Self-adaptive power grid state transformer area intelligent fusion terminal and control method thereof

Abstract

The invention discloses a self-adaptive power grid state court intelligent fusion terminal and a control method thereof, relates to the technical field of intelligent power grids, and solves the defects that an existing fusion terminal is insufficient in self-adaptive capability and limited in data processing and analysis capability. Learning power grid state change in real time through a deep power grid dynamic control algorithm and a prediction analysis engine combining LSTM and random forest, and autonomously adjusting a control strategy; the edge calculation and distributed processing module is utilized, so that the data processing flow is optimized, and the data processing speed and efficiency are greatly improved; the application of time sequence analysis and a machine learning model further excavates and understands a complex data mode; the resource configuration efficiency is improved through an intelligent scheduling algorithm, and a cross-layer coordination mechanism is established through a distributed coordination algorithm, so that rapid issuing and feedback collection of a control instruction are ensured; according to the invention, the intelligent degree of the existing fusion terminal is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of smart grid technology, and more specifically to an adaptive grid status substation smart fusion terminal and a control method thereof. Background Art

[0002] As the pace of global energy transformation accelerates, smart grids have become an important cornerstone for promoting green and low-carbon development. At the beginning of the 21st century, the deep integration of information technology and traditional power grids gave rise to the concept of smart grids, which aims to achieve a high degree of automation, informatization and interactivity in the power system. As a micro unit of the smart grid, the level of intelligence of the substation is directly related to the efficiency and reliability of the entire power grid. Against this background, the substation intelligent fusion terminal came into being and became the key link between the distribution network and user-side equipment. With the large-scale access of new loads such as distributed energy, electric vehicles, and smart homes, higher requirements have been put forward for the flexibility, real-time and intelligence of substation management, which has promoted the rapid development of substation intelligent fusion terminal technology, aiming to improve the operation efficiency and service quality of the power grid by integrating measurement, control, communication and other functions, and meet the needs of the energy Internet era.

[0003] The intelligent fusion terminal in the substation combines modern information technology and power system technology, and integrates multiple functions such as advanced measurement, remote control, two-way communication, and data processing. At the hardware level, high-performance microprocessors and embedded operating systems are used to ensure the speed and stability of data processing; at the software level, the Internet of Things technology, cloud computing platform, and big data analysis technology are used to achieve efficient collection, storage, and analysis of massive data. In addition, in order to adapt to the complex power grid environment, the terminal also adopts anti-interference design and a wide temperature operating range to ensure stable operation under harsh conditions. With the rise of emerging technologies such as 5G, artificial intelligence, and edge computing, the intelligent fusion terminal in the substation is gradually evolving towards a more efficient and intelligent direction.

[0004] Although the technology of intelligent fusion terminals in substations has made significant progress, it still faces several challenges in practical applications. First, insufficient adaptive capability is one of the key issues. Currently, most terminals rely on preset rules to respond to changes in grid status, lacking real-time response and adaptive adjustment capabilities to complex and dynamic grid environments, affecting the flexibility and efficiency of grid operation. Secondly, data processing and analysis capabilities are limited, especially when faced with large-scale, high-dimensional data. The processing speed and analysis depth of existing terminals are not enough to support deep insights and accurate decision-making, limiting their application potential in intelligent optimization, fault prediction, and other aspects.

[0005] Therefore, the present invention discloses a station area intelligent fusion terminal and a control method thereof; Summary of the invention

[0006] In view of the deficiencies in the prior art, the present invention discloses an adaptive grid status substation intelligent fusion terminal and a control method thereof, which solves the shortcomings of the existing substation intelligent fusion terminal having insufficient adaptive capability and limited data processing and analysis capabilities.

[0007] The present invention adopts the following technical solution: an adaptive grid status area intelligent fusion terminal, comprising:

[0008] Terminal layer, edge layer and cloud platform layer;

[0009] The terminal layer includes a data acquisition module and a preprocessing module. The perception and data acquisition module collects real-time data from the power grid through the Internet of Things sensor network, including but not limited to voltage, current and frequency indicators; the preprocessing module preprocesses the data through abnormal value detection, time series analysis and data formatting methods, and the preprocessed data is uploaded to the adaptive learning and control module and the edge computing and distributed processing module of the edge layer through a standardized communication interface;

[0010] Based on the real-time data uploaded by the terminal layer; the adaptive learning and regulation module dynamically adjusts the control strategy according to the power grid state through the deep power grid dynamic control algorithm, and predicts the future state of the power grid through the power grid prediction and analysis model; the power grid prediction and analysis model predicts the time series through the long short-term memory network, and evaluates the prediction content through the random forest to provide a decision-making basis for regulation;

[0011] The edge computing and distributed processing module realizes real-time data processing and load balancing through dynamic container orchestration and chain services, and reduces data transmission volume and storage costs through Huffman coding and incremental update strategies;

[0012] The processed data is uploaded to the cloud platform layer through the standardized communication IEC 61850 protocol; the cloud platform layer includes a multi-layer communication and information fusion module and a global optimization and collaborative control module; the multi-layer communication and information fusion module realizes data aggregation, fusion and distribution through the data lake and message queue service; the global optimization and collaborative control module receives and analyzes the comprehensive data of the data lake through the intelligent scheduling model, and ensures that the upper-level decision-making information is timely delivered to the terminal layer through the cross-layer collaboration mechanism, and collects feedback to form a closed-loop control; the intelligent scheduling model performs global resource optimization through a genetic algorithm, and copes with uncertainty through fuzzy logic control; the cross-layer collaboration mechanism establishes information transmission and task collaboration between the upper and lower layers through a distributed coordination algorithm.

[0013] As a further technical solution of the present invention, the distributed coordination algorithm includes a communication layer, a consensus layer, a state management layer, a task allocation layer, a data synchronization layer, a security layer and a fault tolerance layer; the distributed coordination algorithm works in the intelligent fusion terminal of the station area with adaptive power grid status as follows:

[0014] 801. Data collection and communication; data transmission and reception are realized between different network nodes through the communication layer; the communication layer realizes decentralized data exchange and communication through the P2P network protocol;

[0015] 802. Data consistency and consensus formation: reaching a consistent state or decision among distributed nodes through the consensus layer, and the consensus layer uses a Byzantine fault tolerance algorithm to ensure that even if some nodes send erroneous or malicious information, the entire network can reach a consensus on data consistency;

[0016] 803. State synchronization and state machine replication; maintaining and updating the system state, including node data and network state, through the state management layer; the state management layer ensures the state consistency of all nodes through the state machine replica;

[0017] 804, task allocation and resource scheduling; the task allocation layer allocates tasks to corresponding nodes according to the current system status and resources; the task allocation layer implements effective task allocation and optimal use of resources through load balancing algorithms and resource scheduling strategies;

[0018] 805. Data synchronization and version control: Synchronize the data in the distributed ledger or database through the data synchronization layer to ensure that the data copies of all nodes are kept up to date; the data synchronization layer realizes rapid data update and historical version management through incremental data synchronization and version control mechanism;

[0019] 806. Security assurance: The security layer protects the security of network communications and data storage to prevent unauthorized access and tampering; the security layer ensures the confidentiality, integrity and non-repudiation of data through symmetric encryption methods and digital signatures;

[0020] 807. Fault-tolerant processing; detecting and recovering system failures through the fault-tolerant layer; the fault-tolerant layer implements rapid response and processing to system anomalies through fault detection algorithms and redundancy mechanisms.

[0021] As a further technical solution of the present invention, the working steps of the deep power grid dynamic control algorithm are as follows:

[0022] Step 1: Multi-dimensional feature extraction of environmental status;

[0023] Construct a multi-dimensional feature vector S t ={s t ,it ,p t ,q t ,f t ,r t}, where s t represents the node voltage, i t Indicates the current load, p t and q t Represent active power and reactive power respectively; f t Indicates frequency; r t represents the phase angle;

[0024] Step 2: Q function approximation and deep neural network construction;

[0025] After the feature vector is constructed, the Q value of each state-action pair is approximated by a deep neural network; the deep neural network is trained by a loss function L(θ), and the loss function L(θ) is defined by the time difference error ρ, and the formula expression of the time difference error ρ is:

[0026] ρ=1-ω t+1 [γ(1-θ)]+γ(ω t+1 -Q(S t ,a t ;θ)) (1)

[0027] In formula (1), ω t+1 is the immediate return, which means the feedback obtained at time t+1, used to evaluate the effect of the action; γ is the discount factor; Q(S t ,a t ; θ) represents the Q function, where a t In state S t The action taken under, θ represents the network parameters;

[0028] Step 3: Design of reward mechanism and strategy evaluation;

[0029] Define a multi-objective reward function to guide the learning direction of the deep neural network. The multi-objective reward function is defined by weight coefficients α, β, γ loss Balancing the importance of different objectives, the formula expression of the multi-objective reward function is:

[0030]

[0031] In formula (2), α, β, γ loss is the weight coefficient, which is used to balance the priorities of different goals; n load represents the load current, β represents the power load intensity of the power system, and n loss Indicates power loss; n dev Indicates voltage deviation;

[0032] Step 4: Online learning and updating of strategies;

[0033] As the state of the power grid changes in real time, the online learning of the model based on real-time data is realized through the gradient descent update function. The formula expression of the gradient descent update function is:

[0034]

[0035] In formula (3), θ new represents the updated network parameters, which are used to calculate the action value in the new state; K represents the learning rate, which is used to control the pace of parameter updating during the learning process; θ is the current network parameter, which represents the current learning state of the deep neural network model; n represents all possible actions that can be taken in a given state; Represents the gradient of the loss function L(θ) to θ; the model parameter θ is adjusted according to the gradient Update with learning rate K;

[0036] Step 5: Implementation and feedback of adaptive control strategy;

[0037] According to the updated policy θ new , through the action selection function to calculate the new state St +1 The best action under The formula expression of the action selection function is:

[0038]

[0039] In formula (4), Indicates state S t+1 The optimal action under the given state; n represents all possible actions to be taken under a given state; execute the action And observe the new state S t+2 and immediate returnω t+2 , collect feedback information for the next round of learning and strategy updating.

