction through a neural network to generate a health diagnosis result; and monitoring the execution result of the collaborative optimization instruction, and generating a collaborative interaction report in combination with the health diagnosis result, the operation parameters and the supply and demand fluctuation characteristics, thereby realizing collaborative control and state monitoring of the source-load-storage system.

Description

Technical Field

[0001] This invention belongs to the field of energy synergistic control technology, and in particular relates to a method for synergistic interaction of multiple twins of energy sources, loads and storage. Background Technology

[0002] With the development of energy collaborative control technology, source-load-storage collaborative management system technology has emerged. This technology, based on the dynamic characteristics of the power system, integrates operational data from source-side power generation equipment, load-side electricity consumption, and energy storage devices to achieve multi-source collaborative optimization control. Its features include the use of fixed scheduling strategies, simple feedback mechanisms, and offline analysis models to improve system stability and energy utilization efficiency. In traditional technologies, source-load-storage collaborative processing mainly relies on predefined rules and static optimization models. By collecting operational parameters from source, load, and storage equipment, a simplified mathematical model is constructed; based on historical data or empirical thresholds, time-series analysis with a fixed time window is used to extract supply and demand fluctuation characteristics, including fluctuation frequency and amplitude; and classical optimization algorithms are applied to allocate the output ratio of source-side power generation equipment and energy storage devices to generate a collaborative scheduling scheme.

[0003] However, current source-load-storage coordinated control methods or traditional approaches suffer from the following problems: Inaccurate fluctuation characteristic analysis: Traditional time-series analysis uses a fixed window length, which cannot adapt to the dynamic response characteristics of the power system, resulting in insufficient accuracy of extracted supply and demand fluctuation characteristics and affecting subsequent optimization effects. Low efficiency of coordinated optimization: Source-side output allocation relies on static cost functions and simple reward mechanisms (such as considering only generation costs), failing to integrate the real-time coupling effects of supply and demand fluctuation characteristics, and thus failing to generate efficient coordinated scheduling schemes. Inaccurate energy flow parameter solutions: Lack of evaluation mechanisms and failure to consider boundary constraints lead to large deviations in solution results and high equipment response deviation rates after the execution of coordinated optimization commands. Lack of equipment health prediction: Traditional methods ignore the real-time fusion of coordinated optimization commands and operating parameters; health status prediction relies on simple statistical models and cannot extract spatiotemporal coupling characteristics through deep neural networks, resulting in inaccurate quantification values ​​of fault risk levels and predicted remaining lifetime, increasing system fault risk. Insufficient communication reliability: Command transmission does not dynamically monitor channel quality parameters, and lacks redundant transmission mechanisms when the channel deteriorates, leading to command data packet loss or delay, affecting the real-time performance of coordinated control. Summary of the Invention

[0004] Therefore, it is necessary to provide a source-load-storage multi-twin collaborative interaction method that can solve the above problems.

[0005] Firstly, this application provides a method for collaborative interaction of multiple twins of source, payload, and storage systems, including:

[0006] Real-time acquisition of operating parameters of source-side power generation equipment, load-side electrical load and energy storage devices to construct a source-load-storage multi-twin;

[0007] Multi-scale time series analysis was performed on the source-load-storage multi-twin to extract supply and demand fluctuation characteristics, including fluctuation frequency, fluctuation amplitude and fluctuation energy ratio.

[0008] Based on the characteristics of supply and demand fluctuations, a reinforcement learning model is used to optimize the output allocation between source-side power generation equipment and energy storage devices, and a source-load-storage coordinated scheduling scheme is generated.

[0009] Based on the source-load-storage coordinated scheduling scheme, an improved particle swarm optimization algorithm is used to solve the energy flow parameters between the load-side electrical load and the energy storage device, and to generate source-load-storage coordinated optimization instructions.

[0010] By integrating real-time operating parameters and collaborative optimization instructions, the system uses deep neural networks to predict the health status of equipment and generate health diagnosis results.

[0011] Monitor the actual execution results of collaborative optimization instructions, and generate collaborative interaction reports by combining health diagnosis results, real-time operating parameters, and supply and demand fluctuation characteristics.

[0012] In one embodiment, multi-scale time-series analysis is performed on the source-load-storage multi-twin to extract supply and demand fluctuation characteristics, including:

[0013] Based on the dynamic response characteristics of power systems using source-load-storage multi-twins, a multi-scale analysis window is defined, where the first-scale window has a length of T1, satisfying... j is the system's moment of inertia, Δp is the maximum power disturbance; the second-scale window length is T2, satisfying... The lower limit is defined by the characteristic switching cycle of the load-side power electronic equipment;

[0014] Within the first scale window, an improved adaptive noise complete set empirical mode decomposition algorithm is used to decompose the source load power sequence and extract the medium- and long-term fluctuation mode components.

[0015] Within the second scale window, the wavelet packet transform algorithm is applied to perform time-frequency joint analysis of the load-side load abrupt change characteristics;

[0016] By integrating the medium- and long-term fluctuation mode components with the time-frequency joint analysis results, supply and demand fluctuation characteristic parameters are generated.

[0017] In one embodiment, based on the characteristics of supply and demand fluctuations, a reinforcement learning model is used to optimize the output allocation between source-side power generation equipment and energy storage devices, generating a source-load-storage coordinated scheduling scheme, including:

[0018] Construct a reinforcement learning model based on the Actor-Critic framework, where:

[0019] The state space is in, Real-time power generation on the source side This represents the real-time load power on the load side. For the real-time state of charge of the energy storage device, Γ freq , Γ amp , Γ energy These are the fluctuation frequency, fluctuation amplitude, and fluctuation energy percentage of the supply and demand fluctuation characteristics, respectively.

[0020] The action space is A t =[α gen α ess ], where α gen ∈[0,1] represents the output ratio coefficient of the source-side power generation equipment, α ess ∈[-1,1] represents the charging and discharging command coefficients of the energy storage device;

[0021] The reward function is:

[0022]

[0023] in, This is a penalty term for the energy storage's state of charge deviating from the optimal range [0.3, 0.7]. Let c be the source-side power generation cost function. k Let be the cost coefficient of the k-th power generation unit. Let w1, w2, and w3 be the net power fluctuation gradient norm, and w1, w2, and w3 be the weighting coefficients.

[0024] A near-end policy optimization algorithm is used to train a reinforcement learning model. The network parameters of the reinforcement learning model are updated by minimizing the policy loss function, and the optimal output allocation policy is output to generate a source-load-storage coordinated scheduling scheme.

[0025] In one embodiment, based on the source-load-storage coordinated scheduling scheme, an improved particle swarm optimization algorithm is used to solve the energy flow parameters between the load-side electrical load and the energy storage device, generating source-load-storage coordinated optimization instructions, including:

[0026] Using the output allocation strategy in the source-load-storage collaborative scheduling scheme as boundary constraints, a particle swarm position vector is constructed. in This represents the power regulation coefficient of the m-th load-side load. and These represent the correction amounts for the discharge / charge power of the energy storage device, respectively. This represents the proportion of the maximum discharge power of the energy storage. This represents the maximum charging power ratio for energy storage.

[0027] Based on the particle swarm position vector, the dual fitness evaluation function is designed using the following formula:

[0028]

[0029] in, To optimize load power as required by the coordinated scheduling scheme, Let τ be the reference power of the m-th load. ess The response time threshold for the energy storage system. This refers to the rated capacity of the energy storage device.

[0030] The particle velocity is updated using the following formula based on the dual fitness evaluation function:

[0031]

[0032] Where σ(·) is the dual-fitness normalized coordination factor, k is the current iteration number, and K max w represents the maximum number of iterations. max w min These are the upper and lower limits of the weight, respectively;

[0033] With the maximum number of iterations K max Or fitness evaluation function As a stopping condition, iterative optimization outputs the optimal particle swarm position vector X. best , where ε is the preset convergence threshold;

[0034] X best In Convert to load-side power regulation command. and Convert into energy storage charging and discharging compensation commands, and generate source-load-storage collaborative optimization commands.

