Intelligent power grid load regulation and control system combined with Internet of Things

Abstract

The invention discloses an intelligent power grid load regulation and control system combined with the Internet of Things, and the system comprises an Internet of Things sensing layer which is used for collecting the real-time data of each user node, and converging the real-time data to an Internet of Things gateway; the edge node load distribution layer is used for receiving real-time data, transmitting the real-time data to the edge computing node, computing a load threshold value of each edge node by adopting a balance initialization strategy, and completing preprocessing of terminal task data; the load prediction layer is used for constructing an LSTM long short-term memory network load prediction model fusing the edge real-time data and generating a load prediction result; the regulation and control strategy execution layer is used for starting a corresponding regulation and control strategy based on the load prediction result; and the feedback and iterative optimization layer is used for feeding back a regulation and control execution result to the edge node and the cloud in real time, and the edge node corrects a workload distribution strategy based on feedback data, iteratively optimizes a load prediction model parameter, and realizes a full-link closed loop. And the energy utilization efficiency is remarkably improved while the power supply stability is ensured.

Description

Technical Field

[0001] This invention belongs to the field of smart grid control, and in particular, a smart grid load control system that integrates the Internet of Things. Background Technology

[0002] In recent years, with the rapid development of the social economy and the continuous growth of electricity demand, the scientific and efficient load regulation of the smart grid, as a core component of the energy internet, has become increasingly important. Currently, traditional grid load regulation methods largely rely on manual experience or single-dimensional data collection and analysis, resulting in problems such as delayed response and insufficient load forecasting accuracy. During peak electricity consumption periods, uneven load distribution and local grid overloads frequently occur, affecting not only power supply stability but also potentially leading to increased equipment wear and energy waste. This makes it difficult to meet the demands of efficient, economical, and reliable operation of the smart grid, thus becoming an urgent technical problem to be solved. Summary of the Invention

[0003] The purpose of this invention is to provide a smart grid load control system that integrates the Internet of Things (IoT) to address the shortcomings of existing technologies. It uses IoT technology as its core support and deploys IoT sensing devices to achieve real-time collection and transmission of load data at each node of the power grid. Combined with a load forecasting model, it solves problems such as response lag, data silos, and single strategies in traditional power grid load control. While ensuring power supply stability, it significantly improves energy utilization efficiency and meets the requirements of smart grids for efficient, economical, and reliable operation.

[0004] One embodiment of this application provides a smart grid load control system integrated with the Internet of Things (IoT). The system includes: an IoT sensing layer, an edge node load allocation layer, a load prediction layer, a control strategy execution layer, and a feedback and iterative optimization layer, all connected by communication. The IoT sensing layer includes smart meters deployed on the user side of the smart grid, line sensors deployed on the distribution network side, and monitoring terminals deployed on the transformer side. The smart meters, line sensors, and monitoring terminals are used to collect real-time data from each user node, aggregate the real-time data to the IoT gateway, and transmit the real-time data through a hybrid communication network. The real-time data includes node voltage, current, power, electricity consumption, equipment temperature, and user electricity consumption pattern data. The user electricity consumption pattern data includes user type identification and electricity consumption time distribution. The edge node load distribution layer is used to receive the real-time data and transmit it to the edge computing nodes. It uses a balanced initialization strategy to calculate the load threshold of each edge node. In response to any edge node's load exceeding the threshold, it adjusts the resource allocation ratio of the virtual machines in the current edge node using a particle swarm algorithm improved by the pheromone strategy. Then, it uses a semidefinite relaxation algorithm to distribute the terminal tasks to the load-balanced edge nodes and completes the preprocessing of the terminal task data. The load prediction layer is used to construct an LSTM (Long Short-Term Memory) network load prediction model that integrates real-time edge data. It takes load data at a preset time as output and trains the load prediction model by dynamically adjusting the number of network layers, neurons, and learning rate through the steepest descent algorithm to generate load prediction results. The input features of the load prediction model include historical load data, average daily temperature, percentage of residential users, and weekday identifiers. The control strategy execution layer is used to activate the corresponding control strategy based on the load forecast results, in response to any regional load exceeding the line carrying capacity threshold. The feedback and iterative optimization layer is used to feed back the control execution results to the edge nodes and the cloud in real time. The load database is updated through the cloud, and the edge nodes correct the workload allocation strategy based on the feedback data and iteratively optimize the load prediction model parameters to achieve a closed loop across the entire link.