[0040] As a further technical solution of the present invention, the power grid prediction and analysis model includes an input layer, a multi-dimensional feature fusion layer, a time series analysis layer, an advanced feature engineering layer, a hybrid prediction model layer, a prediction calibration and optimization layer, an adaptive dynamic adjustment layer, a control strategy optimization layer and a feedback loop layer; the working steps of the power grid prediction and analysis model in the station area intelligent fusion terminal of the adaptive power grid state are:

[0041] T1. Obtaining power grid data, including real-time monitoring data and historical records, through the input layer; the input layer integrates data from different sources and sensors through a Kalman filter to achieve data consistency and integrity;

[0042] T2. Extracting and selecting features from the integrated data through the multidimensional feature fusion layer; the multidimensional feature fusion layer evaluates the feature importance of the power grid historical data through Gaussian process regression, and the Gaussian process regression captures the complex relationship between features through the kernel method to provide a basis for feature selection;

[0043] T3. The output of the multi-dimensional feature fusion layer is received through the time series analysis layer, and the time series data of the power grid state is analyzed; the time series analysis layer captures the long-term dependency of the power grid parameters through the long short-term memory network; the long short-term memory network predicts the short-term change of the power grid state through the state update and gating mechanism; the formula expression of the state update is:

[0044]

[0045] In formula (5), h t is the hidden state of the current time step, containing the information of the time series; h t-1 is the hidden state of the previous time step, x t is the input feature of the current time step, W and U are weight matrices, corresponding to the conversion of input and hidden states respectively, and b is the bias term;

[0046] T4, further extracting and constructing features that have a significant impact on prediction through the advanced feature engineering layer; the advanced feature engineering layer reduces feature dimensions through linear transformation and reveals the intrinsic pattern of the signal through local time-frequency analysis;

[0047] T5. Predictions are made through the hybrid prediction model layer; the hybrid prediction model layer processes and predicts complex time series data through a hybrid prediction weighted formula; the hybrid prediction weighted formula predicts through a long short-term memory network LSTM, and weights and corrects the output of the LSTM through the integration characteristics of the random forest RF; the formula expression of the hybrid prediction weighted formula is

[0048]

[0049] In formula (6), is the mixed prediction result, z is the weighting coefficient, which is used to balance the importance of LSTM and RF prediction results. is the LSTM prediction result, is the prediction result of random forest; c t is the cell state at the current time step;

[0050] T6. Optimizing the performance of the prediction model through the prediction calibration and optimization layer; the prediction calibration and optimization layer tunes the model hyperparameters through Bayesian optimization; the Bayesian optimization finds the optimal hyperparameter combination by constructing a probability model of the hyperparameters and using a sampling strategy;

[0051] T7. The prediction model is dynamically adjusted according to the real-time feedback of the power grid state by the adaptive dynamic adjustment layer; the adaptive dynamic adjustment layer dynamically adjusts the prediction model by the Kalman filter prediction update formula: the Kalman filter prediction update formula uses the difference between the observed value and the prior prediction to dynamically update the prediction state by gain calculation; the formula expression of the Kalman filter prediction update formula is:

[0052]

[0053] In formula (7), V k is the Kalman gain, which is used to measure the impact of the observation value on the predicted state. is the predicted state based on prior knowledge, is the state estimation based on prior prediction, and B is the actual observation value;

[0054] T8. Based on the prediction results, the control strategy optimization layer optimizes the control strategy of the power grid; the control strategy optimization layer optimizes the power grid control strategy through the deep Q network; the deep Q network updates the strategy through the strategy optimization objective function, and the formula expression of the strategy optimization objective function is:

[0055]

[0056] In formula (8), P new is the updated policy network parameter, P is the current policy network parameter, m is the learning rate, which is used to control the speed of parameter update, Q is the expected value, which represents the statistical average of state-action pairs; p is the state-action distribution under the current policy; y is the policy network, which is used to measure the probability of taking an action in a given state;

[0057] T9, based on the prediction error and the actual response of the power grid; a closed-loop control system is constructed through the feedback loop layer, and the feedback loop layer continuously updates the model parameters through the recursive least squares method; the parameter update formula of the recursive least squares method is:

[0058]

[0059] In formula (9), F t+1 is the updated parameter vector, F tis the parameter vector at time t, q is the prediction error, which is used to measure the difference between the model prediction and the actual observation, l is the forgetting factor, which is used to control the weight of historical data, and θ is the input data; f t A weight vector representing the parameter vector.

[0060] As a further technical solution of the present invention, the edge computing and distributed processing module realizes lightweight encapsulation of applications through a containerization method; the containerization method enhances the automation and flexibility of container management through container orchestration; the container orchestration allows users to define the expected cluster state through declarative configuration management, and dynamically adjusts the number of container instances according to real-time load through an automatic expansion mechanism; the chained service splits complex tasks into independent microservices through a microservice architecture; these microservices perform unified access control and routing through an API gateway to ensure the decoupling and scalability of services; the API gateway evenly distributes requests to each service instance through a load balancing algorithm; the containerization method provides intelligent routing, traffic management, policy execution and elastic control between microservices through a service grid, and realizes rapid iteration and deployment of services through a continuous integration / continuous deployment mechanism; the incremental update strategy improves update efficiency through a difference detection algorithm and a data synchronization mechanism; the difference detection algorithm identifies and records differences between data sets through hash summaries, data chunking, timestamps and difference encoding.

[0061] As a further technical solution of the present invention, the multi-layer communication and information fusion module includes a data aggregation submodule, a data fusion submodule, a message queue service submodule, a data distribution submodule, a cross-layer information interaction submodule and a collaborative control feedback submodule; the data aggregation submodule realizes data interaction with the terminal layer through a standardized communication protocol, and performs standardized processing on the accessed data through data cleaning and format conversion methods; the data fusion submodule integrates data from different sources and types through a timestamp-based fusion strategy; and realizes the storage and management of large-scale unstructured data through a distributed file system and a NoSQL database; the timestamp-based fusion strategy fuses and transforms the original data through data abstraction and pattern recognition to provide data support for the intelligent scheduling model; the distributed file system optimizes data access efficiency through data partitioning and indexing mechanisms; the message queue service submodule publishes / The subscription mode realizes asynchronous transmission and decoupling of data, and ensures the reliability of message transmission through the data persistence mechanism; the data distribution submodule includes a data routing unit and a data access control unit; the data routing unit distributes the fused data to different processing modules or analysis engines through routing algorithms and load balancing methods; the data access control unit ensures the security and compliance of the data distribution process through access control lists and identity authentication mechanisms; the cross-layer information interaction submodule realizes data conversion through abstract data models, and realizes the consistency of data formats and semantics between different levels through data mapping; the collaborative control feedback submodule includes a feedback collection unit and a decision feedback unit; the feedback collection unit collects the execution results and status information of the terminal layer through a feedback mechanism; the decision feedback unit transmits the optimization decisions of the upper layer to the terminal layer through an event-driven mechanism and a message queue, and adjusts the decisions based on the feedback.

[0062] As a further technical solution of the present invention, the intelligent dispatching model includes a data input layer, a feature extraction layer, a feature encoding layer, a genetic algorithm population initialization layer, a fitness function calculation layer, a selection layer, a genetic operation layer, an elite retention and population update layer, a decision output layer and an execution and feedback layer; the working method steps of the intelligent dispatching model in the intelligent fusion terminal of the station area with adaptive power grid status are as follows:

[0063] S1. Obtaining raw data from sensors and databases through the data input layer, including but not limited to equipment status, environmental parameters and demand forecasts;

[0064] S2, extracting features through the feature extraction layer; the feature extraction layer removes noise through time series analysis and wavelet transform, and refines effective features through principal component analysis;

[0065] S3, converting the high-dimensional feature vector generated by the feature extraction layer into binary code through the feature coding layer;

[0066] S4, based on the coding features, generating an initial solution set through the genetic algorithm population initialization layer as the starting point of the optimization search;

[0067] S5. Evaluate the quality of each solution in the population through a fitness function calculation layer, wherein the fitness function calculation layer considers factors including energy allocation efficiency, system stability, and cost-effectiveness;

[0068] S6. Based on the individual fitness values ​​output by the upper layer, the selection layer selects individuals to participate in reproduction, and the selection layer makes selection decisions through a proportional selection method;

[0069] S7, simulating natural selection and genetic mechanism through the genetic operation layer; the genetic operation layer improves population diversity and enhances global search capability through crossover and mutation;

[0070] S8. At the end of each generation, the optimal solution in each generation is retained through the elite retention and population update layer, and the population is updated in combination with the newly generated offspring;

[0071] S9, converting the optimal solution obtained by optimization into actual operation instructions through the decision output layer to guide power grid control and resource allocation; the decision output layer decodes the optimal solution back into decision instructions or strategy recommendations through reverse conversion coding;

[0072] S10, formulating a strategy based on the decision output layer; implementing it through the execution and feedback layer and collecting feedback data on the implementation effect for adjusting model parameters or optimizing strategies.

[0073] As a further technical solution of the present invention, the data acquisition module ensures the time consistency of sensor data through a timestamp synchronization mechanism, and realizes temporary storage of data through a circular queue to cope with the problem of mismatch between data acquisition and processing speed; the circular queue tracks the data currently stored in the buffer through a circular index mechanism to optimize space utilization, and ensures data consistency and thread safety through mutex locks and semaphores; the outlier detection identifies and processes outliers in the data through a statistical threshold detection method; the time series analysis performs time series prediction and trend analysis through an autoregressive integrated sliding average model; the data formatting method converts raw data into a standardized format through a data encapsulation protocol JSON.

[0074] As a further technical solution of the present invention, a method for controlling an adaptive power grid status substation intelligent fusion terminal includes the following steps:

[0075] Step 1: Intelligent perception and data collection: Real-time monitoring of grid operation parameters through sensor networks, smart meters and environmental monitoring equipment, and real-time elimination of abnormal data through threshold filtering methods; grid operation parameters include but are not limited to voltage, current, frequency, temperature and humidity parameters;

[0076] Step 2: Anomaly detection and data preprocessing: Receive preliminary clean data from the perception layer and further filter it through the outlier detection algorithm to ensure data quality; then, process the data through time series analysis and data formatting methods; after data formatting, transmit the data to the edge computing node through the wireless transmission protocol MQTT;

[0077] Step 3: Adaptive learning and strategy generation: On edge nodes, a deep reinforcement learning framework is deployed through a Docker container. The deep reinforcement learning framework dynamically learns and adjusts the control strategy according to the real-time power grid status through a deep power grid dynamic control algorithm. After the strategy is updated, it is fed back to the decision-making layer.