[0035] In one embodiment, real-time operating parameters and collaborative optimization instructions are integrated, and a deep neural network is used to predict the device's health status, generating a health diagnosis result, including:

[0036] The operating parameter time series data of the source-side power generation equipment, the operating parameter time series data of the load-side electrical load, and the charging and discharging power time series data of the energy storage device are fused with the power allocation parameters and frequency control parameters in the collaborative optimization instruction to generate a multimodal input vector.

[0037] By using a spatiotemporal attention mechanism to extract features from the input vector, the coupled features of device operating status and scheduling instructions are obtained.

[0038] A deep neural network model containing residual connections is constructed. The input layer of the deep neural network model receives coupling features, the hidden layer calculates health status weights through nonlinear transformation, and the output layer generates the equipment health degradation index.

[0039] Based on the mapping relationship between the health degradation index and the equipment's historical failure data, a health diagnosis result is generated, which includes a quantitative value of the failure risk level and a predicted value of the remaining life.

[0040] In one embodiment, after generating the source-load-storage collaborative optimization instruction, the method further includes:

[0041] The collaborative optimization instructions are encapsulated into an instruction data packet, and verification information is attached.

[0042] Based on data packets, channel quality parameters of the communication link are monitored in real time. These parameters include packet loss rate δ and transmission delay. and signal-to-noise ratio

[0043] When δ>0.05 or At that time, the multi-path redundancy transmission mode is activated, and the number of paths K is dynamically calculated based on δ to select the path that meets the requirements. and K optimal paths are used to transmit instruction data packets in parallel, where The minimum signal-to-noise ratio threshold. For the maximum allowable delay, Δδ is the baseline packet loss rate;

[0044] The integrity of the instruction data packet after transmission is verified by checking the verification information. If the verification fails, a retransmission instruction is generated, which is used to reacquire the instruction data packet.

[0045] In one embodiment, the actual execution results of the collaborative optimization instructions are monitored, and a collaborative interaction report is generated by combining health diagnosis results, real-time operating parameters, and supply and demand fluctuation characteristics, including:

[0046] Monitor the actual execution results of the collaborative optimization instructions and calculate the deviation rate between the actual response parameters of the equipment and the parameters of the collaborative optimization instructions;

[0047] Extract the status parameters of power generation equipment, energy storage devices, and electrical load from the health diagnosis results;

[0048] Extract parameters such as fluctuation frequency, fluctuation amplitude, and fluctuation energy ratio from the characteristics of supply and demand fluctuations;

[0049] The output ratio coefficient of the source-side power generation equipment, the charge and discharge command coefficient of the energy storage device, and the energy flow parameters are obtained from the collaborative optimization instructions.

[0050] The deviation rate, state parameters, fluctuation frequency parameters, fluctuation amplitude parameters, fluctuation energy ratio parameters, output ratio coefficient, charge / discharge command coefficient, and energy flow parameters are integrated into a collaborative interactive report.

[0051] Secondly, this application also provides a source-load-storage multi-twin collaborative interaction device, comprising:

[0052] The twin construction module is used to collect operating parameters of source-side power generation equipment, load-side electrical load and energy storage device in real time, and construct source-load-storage multi-twin;

[0053] The multi-scale time series analysis module is used to perform multi-scale time series analysis on source-load-storage multi-twins and extract supply and demand fluctuation characteristics, including fluctuation frequency, fluctuation amplitude and fluctuation energy ratio.

[0054] The learning optimization scheduling module is used to optimize the output allocation of source-side power generation equipment and energy storage devices based on the characteristics of supply and demand fluctuations, and generate source-load-storage coordinated scheduling schemes.

[0055] The particle swarm optimization module is used to solve the energy flow parameters between the load-side electrical load and the energy storage device based on the source-load-storage coordinated scheduling scheme, and to generate source-load-storage coordinated optimization instructions.

[0056] The equipment health prediction module is used to integrate real-time operating parameters and collaborative optimization instructions, and predict the health status of equipment through a deep neural network to generate health diagnosis results.

[0057] The collaborative interaction report module is used to monitor the actual execution results of collaborative optimization instructions and generate collaborative interaction reports by combining health diagnosis results, real-time operating parameters and supply and demand fluctuation characteristics.

[0058] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the aforementioned source-load-storage multi-twin collaborative interaction method.

[0059] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described source-load-storage multi-twin collaborative interaction method.

[0060] The aforementioned method, device, computer equipment, and storage medium for collaborative interaction among multiple twins of source-load-storage systems construct a multi-twin system by real-time acquisition of operating parameters from source-side power generation equipment, load-side electrical loads, and energy storage devices. Multi-scale time-series analysis is performed on the twins to extract supply and demand fluctuation characteristics. Adaptive partitioning of the analysis window captures the dynamic response characteristics of the system, improving the accuracy of feature extraction and overcoming the shortcomings of traditional methods where insufficient accuracy of fluctuation characteristics affects optimization results. A reinforcement learning model is employed to optimize the output allocation between source-side power generation equipment and energy storage devices, generating a collaborative scheduling scheme. The real-time coupling effect of supply and demand fluctuation characteristics is integrated, and a reward function is designed to efficiently generate dynamic scheduling strategies, improving the low efficiency of collaborative optimization caused by reliance on static cost functions and simple reward mechanisms. Based on the scheduling scheme, an improved particle swarm optimization algorithm is applied to solve for the energy flow parameters between load-side electrical loads and energy storage devices to generate collaborative optimization instructions. By introducing boundary constraints and designing a dual-fitness evaluation function, iterative optimization outputs the optimal parameters, eliminating the shortcomings of inaccurate energy flow parameter solutions and high execution deviation rates in traditional methods. By integrating real-time operating parameters and collaborative optimization commands, a deep neural network is used to predict equipment health status and generate health diagnosis results. A spatiotemporal attention mechanism is used to extract the coupling characteristics between equipment status and scheduling commands, and a health degradation index is calculated based on the residual connection structure, achieving high-precision prediction of fault risk level and remaining life, thus compensating for the lack of equipment health prediction. The actual execution results of collaborative optimization commands are monitored, and collaborative interactive reports are generated by combining health diagnosis results, operating parameters, and supply and demand fluctuation characteristics. By integrating key indicators such as deviation rate, status parameters, and fluctuation characteristics, a closed-loop feedback mechanism for system collaborative control and status monitoring is achieved, comprehensively improving the stability and energy utilization efficiency of the source-load-storage system. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 This is a flowchart of a source-load-storage multi-twin collaborative interaction method according to the present invention;

[0063] Figure 2 This is a structural diagram of a source-load-storage multi-twin collaborative interaction device according to the present invention. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0065] In one embodiment, such as Figure 1 As shown, a method for collaborative interaction among multiple source-load-storage twins is provided. This embodiment illustrates the application of this method to terminal devices (such as edge computing nodes in power plants). The terminal device collects real-time operating parameters (including power generation, load power, and SOC status) of the source-side photovoltaic inverter, the load-side adjustable load controller, and the energy storage converter through an industrial IoT interface to construct a localized source-load-storage multi-twin twin. Simultaneously, the terminal device executes a multi-scale time series analysis algorithm to extract supply and demand fluctuation characteristics and uploads them to a cloud server. It is understood that this method can also be applied to cloud servers. The server receives regional-level source-load-storage data uploaded by a wide-area measurement system, constructs a cross-regional twin cluster, and utilizes distributed computing resources to execute reinforcement learning model optimization and particle swarm optimization algorithms to generate collaborative scheduling schemes and optimization instructions. It can also be applied to collaborative systems including terminals and servers: In areas with high penetration of new energy, when fluctuations in wind and solar power cause the net power fluctuation gradient norm to exceed the limit, the terminal equipment captures the disturbance characteristics in real time and triggers multi-scale analysis. The server dynamically optimizes the power allocation strategy based on the proportion of fluctuating energy, and then generates load adjustment coefficients and charge / discharge compensation commands through an improved particle swarm optimization algorithm. After receiving the commands, the terminal controls the load switch and energy storage PCS to perform adjustments. At the same time, it fuses the command parameters and equipment vibration and temperature data through a spatiotemporal attention mechanism, completes the prediction of the health degradation index in a local deep neural network, and uploads the execution deviation rate and health diagnosis results to the server to generate a cross-domain collaborative interaction report, realizing the full-domain collaborative control of source, grid, load and storage. In this embodiment, the method includes the following steps:

[0066] S01 collects real-time operating parameters of source-side power generation equipment, load-side electrical load, and energy storage devices to construct a source-load-storage multi-twin system.

