Intelligent power supply system data comprehensive management method and system

Abstract

This invention provides an intelligent power supply system data integrated management method and system. The method includes steps such as dynamic monitoring point deployment, multi-source data acquisition, data fusion and analysis, optimized control command generation, and real-time visualization and feedback. The system includes modules corresponding to dynamic monitoring, data acquisition and transmission, AI analysis, optimized control, visualization interaction, and a central processing unit. Each part employs specific technical means, such as reinforcement learning algorithms and specific machine learning models. This invention achieves dynamic optimization of monitoring point layout, accurate data acquisition and analysis, generation of reasonable control commands, and realization of data visualization and interaction. It can improve the accuracy, reliability, and intelligence level of power supply system data management, while also considering economic benefits, carbon emission indicators, and power supply reliability.

Description

Technical Field

[0001] This invention relates to the field of power supply system data management technology, and in particular to an intelligent power supply system data integrated management method and system. Background Technology

[0002] With societal development, power supply systems play a crucial role in the national economy and daily life. Their stable operation directly impacts the continuity of industrial production, the convenience of residents' lives, and the normal functioning of society. In recent years, power supply system technologies have continuously evolved, moving from traditional, simple power supply models towards intelligent and information-based approaches. The introduction of intelligent management technologies enables power supply systems to better cope with complex and ever-changing electricity demands and the influence of external environmental factors, improving power supply efficiency and reliability and providing strong support for social development. Simultaneously, the fusion of multi-source data, the application of artificial intelligence technologies, and the realization of real-time visualization have brought new opportunities and challenges to power supply system management, driving continuous innovation and progress in power supply system management technology.

[0003] In existing technologies, various methods are typically employed to address issues related to power supply system management. For example, some technologies focus on the static layout and threshold clustering of power supply parameter monitoring points, monitoring and analyzing power supply parameters by pre-setting the locations and thresholds of monitoring points. Other technologies use grey models and genetic algorithms for demand forecasting and energy optimization, using mathematical models to predict power supply demand and optimizing energy allocation based on the prediction results. Additionally, visual maintenance management based on Building Information Modeling (BIM) is also a common approach, using 3D models to intuitively display the structure and equipment information of the power supply system, facilitating troubleshooting and maintenance by maintenance personnel.

[0004] However, existing technologies have obvious shortcomings. For example, technologies that focus on the static layout of power supply parameter monitoring points and threshold clustering lack dynamic adaptability and cannot adjust the layout of monitoring points according to real-time load changes and environmental factors. Technologies using gray models and genetic algorithms have insufficient robustness, rely on preset thresholds, lack real-time adjustment capabilities, and are prone to false alarms or missed alarms. Technologies based on BIM to achieve visualized maintenance management have problems with data synchronization delays and limited system scalability, making it difficult to meet the increasingly complex management needs of power supply systems. Summary of the Invention

[0005] This invention addresses the problems of insufficient dynamic adaptability and robustness in existing technologies, and proposes an intelligent power supply system data integrated management method and system.

[0006] To achieve the above objectives, the following technical solution is proposed: A method for integrated data management of an intelligent power supply system includes the following steps: S1. Deploy dynamic monitoring points, dynamically optimize the layout of power supply parameter monitoring points based on geographic information system, real-time load data and environmental parameters, and assign a unique device identifier to each monitoring point; S2. Collect multi-source data, collect power supply system data and environmental data in real time, and transmit them to the data center through a wireless communication network; S3. Data integration and analysis: The collected data is correlated with external data and machine learning models are used for anomaly detection, load forecasting and equipment health assessment. S4. Generate optimized control commands: Based on the analysis results, generate power supply control commands using a multi-objective optimization algorithm. S5. Real-time visualization and feedback: Data visualization is achieved through building information modeling and augmented reality interface, and an adaptive feedback loop is established.

[0007] By adopting the above technical solutions, dynamic monitoring point deployment can optimize the layout and assign unique identifiers based on geographic information systems, real-time load data, and environmental parameters, enabling dynamic monitoring of power supply parameters; multi-source data acquisition can acquire power supply system and environmental data in real time and transmit them to the data center for subsequent analysis; data fusion and analysis can correlate the collected data with external data, and use machine learning models for anomaly detection, load forecasting, and equipment health evaluation, providing a basis for power supply system management; generating optimized control commands can generate power supply control commands based on the analysis results, optimizing the operation of the power supply system; real-time visualization and feedback can achieve data visualization and establish a feedback loop through building information models and augmented reality interfaces, facilitating real-time adjustment of control strategies and achieving efficient and reliable management of the power supply system.

[0008] Preferably, the dynamic monitoring point layout optimization adopts a reinforcement learning algorithm and automatically adjusts the monitoring point density according to real-time load changes.

[0009] By adopting the above technical solution, the dynamic monitoring point layout optimization uses a reinforcement learning algorithm, which can automatically adjust the density of monitoring points according to real-time load changes. This enables the monitoring points to automatically adjust their position and density based on load changes and environmental factors, thereby achieving more accurate dynamic monitoring of the power supply system.

[0010] Preferably, the machine learning model includes a long short-term memory network for load prediction and a graph neural network for equipment health assessment, and the machine learning model incorporates a transfer learning mechanism to adapt to small sample scenarios.

[0011] By adopting the above technical solutions, using long short-term memory networks for load forecasting can more accurately predict the load situation of the power supply system; using graph neural networks for equipment health evaluation can effectively analyze the health status of equipment and generate evaluation indicators; introducing transfer learning mechanisms can enable machine learning models to adapt well to small sample scenarios, improving the adaptability and generalization ability of the models.