[0078] Step 4: Future state prediction and optimization decision-making: The preprocessed data is used as input to predict the future state of the power grid through the power grid prediction analysis model; the power grid prediction analysis model performs time series analysis through the long short-term memory network to predict the future load demand or failure probability of the power grid, and evaluates the effects of different control strategies through the random forest algorithm to determine the best operation, optimize the decision output to the global dispatch, and guide resource allocation;

[0079] Step 5: Data compression and efficient storage: The data generated from the preprocessing stage to the prediction and decision stage is compressed through Huffman coding to reduce storage and bandwidth requirements; the compressed data is stored in the distributed file system through the columnar storage format Parquet to achieve efficient storage and query;

[0080] Step 6: Blockchain data recording: Building a distributed ledger based on Ethereum, which automatically records key events through smart contracts and ensures data consistency and security through consensus mechanisms and elliptic encryption algorithms; key events include policy updates, abnormal events, and optimized decision outputs;

[0081] Step 7: Cross-layer communication and information fusion: Integrate multi-source data through data lake and message queue services to form a unified view, realize data aggregation, processing and efficient distribution for global decision-making;

[0082] Step 8: Global optimization scheduling and resource allocation; based on the integrated data, optimize the grid resource allocation through the intelligent scheduling model; the intelligent scheduling model realizes data exchange with the distributed system through the remote call protocol; the strategy is sent to the terminal for execution, and the grid resources are adjusted to achieve global efficiency and stability optimization.

[0083] As a further technical solution of the present invention, the Ethereum-based distributed ledger includes a smart contract module, a consensus mechanism module, a data encryption module, a distributed storage module and an audit update module; the smart contract module realizes the encoding and logical processing of policy updates, abnormal events and optimization decision output events through the event listening mechanism and rule engine of the smart contract; the consensus mechanism module ensures the consistent recognition of the newly added blocks by all nodes in the network through proof of work, and realizes data synchronization and state sharing among nodes in the whole network through the P2P network protocol; the data encryption module ensures data security and integrity through elliptic curve encryption algorithm and digital signature; the distributed storage module stores the ledger data in multiple nodes in the network through a data sharding method, and realizes rapid retrieval and access to distributed storage data through a distributed hash table; the audit update module includes a contract audit unit and a contract update unit; the contract audit unit strictly verifies the logic of the smart contract through a formal verification method; the contract update unit realizes dynamic update and maintenance of the contract logic through the self-upgrade mechanism of the smart contract to adapt to changes in the power grid control strategy.

[0084] Positive beneficial effects

[0085] The present invention discloses an adaptive grid state area intelligent fusion terminal and a control method thereof. By introducing a deep grid dynamic control algorithm, the present invention enables the terminal to learn grid state changes in real time and autonomously adjust the control strategy, effectively responding to the uncertainty and dynamics in grid operation, and solving the problem of insufficient adaptive capabilities of traditional terminals. The prediction and analysis engine combining LSTM and random forest not only improves the accuracy of future grid state predictions, but also enhances the foresight and flexibility of the regulation strategy by evaluating a variety of possible scenarios, thereby achieving more refined and intelligent control. The use of edge computing and distributed processing modules, combined with Docker containerization technology and efficient data compression and storage mechanisms, not only optimizes the data processing process, but also greatly improves the data processing speed and efficiency, reduces the demand for bandwidth, and solves the bottleneck of limited data processing capabilities of traditional terminals. The application of time series analysis and machine learning models further explores and understands complex data patterns and improves the depth and breadth of data analysis.

[0086] At the same time, by adopting standardized communication interfaces and advanced data lake technology, combined with message queue services (Apache Kafka), an efficient and unified information exchange platform is built, which realizes the seamless flow and high integration of data between different levels. This multi-layer communication and information fusion mechanism breaks the information island phenomenon between traditional terminals, promotes rapid response and collaborative work between upper and lower layers, and improves the overall response speed and decision-making efficiency of the system. With the global optimization ability of genetic algorithms and the good handling of uncertainty by fuzzy logic control, the intelligent scheduling algorithm can formulate the most optimized scheduling strategy based on comprehensive data analysis, thereby improving the efficiency of resource allocation. The cross-layer coordination mechanism established by distributed coordination algorithms (such as Raft) ensures the rapid issuance of control instructions and feedback collection, forming a tightly coupled and efficiently operated closed-loop control system, which further enhances the stability and adaptability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 It is a schematic diagram of the overall architecture of an adaptive grid status area intelligent fusion terminal of the present invention;

[0088] Figure 2 A working step diagram of the deep power grid dynamic control algorithm of the present invention;

[0089] Figure 3 It is a schematic diagram of the working method of the intelligent scheduling model of the present invention. DETAILED DESCRIPTION

[0090] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0091] In the above embodiment, if Figure 1-Figure 3 As shown: an adaptive grid status area intelligent fusion terminal, including: a terminal layer, an edge layer and a cloud platform layer;

[0092] The terminal layer includes a data acquisition module and a preprocessing module. The perception and data acquisition module collects real-time data from the power grid through the Internet of Things sensor network, including but not limited to voltage, current and frequency indicators; the preprocessing module preprocesses the data through abnormal value detection, time series analysis and data formatting methods, and the preprocessed data is uploaded to the adaptive learning and control module and the edge computing and distributed processing module of the edge layer through a standardized communication interface;

[0093] Based on the real-time data uploaded by the terminal layer; the adaptive learning and regulation module dynamically adjusts the control strategy according to the power grid state through the deep power grid dynamic control algorithm, and predicts the future state of the power grid through the power grid prediction and analysis model; the power grid prediction and analysis model predicts the time series through the long short-term memory network, and evaluates the prediction content through the random forest to provide a decision-making basis for regulation;

[0094] The edge computing and distributed processing module realizes real-time data processing and load balancing through dynamic container orchestration and chain services, and reduces data transmission volume and storage costs through Huffman coding and incremental update strategies;

[0095] The processed data is uploaded to the cloud platform layer through the standardized communication IEC 61850 protocol; the cloud platform layer includes a multi-layer communication and information fusion module and a global optimization and collaborative control module; the multi-layer communication and information fusion module realizes data aggregation, fusion and distribution through the data lake and message queue service; the global optimization and collaborative control module receives and analyzes the comprehensive data of the data lake through the intelligent scheduling model, and ensures that the upper-level decision-making information is timely delivered to the terminal layer through the cross-layer collaboration mechanism, and collects feedback to form a closed-loop control; the intelligent scheduling model performs global resource optimization through a genetic algorithm, and copes with uncertainty through fuzzy logic control; the cross-layer collaboration mechanism establishes information transmission and task collaboration between the upper and lower layers through a distributed coordination algorithm.

[0096] In a specific embodiment, the intelligent sensing component is deployed at the terminal layer, and real-time data is collected from the power grid through IoT technology, including but not limited to key indicators such as voltage, current, and frequency. The collected data is then passed to the data cleaning and preprocessing module. At this stage, an abnormal monitoring algorithm is used to remove outliers, and time series analysis technology is used to smooth the data sequence, reduce noise, and ensure data quality. The preprocessed data is sent to the adaptive learning and control module, which uses a deep power grid dynamic control algorithm to learn and adjust the control strategy in a timely manner according to the current state of the power grid. At the same time, these data also enter the predictive analysis engine, combined with LSTM for time series prediction, and random forest to evaluate various future possibilities, providing a decision-making basis for regulation.

[0097] At the same time, the pre-processed data is synchronously transmitted to the edge computing and distributed processing module of the edge computing layer, where Docker containerization technology is used to dynamically allocate computing resources according to data processing requirements. In this process, data compression and efficient storage mechanism uses Huffman coding to compress data, combined with incremental update strategy to reduce unnecessary data transmission, and optimize storage and bandwidth usage.

[0098] The processed data is uploaded to the cloud platform through a standardized communication interface (IEC 61850 protocol) to ensure data format consistency and communication efficiency. In the information fusion and data exchange center of the cloud platform, data lake technology centrally stores and manages data from various terminals. Apache Kafka, as a message queue service, efficiently distributes this data to the required services or modules, realizing cross-layer data circulation and fusion.

[0099] The final decision and feedback are executed by the global optimization and collaborative control module; the core of the intelligent scheduling algorithm receives and analyzes the comprehensive data from the data lake, uses genetic algorithms to optimize global resources, and fuzzy logic control to deal with uncertainty, and formulates the best scheduling strategy. The cross-layer coordination mechanism is based on distributed coordination algorithms such as Raft to ensure that the upper-level decision information is timely delivered to the terminal layer, while collecting feedback to form a closed-loop control and achieve efficient collaborative work of the entire system.

[0100] In the above embodiment, the distributed coordination algorithm includes a communication layer, a consensus layer, a state management layer, a task allocation layer, a data synchronization layer, a security layer and a fault tolerance layer; the distributed coordination algorithm works in the intelligent fusion terminal of the station area with adaptive power grid status as follows:

[0101] 801. Data collection and communication; data transmission and reception are realized between different network nodes through the communication layer; the communication layer realizes decentralized data exchange and communication through the P2P network protocol;

[0102] 802. Data consistency and consensus formation: reaching a consistent state or decision among distributed nodes through the consensus layer, and the consensus layer uses a Byzantine fault tolerance algorithm to ensure that even if some nodes send erroneous or malicious information, the entire network can reach a consensus on data consistency;

[0103] 803. State synchronization and state machine replication; maintaining and updating the system state, including node data and network state, through the state management layer; the state management layer ensures the state consistency of all nodes through the state machine replica;

[0104] 804, task allocation and resource scheduling; the task allocation layer allocates tasks to corresponding nodes according to the current system status and resources; the task allocation layer implements effective task allocation and optimal use of resources through load balancing algorithms and resource scheduling strategies;

[0105] 805. Data synchronization and version control: Synchronize the data in the distributed ledger or database through the data synchronization layer to ensure that the data copies of all nodes are kept up to date; the data synchronization layer realizes rapid data update and historical version management through incremental data synchronization and version control mechanism;

[0106] 806. Security assurance: The security layer protects the security of network communications and data storage to prevent unauthorized access and tampering; the security layer ensures the confidentiality, integrity and non-repudiation of data through symmetric encryption methods and digital signatures;

[0107] 807. Fault-tolerant processing; detecting and recovering system faults through the fault-tolerant layer; the fault-tolerant layer implements rapid response and processing to system anomalies through fault detection algorithms and redundancy mechanisms.

[0108] In a specific embodiment, the distributed coordination algorithm first determines a leader through an election mechanism, which is responsible for coordinating and managing the task allocation and information synchronization of the entire system. The leader is responsible for receiving decision information from the upper layer and copying it to all follower nodes as log entries. In addition, ensuring that the logs of all nodes remain consistent is the key to reaching consensus in a distributed system.