[0067] This involves real-time acquisition of operating parameters from source-side power generation equipment, load-side electrical consumption, and energy storage devices to construct a multi-twin system encompassing source, load, and storage. A virtual model is created through a data-driven approach, laying the foundation for collaborative monitoring and optimization of the source-grid-load-storage system. Source-side power generation equipment comprises physical entities such as photovoltaic inverters and wind turbines, responsible for generating electricity. Load-side electrical consumption includes adjustable load controllers and other electrical equipment, representing the energy consumption terminals. Energy storage devices are charging and discharging units such as battery storage systems, used for energy buffering. Operating parameters include dynamic indicators such as real-time power generation, load power, and state of charge (SOC), which can be obtained through industrial IoT interfaces or other data acquisition methods, reflecting the equipment's operating status. The construction of this multi-twin system involves integrating the acquired operating parameters and establishing a virtual mapping model using digital twin technology. This merges real-time data from the source, load, and energy storage devices into a unified entity, facilitating subsequent multi-scale time-series analysis and collaborative optimization, and supporting system-level response prediction and control decisions.

[0068] S02, multi-scale time series analysis of source-load-storage multi-twins is performed to extract supply and demand fluctuation characteristics, including fluctuation frequency, fluctuation amplitude and fluctuation energy ratio.

[0069] Among these methods, dynamic response characteristics of the power system can be captured by dividing the analysis window. Multi-scale time series analysis is a window design strategy based on core parameters such as system rotational inertia and maximum power disturbance. It can achieve dynamic characterization of source-load power sequences through data decomposition and fusion at different time scales. Supply and demand fluctuation characteristics include fluctuation frequency (characterizing the periodic rate of power change), fluctuation amplitude (quantifying the intensity range of power deviation), and fluctuation energy ratio (reflecting the proportion and weight of a specific fluctuation mode in the total system energy). The extraction of these characteristics can be achieved by fusing the results of medium- and long-term modal components and short-term abrupt change analysis, without relying on a fixed window, and can improve the accuracy of the characteristics, supporting subsequent collaborative optimization.

[0070] S03, based on the characteristics of supply and demand fluctuations, uses a reinforcement learning model to optimize the output allocation of source-side power generation equipment and energy storage devices, and generates a source-load-storage coordinated scheduling scheme.

[0071] Specifically, the supply and demand fluctuation characteristics (key parameters extracted from multi-scale time series analysis, including fluctuation frequency, fluctuation amplitude, and fluctuation energy ratio) can be input into a reinforcement learning framework. This framework is then combined with operational parameters such as the real-time power generation of the source-side power generation equipment, the real-time load power of the load-side electrical load, and the state of charge of the energy storage device to construct an intelligent agent decision-making environment. The reinforcement learning model adopts a policy optimization-based mechanism, which can drive the agent to explore the optimal action strategy through a reward function. This dynamically adjusts the output ratio coefficient of the power generation equipment and the charging and discharging command coefficient of the energy storage device, achieving global optimization of power output allocation (i.e., coordinated control of power output between the source side and the energy storage unit). This generates a source-load-storage coordinated scheduling scheme (including a dynamic strategy set that optimizes the output ratio and charging and discharging commands), which serves as the input basis for subsequent coordinated optimization, ensuring the efficiency and stability of the system response.

[0072] S04. Based on the source-load-storage coordinated scheduling scheme, an improved particle swarm optimization algorithm is used to solve the energy flow parameters between the load-side electrical load and the energy storage device, and to generate source-load-storage coordinated optimization instructions.

[0073] Among them, based on the source-load-storage coordinated scheduling scheme, an improved particle swarm optimization algorithm can be used to solve the energy flow parameters (including the load-side power adjustment coefficient and the energy storage charging and discharging power correction amount) between the load-side electrical load and the energy storage device. The particle swarm position vector can be constructed by using the coordinated scheduling scheme as the boundary constraint condition, and a dual fitness evaluation function can be designed. Combined with a dynamic weight update mechanism, iterative optimization is carried out. The optimal position vector is output with a preset convergence threshold or the maximum number of iterations as the termination condition. The parameters in the vector are converted into load-side power adjustment commands and energy storage charging and discharging compensation commands to generate source-load-storage coordinated optimization commands (including a dynamic control command set of load adjustment coefficient and charging and discharging correction amount), which serve as the input basis for subsequent equipment control and health prediction, so as to achieve efficient coordination of energy flow.

[0074] S05 integrates real-time operating parameters and collaborative optimization instructions, uses a deep neural network to predict the health status of equipment, and generates health diagnosis results.

[0075] The process of predicting equipment health status by integrating real-time operating parameters (including time-series data of power generation from source-side generators, time-series data of load characteristics from load-side electrical loads, and time-series data of charging and discharging power from energy storage devices) with collaborative optimization instructions (load power adjustment instructions and charging and discharging compensation instructions generated in step S04) through a deep neural network involves multi-dimensional feature fusion of dynamic indicators from operating parameters with power allocation parameters and frequency control parameters from collaborative optimization instructions to generate a multi-modal input vector (a feature set reflecting the coupling relationship between equipment operating status and scheduling instructions). A spatiotemporal attention mechanism is then used to weight and focus the input vector to extract the spatiotemporal coupling features between equipment operating status and scheduling control strategy. Based on these coupling features, a deep neural network model containing residual connections (a multi-layer computational structure with nonlinear transformation capabilities) is constructed. The hidden layer calculates the health status weights of the features, and the output layer generates an equipment health degradation index (a continuous variable that quantifies the degree of equipment performance degradation). Based on the mapping relationship between historical fault data and the health degradation index, a health diagnosis result (an assessment report containing quantitative values ​​of fault risk level and predicted remaining life) is generated, providing key equipment status basis for subsequent collaborative interaction reports.

[0076] S06 monitors the actual execution results of collaborative optimization instructions and generates a collaborative interaction report by combining health diagnosis results, real-time operating parameters, and supply and demand fluctuation characteristics.

[0077] The process of monitoring the actual execution results of the collaborative optimization command (the actual equipment response data generated after the execution of the source-load-storage collaborative optimization command generated in step S04, including the actual values ​​of load-side power adjustment and energy storage charging and discharging compensation) and combining them with health diagnosis results (the quantitative value of fault risk level and the predicted value of remaining life derived from the equipment health degradation index generated in step S05), real-time operating parameters (including dynamic indicators such as real-time power generation of source-side power generation equipment, real-time load power of load-side power consumption, and real-time state of charge of energy storage devices), and supply and demand fluctuation characteristics (parameters such as fluctuation frequency, fluctuation amplitude, and fluctuation energy ratio extracted in step S02) to generate a collaborative interaction report (a comprehensive data set integrating deviation rate, state parameters, fluctuation characteristic parameters, output ratio coefficient, and energy flow parameters) is carried out through actual... The system calculates the deviation rate (a ratio indicator that quantifies the execution deviation) between the actual response parameters of the equipment (such as the actual values ​​of the load power adjustment coefficient and the charge / discharge correction amount) and the parameters of the collaborative optimization command (such as the set values ​​of the load power adjustment command and the charge / discharge compensation command). It also extracts the status parameters of the power generation equipment, energy storage device, and power load from the health diagnosis results (such as the fault risk level and the predicted value of the remaining life). It obtains the fluctuation frequency, fluctuation amplitude, and fluctuation energy ratio parameters from the supply and demand fluctuation characteristics. Furthermore, it analyzes the output ratio coefficient of the power generation equipment on the source side, the charge / discharge command coefficient of the energy storage device, and the energy flow parameters (such as the load power adjustment coefficient on the load side) from the collaborative optimization command. It dynamically integrates these multi-dimensional data to generate a collaborative interactive report, realizing closed-loop monitoring and status feedback optimization of the source-load-storage system, and providing a decision support basis for the collaborative control of the system.