[0012] Preferably, the machine learning model provides interpretability assessment through the SHAP analysis tool to verify the credibility of the anomaly detection and load prediction results.

[0013] By adopting the above technical solutions, in the intelligent power supply system data comprehensive management method, combining steps such as dynamic monitoring point deployment, multi-source data acquisition, data fusion and analysis, optimized control command generation, real-time visualization and feedback, when the machine learning model introduces long short-term memory networks, graph neural networks and transfer learning mechanisms, the interpretability evaluation of the model can be carried out using SHAP analysis tools, which can verify the credibility of anomaly detection and load prediction results and improve the reliability and trustworthiness of the model prediction results.

[0014] Preferably, the wireless communication network adopts low-power wide-area network technology and adds redundancy checks to the transmitted data to improve reliability.

[0015] By adopting the above technical solutions, the wireless communication network using low-power wide-area network technology can meet the data transmission requirements of the power supply system while reducing power consumption. Adding redundancy verification to the transmitted data can improve the reliability of data transmission. Combined with other steps in the overall solution, it can achieve efficient acquisition and transmission of power supply system data, thereby ensuring the smooth progress of subsequent data fusion analysis, control command generation, and visualization feedback, and ultimately achieving efficient and reliable management of the power supply system.

[0016] An intelligent power supply system data management system includes: The dynamic monitoring module integrates geographic information systems and IoT sensors to deploy monitoring points; The data acquisition and transmission module collects voltage, current, power, harmonics, ambient temperature and humidity, and user-end load data and transmits them through a wireless communication network. The AI ​​analytics module includes a library of machine learning models to perform data fusion, anomaly detection, load forecasting, and equipment health assessment. The control module is optimized by embedding a multi-objective optimization algorithm to generate power supply control commands. The visualization and interaction module combines building information modeling and augmented reality technology to achieve data visualization and interaction. The central processing unit coordinates the operation of various modules and provides integration interfaces for third-party systems.

[0017] By adopting the above technical solutions, the dynamic monitoring module integrates geographic information systems and IoT sensors to deploy monitoring points, enabling dynamic optimization of the power supply parameter monitoring point layout based on geographic information systems, real-time load data, and environmental parameters; the data acquisition and transmission module collects multi-source data and transmits it through wireless communication networks, enabling real-time acquisition of power supply system and environmental data; the AI ​​analysis module includes a machine learning model library, which can correlate collected data with external data and use machine learning models for anomaly detection, load forecasting, and equipment health assessment; the optimization control module embeds multi-objective optimization algorithms, which can generate power supply control commands based on analysis results; the visualization and interaction module combines building information modeling and augmented reality technology to achieve data visualization and interaction; and the central processing unit coordinates the operation of each module and provides third-party system integration interfaces, enabling the system to exchange data with third-party systems and achieve collaborative operation between systems.

[0018] Preferably, the optimization control module uses a non-dominated sorting genetic algorithm II for multi-objective optimization and integrates a model predictive control framework to achieve rolling optimization.

[0019] By adopting the above technical solutions, the optimization control module uses the non-dominated sorting genetic algorithm II for multi-objective optimization, which can simultaneously optimize economic benefits, carbon emission indicators, and power supply reliability. By incorporating carbon emissions as a constraint into the optimization objective, it can better balance economic, environmental, and reliability objectives. The integrated model predictive control framework realizes rolling optimization, which can dynamically adjust the power supply system control strategy, making the power supply system control more real-time and adaptable, thereby achieving efficient and reliable management of the power supply system.

[0020] Preferably, the non-dominated sorting genetic algorithm II simultaneously optimizes economic benefits, carbon emission indicators, and power supply reliability, and incorporates carbon emission as a constraint into the optimization objective.

[0021] By adopting the above technical solutions, in the comprehensive data management of intelligent power supply systems, the non-dominated sorting genetic algorithm II can comprehensively consider economic benefits, carbon emission indicators and power supply reliability, and incorporate carbon emission as a constraint into the optimization objective. This enables the power supply system to improve economic benefits and effectively control carbon emissions while ensuring power supply reliability, thus achieving a multi-faceted optimization balance.

[0022] Preferably, the visualization interaction module achieves bidirectional real-time synchronization with the physical device through digital thread technology, and supports interaction with augmented reality glasses to guide on-site repairs.

[0023] By adopting the above technical solutions, the visualization interaction module utilizes digital thread technology to achieve bidirectional real-time synchronization with physical devices, ensuring that the virtual model is consistent with the actual device status, enabling maintenance personnel to obtain accurate device information in a timely manner; it supports interaction with augmented reality glasses, providing intuitive and precise guidance for on-site maintenance personnel, improving maintenance efficiency and accuracy, and thus enhancing the reliability and management efficiency of the entire intelligent power supply system.

[0024] Preferably, the central processing unit exchanges data with third-party systems through an application programming interface and coordinates the collaborative operation of the dynamic monitoring module, the data acquisition and transmission module, and the optimization control module.

[0025] By adopting the above technical solutions, the central processing unit can exchange data with third-party systems through application programming interfaces, realizing data sharing and interaction between systems; at the same time, it coordinates the collaborative operation of the dynamic monitoring module, data acquisition and transmission module, and optimization control module, so that each module works closely together to ensure the efficient and stable operation of the intelligent power supply system data integrated management system.