[0109] In the intelligent fusion terminal of the substation with adaptive grid status, the leader regularly sends heartbeat signals to the followers to maintain its leadership and detect the status of the followers. Then, the leader sends the new decision information as log entries to the followers, and the followers receive and store these log entries. If the follower receives a log entry that is inconsistent with the leader, the conflict will be resolved through a voting mechanism to ensure eventual consistency.

[0110] In specific implementation, when the system starts, all nodes start running the distributed coordination algorithm to elect a leader. Then, the leader receives the upper-level decision information, converts it into log entries, and then broadcasts it to all followers. The follower receives the log entry, stores it, and sends a confirmation message to the leader. Once the log entry is confirmed by the majority of nodes, the leader marks the log entry as committed and applies it to the state machine. After the follower receives the committed log entry, it also applies it to the local state machine to ensure that all nodes are in the same state.

[0111] Through leader election and heartbeat mechanisms, the system can quickly recover and continue to operate even in the event of partial node failure. Log replication and consistency guarantee mechanisms ensure that all nodes understand and execute decision information in a consistent manner, improving the stability and reliability of the system. In addition, the distributed coordination algorithm allows multiple nodes to process tasks in parallel, improving the system's processing power and response speed. Secondly, fuzzy logic control provides flexibility in the face of uncertainty, allowing the system to make reasonable decisions with incomplete information.

[0112] In the above embodiment, the working steps of the deep power grid dynamic control algorithm are as follows:

[0113] Step 1: Multi-dimensional feature extraction of environmental status;

[0114] Construct a multi-dimensional feature vector S t ={s t ,i t ,p t ,q t ,f t ,r t}, where s t represents the node voltage, i t Indicates the current load, p t and q t Represent active power and reactive power respectively; f t Indicates frequency; r t represents the phase angle;

[0115] Step 2: Q function approximation and deep neural network construction;

[0116] After the feature vector is constructed, the Q value of each state-action pair is approximated by a deep neural network; the deep neural network is trained by a loss function L(θ), and the loss function L(θ) is defined by the time difference error ρ, and the formula expression of the time difference error ρ is:

[0117] ρ=1-ω t+1 [γ(1-θ)]+γ(ω t+1 -Q(S t ,a t ;θ)) (1)

[0118] In formula (1), ω t+1 is the immediate return, which means the feedback obtained at time t+1, used to evaluate the effect of the action; γ is the discount factor; Q(S t ,a t ; θ) represents the Q function, where a t In state S t The action taken under, θ represents the network parameters;

[0119] Step 3: Design of reward mechanism and strategy evaluation;

[0120] Define a multi-objective reward function to guide the learning direction of the deep neural network. The multi-objective reward function is defined by weight coefficients α, β, γ loss Balancing the importance of different objectives, the formula expression of the multi-objective reward function is:

[0121]

[0122] In formula (2), α, β, γ loss is the weight coefficient, which is used to balance the priorities of different goals; n loadrepresents the load current, β represents the power load intensity of the power system, and n loss Indicates power loss; n dev Indicates voltage deviation;

[0123] Step 4: Online learning and updating of strategies;

[0124] As the state of the power grid changes in real time, the online learning of the model based on real-time data is realized through the gradient descent update function. The formula expression of the gradient descent update function is:

[0125]

[0126] In formula (3), θ new represents the updated network parameters, which are used to calculate the action value in the new state; K represents the learning rate, which is used to control the pace of parameter updating during the learning process; θ is the current network parameter, which represents the current learning state of the deep neural network model; n represents all possible actions that can be taken in a given state; Represents the gradient of the loss function L(θ) to θ; the model parameter θ is adjusted according to the gradient Update with learning rate K;

[0127] Step 5: Implementation and feedback of adaptive control strategy;

[0128] According to the updated policy θ new , through the action selection function to calculate in the new state S t+1 The best action under The formula expression of the action selection function is:

[0129]

[0130] In formula (4), Indicates state S t+1 The optimal action under the given state; n represents all possible actions to be taken under a given state; execute the action And observe the new state S t+2 and immediate returnω t+2 , collect feedback information for the next round of learning and strategy updating.

[0131] In a specific embodiment, in the intelligent fusion terminal of the substation that adapts to the grid state, experiment 1 is designed to evaluate and verify the adaptability, accuracy and effectiveness of the deep grid dynamic control algorithm under different grid states; the specific experimental method is: first, high-precision sensors are used to collect grid parameters in real time, including voltage, current, power, load, frequency and temperature. Next, the collected data is input into the deep learning model, which is trained to identify the grid state and generate corresponding control strategies. Then, according to the control strategy generated by the model, the grid operation parameters such as load, power generation, voltage and frequency are adjusted. Finally, the predicted state and the actual state are compared, and the error rate is calculated to evaluate the accuracy of the control strategy; the experimental test data records are shown in Table 1:

[0132] Table 1 Experimental test data record table

[0133]

[0134]

[0135] In Data Table 1, the control strategy types include load adjustment, generation adjustment, voltage adjustment, frequency adjustment, load forecasting, energy optimization, demand response, fault prediction, voltage stability control, and load balancing. The strategy parameters are specific parameters of the control strategy, such as the percentage of increase or decrease in load demand, the percentage of power generation adjustment, etc. By comparing the predicted state with the actual state, calculating the error rate, and evaluating the accuracy and effectiveness of the control strategy, the effectiveness of the deep power grid dynamic control algorithm can be verified.

[0136] In addition, Experiment 2 is designed to compare the performance differences between the deep power grid dynamic control algorithm and the traditional control algorithm to verify the effectiveness and advantages of the deep power grid dynamic control algorithm. The experimental environment includes high-performance servers, GPU acceleration devices, high-speed network connections, etc. to ensure real-time data processing and the issuance of control instructions. The deep power grid dynamic control algorithm and the traditional control algorithm are deployed on the same hardware and operating system platform to ensure the fairness of the test.

[0137] The specific experimental method is: first, collect real-time power grid data from the terminal layer, including voltage, current, frequency, etc. Next, deploy the deep power grid dynamic control algorithm and the traditional control algorithm respectively. Then, simulate different power grid operation states, including steady-state operation, load changes, fault states, etc. Finally, record the response time, prediction accuracy, system stability and other indicators of the two algorithms in different scenarios. The experimental indicators include: Response time: the time it takes for the algorithm to respond to changes in the power grid state. Prediction accuracy: the accuracy of the algorithm's prediction of the future state of the power grid. System stability: the performance of the algorithm in maintaining stable power grid operation. Fault recovery time: the time it takes for the algorithm to resume normal operation when a power grid fault occurs. Control strategy adjustment time: the time it takes for the algorithm to adjust the control strategy according to the power grid state. Computing resource utilization: the efficiency of the algorithm in utilizing computing resources during operation. Energy consumption: the energy consumption of the algorithm during operation. Maintenance cost: the maintenance cost required for the long-term operation of the algorithm. The specific data comparison is shown in Table 2:

[0138] Table 2 Algorithm comparison experiment record table

[0139] index Deep Power Grid Dynamic Control Algorithm Traditional control algorithm Response time 100ms 200ms Prediction accuracy 95％ 85％ System stability high middle Fault recovery time 50ms 100ms Control strategy adjustment time 5s 10s Computing resource utilization 75％ 50％ Energy consumption Low high Maintenance costs Lower Higher

[0140] It can be seen from Table 2 that the deep power grid dynamic control algorithm shows better performance than the traditional control algorithm in practical applications, especially in terms of response time, prediction accuracy and system stability. This shows that the deep power grid dynamic control algorithm has higher practical value and development potential, and is worthy of promotion and application in power grid control systems.

[0141] In the above embodiment, the power grid prediction and analysis model includes an input layer, a multi-dimensional feature fusion layer, a time series analysis layer, an advanced feature engineering layer, a hybrid prediction model layer, a prediction calibration and optimization layer, an adaptive dynamic adjustment layer, a control strategy optimization layer and a feedback loop layer; the working steps of the power grid prediction and analysis model in the station area intelligent fusion terminal of the adaptive power grid state are:

[0142] T1. Obtaining power grid data, including real-time monitoring data and historical records, through the input layer; the input layer integrates data from different sources and sensors through a Kalman filter to achieve data consistency and integrity;

[0143] T2. Extracting and selecting features from the integrated data through the multidimensional feature fusion layer; the multidimensional feature fusion layer evaluates the feature importance of the power grid historical data through Gaussian process regression, and the Gaussian process regression captures the complex relationship between features through the kernel method to provide a basis for feature selection;

[0144] T3. The output of the multi-dimensional feature fusion layer is received through the time series analysis layer, and the time series data of the power grid state is analyzed; the time series analysis layer captures the long-term dependency of the power grid parameters through the long short-term memory network; the long short-term memory network predicts the short-term change of the power grid state through the state update and gating mechanism; the formula expression of the state update is:

[0145]

[0146] In formula (5), h t is the hidden state of the current time step, containing the information of the time series; h t-1 is the hidden state of the previous time step, x t is the input feature of the current time step, W and U are weight matrices, corresponding to the conversion of input and hidden states respectively, and b is the bias term;

[0147] T4, further extracting and constructing features that have a significant impact on prediction through the advanced feature engineering layer; the advanced feature engineering layer reduces feature dimensions through linear transformation and reveals the intrinsic pattern of the signal through local time-frequency analysis;

[0148] T5. Predictions are made through the hybrid prediction model layer; the hybrid prediction model layer processes and predicts complex time series data through a hybrid prediction weighted formula; the hybrid prediction weighted formula predicts through a long short-term memory network LSTM, and weights and corrects the output of the LSTM through the integration characteristics of the random forest RF; the formula expression of the hybrid prediction weighted formula is

[0149]

[0150] In formula (6), is the mixed prediction result, z is the weighting coefficient, which is used to balance the importance of LSTM and RF prediction results. is the LSTM prediction result, is the prediction result of random forest; c t is the cell state at the current time step;

[0151] T6. Optimizing the performance of the prediction model through the prediction calibration and optimization layer; the prediction calibration and optimization layer tunes the model hyperparameters through Bayesian optimization; the Bayesian optimization finds the optimal hyperparameter combination by constructing a probability model of the hyperparameters and using a sampling strategy;

[0152] T7. The prediction model is dynamically adjusted according to the real-time feedback of the power grid state by the adaptive dynamic adjustment layer; the adaptive dynamic adjustment layer dynamically adjusts the prediction model by the Kalman filter prediction update formula: the Kalman filter prediction update formula uses the difference between the observed value and the prior prediction to dynamically update the prediction state by gain calculation; the formula expression of the Kalman filter prediction update formula is:

[0153]

[0154] In formula (7), V k is the Kalman gain, which is used to measure the impact of the observation value on the predicted state. is the predicted state based on prior knowledge, is the state estimation based on prior prediction, and B is the actual observation value;

[0155] T8. Based on the prediction results, the control strategy optimization layer optimizes the control strategy of the power grid; the control strategy optimization layer optimizes the power grid control strategy through the deep Q network; the deep Q network updates the strategy through the strategy optimization objective function, and the formula expression of the strategy optimization objective function is:

[0156]

[0157] In formula (8), P new is the updated policy network parameter, P is the current policy network parameter, m is the learning rate, which is used to control the speed of parameter update, Q is the expected value, which represents the statistical average of state-action pairs; p is the state-action distribution under the current policy; y is the policy network, which is used to measure the probability of taking an action in a given state;

[0158] T9, based on the prediction error and the actual response of the power grid; a closed-loop control system is constructed through the feedback loop layer, and the feedback loop layer continuously updates the model parameters through the recursive least squares method; the parameter update formula of the recursive least squares method is:

[0159]

[0160] In formula (9), F t+1 is the updated parameter vector, F t is the parameter vector at time t, q is the prediction error, which is used to measure the difference between the model prediction and the actual observation, l is the forgetting factor, which is used to control the weight of historical data, is the input data; f t A weight vector representing the parameter vector.