[0078] The aforementioned multi-twin collaborative interaction method for source-load-storage systems constructs a multi-twin system by real-time acquisition of operating parameters from source-side power generation equipment, load-side electricity consumption, and energy storage devices. Multi-scale time-series analysis is performed on this twin to generate supply and demand fluctuation characteristics, including fluctuation frequency, fluctuation amplitude, and fluctuation energy proportion. A dynamic window is used to capture system response characteristics, addressing the inaccuracy of fluctuation characteristic analysis caused by traditional fixed windows. Based on the supply and demand fluctuation characteristics, a reinforcement learning model is employed to optimize power allocation and generate a collaborative scheduling scheme. By integrating the real-time coupling effects of fluctuation characteristics and a dynamic reward mechanism, a scheduling strategy is efficiently generated, overcoming the low efficiency of collaborative optimization caused by relying on static cost functions. Based on the scheduling scheme, an improved particle swarm optimization algorithm is applied to solve for energy. Flow parameters ensure energy storage response time, and dynamic weight update formulas are used for iterative optimization to generate collaborative optimization instructions. Boundary constraints and a dual fitness mechanism improve solution accuracy, eliminating the shortcomings of traditional methods such as large deviations in energy flow parameters and high execution deviation rates. Real-time operating parameters and collaborative optimization instructions are integrated, and deep neural networks are used to predict equipment health status, generating health diagnosis results including quantitative values ​​of fault risk levels and predicted remaining lifespan. Deep feature extraction is used to address the problem of missing equipment health predictions. The actual execution results of collaborative optimization instructions are monitored, deviation rates are calculated, and collaborative interactive reports are generated by combining health diagnosis results, operating parameters, and supply and demand fluctuation characteristics to achieve closed-loop feedback control, thereby improving the overall stability and energy utilization efficiency of the source-load-storage system.

[0079] In one embodiment, multi-scale time-series analysis is performed on the source-load-storage multi-twin to extract supply and demand fluctuation characteristics, including:

[0080] S11, based on the dynamic response characteristics of the power system using source-load-storage multi-twins, divides the analysis window into multiple scales, where the length of the first scale window is T1, satisfying... j is the system's moment of inertia, Δp is the maximum power disturbance; the second-scale window length is T2, satisfying... The lower limit is defined by the characteristic switching cycle of the load-side power electronic equipment;

[0081] S12, within the first scale window, the improved adaptive noise complete set empirical mode decomposition algorithm is used to decompose the source load power sequence and extract the medium and long-term fluctuation mode components.

[0082] S13, within the second scale window, the wavelet packet transform algorithm is applied to perform time-frequency joint analysis of the load-side load mutation characteristics;

[0083] S14 integrates the medium- and long-term fluctuation mode components with the time-frequency joint analysis results to generate supply and demand fluctuation characteristic parameters.

[0084] Specifically, based on the dynamic response characteristics of the power system in the aforementioned source-load-storage multi-twin, a multi-scale analysis window is defined, where the length of the first-scale window is the time-domain range that adapts to the system's inertial and disturbance responses, satisfying... j is the system's moment of inertia (a physical parameter characterizing the inertial response of a power system), Δp is the maximum power disturbance (the threshold of power surges the system can withstand), and the length of the second-scale window satisfies... The lower limit is defined by the characteristic switching cycle of the load-side power electronic equipment (such as the reciprocal of the switching frequency of IGBT devices) to ensure that the window covers high-frequency dynamic processes. Within the first-scale window, an improved adaptive noise complete set empirical mode decomposition (such as CEEMDAN) algorithm can be used to decompose the source-load power time series. Medium- and long-term fluctuation mode components (reflecting energy trend changes from minutes to hours) are extracted through adaptive noise injection and mode screening. Within the second-scale window, a wavelet packet transform algorithm is applied to perform time-frequency joint analysis of load abrupt change characteristics on the load side. Millisecond to second-level load abrupt change points are located through multi-resolution time-frequency energy distribution. The medium- and long-term fluctuation mode components (characterizing slow dynamics) are fused with the time-frequency joint analysis results (characterizing fast dynamics) to generate supply-demand fluctuation characteristic parameters, including: fluctuation frequency (the number of power periodic changes per unit time); fluctuation amplitude (the maximum deviation of power from the benchmark value); and fluctuation energy proportion (the proportion of fluctuation energy in the total power within a specific frequency band). Through dynamic window partitioning and multi-algorithm collaboration, the full-scale fluctuation characteristics of the power system from slow to fast dynamics are captured, providing high-precision input for subsequent collaborative optimization.

[0085] In one embodiment, based on the characteristics of supply and demand fluctuations, a reinforcement learning model is used to optimize the output allocation between source-side power generation equipment and energy storage devices, generating a source-load-storage coordinated scheduling scheme, including:

[0086] S21, Construct a reinforcement learning model based on the Actor-Critic framework, where:

[0087] S21.1, the state space is in, Real-time power generation on the source side This represents the real-time load power on the load side. For the real-time state of charge of the energy storage device, Γ freq , Γ amp , Γ energy These are the fluctuation frequency, fluctuation amplitude, and fluctuation energy percentage of the supply and demand fluctuation characteristics, respectively.

[0088] S21.2, the action space is A t =[α gen α ess ], where α gen∈[0,1] represents the output ratio coefficient of the source-side power generation equipment, α ess ∈[-1,1] represents the charging and discharging command coefficients of the energy storage device;

[0089] S21.3, the reward function is:

[0090]

[0091] in, This is a penalty term for the energy storage's state of charge deviating from the optimal range [0.3, 0.7]. Let c be the source-side power generation cost function. k Let be the cost coefficient of the k-th power generation unit. Let w1, w2, and w3 be the net power fluctuation gradient norm, and w1, w2, and w3 be the weighting coefficients.

[0092] S22 uses a near-end policy optimization algorithm to train a reinforcement learning model, updates the network parameters of the reinforcement learning model by minimizing the policy loss function, outputs the optimal output allocation policy, and generates a source-load-storage collaborative scheduling scheme.

[0093] For example, a reinforcement learning model based on the Actor-Critic architecture is constructed, whose state space is a six-dimensional vector. in This refers to the real-time power generation on the source side (reflecting the instantaneous output capability of the power generation unit). This represents the real-time load power on the load side (characterizing dynamic electricity demand). For real-time state of charge of energy storage devices (quantifying the energy reserves of energy storage units), Γ freq , Γ amp , Γ energy These are the fluctuation frequency (rate of periodic power change), fluctuation amplitude (power deviation extremes), and fluctuation energy proportion (energy weighting for specific frequency bands) of supply and demand fluctuation characteristics; action space.

[0094] Let A be a two-dimensional vector. t =[α gen α ess ], where α gen ∈[0,1] represents the output ratio coefficient of the source-side power generation equipment (a normalized parameter for regulating the total output power of the power generation unit), α ess ∈[-1,1] represents the charging and discharging command coefficients of the energy storage device (power adjustment parameters for charging commands with negative values ​​and discharging commands with positive values); the reward function adopts a multi-objective weighted fusion mechanism: in This constitutes a constraint for the safe operation of energy storage (a penalty is triggered when the state of charge deviates from the optimal range of [0.3, 0.7]). Characterizing the economic cost of source-side power generation (c kThe cost coefficient for the k-th generating unit is represented by the squared term (which enhances the high-power operating cost). The net power fluctuation gradient norm (a stability index for suppressing dynamic deviations in source-load power) is defined, and w1, w2, and w3 are configurable weight coefficients. These are calibrated using historical data to dynamically balance safety, economy, and stability optimization objectives. The network parameters are updated by minimizing the policy loss function using the Proximal Policy Optimization (PPO) algorithm, outputting the optimal power allocation policy. A source-load-storage coordinated scheduling scheme is generated. This scheme will incorporate the fluctuation characteristic parameter Γ. freq , Γ amp , Γ energy As an implicit constraint, the power allocation ratio between power generation and energy storage is dynamically adjusted to achieve coordinated control of source, grid, load and storage under a high proportion of new energy access.