[0005] Optionally, the transmission rate of the hybrid communication network is not less than 300kbps, the communication bandwidth is not less than 200Mbps, and the real-time data acquisition cycle does not exceed 1s.

[0006] Optionally, in the improved particle swarm optimization algorithm for the pheromone strategy, the pheromone evaporation factor ranges from 0.05 to 0.45, the pheromone intensity ranges from 0.4 to 1.4, the learning factor of the particle swarm optimization algorithm ranges from 1.4 to 2.0, and the inertia weight decreases linearly from 0.5 to 0.2 with the number of iterations.

[0007] Optionally, the training process of the load prediction model adopts an adaptive learning rate. The initial learning rate is set to 0.001. When the validation set loss does not decrease for 6 consecutive rounds, the learning rate is reduced to 0.5 times the original value. The training batch size is set to 64, and the number of iterations is not less than 200 rounds. The load prediction model uses a Dropout layer to suppress overfitting during training, and the Dropout probability is 0.2.

[0008] Optionally, the preprocessing of the terminal task data adopts the sequence probability proportional test theory to calculate the probability that the terminal task data is accurately learned, establishes a mapping relationship between the data sample size and the privacy leakage risk, and when the privacy leakage risk exceeds a preset threshold, the terminal task data is diverted to the encrypted edge node, and the privacy protection in the terminal task data processing process is achieved through homomorphic encryption technology.

[0009] Optionally, the load forecasting results also include constructing a digital twin of the power grid to map the topology and load distribution of the physical power grid, inputting the feature parameters uploaded from the edge side, historical load data, and simulated operating condition data output by the digital twin into the load forecasting model, and dynamically adjusting the attention mechanism and hidden layer dimension of the load forecasting model through a Bayesian optimization algorithm to achieve load forecasting for a preset time period.

[0010] Optionally, the power grid digital twin is constructed using a hybrid modeling method, wherein the physical layer of the power grid digital twin uses Kirchhoff's laws to construct the power grid topology model, and the behavioral layer of the power grid digital twin uses a long short-term memory network to simulate the dynamic characteristics of the load.

[0011] Optionally, the regulation strategy is generated using a multi-objective optimization algorithm, with the dual objectives of minimizing regulation delay and minimizing grid energy consumption, and the resource allocation ratio is optimized through the steepest descent algorithm.

[0012] Optionally, the control strategy includes: first-level control pushes real-time electricity price signals to industrial user terminals through edge nodes to guide production during off-peak hours; second-level control remotely controls commercial building edge nodes through the cloud to limit the short-term power of non-core loads such as air conditioning and lighting; third-level control uses a graph game-related equilibrium strategy for interruptible residential loads through smart switches to plan charging periods and power, and balance the load of charging stations.

[0013] Optionally, the graph game-related equilibrium strategy constructs an undirected graph game model with the goal of minimizing charging delay, and solves the joint probability distribution through linear programming to achieve load balancing.

[0014] Compared with existing technologies, this invention provides a smart grid load control system integrated with the Internet of Things (IoT), comprising: an IoT sensing layer, an edge node load allocation layer, a load prediction layer, a control strategy execution layer, and a feedback and iterative optimization layer, all connected by communication. The IoT sensing layer includes smart meters deployed on the user side of the smart grid, line sensors deployed on the distribution network side, and monitoring terminals deployed on the transformer side. The smart meters, line sensors, and monitoring terminals are used to collect real-time data from each user node, aggregate the real-time data to an IoT gateway, and transmit the real-time data through a hybrid communication network. The edge node load allocation layer receives the real-time data and transmits it to edge computing nodes. It calculates the load threshold for each edge node using a balanced initialization strategy, and in response to any edge node's load exceeding the threshold, adjusts the load of the current edge node using a particle swarm optimization algorithm improved with a pheromone strategy. The resource allocation ratio of virtual machines within the edge nodes is determined, and then the semidefinite relaxation algorithm is used to distribute terminal tasks to load-balanced edge nodes, and the terminal task data is preprocessed. The load prediction layer is used to construct an LSTM long short-term memory network load prediction model that integrates real-time edge data. With load data at a preset time as the output, the model is trained by dynamically adjusting the number of network layers, neurons, and learning rate using the steepest descent algorithm to generate load prediction results. The regulation strategy execution layer is used to initiate corresponding regulation strategies based on the load prediction results, in response to any area load exceeding the line carrying capacity threshold. The feedback and iterative optimization layer is used to feed back the regulation execution results to the edge nodes and the cloud in real time. The load database is updated through the cloud, and the edge nodes correct the workload allocation strategy based on the feedback data, iteratively optimizing the load prediction model parameters to achieve a closed-loop end-to-end system. With IoT technology as the core support, the real-time collection and transmission of load data of each node in the power grid is achieved by deploying IoT sensing devices. Combined with the load prediction model, this solves the problems of response lag, data silos, and single strategies in traditional power grid load regulation. While ensuring power supply stability, it significantly improves energy utilization efficiency and meets the requirements of smart grids for efficient, economical, and reliable operation. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a smart grid load control system that integrates the Internet of Things, provided as an embodiment of the present invention. Detailed Implementation