[0161] In a specific embodiment, the power grid prediction and analysis model is based on a long short-term memory network (LSTM) and a random forest. LSTM is a special recurrent neural network (RNN) that can learn long-term dependencies in time series data and is very suitable for processing time series prediction tasks of power grid parameters such as power load, frequency, and voltage. Random forest is an integrated learning method that improves the accuracy and robustness of predictions by constructing multiple decision trees and taking their average or majority vote. Specifically, Experiment 3 is designed to verify the accuracy and stability of the power grid prediction and analysis model, and the prediction performance of the model under different environmental conditions is evaluated by comparing the actual power grid load with the model prediction value. The experimental method is as follows: First, multidimensional data including load demand, meteorological conditions, etc. are collected from the power grid system. Then, features that have a significant impact on the power grid load, such as historical load data, temperature, humidity, wind speed, and solar radiation, are selected through feature engineering. Then, LSTM is used for deep learning of time series data to capture long-term dependencies; random forest is used to correct LSTM output to improve prediction accuracy. Finally, the error percentage between the predicted value and the actual value is calculated to evaluate the accuracy of the model prediction. The specific experimental data records are shown in Table 3:

[0162] Table 3 Model experiment record

[0163]

[0164] In data table 3, actual load refers to the actual power demand of the power grid at a specific point in time. LSTM predicted load refers to the future power grid load predicted by the LSTM model based on historical data. Random forest corrected prediction refers to the final predicted load value after correction by the random forest model. Meteorological parameters such as temperature, humidity, wind speed, and solar radiation have a significant impact on the power grid load and are included in the model training and prediction process. Error refers to the difference between the actual value and the predicted value, which is used to evaluate the performance of the prediction model. By comparing the LSTM direct prediction value with the random forest corrected prediction value, it can be observed that the random forest correction helps to improve the accuracy of the prediction and reduce the error. Especially when the meteorological conditions change greatly, the corrected prediction value is closer to the actual load, showing the good adaptability and accuracy of the model.

[0165] In addition, Experiment 4 is designed to evaluate the performance difference between the power grid prediction analysis model (a combined model of long short-term memory network LSTM and random forest) and traditional algorithms (such as ARIMA and linear regression) in power load forecasting, and to demonstrate the superiority of the power grid prediction analysis model. The experimental environment includes: high-performance servers equipped with NVIDIA GPU, multi-core CPU, and high-speed SSD storage. The operating system is Linux, the database management system is PostgreSQL, and Python 3.8 is used for data analysis and model training. The relevant libraries include NumPy, Pandas, Scikit-learn, Keras, and XGBoost. The experimental data comes from the historical power load records of a regional power grid, with a time span of one year and a sampling frequency of every 15 minutes. The specific experimental method is as follows: First, data preprocessing is performed: including data cleaning, missing value processing, normalization and other steps. Then, long short-term memory network (LSTM), random forest (RF), ARIMA model and traditional linear regression model are selected. All models are trained using the same training data set, and the model stability is evaluated by cross-validation method. Finally, an independent test data set is used to predict each model, and indicators such as MAE, RMSE, and MAPE are calculated. The experimental data record table is shown in Table 4:

[0166] Table 4 Experimental test data comparison table

[0167]

[0168] In data table 4, the "time series length" in the table refers to the length of the time series data used for prediction, the mean absolute error represents the average of the absolute value of the difference between the predicted value and the actual value, the root mean square error is the square root of the average of the square of the prediction error, and the mean absolute percentage error is the average percentage of the absolute value of the ratio of the prediction error to the actual value. The smaller the index, the higher the prediction accuracy of the model; from the table data, it can be seen that the combined model of LSTM and RF can effectively capture the nonlinear characteristics and long-term dependencies in the time series data, thereby providing more accurate predictions. Compared with traditional algorithms, the power grid prediction analysis model has obvious advantages in reducing prediction errors and improving prediction accuracy.

[0169] In the above embodiment, the edge computing and distributed processing module realizes lightweight encapsulation of applications through a containerization method; the containerization method enhances the automation and flexibility of container management through container orchestration; the container orchestration allows users to define the desired cluster state through declarative configuration management, and dynamically adjusts the number of container instances according to real-time load through an automatic expansion mechanism; the chained service splits complex tasks into independent microservices through a microservice architecture; these microservices perform unified access control and routing through an API gateway to ensure the decoupling and scalability of services; the API gateway distributes requests evenly to each service instance through a load balancing algorithm; the containerization method provides intelligent routing, traffic management, policy execution and elasticity control between microservices through a service grid, and realizes rapid iteration and deployment of services through a continuous integration / continuous deployment mechanism; the incremental update strategy improves update efficiency through a difference detection algorithm and a data synchronization mechanism; the difference detection algorithm identifies and records differences between data sets through hash summaries, data chunking, timestamps and difference encoding.

[0170] In a specific embodiment, in an intelligent fusion terminal for a substation that is adaptive to the state of the power grid, edge computing reduces the data transmission distance and time by migrating computing tasks from the cloud or central server to devices at the edge of the network. In a power grid monitoring system, this means distributing data processing tasks on edge devices near intelligent sensing components close to the data source. In this way, only processed data or key information needs to be transmitted to the central server or cloud platform, greatly reducing the demand and delay of communication bandwidth. Container technology (such as Docker) allows applications and their dependent environments to be packaged into a lightweight, portable container unit. Dynamic container orchestration can dynamically start, stop, expand or shrink service containers on edge devices according to real-time needs. This elastic resource management method enables the system to automatically adjust resource allocation according to changes in workload, improving resource utilization efficiency and system processing capabilities. Chain service design is to connect a series of related data processing services in a specific order to form a processing link. Each service is responsible for performing a part of a specific task, such as data cleaning, feature extraction, anomaly detection, etc. Data is transmitted between these services in sequence, and each service passes the result to the next service after processing until the entire processing flow is completed. Huffman coding is an effective data compression algorithm that constructs the optimal prefix code based on the frequency of each character in the data to reduce the size of the encoded data. The incremental update strategy only transmits the changed part of the data, rather than the entire data set, which can reduce the network bandwidth usage and improve data processing efficiency. Specifically, according to the data processing requirements, the dynamic container orchestration system will automatically start the corresponding service container to form a processing link. The data is gradually transmitted along this link, and each service container performs its specific task until the data processing is completed. The processed data is compressed by Huffman coding to reduce the data size. Using the incremental update strategy, only the changed part of the data is transmitted to the cloud or central server, reducing the network bandwidth usage.

[0171] In the above embodiment, the multi-layer communication and information fusion module includes a data aggregation submodule, a data fusion submodule, a message queue service submodule, a data distribution submodule, a cross-layer information interaction submodule and a collaborative control feedback submodule; the data aggregation submodule realizes data interaction with the terminal layer through a standardized communication protocol, and standardizes the accessed data through data cleaning and format conversion methods; the data fusion submodule integrates data from different sources and types through a timestamp-based fusion strategy; and realizes the storage and management of large-scale unstructured data through a distributed file system and a NoSQL database; the timestamp-based fusion strategy fuses and transforms the original data through data abstraction and pattern recognition to provide data support for the intelligent scheduling model; the distributed file system optimizes data access efficiency through data partitioning and indexing mechanisms; the message queue service submodule realizes data access through a publish / subscribe model The method realizes asynchronous transmission and decoupling of data, and ensures the reliability of message transmission through a data persistence mechanism; the data distribution submodule includes a data routing unit and a data access control unit; the data routing unit distributes the fused data to different processing modules or analysis engines through a routing algorithm and a load balancing method; the data access control unit ensures the security and compliance of the data distribution process through an access control list and an identity authentication mechanism; the cross-layer information interaction submodule realizes data conversion through an abstract data model, and realizes the consistency of data formats and semantics between different levels through data mapping; the collaborative control feedback submodule includes a feedback collection unit and a decision feedback unit; the feedback collection unit collects the execution results and status information of the terminal layer through a feedback mechanism; the decision feedback unit transmits the optimization decision of the upper layer to the terminal layer through an event-driven mechanism and a message queue, and adjusts the decision according to the feedback.

[0172] In the multi-layer communication and information fusion module, first, all kinds of real-time data collected by the terminal layer are collected to the data aggregation submodule through standardized interfaces (such as MQTT and CoAP). The aggregation submodule performs preliminary data sorting, including format unification, deduplication, and timestamp synchronization, in preparation for subsequent processing. The data fusion submodule deeply integrates the aggregated data and applies data fusion algorithms to identify and merge information from different sources but related to each other. Through data lake or data warehouse technology, these fused data are stored and indexed for fast query and analysis. The message queue service submodule uses message queue technology to realize asynchronous transmission of data and control instructions, and improve the decoupling and scalability of the system. Messages are classified by topic, supporting high-throughput, low-latency message publishing and subscription, and ensuring the real-time and sequential nature of data distribution. The data distribution submodule distributes the processed data to the required upper-layer applications or lower-layer terminals according to predefined rules or dynamic requirements. It realizes the directional push of data to specific services, algorithm models or decision systems, and optimizes the efficiency of data flow. The cross-layer information interaction submodule realizes flexible communication and data exchange between different levels through a unified API interface or service bus technology. It supports the downward penetration of upper-level strategy instructions and the upward feedback of underlying status information, ensuring the accurate execution of control instructions and real-time monitoring of status. The collaborative control feedback submodule integrates the feedback mechanism of control instruction execution, monitors the execution effect of instructions, and adjusts the control strategy in time. Through closed-loop control logic, the decision algorithm is optimized based on feedback information to form a self-learning and dynamically adjusted control loop.