[0095] In one embodiment, based on the source-load-storage coordinated scheduling scheme, an improved particle swarm optimization algorithm is used to solve the energy flow parameters between the load-side electrical load and the energy storage device, generating source-load-storage coordinated optimization instructions, including:

[0096] S31, using the output allocation strategy in the source-load-storage collaborative scheduling scheme as a boundary constraint, constructs the particle swarm position vector. in This represents the power regulation coefficient of the m-th load-side load. and These represent the correction amounts for the discharge / charge power of the energy storage device, respectively. This represents the proportion of the maximum discharge power of the energy storage. This represents the maximum charging power ratio for energy storage.

[0097] S32, based on the particle swarm position vector, uses the following formula to design a dual fitness evaluation function:

[0098]

[0099] in, To optimize load power as required by the coordinated scheduling scheme, Let τ be the reference power of the m-th load. ess The response time threshold for the energy storage system. This refers to the rated capacity of the energy storage device.

[0100] S33, based on the dual fitness evaluation function, updates the particle velocity using the following formula:

[0101]

[0102] Where σ(·) is the dual-fitness normalized coordination factor, k is the current iteration number, and K max w represents the maximum number of iterations.max w min These are the upper and lower limits of the weight, respectively;

[0103] S34, with the maximum number of iterations K max Or fitness evaluation function As a stopping condition, iterative optimization outputs the optimal particle swarm position vector X. best , where ε is the preset convergence threshold;

[0104] S35, X best In Convert to load-side power regulation command. and Convert into energy storage charging and discharging compensation commands, and generate source-load-storage collaborative optimization commands.

[0105] Specifically, the operation of generating collaborative optimization instructions based on the source-load-storage collaborative scheduling scheme is implemented by improving the particle swarm optimization algorithm. In the implementation, the output allocation strategy in the collaborative scheduling scheme is used as the boundary constraint condition to construct the particle swarm position vector. in This represents the power regulation factor for the m-th load-side load (characterizing the scaling ratio of the load power relative to the reference power). and These represent the correction amount of the energy storage device's discharge / charge power (quantifying the dynamic compensation range of the discharge / charge power against the collaborative scheduling baseline value). This represents the proportion of the maximum discharge power of the energy storage device (a boundary parameter determined by the physical characteristics of the energy storage equipment). The maximum charging power ratio for energy storage is negative, opposite to the discharge direction; a dual-fitness evaluation function is designed based on this position vector: and in, To achieve the optimized load power (target total load power value) required by the coordinated scheduling scheme, τ is the reference power of the m-th load (historical operating statistics). ess This is the response time threshold for the energy storage system (a preset millisecond-level time constraint). The rated capacity of the energy storage device (equipment nameplate parameters) is updated using a dynamic weighting formula. Adjust the particle velocity, where σ(·) is the dual-fitness normalization coordination factor (the transformation function that maps F1 / F2 to the interval [0, 1]), k is the current iteration number, and K max w represents the maximum number of iterations. max w min These are the upper and lower limits of the weights, respectively; with K max Or fitness gradient convergence condition (ε is the preset convergence threshold) is the stopping condition, and the optimal position vector X is output. best ,Will Convert to load-side power regulation command. It is converted into energy storage charging and discharging compensation commands, and source-load-storage collaborative optimization commands are generated to achieve millisecond-level power coordination between the load-side adjustable load and energy storage.

[0106] In one embodiment, real-time operating parameters and collaborative optimization instructions are integrated, and a deep neural network is used to predict the device's health status, generating a health diagnosis result, including:

[0107] S41, the operating parameter time series data of the source-side power generation equipment, the operating parameter time series data of the load-side electrical load, and the charging and discharging power time series data of the energy storage device are fused with the power allocation parameters and frequency control parameters in the collaborative optimization instruction to generate a multi-modal input vector;

[0108] S42 extracts features from the input vector through a spatiotemporal attention mechanism to obtain the coupled features of device operating status and scheduling instructions;

[0109] S43, Construct a deep neural network model containing residual connections. The input layer of the deep neural network model receives coupling features, the hidden layer calculates health status weights through nonlinear transformation, and the output layer generates the equipment health degradation index.

[0110] S44 generates health diagnosis results based on the mapping relationship between the health degradation index and the equipment's historical failure data. The health diagnosis results include the quantitative value of the failure risk level and the predicted value of the remaining life.

[0111] For example, the time-series data of the operating parameters of the source-side power generation equipment (including dynamic monitoring values ​​such as voltage, current, and temperature), the time-series data of the operating parameters of the load-side electrical load (such as harmonic distortion rate and power factor), and the charging and discharging power time-series curves of the energy storage device are combined with the power allocation parameters (output ratio coefficient α) in the collaborative optimization instruction. gen ) and frequency control parameters (fluctuation frequency Γ) freqThe system performs spatiotemporal alignment and feature fusion to generate a multimodal input vector (integrating heterogeneous data streams of equipment operating status and scheduling commands). This input vector is then dynamically weighted using a spatiotemporal attention mechanism, where the temporal attention layer focuses on key time slices (e.g., moments of sudden load changes), and the spatial attention layer locates key equipment nodes (e.g., energy storage converters), extracting the coupling features between equipment operating status and scheduling commands (characterizing the control correlation between mechanical stress, heat accumulation, and electrical parameters). A deep neural network model with residual connections (preferably a structure with 8 or more hidden layers) is constructed. After the input layer receives the coupling features, the hidden layers undergo nonlinear transformation using the ReLU activation function and batch normalization. The system calculates health status weights (quantifying the contribution of different features to equipment degradation), and the output layer generates an equipment health degradation index (continuous value range [0,1], where 0 represents a brand new state and 1 represents complete failure) through the Sigmoid function. Based on the mapping relationship between this health degradation index and historical equipment failure data (establishing a probabilistic correlation between the index and the failure rate through Gaussian process regression), a health diagnosis result is generated. The failure risk level quantification value (divided into high / medium / early levels) is classified according to the index threshold (>0.8 is a high risk level), and the remaining life prediction value (unit: hours) is derived through the Weibull distribution model, realizing the full life cycle status assessment of power generation equipment, energy storage units, and electrical loads.

[0112] In one embodiment, after generating the source-load-storage collaborative optimization instruction, the method further includes:

[0113] S51 encapsulates collaborative optimization instructions into instruction data packets and adds verification information;

[0114] S52, based on data packets, monitors the channel quality parameters of the communication link in real time. These parameters include packet loss rate δ and transmission delay. and signal-to-noise ratio

[0115] S53, when δ>0.05 or At that time, the multi-path redundancy transmission mode is activated, and the number of paths K is dynamically calculated based on δ to select the path that meets the requirements. and K optimal paths are used to transmit instruction data packets in parallel, where The minimum signal-to-noise ratio threshold. For the maximum allowable delay,

[0116] Δδ is the baseline packet loss rate;

[0117] S54 verifies the integrity of the transmitted instruction data packet by checking the verification information. If the verification fails, a retransmission instruction is generated, which is used to reacquire the instruction data packet.