[0026] The beneficial effects of this invention are: 1. By using dynamic monitoring point deployment and reinforcement learning algorithms, the location and density of monitoring points can be automatically adjusted according to load changes and environmental factors, solving the problem of lack of dynamic adaptability in the static layout of monitoring points in the existing technology; 2. By adopting multi-source data fusion technology, the collected data is correlated with external data, and machine learning models are used for anomaly detection, load prediction and equipment health assessment. This avoids the problems of insufficient model robustness, reliance on preset thresholds and lack of real-time adjustment capabilities in existing technologies, which lead to false alarms or missed alarms. 3. By combining Building Information Modeling (BIM) with Augmented Reality (AR) technology to achieve data visualization and interaction, and by using digital thread technology to achieve bidirectional real-time synchronization with physical devices, the problems of data synchronization delay and limited system scalability in existing BIM-based visualized maintenance management technologies are solved. Attached Figure Description

[0027] Figure 1 This is a flowchart of the method of the present invention.

[0028] Figure 2 This is a system module framework diagram of the present invention. Detailed Implementation

[0029] The technical solutions in the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. The described embodiments are only possible technical implementations of the present invention, but are not limited thereto. Other embodiments obtained by those skilled in the art in conjunction with the embodiments of the present invention without creative effort are also within the protection scope of the present invention.

[0030] This application mainly adopts dynamic monitoring, multi-source fusion and intelligent control management of power supply data, which achieves the effect of efficient and reliable management of power supply system. The following is a further detailed description of this application.

[0031] Example 1 This application provides an intelligent power supply system data comprehensive management method. First, monitoring points are rationally laid out and labeled. Then, various types of data are collected. Next, the data is deeply integrated and analyzed. Then, control commands are generated based on the analysis results. Finally, real-time visualization and feedback adjustments are performed, achieving comprehensive and efficient management of power supply system data and improving the reliability and operating efficiency of the power supply system. The method includes the following steps: S1. Deploy dynamic monitoring points, dynamically optimize the layout of power supply parameter monitoring points based on geographic information system, real-time load data and environmental parameters, and assign a unique device identifier to each monitoring point.

[0032] The dynamic monitoring point layout optimization adopts a reinforcement learning algorithm and automatically adjusts the monitoring point density according to real-time load changes.

[0033] Geographic Information Systems (GIS) can be common commercial GIS software platforms that provide detailed geographic location information and topographic data to help determine the initial layout of monitoring points. Besides commercial GIS software, open-source GIS systems, such as QGIS, can also be used. QGIS offers a wealth of plugins and functionalities and can be customized to meet different needs.

[0034] Real-time load data needs to be acquired through load sensors installed on the power supply lines, which can accurately measure instantaneous power load.

[0035] Obtaining environmental parameters requires various environmental monitoring devices, such as temperature sensors, humidity sensors, and light sensors, which record surrounding environmental information in real time.

[0036] Reinforcement learning algorithms target key nodes such as substations and densely populated user areas, and automatically adjust the density of monitoring points based on real-time load changes. A Deep Q-Network (DQN) algorithm can be used, which can find the optimal monitoring point layout strategy through continuous learning and trial and error. When the real-time load increases, the algorithm increases the density of monitoring points to more accurately grasp the power supply situation; conversely, when the load decreases, it appropriately reduces the density of monitoring points to avoid resource waste. These factors work together to dynamically optimize the layout of monitoring points according to the actual situation, improving the accuracy and effectiveness of monitoring.

[0037] Detailed algorithm process: Problem modeling (Markov decision process): State space (S): Defined as a set of geographic grids covering the entire power supply area. The state feature vector of each grid includes: geographic location coordinates (from GIS), real-time load values, environmental parameters (temperature, humidity), distribution of existing monitoring points, and identifiers of key nodes (such as substations and important users).

[0038] Action Space (A): "Add", "Remove", or "Maintain" a monitoring point within a geographic grid.

[0039] Reward function (R): Designed as a composite reward to guide the agent to learn the optimal layout strategy.

[0040] R(s,a)=α*Coverage(s')+β*LoadMatch(s')+γ*Cost(s,a) In the formula: Coverage(s') is the coverage of the monitoring point to the key node after the execution of action a to reach the new state s'; LoadMatch(s') is the matching degree between the monitoring point density and the real-time load distribution; Cost(s, a) is the action cost, which penalizes unnecessary additions or subtractions of monitoring points to save resources; α, β, and γ are the weighting coefficients for balancing different objectives.

[0041] Algorithm execution, taking Deep Q-Network (DQN) as an example: Initialization: Based on GIS data, such as terrain, buildings, route distribution, and expert rules, an initial monitoring point layout is generated as the initial state s0.

[0042] Loop iteration: a. Policy selection: In the current state s_t, the agent selects an action a_t according to the ε-greedy policy, for example, "adding" monitoring points in a grid with a surge in load but sparse monitoring.

[0043] b. Environmental interaction: Execute action a_t, update the power grid environment to the new monitoring layout state s_{t+1}, and calculate the immediate reward r_t.

[0044] c. Experience storage: Store the transition (s_t, a_t, r_t, s_{t+1}) into the experience replay buffer.

[0045] d. Network training: Sample a batch of experience from the buffer and update the Q-network parameters θ using the following loss function: L(θ)=E[(r+γ*max_{a'}Q(s',a';θ^-)-Q(s,a;θ))^2] In the formula: θ^- represents the target network parameter (periodically copied from θ); γ is the discount factor.