[0173] In the specific implementation, through efficient data aggregation and fusion, the delay of information processing is reduced and the overall response speed of the system is improved. In addition, the message queue and cross-layer interaction mechanism enhance the decoupling of the system, facilitate the independent upgrade and expansion of modules, and improve the stability and flexibility of the system. Secondly, the collaborative control feedback mechanism ensures the effective execution of control instructions and the rational allocation of resources, and improves the operation efficiency and fault response capability of the power grid. At the same time, the transparent cross-layer flow of information promotes knowledge sharing and collaborative decision-making among various levels, which is conducive to the formulation and implementation of the global optimal strategy and improves the intelligence level and operation and maintenance efficiency of the power grid.

[0174] In the above embodiment, the intelligent scheduling model includes a data input layer, a feature extraction layer, a feature encoding layer, a genetic algorithm population initialization layer, a fitness function calculation layer, a selection layer, a genetic operation layer, an elite retention and population update layer, a decision output layer and an execution and feedback layer; the working method steps of the intelligent scheduling model in the intelligent fusion terminal of the station area with adaptive power grid status are as follows:

[0175] S1. Obtaining raw data from sensors and databases through the data input layer, including but not limited to equipment status, environmental parameters and demand forecasts;

[0176] S2, extracting features through the feature extraction layer; the feature extraction layer removes noise through time series analysis and wavelet transform, and refines effective features through principal component analysis;

[0177] S3, converting the high-dimensional feature vector generated by the feature extraction layer into binary code through the feature coding layer;

[0178] S4, based on the coding features, generating an initial solution set through the genetic algorithm population initialization layer as the starting point of the optimization search;

[0179] S5. Evaluate the quality of each solution in the population through a fitness function calculation layer, wherein the fitness function calculation layer considers factors including energy allocation efficiency, system stability, and cost-effectiveness;

[0180] S6. Based on the individual fitness values ​​output by the upper layer, the selection layer selects individuals to participate in reproduction, and the selection layer makes selection decisions through a proportional selection method;

[0181] S7, simulating natural selection and genetic mechanism through the genetic operation layer; the genetic operation layer improves population diversity and enhances global search capability through crossover and mutation;

[0182] S8. At the end of each generation, the optimal solution in each generation is retained through the elite retention and population update layer, and the population is updated in combination with the newly generated offspring;

[0183] S9, converting the optimal solution obtained by optimization into actual operation instructions through the decision output layer to guide power grid control and resource allocation; the decision output layer decodes the optimal solution back into decision instructions or strategy recommendations through reverse conversion coding;

[0184] S10, formulating a strategy based on the decision output layer; implementing it through the execution and feedback layer and collecting feedback data on the implementation effect for adjusting model parameters or optimizing strategies.

[0185] In a specific embodiment, the intelligent dispatch model integrates data from different sources, including real-time monitoring data, historical operation data, meteorological information, market supply and demand, etc., to obtain a comprehensive view of the power grid operation. At the same time, the natural selection, crossover and mutation mechanisms of the genetic algorithm are used to simulate the evolutionary process in nature and find the optimal power grid resource configuration plan. The algorithm continuously approaches the global optimal solution during the iteration process. Fuzzy logic is used to deal with fuzziness and uncertainty problems in the power grid, such as load forecasting errors, volatility of renewable energy, etc., and precise control instructions are converted into fuzzy sets through the fuzzy rule base to adapt to the dynamic changes of the power grid.

[0186] Specifically, in the genetic algorithm, the intelligent scheduling model first encodes the scheduling problem of the power grid into chromosomes, and each chromosome represents a potential scheduling solution. The chromosome is converted into an actual scheduling strategy through a decoding process. Next, a fitness function is defined to evaluate the performance of each chromosome (scheduling solution), such as cost minimization, loss minimization, reliability maximization, etc. Then, cloud computing and edge computing resources are used to realize parallel computing of the genetic algorithm and accelerate the optimization process. In the specific implementation, through the global search capability of the genetic algorithm, the intelligent scheduling model can find a resource allocation solution that is closer to the global optimum and improve the overall performance of the power grid. In addition, fuzzy logic control provides flexibility for scheduling decisions, enabling the system to maintain stable operation in the face of uncertainty and change. Secondly, parallel processing and real-time scheduling capabilities ensure that the intelligent scheduling model can respond quickly to changes in the state of the power grid and make quick decisions.

[0187] In the above embodiment, the data acquisition module ensures the time consistency of sensor data through a timestamp synchronization mechanism, and realizes temporary storage of data through a circular queue to cope with the problem of mismatch between data acquisition and processing speed; the circular queue tracks the data currently stored in the buffer through a circular index mechanism, optimizes space utilization, and ensures data consistency and thread safety through mutex locks and semaphores; the outlier detection identifies and processes outliers in the data through a statistical threshold detection method; the time series analysis performs time series prediction and trend analysis through an autoregressive integral moving average model; the data formatting method converts the original data into a standardized format through the data encapsulation protocol JSON.

[0188] In a specific embodiment, the terminal layer is mainly composed of intelligent sensing components deployed on each node, which are connected to various devices and sensors in the power grid through the Internet of Things (IoT) technology. The intelligent sensing components can monitor the key indicators of the power grid in real time, such as voltage, current, frequency, etc., and transmit the collected data to the data cleaning and preprocessing module in real time. The intelligent sensing components include: sensors: used to capture various physical quantities of the power grid, such as voltage, current, power, etc. These sensors generally include voltage sensors, current sensors, power sensors, etc. Data collectors: responsible for collecting data captured by sensors and converting them into a format suitable for transmission and processing. Communication module: Transmits the collected data to the data preprocessing and analysis system by wireless or wired means; including Wi-Fi, LoRa, 4G / 5G, etc. Microcontroller or processor: As the core of the intelligent sensing component, it is responsible for coordinating the work of sensors, data collectors and communication modules to realize local processing and preliminary analysis of data. Power management unit: Ensures stable power supply of intelligent sensing components, which may include batteries, energy harvesting devices or power supplies directly connected to the power grid. Embedded software: runs on a microcontroller or processor and is responsible for controlling the operation of the intelligent sensing components, including data sampling frequency, data transmission time interval, data preprocessing algorithm, etc. Indicators for real-time monitoring of the power grid include: voltage level in the grid, including phase voltage and line voltage. Current value flowing through the grid equipment, divided into phase current and neutral current. The frequency of the grid AC power, usually 50Hz or 60Hz. Active power, reactive power and apparent power. Power factor; the phase difference between voltage and current in the grid. Harmonics, load, transformer temperature and environmental parameters;

[0189] In the above embodiment, a method for controlling an adaptive grid status substation intelligent fusion terminal includes the following steps:

[0190] Step 1: Intelligent perception and data collection: Real-time monitoring of grid operation parameters through sensor networks, smart meters and environmental monitoring equipment, and real-time elimination of abnormal data through threshold filtering methods; grid operation parameters include but are not limited to voltage, current, frequency, temperature and humidity parameters;

[0191] Step 2: Anomaly detection and data preprocessing: Receive preliminary clean data from the perception layer and further filter it through the outlier detection algorithm to ensure data quality; then, process the data through time series analysis and data formatting methods; after data formatting, transmit the data to the edge computing node through the wireless transmission protocol MQTT;

[0192] Step 3: Adaptive learning and strategy generation: On edge nodes, a deep reinforcement learning framework is deployed through a Docker container. The deep reinforcement learning framework dynamically learns and adjusts the control strategy according to the real-time power grid status through a deep power grid dynamic control algorithm. After the strategy is updated, it is fed back to the decision-making layer.

[0193] Step 4: Future state prediction and optimization decision-making: The preprocessed data is used as input to predict the future state of the power grid through the power grid prediction analysis model; the power grid prediction analysis model performs time series analysis through the long short-term memory network to predict the future load demand or failure probability of the power grid, and evaluates the effects of different control strategies through the random forest algorithm to determine the best operation, optimize the decision output to the global dispatch, and guide resource allocation;

[0194] Step 5: Data compression and efficient storage: The data generated from the preprocessing stage to the prediction and decision stage is compressed through Huffman coding to reduce storage and bandwidth requirements; the compressed data is stored in the distributed file system through the columnar storage format Parquet to achieve efficient storage and query;

[0195] Step 6: Blockchain data recording: Building a distributed ledger based on Ethereum, which automatically records key events through smart contracts and ensures data consistency and security through consensus mechanisms and elliptic encryption algorithms; key events include policy updates, abnormal events, and optimized decision outputs;

[0196] Step 7: Cross-layer communication and information fusion: Integrate multi-source data through data lake and message queue services to form a unified view, realize data aggregation, processing and efficient distribution for global decision-making;

[0197] Step 8: Global optimization scheduling and resource allocation; based on the integrated data, optimize the grid resource allocation through the intelligent scheduling model; the intelligent scheduling model realizes data exchange with the distributed system through the remote call protocol; the strategy is sent to the terminal for execution, and the grid resources are adjusted to achieve global efficiency and stability optimization.

[0198] In the above embodiment, the Ethereum-based distributed ledger includes a smart contract module, a consensus mechanism module, a data encryption module, a distributed storage module and an audit update module; the smart contract module implements the encoding and logic processing of policy updates, abnormal events and optimization decision output events through the event listening mechanism and rule engine of the smart contract; the consensus mechanism module ensures the consistent recognition of the newly added blocks by all nodes in the network through proof of work, and realizes data synchronization and state sharing among nodes in the whole network through the P2P network protocol; the data encryption module ensures data security and integrity through elliptic curve encryption algorithm and digital signature; the distributed storage module stores the ledger data in multiple nodes in the network through the data sharding method, and realizes fast retrieval and access to distributed storage data through a distributed hash table; the audit update module includes a contract audit unit and a contract update unit; the contract audit unit strictly verifies the logic of the smart contract through a formal verification method; the contract update unit realizes dynamic update and maintenance of the contract logic through the self-upgrade mechanism of the smart contract to adapt to changes in the power grid control strategy.