[0118] Specifically, after generating the source-load-storage collaborative optimization command, a communication reliability guarantee mechanism is also included: the collaborative optimization command (including the load-side load power adjustment coefficient) will be used to ensure the communication reliability of the source-load-storage ... and energy storage charging and discharging compensation amount The optimized parameter set is encapsulated into an instruction data packet (using a binary encoding structure) and supplemented with verification information (such as CRC cyclic redundancy check code or MD5 hash value, used for data integrity verification); based on this data packet, the channel quality parameters of the communication link are monitored in real time, including packet loss rate δ (the proportion of data packets lost per unit time to the total number of transmitted packets), transmission delay, etc. (One-way transmission delay of data packets from the sender to the receiver), signal-to-noise ratio (The ratio of signal power to noise power, characterizing the channel's anti-interference capability), δ>0.05 or At the time (channel quality degradation trigger threshold), the multi-path redundancy transmission mode is activated, and the number of paths is dynamically calculated using δ. (Where Δδ is the baseline packet loss rate, preset to 0.01), and the path selection algorithm is used to filter those that meet the requirements. (Minimum signal-to-noise ratio threshold, usually set to 20dB) and K optimal paths (maximum allowable delay set to 150ms) are used to transmit instruction data packets in parallel (using multipath transmission to distribute data streams and reduce the risk of packet loss); the receiving end verifies the integrity of the transmitted data packets by checking the verification information (comparing the original hash value with the hash value calculated from the received data). If the verification fails, a retransmission instruction is generated (triggering the data packet re-request and transmission process), ensuring the reliability and real-time performance of the collaboratively optimized instructions in harsh communication environments.

[0119] In one embodiment, the actual execution results of the collaborative optimization instructions are monitored, and a collaborative interaction report is generated by combining health diagnosis results, real-time operating parameters, and supply and demand fluctuation characteristics, including:

[0120] S61, monitor the actual execution results of the collaborative optimization instructions and calculate the deviation rate between the actual response parameters of the equipment and the parameters of the collaborative optimization instructions;

[0121] S62, extract the status parameters of power generation equipment, energy storage devices and electrical load from the health diagnosis results;

[0122] S63, extract parameters such as fluctuation frequency, fluctuation amplitude and fluctuation energy ratio from the characteristics of supply and demand fluctuations;

[0123] S64, obtain the power output ratio coefficient of the source-side power generation equipment, the charge and discharge command coefficient of the energy storage device, and the energy flow parameters from the collaborative optimization instructions;

[0124] S65 integrates deviation rate, state parameters, fluctuation frequency parameters, fluctuation amplitude parameters, fluctuation energy ratio parameters, output ratio coefficient, charge / discharge command coefficient, and energy flow parameters into a collaborative interactive report.

[0125] For example, during implementation, the actual execution results of the collaborative optimization instructions (including the load-side load power regulation coefficient) are monitored. and energy storage charging and discharging compensation amount The optimized parameter set is used to calculate the actual response parameters of the equipment (such as the actual adjustment value of the load power). Actual values ​​of energy storage charging and discharging power ) and collaborative optimization instruction parameters (set values) and deviation rate Quantify the precision of command execution; extract status parameters from health diagnosis results, including quantified values ​​of power generation equipment fault risk levels (such as high / medium / low warning levels based on health degradation index), predicted remaining lifespan of energy storage devices (unit: hours, derived through the Weibull distribution model), and health status parameters of electrical loads (such as load insulation degradation index); extract the fluctuation frequency Γ from supply and demand fluctuation characteristics. freq (Unit: Hz, reflecting the rate of periodic change in power), fluctuation amplitude Γ amp (Unit: kW, representing extreme power deviation) and percentage of fluctuating energy Γ energy (Dimensionless, the proportion of energy in a specific frequency band in the total power); obtain the output ratio coefficient α of the source-side power generation equipment from the collaborative optimization instructions. gen (Normalized parameters for adjusting the total output of the power generation unit), energy storage device charge / discharge command coefficient α ess (Power regulation parameters for negative charging and positive discharging) and energy flow parameters (such as...) and γ ess Define the power distribution relationship between the load and energy storage); and combine the deviation rate η, state parameters (fault risk level, remaining lifetime, etc.), and fluctuation characteristic parameters (Γ). freq ,Γ amp ,Γ energy ), output ratio coefficient α gen Charge / discharge command coefficient α ess The energy flow parameters are integrated into a structured collaborative interaction report. This report, as a data set in JSON or XML format, includes timestamps and is used to drive the closed-loop feedback optimization and status monitoring of the source-load-storage system, realizing collaborative control throughout the entire process from command execution to system diagnosis.

[0126] The aforementioned multi-twin collaborative interaction method for source-load-storage systems constructs a multi-twin system by real-time acquisition of operating parameters from source-side power generation equipment, load-side electricity consumption, and energy storage devices. Based on the dynamic response characteristics of the power system, a multi-scale analysis window is defined. Within the first-scale window, an improved adaptive noise-complete set empirical mode decomposition algorithm is used to extract medium- and long-term fluctuation mode components. Within the second-scale window, a wavelet packet transform algorithm is applied for time-frequency joint analysis to capture load abrupt changes. The two methods are then fused to generate supply and demand fluctuation characteristic parameters, including fluctuation frequency, fluctuation amplitude, and fluctuation energy proportion. The adaptive window mechanism captures the dynamic response characteristics of the system, improving feature extraction accuracy and addressing the deficiency of poor optimization results caused by inaccurate fluctuation feature analysis. Based on these supply and demand fluctuation characteristics, a reinforcement learning model based on the Actor-Critic framework is constructed, and a near-end policy optimization algorithm is used to train the model to optimize the output allocation between source-side power generation equipment and energy storage devices, generating a source-load-storage collaborative scheduling scheme. By integrating the real-time coupling effect of supply and demand fluctuation characteristics and a dynamic reward mechanism, a scheduling strategy is efficiently generated, solving the problem of low efficiency in traditional collaborative optimization relying on static cost functions and simple reward mechanisms. This scheduling scheme constructs a particle swarm position vector using the power allocation strategy as a boundary constraint, designs a dual-fitness evaluation function, and employs a dynamic weight update method to iteratively optimize and output the optimal position vector. It then generates source-load-storage collaborative optimization instructions, ensuring the energy storage response time threshold is met through boundary constraints and a dual-fitness mechanism. This improves parameter solution accuracy and eliminates problems of large deviations in energy flow parameters and high execution deviation rates. By integrating real-time operating parameters and collaborative optimization instructions, a spatiotemporal attention mechanism is used to extract the coupling features between equipment operating status and scheduling instructions. A deep neural network model containing residual connections is constructed to generate an equipment health degradation index. Based on the mapping relationship with historical fault data, a health diagnosis result including a quantitative value of fault risk level and a predicted remaining lifespan is output. Deep feature extraction is used to address the lack of equipment health prediction. After generating collaborative optimization instructions, data packets are encapsulated and additional verification information is added. Channel quality parameters are monitored in real-time and transmitted in parallel to address insufficient communication reliability. The actual execution results of the collaborative optimization instructions are monitored to calculate the deviation rate. Combined with health diagnosis results, real-time operating parameters, and supply-demand fluctuation characteristics, a collaborative interaction report is generated, achieving closed-loop feedback control from instruction execution to system diagnosis.

[0127] By improving the accuracy of fluctuation characteristics to ensure the accuracy of subsequent optimization, enhancing the efficiency of collaborative optimization to improve the dynamic response of energy allocation, accurately solving energy flow parameters to reduce the execution deviation rate, improving equipment health prediction to reduce the risk of failure, and ensuring communication reliability to improve the real-time performance of command transmission, the stability and energy utilization efficiency of the source-load-storage system are comprehensively improved.

[0128] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0129] Based on the same inventive concept, this application also provides a source-load-storage multi-twin collaborative interaction device for implementing the aforementioned source-load-storage multi-twin collaborative interaction method. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more source-load-storage multi-twin collaborative interaction device embodiments provided below can be found in the above-described limitations of the source-load-storage multi-twin collaborative interaction method, and will not be repeated here.

[0130] In one exemplary embodiment, such as Figure 2 As shown, a source-load-storage multi-twin collaborative interaction device is provided, comprising:

[0131] The twin construction module 101 is used to collect the operating parameters of the source-side power generation equipment, the load-side power consumption and energy storage device in real time, and construct a source-load-storage multi-twin;

[0132] The multi-scale time series analysis module 102 is used to perform multi-scale time series analysis on the source-load-storage multi-twin and extract supply and demand fluctuation characteristics, including fluctuation frequency, fluctuation amplitude and fluctuation energy ratio.