[0046] Layout output: After training, for any given real-time state, by selecting the action sequence that can obtain the maximum long-term reward through the Q network, a dynamically optimized monitoring point layout plan can be output.

[0047] S2. Collect multi-source data, collect power supply system data and environmental data in real time and transmit them to the data center through a wireless communication network.

[0048] The multi-source data collection step includes elements such as power supply system data, environmental data, and wireless communication network.

[0049] The power supply system data covers aspects such as voltage, current, power, and harmonics, and these data can be collected through various meters and monitoring devices installed on power supply equipment.

[0050] The environmental data includes environmental temperature, humidity, and user-side load data, etc., and is also collected by relying on corresponding sensors.

[0051] The wireless communication network uses low-power wide-area network technology or the fifth-generation mobile communication technology, such as LoRa, NB-IoT, etc.

[0052] The 5G network has the characteristics of high speed and low latency, which is suitable for the rapid transmission of a large amount of data; while the low-power wide-area network technology has the advantages of low power consumption and long-distance communication, and is suitable for the data transmission of relatively scattered monitoring points. In order to improve the reliability of data transmission, redundancy checks are added to the transmitted data, such as parity checks, cyclic redundancy checks (CRC), etc., to ensure that the data does not have errors or losses during the transmission process.

[0053] Specific algorithm: The data of each monitoring point is encapsulated into a data frame before being sent. The frame structure includes: frame header, device unique identifier (UID), timestamp, data payload (voltage, current, temperature, etc.), cyclic redundancy check (CRC) code. The sender performs a specific polynomial division operation on the data payload and appends the remainder as the CRC code to the frame tail. The receiver performs the same operation on the received complete frame. If the remainder is zero, it is determined that the data is correct.

[0054] Define a decision function to select the transmission path based on data characteristics and network status: Network=f(Data_Urgency,Data_Volume,Channel_Condition) Logical judgment process: If Data_Urgency==HIGH and Channel_Condition(5G).RTT<Threshold, then preferentially select the 5G network for low-latency transmission.

[0055] If Data_Volume==LARGE and Channel_Condition(5G).Bandwidth>Threshold, then the 5G network will be selected for high-speed transmission.

[0056] If Data_Urgency==LOW and Data_Volume==SMALL, then select LPWAN to reduce power consumption.

[0057] S3. Data integration and analysis: The collected data is correlated with external data and machine learning models are used for anomaly detection, load forecasting, and equipment health assessment.

[0058] The data fusion and analysis process includes elements such as collected data, external data, and machine learning models.

[0059] The collected data consists of power supply system data and environmental data obtained in the previous steps, while external data includes meteorological data, power grid policy information, etc.

[0060] The machine learning models include long short-term memory networks for load prediction, graph neural networks for equipment health assessment, and a transfer learning mechanism to adapt to small sample scenarios.

[0061] Long Short-Term Memory (LSTM) networks can process time-series data and accurately predict future load conditions by learning patterns and regularities from historical load data.

[0062] Graph Neural Networks (GNNs) can model the relationships between devices in a power supply system, analyze the topology between devices, and thus assess the health status of the devices.

[0063] Transfer learning mechanisms can leverage knowledge learned in other related fields or on large datasets and transfer it to current small-sample power supply system data, improving model performance and generalization ability. Furthermore, machine learning models provide interpretability assessments through SHAP analysis tools, validating the reliability of anomaly detection and load forecasting results, and helping operations personnel better understand the model's decision-making process.

[0064] Detailed algorithm process: Data fusion and correlation: Time alignment: For asynchronous data streams from different sources, interpolation is used to align the data to the same timestamp sequence based on a unified clock source.

[0065] Feature association: Construct an association matrix M, where the element M_{i,j} represents the Pearson correlation coefficient between the operating parameters of device i and environmental factor j, which is used to quantify the impact of external factors on the device.

[0066] Load forecasting (based on LSTM): Model input and output: The input is a historical load sequence X of a time window [tn, ..., t-1], and the output is the load forecast y_t at time t.

[0067] LSTM cell internal logic (simplified): Forget gate: f_t =σ(W_f·[h_{t-1},x_t]+b_f), determines which old information to discard.

[0068] Input gate: i_t = σ(W_i·[h_{t-1},x_t] + b_i), = tanh(W_C·[h_{t-1},x_t]+b_C), determines which new information to update.

[0069] Cell state update: C_t = f_t⊙C_{t-1} + i_t⊙ .

[0070] Output gate: o_t=σ(W_o·[h_{t-1},x_t]+b_o), h_t=o_t⊙tanh(C_t).

[0071] Prediction process: The trained LSTM network takes the aligned historical load sequence as input, captures long-term dependencies through the gating mechanism, and finally outputs the predicted value ŷ_t through a fully connected layer.

[0072] Equipment health assessment (based on graph neural network, GNN): Graph construction: The power supply network is modeled as a graph G=(V, E), where nodes v_i∈V represent devices, their feature vectors are the multi-source monitoring data of the devices, and edges e_{ij}∈E represent the physical connections or electrical relationships between devices.

[0073] Message passing and aggregation: A graph convolutional network (GCN) layer is used, where the representation of each node is updated by aggregating its neighbor information. H^{(l+1)}=σ(ÃH^{(l)}W^{(l)}) In the formula: Ã is the normalized adjacency matrix, H^{(l)} is the node feature matrix of the l-th layer, W^{(l)} is the trainable weight matrix, and σ is the activation function.