[0199] In a specific embodiment, the smart contract module uses smart contracts to automatically execute contract terms, ensuring the automatic execution and non-tamperability of power grid transactions and dispatch instructions. The consensus mechanism module ensures that all participating nodes reach a consensus on the state of the ledger through Ethereum's consensus algorithm (such as Proof of Work or Proof of Stake). The data encryption module uses encryption technology to protect the security and privacy of data, ensuring that only authorized users can access sensitive information. The distributed storage module stores data in a decentralized manner on multiple nodes in the network to improve the reliability and anti-attack capabilities of the data. The audit update module provides auditability and update mechanisms for the ledger to ensure the accuracy and traceability of the data.

[0200] In the intelligent fusion terminal of the substation area that adapts to the state of the power grid, the specific implementation process of the smart contract module is as follows: First, deploy the Ethereum client at each key node of the power grid to form a distributed ledger network. Next, deploy the written smart contract to the Ethereum network for power grid transactions and scheduling. Then, encrypt the key data and transaction records of the power grid and store them on the chain to ensure that the data cannot be tampered with. Fourth, the nodes in the network reach a consensus on the update of the ledger through a consensus mechanism to ensure data synchronization of all nodes. Fifth, audit the ledger regularly to ensure the accuracy of the data, and make necessary updates through smart contracts.

[0201] In specific implementation, data encryption and distributed storage improve the security of power grid data and prevent data tampering and single point failure. All transactions and dispatch instructions are publicly recorded on the blockchain, which improves the transparency of power grid operations. In addition, the automatic execution of smart contracts reduces manual intervention and improves the efficiency of power grid transactions and dispatch. Secondly, decentralized ledgers reduce intermediary links and reduce transaction and maintenance costs.

[0202] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that these specific embodiments are only illustrative, and those skilled in the art may omit, replace, and change the details of the above methods and systems in various ways without departing from the principles and essence of the present invention. For example, merging the above method steps so as to perform substantially the same functions in substantially the same manner to achieve substantially the same results is within the scope of the present invention. Therefore, the scope of the present invention is limited only by the appended claims.

Claims

1. An adaptive grid status area intelligent fusion terminal, characterized by: include: Terminal layer, edge layer and cloud platform layer; The terminal layer includes a data acquisition module and a preprocessing module. The perception and data acquisition module collects real-time data from the power grid through the Internet of Things sensor network, including but not limited to voltage, current and frequency indicators; the preprocessing module preprocesses the data through abnormal value detection, time series analysis and data formatting methods, and the preprocessed data is uploaded to the adaptive learning and control module and the edge computing and distributed processing module of the edge layer through a standardized communication interface; Based on the real-time data uploaded by the terminal layer; the adaptive learning and regulation module dynamically adjusts the control strategy according to the power grid state through the deep power grid dynamic control algorithm, and predicts the future state of the power grid through the power grid prediction and analysis model; the power grid prediction and analysis model predicts the time series through the long short-term memory network, and evaluates the prediction content through the random forest to provide a decision-making basis for regulation; The edge computing and distributed processing module realizes real-time data processing and load balancing through dynamic container orchestration and chain services, and reduces data transmission volume and storage costs through Huffman coding and incremental update strategies; The processed data is uploaded to the cloud platform layer through the standardized communication IEC 61850 protocol; the cloud platform layer includes a multi-layer communication and information fusion module and a global optimization and collaborative control module; the multi-layer communication and information fusion module realizes data aggregation, fusion and distribution through the data lake and message queue service; the global optimization and collaborative control module receives and analyzes the comprehensive data of the data lake through the intelligent scheduling model, and ensures that the upper-level decision-making information is timely delivered to the terminal layer through the cross-layer collaboration mechanism, and collects feedback to form a closed-loop control; the intelligent scheduling model performs global resource optimization through a genetic algorithm, and copes with uncertainty through fuzzy logic control; the cross-layer collaboration mechanism establishes information transmission and task collaboration between the upper and lower layers through a distributed coordination algorithm.

2. The adaptive grid status area intelligent fusion terminal according to claim 1 is characterized by: The distributed coordination algorithm includes a communication layer, a consensus layer, a state management layer, a task allocation layer, a data synchronization layer, a security layer and a fault tolerance layer; the distributed coordination algorithm works in the intelligent fusion terminal of the station area with adaptive power grid status as follows:

801. Data collection and communication; The communication layer realizes data transmission and reception between different network nodes; the communication layer realizes decentralized data exchange and communication through the P2P network protocol; 802. Data consistency and consensus formation; A state or decision that is consistent among distributed nodes is reached through the consensus layer, which uses a Byzantine fault-tolerant algorithm to ensure that the entire network can reach a consensus on data consistency even if some nodes send erroneous or malicious information; 803. State synchronization and state machine replication; The state management layer maintains and updates the system state, including node data and network state; the state management layer ensures the state consistency of all nodes through state machine replicas; 804. Task allocation and resource scheduling; The task allocation layer allocates tasks to corresponding nodes according to the current system status and resources; the task allocation layer realizes effective allocation of tasks and optimal use of resources through load balancing algorithm and resource scheduling strategy; 805. Data synchronization and version control; The data synchronization layer synchronizes the data in the distributed ledger or database to ensure that the data copies of all nodes are kept up to date; the data synchronization layer realizes rapid data update and historical version management through incremental data synchronization and version control mechanism; 806. Security assurance; The security layer protects the security of network communications and data storage to prevent unauthorized access and tampering; the security layer ensures the confidentiality, integrity and non-repudiation of data through symmetric encryption methods and digital signatures; 807. Fault-tolerant processing; The fault-tolerant layer detects and recovers system failures; the fault-tolerant layer implements rapid response and processing of system anomalies through fault detection algorithms and redundancy mechanisms.

3. The adaptive grid status area intelligent fusion terminal according to claim 1 is characterized by: The working steps of the deep power grid dynamic control algorithm are as follows: Step 1: Multi-dimensional feature extraction of environmental status; Construct a multi-dimensional feature vector S t ={s t ,i t ,p t ,q t ,f t ,r t }, where s t represents the node voltage, i t Indicates the current load, p t and q t Represent active power and reactive power respectively; f t Represents frequency; r t represents the phase angle; Step 2: Q function approximation and deep neural network construction; After the feature vector is constructed, the Q value of each state-action pair is approximated by a deep neural network; the deep neural network is trained by a loss function L(θ), and the loss function L(θ) is defined by the time difference error ρ, and the formula expression of the time difference error ρ is: p=1-ω t+1 [γ(1-θ)]+γ(ω t+1 -Q(S t ,a t (i)) (1) In formula (1), ω t+1 is the immediate return, which means the feedback obtained at time t+1, used to evaluate the effect of the action; γ is the discount factor; Q(S t ,a t ; θ) represents the Q function, where a t In state S t The action taken under, θ represents the network parameters; Step 3: Design of reward mechanism and strategy evaluation; Define a multi-objective reward function to guide the learning direction of the deep neural network. The multi-objective reward function is defined by weight coefficients α, β, γ loss Balancing the importance of different objectives, the formula expression of the multi-objective reward function is: In formula (2), α, β, γ loss is the weight coefficient, which is used to balance the priorities of different goals; n load represents the load current, β represents the power load intensity of the power system, and n loss Indicates power loss; n dev Indicates voltage deviation; Step 4: Online learning and updating of strategies; As the state of the power grid changes in real time, the online learning of the model based on real-time data is realized through the gradient descent update function. The formula expression of the gradient descent update function is: In formula (3), θ new represents the updated network parameters, which are used to calculate the action value in the new state. K represents the learning rate, which is used to control the pace of parameter updating during the learning process. θ is the current network parameter, which indicates the current learning state of the deep neural network model; n represents all possible actions that can be taken in a given state; Represents the gradient of the loss function L(θ) to θ; the model parameter θ is adjusted according to the gradient Update with learning rate K; Step 5: Implementation and feedback of adaptive control strategy; According to the updated policy θ new , through the action selection function to calculate in the new state S t+1 The best action under The formula expression of the action selection function is: In formula (4), Indicates state S t+1 The optimal action under the given state; n represents all possible actions to be taken under a given state; execute the action And observe the new state S t+2 and immediate returnω t+2 , collect feedback information for the next round of learning and strategy updating.

4. The adaptive grid status area intelligent fusion terminal according to claim 1 is characterized by: The power grid prediction and analysis model includes an input layer, a multi-dimensional feature fusion layer, a time series analysis layer, an advanced feature engineering layer, a hybrid prediction model layer, a prediction calibration and optimization layer, an adaptive dynamic adjustment layer, a control strategy optimization layer and a feedback loop layer; the working steps of the power grid prediction and analysis model in the station area intelligent fusion terminal of the adaptive power grid state are as follows: T1. Obtaining power grid data, including real-time monitoring data and historical records, through the input layer; the input layer integrates data from different sources and sensors through a Kalman filter to achieve data consistency and integrity; T2. Extracting and selecting features from the integrated data through the multidimensional feature fusion layer; the multidimensional feature fusion layer evaluates the feature importance of the power grid historical data through Gaussian process regression, and the Gaussian process regression captures the complex relationship between features through the kernel method to provide a basis for feature selection; T3. The output of the multi-dimensional feature fusion layer is received through the time series analysis layer, and the time series data of the power grid state is analyzed; the time series analysis layer captures the long-term dependency of the power grid parameters through the long short-term memory network; the long short-term memory network predicts the short-term change of the power grid state through the state update and gating mechanism; the formula expression of the state update is: In formula (5), h t is the hidden state of the current time step, containing the information of the time series; h t-1 is the hidden state of the previous time step, x t is the input feature of the current time step, W and U are weight matrices, corresponding to the conversion of input and hidden states respectively, and b is the bias term; T4, further extracting and constructing features that have a significant impact on prediction through the advanced feature engineering layer; the advanced feature engineering layer reduces feature dimensions through linear transformation and reveals the intrinsic pattern of the signal through local time-frequency analysis; T5. Predictions are made through the hybrid prediction model layer; the hybrid prediction model layer processes and predicts complex time series data through a hybrid prediction weighted formula; the hybrid prediction weighted formula predicts through a long short-term memory network LSTM, and weights and corrects the output of the LSTM through the integration characteristics of the random forest RF; the formula expression of the hybrid prediction weighted formula is In formula (6), is the mixed prediction result, z is the weighting coefficient, which is used to balance the importance of LSTM and RF prediction results. is the LSTM prediction result, is the prediction result of random forest; c t is the cell state at the current time step; T6. Optimizing the performance of the prediction model through the prediction calibration and optimization layer; the prediction calibration and optimization layer tunes the model hyperparameters through Bayesian optimization; the Bayesian optimization finds the optimal hyperparameter combination by constructing a probability model of the hyperparameters and using a sampling strategy; T7. The prediction model is dynamically adjusted according to the real-time feedback of the power grid state by the adaptive dynamic adjustment layer; the adaptive dynamic adjustment layer dynamically adjusts the prediction model by the Kalman filter prediction update formula: the Kalman filter prediction update formula uses the difference between the observed value and the prior prediction to dynamically update the prediction state by gain calculation; the formula expression of the Kalman filter prediction update formula is: In formula (7), V k is the Kalman gain, which is used to measure the impact of the observation value on the predicted state. is the predicted state based on prior knowledge, is the state estimation based on prior prediction, and B is the actual observation value; T8. Based on the prediction results, the control strategy optimization layer optimizes the control strategy of the power grid; the control strategy optimization layer optimizes the power grid control strategy through the deep Q network; the deep Q network updates the strategy through the strategy optimization objective function, and the formula expression of the strategy optimization objective function is: In formula (8), P new is the updated policy network parameter, P is the current policy network parameter, m is the learning rate, which is used to control the speed of parameter update, Q is the expected value, which represents the statistical average of state-action pairs; p is the state-action distribution under the current policy; y is the policy network, which is used to measure the probability of taking an action in a given state; T9, based on prediction error and actual grid response; A closed-loop control system is constructed through the feedback loop layer, and the feedback loop layer continuously updates the model parameters through the recursive least squares method; the parameter update formula of the recursive least squares method is: In formula (9), F t+1 is the updated parameter vector, F t is the parameter vector at time t, q is the prediction error, which is used to measure the difference between the model prediction and the actual observation, l is the forgetting factor, which is used to control the weight of historical data, is the input data; f t A weight vector representing the parameter vector.