[0133] The learning optimization scheduling module 103 is used to optimize the output allocation of source-side power generation equipment and energy storage devices based on the characteristics of supply and demand fluctuations and to generate a source-load-storage coordinated scheduling scheme.

[0134] The particle swarm optimization module 104 is used to solve the energy flow parameters between the load-side electrical load and the energy storage device based on the source-load-storage coordinated scheduling scheme, and to generate source-load-storage coordinated optimization instructions by using an improved particle swarm optimization algorithm.

[0135] The equipment health prediction module 105 is used to integrate real-time operating parameters and collaborative optimization instructions, predict the health status of equipment through a deep neural network, and generate health diagnosis results.

[0136] The collaborative interaction report module 106 is used to monitor the actual execution results of collaborative optimization instructions and generate collaborative interaction reports by combining health diagnosis results, real-time operating parameters and supply and demand fluctuation characteristics.

[0137] In one embodiment, the multi-scale time series analysis module 102 is further configured to:

[0138] Based on the dynamic response characteristics of power systems using source-load-storage multi-twins, a multi-scale analysis window is defined, where the first-scale window has a length of T1, satisfying... j is the system's moment of inertia, Δp is the maximum power disturbance; the second-scale window length is T2, satisfying... The lower limit is defined by the characteristic switching cycle of the load-side power electronic equipment;

[0139] Within the first scale window, an improved adaptive noise complete set empirical mode decomposition algorithm is used to decompose the source load power sequence and extract the medium- and long-term fluctuation mode components.

[0140] Within the second scale window, the wavelet packet transform algorithm is applied to perform time-frequency joint analysis of the load-side load abrupt change characteristics;

[0141] By integrating the medium- and long-term fluctuation mode components with the time-frequency joint analysis results, supply and demand fluctuation characteristic parameters are generated.

[0142] In one embodiment, the learning optimization scheduling module 103 is further configured to:

[0143] Construct a reinforcement learning model based on the Actor-Critic framework, where:

[0144] The state space is in, Real-time power generation on the source side This represents the real-time load power on the load side. For the real-time state of charge of the energy storage device, Γ freq , Γ amp , Γ energy These are the fluctuation frequency, fluctuation amplitude, and fluctuation energy percentage of the supply and demand fluctuation characteristics, respectively.

[0145] The action space is A t =[α gen α ess ], where α gen ∈[0,1] represents the output ratio coefficient of the source-side power generation equipment, α ess ∈[-1,1] represents the charging and discharging command coefficients of the energy storage device;

[0146] The reward function is:

[0147]

[0148] in, This is a penalty term for the energy storage state of charge deviating from the optimal range of 0.3, 0.7. Let c be the source-side power generation cost function. k Let be the cost coefficient of the k-th power generation unit. Let w1, w2, and w3 be the net power fluctuation gradient norm, and w1, w2, and w3 be the weighting coefficients.

[0149] A near-end policy optimization algorithm is used to train a reinforcement learning model. The network parameters of the reinforcement learning model are updated by minimizing the policy loss function, and the optimal output allocation policy is output to generate a source-load-storage coordinated scheduling scheme.

[0150] In one embodiment, the particle swarm cooperative optimization module 104 is further configured to:

[0151] Using the output allocation strategy in the source-load-storage collaborative scheduling scheme as boundary constraints, a particle swarm position vector is constructed. in This represents the power regulation coefficient of the m-th load-side load. and These represent the correction amounts for the discharge / charge power of the energy storage device, respectively. This represents the proportion of the maximum discharge power of the energy storage. This represents the maximum charging power ratio for energy storage.

[0152] Based on the particle swarm position vector, the dual fitness evaluation function is designed using the following formula:

[0153]

[0154] in, To optimize load power as required by the coordinated scheduling scheme, Let τ be the reference power of the m-th load. ess The response time threshold for the energy storage system. This refers to the rated capacity of the energy storage device.

[0155] The particle velocity is updated using the following formula based on the dual fitness evaluation function:

[0156]

[0157] Where σ(·) is the dual-fitness normalized coordination factor, k is the current iteration number, and K max w represents the maximum number of iterations. max w min These are the upper and lower limits of the weight, respectively;

[0158] With the maximum number of iterations K max Or fitness evaluation function As a stopping condition, iterative optimization outputs the optimal particle swarm position vector X. best , where ε is the preset convergence threshold;

[0159] X best In Convert to load-side power regulation command. and Convert into energy storage charging and discharging compensation commands, and generate source-load-storage collaborative optimization commands.

[0160] In one embodiment, the device health prediction module 105 is further configured to:

[0161] The operating parameter time series data of the source-side power generation equipment, the operating parameter time series data of the load-side electrical load, and the charging and discharging power time series data of the energy storage device are fused with the power allocation parameters and frequency control parameters in the collaborative optimization instruction to generate a multimodal input vector.

[0162] By using a spatiotemporal attention mechanism to extract features from the input vector, the coupled features of device operating status and scheduling instructions are obtained.

[0163] A deep neural network model containing residual connections is constructed. The input layer of the deep neural network model receives coupling features, the hidden layer calculates health status weights through nonlinear transformation, and the output layer generates the equipment health degradation index.

[0164] Based on the mapping relationship between the health degradation index and the equipment's historical failure data, a health diagnosis result is generated, which includes a quantitative value of the failure risk level and a predicted value of the remaining life.

[0165] In one embodiment, the particle swarm cooperative optimization module 104 is further configured to:

[0166] The collaborative optimization instructions are encapsulated into an instruction data packet, and verification information is attached.

[0167] Based on data packets, channel quality parameters of the communication link are monitored in real time. These parameters include packet loss rate δ and transmission delay. and signal-to-noise ratio

[0168] When δ>0.05 or At that time, the multi-path redundancy transmission mode is activated, and the number of paths K is dynamically calculated based on δ to select the path that meets the requirements. and K optimal paths are used to transmit instruction data packets in parallel, where The minimum signal-to-noise ratio threshold. For the maximum allowable delay, Δδ is the baseline packet loss rate;

[0169] The integrity of the instruction data packet after transmission is verified by checking the verification information. If the verification fails, a retransmission instruction is generated, which is used to reacquire the instruction data packet.

[0170] In one embodiment, the collaborative interaction reporting module 106 is further configured to:

[0171] Monitor the actual execution results of the collaborative optimization instructions and calculate the deviation rate between the actual response parameters of the equipment and the parameters of the collaborative optimization instructions;

[0172] Extract the status parameters of power generation equipment, energy storage devices, and electrical load from the health diagnosis results;

[0173] Extract parameters such as fluctuation frequency, fluctuation amplitude, and fluctuation energy ratio from the characteristics of supply and demand fluctuations;

[0174] The output ratio coefficient of the source-side power generation equipment, the charge and discharge command coefficient of the energy storage device, and the energy flow parameters are obtained from the collaborative optimization instructions.

[0175] The deviation rate, state parameters, fluctuation frequency parameters, fluctuation amplitude parameters, fluctuation energy ratio parameters, output ratio coefficient, charge / discharge command coefficient, and energy flow parameters are integrated into a collaborative interactive report.

[0176] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the source-load-storage multi-twin collaborative interaction method as described above.

[0177] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0178] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0179] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for collaborative interaction among multiple twins of source, load, and storage systems, characterized in that, The method includes: Real-time acquisition of operating parameters of source-side power generation equipment, load-side electrical load and energy storage devices to construct a source-load-storage multi-twin; Multi-scale time series analysis was performed on the source-load-storage multi-twin to extract supply and demand fluctuation characteristics, including fluctuation frequency, fluctuation amplitude and fluctuation energy ratio. Based on the supply and demand fluctuation characteristics, a reinforcement learning model is used to optimize the output allocation of the source-side power generation equipment and energy storage device, and a source-load-storage coordinated scheduling scheme is generated. Based on the source-load-storage coordinated scheduling scheme, an improved particle swarm optimization algorithm is used to solve the energy flow parameters between the load-side electrical load and the energy storage device, and to generate source-load-storage coordinated optimization instructions. By integrating the real-time operating parameters and collaborative optimization instructions, a deep neural network is used to predict the health status of the equipment and generate a health diagnosis result. The actual execution results of the collaborative optimization instructions are monitored, and a collaborative interaction report is generated by combining the health diagnosis results, real-time operating parameters, and supply and demand fluctuation characteristics.