[0074] Health Score: After passing through multiple layers of GCN, each node obtains a final representation h_i containing topological relationships. This is input into a classifier, which outputs the device's health score Health_Score_i∈ [0,1].

[0075] Transfer learning adapts to small samples: Process: First, pre-train an LSTM or GNN model on a large, general-purpose device running dataset to learn general temporal patterns or graph structure features. Then, retain most of the model's low-level parameters, replace and retrain only the last task-specific layer, and fine-tune it using small sample data from the target power supply system.

[0076] Interpretability Assessment (SHAP): Calculate feature contribution: For any prediction result, the SHAP value φ_i is obtained by calculating the weighted average of the marginal contributions of feature i when it appears in all feature subsets, which satisfies Σφ_i=f(x)- E[f(x)], that is, the SHAP value of each feature explains how much the feature pushes the model prediction from the baseline value.

[0077] S4. Generate optimized control commands: Based on the analysis results, generate power supply control commands using a multi-objective optimization algorithm.

[0078] The process of optimizing control command generation includes elements such as analysis results and multi-objective optimization algorithms.

[0079] The analysis results are an assessment and prediction of the power supply system's operating status derived from the preceding data fusion and analysis steps. The multi-objective optimization algorithm generally employs the Non-Dominated Sorting Genetic Algorithm II (NSGA-II), which can simultaneously optimize economic benefits, carbon emission indicators, and power supply reliability, incorporating carbon emissions as a constraint into the optimization objective.

[0080] Meanwhile, the integrated model predictive control (MPC) framework enables rolling optimization, continuously adjusting control commands based on real-time data to ensure that the power supply system is always in optimal operating condition.

[0081] Multi-objective optimization algorithms comprehensively consider the weights of different factors when balancing various objectives, and flexibly adjust them according to actual needs to achieve sustainable development and efficient operation of the power supply system.

[0082] Detailed algorithm process: Multi-objective optimization modeling (NSGA-II): Decision variables: x=[P_1,P_2, ...,P_n], representing the output or consumption plans of n controllable power generation units or adjustable loads.

[0083] Objective function (to be minimized simultaneously): Economic cost: f1(x)=Σ(C_i(P_i)), where C_i is the power generation cost function of the i-th unit.

[0084] Carbon emissions: f2(x)=Σ(E_i(P_i)), where E_i is the carbon emission function of the i-th unit.

[0085] Power supply reliability risk: f3(x)=g(Reserve_Margin(x), ...), where g is a function for calculating the risk of insufficient system reserve margin.

[0086] Constraints include power balance ΣP_i = Load, upper and lower limits of equipment output P_i_min ≤ P_i ≤ P_i_max, and total carbon emission constraint f2(x) ≤ E_max.

[0087] NSGA-II solution process: a. Initialization: Randomly generate the initial population.

[0088] b. Fast non-dominated ranking: stratifying the population based on the individual's performance across all objectives.

[0089] c. Crowding Calculation: Within the same non-dominated layer, calculate the crowding distance of each individual in the target space to maintain population diversity.

[0090] d. Selection, crossover, and mutation: Based on sorting and crowding, the parent generation is selected through a tournament, and crossover and mutation operations are performed to generate offspring.

[0091] e. Elite Preservation: Merge the parent and offspring generations, perform non-dominated ranking and crowding comparison again, and select the best individuals to form a new generation of population.

[0092] f. Iteration: Repeat be until convergence, and obtain a set of Pareto optimal solutions, that is, solutions that cannot improve one objective without harming other objectives.

[0093] Model Predictive Control (MPC) Rolling Optimization: In each control cycle k: a. Measurement: Obtain the current system state x(k) (from the analysis results of step S3, such as actual load and equipment status).

[0094] b. Prediction: Predicting the system behavior N steps ahead based on the model.

[0095] c. Optimization: Using the current state as the initial condition, solve the above multi-objective optimization problem in a finite time domain (k to k+N) to obtain the optimal control sequence U*={u*(k), u*(k+1),...,u*(k+N-1)}.

[0096] d. Execution: Only the first control instruction u*(k) in the sequence is executed.

[0097] e. Feedback: In the next period k+1, this process is repeated to correct the prediction error with the new measurement value, thus achieving closed-loop optimization.

[0098] S5. Real-time visualization and feedback: Data visualization is achieved through building information modeling and augmented reality interface, and an adaptive feedback loop is established.

[0099] Real-time visualization and feedback steps include elements such as building information modeling, augmented reality interface, and adaptive feedback loop.

[0100] Building Information Modeling (BIM) and physical equipment are synchronized in two directions in real time through digital thread technology, which can accurately reflect the actual operation of the power supply system. The augmented reality interface supports AR glasses interaction for on-site maintenance guidance. By wearing AR glasses, staff can intuitively see the status information of the equipment, maintenance suggestions, etc.

[0101] The adaptive feedback closed loop dynamically adjusts the control commands based on actual operating data, forming a continuous optimization process.

[0102] When real-time operating data deviates from the expected target, the system will automatically analyze the cause and correct the deviation by adjusting control commands to ensure the stability and reliability of the power supply system.

[0103] Specific algorithm and process: BIM Model Synchronization (Digital Thread): Data mapping: Establish a one-to-one mapping table between physical device UIDs and component IDs in the BIM model.