5. The adaptive grid status area intelligent fusion terminal according to claim 1 is characterized by: The edge computing and distributed processing module achieves lightweight encapsulation of applications through containerization; the containerization method enhances the automation and flexibility of container management through container orchestration; the container orchestration allows users to define the desired cluster state through declarative configuration management, and dynamically adjusts the number of container instances according to real-time load through an automatic expansion mechanism; the chained service splits complex tasks into independent microservices through a microservice architecture; these microservices perform unified access control and routing through an API gateway to ensure the decoupling and scalability of services; The API gateway evenly distributes requests to each service instance through a load balancing algorithm; the containerization method provides intelligent routing, traffic management, policy execution and elastic control between microservices through a service grid, and realizes rapid iteration and deployment of services through a continuous integration / continuous deployment mechanism; the incremental update strategy improves update efficiency through a difference detection algorithm and a data synchronization mechanism; the difference detection algorithm identifies and records differences between data sets through hash summaries, data chunking, timestamps and difference encoding.

6. The adaptive grid status area intelligent fusion terminal according to claim 1 is characterized by: The multi-layer communication and information fusion module includes a data aggregation submodule, a data fusion submodule, a message queue service submodule, a data distribution submodule, a cross-layer information interaction submodule and a collaborative control feedback submodule; the data aggregation submodule realizes data interaction with the terminal layer through a standardized communication protocol, and performs standardized processing on the accessed data through data cleaning and format conversion methods; the data fusion submodule integrates data from different sources and types through a timestamp-based fusion strategy; And through the distributed file system and NoSQL database, the storage and management of large-scale unstructured data are realized; the timestamp-based fusion strategy fuses and transforms the original data through data abstraction and pattern recognition to provide data support for the intelligent scheduling model; the distributed file system optimizes the access efficiency of data through data partitioning and indexing mechanism; the message queue service submodule realizes asynchronous transmission and decoupling of data through the publish / subscribe mode, and ensures the reliability of message transmission through the data persistence mechanism; the data distribution submodule includes a data routing unit and a data access control unit; the data routing unit distributes the fused data to different processing modules or analysis engines through routing algorithms and load balancing methods; the data access control unit ensures the security and compliance of the data distribution process through access control lists and identity authentication mechanisms; the cross-layer information interaction submodule realizes data conversion through abstract data models, and realizes the consistency of data formats and semantics between different levels through data mapping; the collaborative control feedback submodule includes a feedback collection unit and a decision feedback unit; the feedback collection unit collects the execution results and status information of the terminal layer through the feedback mechanism; The decision feedback unit transmits the optimization decision of the upper layer to the terminal layer through the event-driven mechanism and the message queue, and adjusts the decision according to the feedback.

7. The adaptive grid status area intelligent fusion terminal according to claim 1 is characterized by: The intelligent dispatching model includes a data input layer, a feature extraction layer, a feature encoding layer, a genetic algorithm population initialization layer, a fitness function calculation layer, a selection layer, a genetic operation layer, an elite retention and population update layer, a decision output layer and an execution and feedback layer; the working method steps of the intelligent dispatching model in the intelligent fusion terminal of the station area with adaptive power grid status are as follows: S1. Obtaining raw data from sensors and databases through the data input layer, including but not limited to equipment status, environmental parameters and demand forecasts; S2, extracting features through the feature extraction layer; The feature extraction layer removes noise through time series analysis and wavelet transform, and extracts effective features through principal component analysis; S3, converting the high-dimensional feature vector generated by the feature extraction layer into binary code through the feature coding layer; S4, based on the coding features, generating an initial solution set through the genetic algorithm population initialization layer as the starting point of the optimization search; S5. Evaluate the quality of each solution in the population through a fitness function calculation layer, wherein the fitness function calculation layer considers factors including energy allocation efficiency, system stability, and cost-effectiveness; S6. Based on the individual fitness values ​​output by the upper layer, the selection layer selects individuals to participate in reproduction, and the selection layer makes selection decisions through a proportional selection method; S7, simulating natural selection and genetic mechanism through the genetic operation layer; the genetic operation layer improves population diversity and enhances global search capability through crossover and mutation; S8. At the end of each generation, the optimal solution in each generation is retained through the elite retention and population update layer, and the population is updated in combination with the newly generated offspring; S9, converting the optimal solution obtained by optimization into actual operation instructions through the decision output layer to guide power grid control and resource allocation; the decision output layer decodes the optimal solution back into decision instructions or strategy recommendations through reverse conversion coding; S10, formulating a strategy based on the decision output layer; implementing it through the execution and feedback layer and collecting feedback data on the implementation effect for adjusting model parameters or optimizing strategies.

8. The adaptive grid status area intelligent fusion terminal according to claim 1 is characterized by: The data acquisition module ensures the time consistency of sensor data through a timestamp synchronization mechanism, and realizes temporary storage of data through a circular queue to cope with the problem of mismatch between data acquisition and processing speed; the circular queue tracks the data currently stored in the buffer through a circular index mechanism to optimize space utilization, and ensures data consistency and thread safety through mutex locks and semaphores; the outlier detection identifies and processes outliers in the data through a statistical threshold detection method; the time series analysis performs time series prediction and trend analysis through an autoregressive integrated moving average model; the data formatting method converts raw data into a standardized format through the data encapsulation protocol JSON.

9. A method for controlling an intelligent fusion terminal in an adaptive grid state area, characterized in that: An adaptive grid status area intelligent fusion terminal as claimed in any one of claims 1 to 8 comprises the following steps: Step 1: Intelligent perception and data collection; Real-time monitoring of grid operation parameters through sensor networks, smart meters and environmental monitoring equipment, and real-time elimination of abnormal data through threshold filtering methods; grid operation parameters include but are not limited to voltage, current, frequency, temperature and humidity parameters; Step 2: Anomaly detection and data preprocessing; Receive preliminary clean data from the perception layer and further filter it through outlier detection algorithms to ensure data quality; then process the data through time series analysis and data formatting methods; after data formatting, transmit the data to the edge computing node through the wireless transmission protocol MQTT; Step 3: Adaptive learning and strategy generation; At the edge node, a deep reinforcement learning framework is deployed through a Docker container. The deep reinforcement learning framework dynamically learns and adjusts the control strategy according to the real-time power grid status through a deep power grid dynamic control algorithm. After the strategy is updated, it is fed back to the decision-making layer. Step 4: Future state prediction and optimization decision; The preprocessed data is used as input to predict the future state of the power grid through the power grid prediction and analysis model; the power grid prediction and analysis model performs time series analysis through a long short-term memory network to predict the future load demand or failure probability of the power grid, and evaluates the effects of different control strategies through a random forest algorithm to determine the best operation, optimize the decision output to the global dispatch, and guide resource allocation; Step 5: Data compression and efficient storage; The data generated from the preprocessing stage to the prediction and decision stage is compressed through Huffman coding to reduce storage and bandwidth requirements; the compressed data is stored in the distributed file system through the columnar storage format Parquet to achieve efficient storage and query; Step 6: Blockchain data recording; Constructing a distributed ledger based on Ethereum, which automatically records key events through smart contracts and ensures data consistency and security through consensus mechanisms and elliptic encryption algorithms; key events include policy updates, abnormal events, and optimized decision outputs; Step 7: Cross-layer communication and information fusion; Integrate multi-source data through data lake and message queue services to form a unified view, realize data aggregation, processing and efficient distribution for global decision-making; Step 8: Global optimization scheduling and resource allocation; Based on the fused data, the allocation of power grid resources is optimized through an intelligent dispatching model; the intelligent dispatching model realizes data exchange with the distributed system through a remote call protocol; the strategy is issued to the terminal for execution, and the power grid resources are adjusted to achieve global efficiency and stability optimization.

10. The adaptive grid status area intelligent fusion terminal according to claim 9, characterized in that: The Ethereum-based distributed ledger includes a smart contract module, a consensus mechanism module, a data encryption module, a distributed storage module and an audit update module; the smart contract module implements the encoding and logic processing of policy updates, abnormal events and optimization decision output events through the event listening mechanism and rule engine of the smart contract; the consensus mechanism module ensures the consistent recognition of the newly added blocks by all nodes in the network through proof of work, and realizes data synchronization and state sharing among nodes in the whole network through the P2P network protocol; the data encryption module ensures data security and integrity through elliptic curve encryption algorithm and digital signature; the distributed storage module stores the ledger data in multiple nodes in the network through the data sharding method, and realizes fast retrieval and access to distributed storage data through a distributed hash table; the audit update module includes a contract audit unit and a contract update unit; the contract audit unit strictly verifies the logic of the smart contract through a formal verification method; the contract update unit realizes dynamic update and maintenance of the contract logic through the self-upgrade mechanism of the smart contract to adapt to changes in the power grid control strategy.

Images