2. The method according to claim 1, characterized in that, The multi-scale time series analysis of the source-load-storage multi-twin to extract supply and demand fluctuation characteristics includes: Based on the dynamic response characteristics of the power system using the aforementioned source-load-storage multi-twin, a multi-scale analysis window is defined, wherein the length of the first-scale window is T1, satisfying... j is the system's moment of inertia, Δp is the maximum power disturbance; the second-scale window length is T2, satisfying... The lower limit is defined by the characteristic switching cycle of the load-side power electronic equipment; Within the first scale window, the improved adaptive noise complete set empirical mode decomposition algorithm is used to decompose the source load power sequence and extract the medium and long-term fluctuation mode components. Within the second scale window, the wavelet packet transform algorithm is applied to perform time-frequency joint analysis of the load-side load abrupt change characteristics; The supply and demand fluctuation characteristic parameters are generated by integrating the medium- and long-term fluctuation mode components with the time-frequency joint analysis results.

3. The method according to claim 2, characterized in that, Based on the supply and demand fluctuation characteristics, a reinforcement learning model is used to optimize the output allocation between the source-side power generation equipment and the energy storage device, generating a source-load-storage coordinated scheduling scheme, including: Construct a reinforcement learning model based on the Actor-Critic framework, where: The state space is in, Real-time power generation on the source side This represents the real-time load power on the load side. For the real-time state of charge of the energy storage device, Γ freq , Γ amp , Γ energy These are the fluctuation frequency, fluctuation amplitude, and fluctuation energy percentage of the supply and demand fluctuation characteristics, respectively. The action space is A t =[α gen α ess ], where α gen ∈[0,1] represents the output ratio coefficient of the source-side power generation equipment, α ess ∈[-1,1] represents the charging and discharging command coefficients of the energy storage device; The reward function is: in, This is a penalty term for the energy storage's state of charge deviating from the optimal range [0.3, 0.7]. Let c be the source-side power generation cost function. k Let be the cost coefficient of the k-th power generation unit. Let w1, w2, and w3 be the net power fluctuation gradient norm, and w1, w2, and w3 be the weighting coefficients. The reinforcement learning model is trained using a near-end policy optimization algorithm. The network parameters of the reinforcement learning model are updated by minimizing the policy loss function, and the optimal output allocation policy is output to generate the source-load-storage collaborative scheduling scheme.

4. The method according to claim 3, characterized in that, The source-load-storage coordinated scheduling scheme employs an improved particle swarm optimization algorithm to solve for the energy flow parameters between the load-side electrical load and the energy storage device, generating source-load-storage coordinated optimization instructions, including: Using the power allocation strategy in the source-load-storage collaborative scheduling scheme as boundary constraints, a particle swarm position vector is constructed. in This represents the power regulation coefficient of the m-th load-side load. and These represent the correction amounts for the discharge / charge power of the energy storage device, respectively. This represents the proportion of the maximum discharge power of the energy storage. This represents the maximum charging power ratio for energy storage. Based on the particle swarm position vector, the dual fitness evaluation function is designed using the following formula: in, To optimize load power as required by the coordinated scheduling scheme, Let τ be the reference power of the m-th load. ess The response time threshold for the energy storage system. This refers to the rated capacity of the energy storage device. The particle velocity is updated using the following formula based on the dual fitness evaluation function: Where σ(·) is the dual-fitness normalized coordination factor, k is the current iteration number, and K max w represents the maximum number of iterations. max w min These are the upper and lower limits of the weight, respectively; With the maximum number of iterations K max Or the fitness evaluation function As a stopping condition, iterative optimization outputs the optimal particle swarm position vector X. best , where ε is the preset convergence threshold; X best In Convert to load-side power regulation command. and The command is converted into an energy storage charging and discharging compensation command, which generates the source-load-storage collaborative optimization command.

5. The method according to claim 4, characterized in that, The process of integrating the real-time operating parameters and collaborative optimization instructions, and using a deep neural network to predict the device's health status, generates a health diagnosis result, including: The operating parameter time series data of the source-side power generation equipment, the operating parameter time series data of the load-side power load, and the charging and discharging power time series data of the energy storage device are fused with the power allocation parameters and frequency control parameters in the collaborative optimization instruction to generate a multimodal input vector. The input vector is used to extract features through a spatiotemporal attention mechanism to obtain the coupling features between the device operating status and the scheduling instructions; A deep neural network model containing residual connections is constructed. The input layer of the deep neural network model receives the coupling features, the hidden layer calculates the health status weights through nonlinear transformation, and the output layer generates the equipment health degradation index. Based on the mapping relationship between the health degradation index and the equipment's historical fault data, a health diagnosis result is generated, which includes a quantitative value of the fault risk level and a predicted value of the remaining life.

6. The method according to claim 1, characterized in that, After generating the source-load-storage collaborative optimization instruction, it also includes: The collaborative optimization instructions are encapsulated into an instruction data packet, and verification information is attached. Based on the data packets, the channel quality parameters of the communication link are monitored in real time. These channel quality parameters include packet loss rate δ and transmission delay. and signal-to-noise ratio When δ>0.05 or At that time, the multi-path redundancy transmission mode is activated, and the number of paths K is dynamically calculated based on δ to select the path that meets the requirements. and K optimal paths are used to transmit instruction data packets in parallel, where The minimum signal-to-noise ratio threshold. For the maximum allowable delay, Δδ is the baseline packet loss rate; The integrity of the transmitted instruction data packet is verified using the verification information. If the verification fails, a retransmission instruction is generated, which is used to reacquire the instruction data packet.

7. The method according to claim 4, characterized in that, The monitoring of the actual execution results of the collaborative optimization instructions, combined with the health diagnosis results, real-time operating parameters, and supply and demand fluctuation characteristics, generates a collaborative interaction report, including: Monitor the actual execution results of the collaborative optimization instructions and calculate the deviation rate between the actual response parameters of the device and the parameters of the collaborative optimization instructions; Extract the status parameters of power generation equipment, energy storage devices, and electrical load from the health diagnosis results; Extract the fluctuation frequency, fluctuation amplitude, and fluctuation energy ratio parameters from the supply and demand fluctuation characteristics; The power output ratio coefficient of the source-side power generation equipment, the charge and discharge command coefficient of the energy storage device, and the energy flow parameters are obtained from the collaborative optimization instructions. The deviation rate, state parameters, fluctuation frequency parameters, fluctuation amplitude parameters, fluctuation energy ratio parameters, output ratio coefficient, charge / discharge command coefficient, and energy flow parameters are integrated into the collaborative interaction report.

8. A source-load-storage multi-twin collaborative interaction device, characterized in that, The device includes: The twin construction module is used to collect operating parameters of source-side power generation equipment, load-side electrical load and energy storage device in real time, and construct source-load-storage multi-twin; The multi-scale time series analysis module is used to perform multi-scale time series analysis on the source-load-storage multi-twin and extract supply and demand fluctuation characteristics, including fluctuation frequency, fluctuation amplitude and fluctuation energy ratio. The learning optimization scheduling module is used to optimize the output allocation of the source-side power generation equipment and energy storage device based on the supply and demand fluctuation characteristics, and generate a source-load-storage coordinated scheduling scheme. The particle swarm optimization module is used to solve the energy flow parameters between the load-side electrical load and the energy storage device based on the source-load-storage coordinated scheduling scheme, and to generate source-load-storage coordinated optimization instructions by using an improved particle swarm optimization algorithm. The equipment health prediction module is used to integrate the real-time operating parameters and collaborative optimization instructions, and predict the health status of the equipment through a deep neural network to generate health diagnosis results. The collaborative interaction report module is used to monitor the actual execution results of the collaborative optimization instructions and generate a collaborative interaction report by combining the health diagnosis results, real-time operating parameters and supply and demand fluctuation characteristics.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images