[0104] State synchronization: Define a state update function Update_BIM(BIM_Model, Physical_Data). When new data Data_i of device i is received from step S2, the system finds the corresponding component j in the BIM according to the mapping table and updates its attribute set in real time, such as: BIM_Model.Component[j].Attribute["Real-time temperature"]=Data_i.Temperature.

[0105] Adaptive feedback closed-loop control: Deviation calculation: In each control cycle, calculate the deviation between the set value V_set and the actual value V_actual (from S2) of the key operating indicators (such as bus voltage V): e(t) = V_set - V_actual(t).

[0106] Control law adjustment: The adjustment amount Δu(t) of the control command is generated using an incremental PID control algorithm. Δu(t)=K_p*[e(t)-e(t-1)]+K_i*e(t)+K_d*[e(t)-2e(t-1)+e(t-2)] In the formula: K_p, K_i, K_d are the proportional, integral, and differential coefficients.

[0107] Parameter self-tuning: The system can optimize PID parameters online based on historical deviation data {e(1), e(2), ..., e(t)}, using methods such as reinforcement learning (state is deviation mode, action is adjustment (K_p, K_i, K_d), reward is reduction of deviation), so that the feedback control loop has adaptiveness.

[0108] The following example, using a power plant in East China, illustrates the implementation steps of the method embodiment of this application.

[0109] It includes a main gas-fired power plant (G1), a supporting photovoltaic power plant (PV1), several key substations (Sub_A, Sub_B) and a mixed commercial-residential load area. The load in this area is significantly affected by weather and holidays, and the aging equipment issues require refined management.

[0110] Step S1: Deploy dynamic monitoring points Problem modeling: State space: The area under the jurisdiction of the city's power grid is divided into 1km x 1km grids (e.g., 100 grids). Example of state feature vector: grid coordinates (x, y), current average load (300kW), ambient temperature (30°C), number of monitoring points in the current grid (1), whether it is a critical node (the grid where Sub_A is located is marked as 1).

[0111] Action space: For each grid, the action set is {+1, 0, -1}, which respectively represent adding, maintaining, and removing a monitoring point in that grid.

[0112] Example of reward function calculation: Suppose that in a certain state, the algorithm performs the action a=+1 (add monitoring points) on grid g23 where the load increases dramatically.

[0113] Coverage(s'): Under the new layout, the coverage of monitoring points to key nodes such as Sub_A and Sub_B increases from 85% to 90%, and Coverage(s') = 0.90.

[0114] LoadMatch(s'): The load density of grid g23 is "high". After adjustment, the matching degree of its monitoring point density is improved, and the calculated spatial correlation coefficient increases from 0.7 to 0.8. LoadMatch(s') = 0.80.

[0115] Cost(s,a): The cost of adding a monitoring point is -0.1.

[0116] Let the weights be α=1.0, β=0.8, and γ=0.5. Then the immediate reward is: R=1.0*0.90+0.8*0.80+0.5*(-0.1)=1.44.

[0117] Algorithm execution: After training, the DQN network outputs an action sequence under the real-time condition of "high temperature in the summer afternoon combined with commercial area promotion". The final decision is to add 2 monitoring points in the commercial area (grids g23, g40) and reduce 1 monitoring point in the lightly loaded residential area (grid g67), thereby achieving dynamic optimal allocation of monitoring resources.

[0118] Step S2: Collect multi-source data Data packet and CRC check: The newly added g23 grid monitoring point collected the following data: voltage 235.4V, current 1200A, temperature 65°C. The data payload is represented in binary as D.

[0119] Sending end: Perform modulo-2 division on D using CRC-16-CCITT polynomial 0x1021 to obtain remainder R=0x3A5F, and append it to the end of the data frame.

[0120] Transmission: Since this is overload alarm data (Data_Urgency = HIGH) and the current 5G network round-trip time (RTT) is 20ms < threshold (50ms), the system chooses to upload it urgently via the 5G network.

[0121] The receiving end performs the same polynomial division operation on the received complete frame containing the CRC code. If the remainder is 0x0000, the data is deemed to be correct and is received.

[0122] S3: Data Integration and Analysis Data fusion: The current data collected per second at monitoring point g23 was aligned with the temperature data per second provided by the meteorological bureau through linear interpolation to the same millisecond time series. The calculation showed that the Pearson correlation coefficient M_{pv,light} = 0.92 between the output power P_pv of the photovoltaic power station PV1 and the ambient light intensity I_light, which is a strong correlation, indicating that the weather has a great impact on the output of new energy.

[0123] Load forecasting (LSTM example): Input the city’s historical load sequence X (e.g., [500, 480, ... 300, 350] MW) for the past 24 hours (t-24 to t-1).

[0124] At time t, the LSTM unit calculates: The forgetting gate f_t=σ(W_f·[h_{t-1},x_t]+b_f) determines to discard some of the nighttime load memories that are not currently relevant.

[0125] Input gate i_t and candidate state _t Learn about the current afternoon load increase pattern.

[0126] Update cell status: .

[0127] The output gate o_t controls the output and obtains the hidden state h_t.

[0128] Finally, the fully connected layer maps h_t to the load forecast for the next hour (t+1): ŷ_{t+1} = 620MW.

[0129] Equipment health assessment (GNN example): Graph construction: Gas turbine G1, transformer T1, and line L1 are nodes, and the connection relationships are edges.

[0130] Message passing: For transformer node T1, its characteristic h_{T1}^{(0)} includes its oil temperature and winding temperature. Within a single GCN layer, it aggregates messages from neighbors G1 (output) and L1 (current): h_{T1}^{(1)}=σ(÷[h_{G1}^{(0)};h_{T1}^{(0)}; h_{L1}^{(0)}]·W^{(0)}) Where à is a normalized adjacency matrix sub-block containing the connection relationships between T1 and G1, L1.

[0131] Health score: After two layers of GCN, the final representation h_{T1}^{(2)} of T1 is input into the classifier, and the output Health_Score_{T1} = 0.65 (low score, warning).

[0132] Interpretability Assessment (SHAP): For a T1 health score of 0.65, SHAP analysis shows that oil temperature characteristics contributed -0.2 points (lowering the score the most), load current characteristics contributed -0.1 points, and ambient temperature contributed -0.05 points. This clearly tells maintenance personnel that excessively high oil temperature is a major risk point and the cooling system should be checked immediately.

[0133] Step S4: Generate optimized control instructions Optimization modeling: Decision variables: x = [P_G1, P_PV1, L_curtail], where P_G1 is the output of the gas-fired power plant, P_PV1 is the output of the photovoltaic power plant, and L_curtail is the interruptible load reduction.

[0134] Objective function (to be minimized): f1(x) = a * P_G1^2 + b * P_G1 + c (Cost of gas-fired power generation, a quadratic function) f2(x) = e * P_G1 (carbon emission cost, linearly related to gas-fired power output) f3(x) = Risk(P_G1, P_PV1) (Power outage risk calculated based on the predicted load of 620MW and reserve margin) Constraints: P_G1 + P_PV1 - L_curtail = 620 (power balance), 0 ≤ P_PV1 ≤ 100 (maximum photovoltaic output), L_curtail ≤ 20 (maximum interruptible load).

[0135] Solving NSGA-II: After the algorithm runs, it yields a Pareto optimal solution set, for example: Option A (Economy Priority): [P_G1=550, P_PV1=70, L_curtail=0], low cost but high carbon emissions.

[0136] Option B (low carbon priority): [P_G1=520, P_PV1=100, L_curtail=0], has the lowest carbon emissions but higher costs.

[0137] Option C (Balanced): [P_G1=530, P_PV1=90, L_curtail=0], which balances the two.

[0138] MPC scrolling optimization: In the current control cycle k, the system administrator selects scheme C. Only u*(k) = {P_G1=530, P_PV1=90} is executed. 15 minutes later (cycle k+1), the measured load is 615MW (lower than predicted), and the enhanced sunlight allows PV1 to achieve an actual output of 95MW. MPC immediately re-optimizes in the new state, and may issue a new instruction in cycle k+1: u*(k+1) = {P_G1=525, P_PV1=95}, achieving dynamic and precise tracking.

[0139] Step S5: Real-time visualization and feedback BIM-AR Synchronization: In the BIM 3D model of the power plant, the component IDBIM_003 of transformer T1 has been mapped to the physical device T1's UIDDT_2023_001. When step S2 collects the real-time oil temperature of T1 as 85°C, the system automatically executes Update_BIM(BIM_Model, Data_T1), updates BIM_Model.Component["BIM_003"].Attribute["Oil Temperature"] to "85°C", and displays it on the model in a highlighted warning color.

[0140] Adaptive feedback control: The bus voltage control circuit setpoint V_set = 10.5kV, and the actual measured value V_actual(t) = 10.3kV.

[0141] Deviation calculation: e(t) = 10.5 - 10.3 = 0.2kV.

[0142] PID adjustment: Based on the incremental PID formula Δu(t)=K_p*[e(t)-e(t-1)]+K_i*e(t)+ K_d*[e(t)-2e(t-1)+e(t-2)], the required additional reactive power compensation device output ΔQ = 15MVar is calculated.

[0143] Parameter self-tuning: Due to frequent voltage fluctuations after the connection of new energy sources to this line, the built-in reinforcement learning module automatically adjusts the PID parameters from (K_p=1.0, Ki=0.5, K_d =0.1) to (K_p=1.2, Ki=0.3, K_d=0.2) by analyzing recent voltage deviation patterns, in order to provide faster and more stable voltage control.

[0144] The implementation principle of this embodiment is as follows: This intelligent power supply system data comprehensive management method, supported by multi-source data and advanced algorithms, realizes intelligent management of the entire power supply system from monitoring to control. Through dynamic monitoring point deployment, various data of the power supply system can be acquired in a timely and accurate manner; multi-source data acquisition and fusion analysis provide a deeper understanding of the operating status of the power supply system; optimized control command generation takes into account multiple objectives such as economy, environmental protection, and reliability based on actual conditions, making the operation of the power supply system more scientific and rational; real-time visualization and feedback enable human-computer interaction and automatic system adjustment, facilitating operation by maintenance personnel and continuous system optimization. Compared with existing technologies, it overcomes problems such as lack of dynamic adaptability in monitoring point layout, insufficient model robustness, and data synchronization delay, greatly improving the management efficiency and reliability of the power supply system. Example

[0145] This embodiment is based on Embodiment 1. The difference between this embodiment and Embodiment 1 is that the intelligent power supply system data integrated management system provided in this embodiment includes a dynamic monitoring module, a data acquisition and transmission module, an AI analysis module, an optimization control module, a visualization interaction module, and a central processing unit.

[0146] The dynamic monitoring module integrates a Geographic Information System (GIS) and IoT sensors to deploy monitoring points. The GIS provides geolocation and spatial analysis capabilities, while the IoT sensors are responsible for real-time data collection from the power supply system and the environment. The data acquisition and transmission module collects data on voltage, current, power, harmonics, ambient temperature and humidity, and user-end load, and transmits it via wireless communication networks such as 5G or Low Power Wide Area Network (LPWAN), while adding redundancy checks to the transmitted data. The AI ​​analysis module includes a machine learning model library, with Long Short-Term Memory (LSTM) networks used for load forecasting, graph neural networks for equipment health assessment, and a transfer learning mechanism to adapt to small sample scenarios. It also provides interpretability assessment through the SHAP analysis tool. The optimization control module embeds a non-dominated sorting genetic algorithm II for multi-objective optimization and integrates a model predictive control framework to achieve rolling optimization, incorporating carbon emissions as a constraint into the optimization objective to generate power supply control commands. The visualization and interaction module combines Building Information Modeling (BIM) and Augmented Reality (AR) technologies, achieving bidirectional real-time synchronization with physical devices through digital thread technology. It supports AR glasses interaction to guide on-site maintenance, realizing data visualization and interaction. The central processing unit coordinates the operation of each module, exchanges data with third-party systems through application programming interfaces, and coordinates the collaborative operation of the dynamic monitoring module, data acquisition and transmission module, and optimization control module.

[0147] The implementation principle of this embodiment is as follows: This intelligent power supply system data management system organically combines various functional modules and coordinates and manages them uniformly through a central processing unit. The dynamic monitoring module provides basic data for the system, the data acquisition and transmission module ensures effective data transmission, the AI ​​analysis module performs in-depth data mining and analysis, the optimization control module generates reasonable control commands based on the analysis results, and the visualization interaction module allows maintenance personnel to intuitively understand the system's operating status and perform operations. The entire system forms a complete closed loop, enabling real-time and efficient management of power supply system data, improving the operating efficiency and reliability of the power supply system. It also possesses good scalability and compatibility, allowing data exchange and collaborative work with third-party systems, adapting to different application scenarios and needs. Compared with existing technologies, it solves the problems of existing systems in terms of dynamic adaptability, model robustness, and data synchronization, providing a more effective solution for the intelligent management of power supply systems.

[0148] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A data integrated management method for intelligent power supply system, characterized in that, Includes the following steps: S1. Deploy dynamic monitoring points, dynamically optimize the layout of power supply parameter monitoring points based on geographic information system, real-time load data and environmental parameters, and assign a unique device identifier to each monitoring point; S2. Collect multi-source data, collect power supply system data and environmental data in real time, and transmit them to the data center through a wireless communication network; S3. Data integration and analysis: The collected data is correlated with external data and machine learning models are used for anomaly detection, load forecasting and equipment health assessment. S4. Generate optimized control commands: Based on the analysis results, generate power supply control commands using a multi-objective optimization algorithm. S5. Real-time visualization and feedback: Data visualization is achieved through building information modeling and augmented reality interface, and an adaptive feedback loop is established.

2. The data integrated management method of the intelligent power supply system according to claim 1, characterized in that, The dynamic monitoring point layout optimization adopts a reinforcement learning algorithm and automatically adjusts the monitoring point density according to real-time load changes.

3. The intelligent power supply system data integrated management method according to claim 1, characterized in that, The machine learning model includes a long short-term memory network for load forecasting and a graph neural network for equipment health assessment. The machine learning model incorporates a transfer learning mechanism to adapt to small sample scenarios.

4. The data integrated management method of the intelligent power supply system according to claim 1 or 3, characterized in that, The machine learning model is evaluated for interpretability using the SHAP analysis tool to verify the reliability of anomaly detection and load prediction results.

5. The data integrated management method of the intelligent power supply system according to claim 1, characterized in that, The wireless communication network employs low-power wide-area network technology and adds redundancy checks to the transmitted data to improve reliability.

6. An intelligent power supply system data integrated management system, which is suitable for the intelligent power supply system data integrated management method of any one of claims 1-5, characterized in that, include: The dynamic monitoring module integrates geographic information systems and IoT sensors to deploy monitoring points; The data acquisition and transmission module collects voltage, current, power, harmonics, ambient temperature and humidity, and user-end load data and transmits them through a wireless communication network. The AI ​​analytics module includes a library of machine learning models to perform data fusion, anomaly detection, load forecasting, and equipment health assessment. The control module is optimized by embedding a multi-objective optimization algorithm to generate power supply control commands. The visualization and interaction module combines building information modeling and augmented reality technology to achieve data visualization and interaction. The central processing unit coordinates the operation of various modules and provides integration interfaces for third-party systems.

7. The intelligent power supply system data integrated management system according to claim 6, characterized in that, The optimization control module uses a non-dominated sorting genetic algorithm II for multi-objective optimization and integrates a model predictive control framework to achieve rolling optimization.

8. The data integrated management system of the intelligent power supply system according to claim 7, characterized in that, The non-dominated sorting genetic algorithm II simultaneously optimizes economic benefits, carbon emission indicators, and power supply reliability, and incorporates carbon emission as a constraint into the optimization objective.

9. The data integrated management system of the intelligent power supply system according to claim 6, characterized in that, The visualization interaction module achieves bidirectional real-time synchronization with physical devices through digital thread technology and supports interaction with augmented reality glasses to guide on-site repairs.

10. The data integrated management system of the intelligent power supply system according to claim 6, characterized in that, The central processing unit exchanges data with third-party systems through an application programming interface and coordinates the collaborative operation of the dynamic monitoring module, data acquisition and transmission module, and optimization control module.

Images