[0016] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0017] See Figure 1 , Figure 1This invention provides an embodiment of a smart grid load control system 100 integrated with the Internet of Things (IoT). The system 100 includes: an IoT sensing layer 101, an edge node load allocation layer 102, a load prediction layer 103, a control strategy execution layer 104, and a feedback and iterative optimization layer 105, all connected by communication. The IoT sensing layer 101 includes smart meters deployed on the user side of the smart grid, line sensors deployed on the distribution network side, and monitoring terminals deployed on the transformer side. The smart meters, line sensors, and monitoring terminals collect real-time data from each user node and aggregate this data to an IoT gateway. Real-time data transmission is achieved through a hybrid communication network. The real-time data includes node voltage, current, power, electricity consumption, equipment temperature, and user electricity consumption pattern data. The user electricity consumption pattern data includes user type identification and electricity consumption time distribution. The hybrid communication network has a transmission rate of at least 300 kbps, a communication bandwidth of at least 200 Mbps, and a real-time data acquisition period of no more than 1 second.

[0018] For example, data from all sensing nodes converges to the IoT gateway, forming a global real-time dataset at time t: in, Represents any sensing node Real-time data collected at time t This represents the total set of IoT sensing nodes. Indicates the time of data collection.

[0019] The edge node load distribution layer 102 receives the real-time data and transmits it to the edge computing nodes. It calculates the load threshold for each edge node using a balanced initialization strategy. If the load of any edge node exceeds the threshold, it adjusts the resource allocation ratio of virtual machines within the current edge node using a particle swarm optimization algorithm improved with a pheromone strategy. Then, it uses a semi-definite relaxation algorithm to distribute terminal tasks to load-balanced edge nodes and completes the preprocessing of the terminal task data. Specifically, the pheromone evaporation factor in the improved particle swarm optimization algorithm ranges from 0.05 to 0.45, the pheromone intensity ranges from 0.4 to 1.4, the learning factor ranges from 1.4 to 2.0, and the inertia weight decreases linearly from 0.5 to 0.2 with each iteration. The preprocessing of the terminal task data uses a sequence probability proportionality test theory to calculate the probability that the terminal task data is accurately learned, establishing a mapping relationship between the data sample size and the privacy leakage risk. When the privacy leakage risk exceeds a preset threshold, the terminal task data is diverted to encrypted edge nodes, and homomorphic encryption technology is used to protect privacy during the terminal task data processing.

[0020] For example, real-time load computing on edge nodes is performed via edge nodes. The load at time t can be defined as the weighted sum of task volume and resource utilization: in, This represents the load weighting coefficient. Represents a node The number of tasks to be processed at time t. Represents a node Resource utilization rate at time t.

[0021] It should be noted that the semidefinite relaxation algorithm distributes terminal tasks to load-balanced edge nodes, which can be achieved by defining a task allocation matrix. ,in, Indicates task Assigned to nodes The constraints and objectives are as follows: Constraints: Single task, single node: Node resource limit: ,in, Indicates task Resource requirements; the optimization objective is expressed as: in, Indicates task At the node Delay handling.

[0022] The load prediction layer 103 is used to construct an LSTM (Long Short-Term Memory) network load prediction model that integrates real-time edge data. Using preset time-based load data as output, the model is trained by dynamically adjusting the number of network layers, neurons, and learning rate through a steepest descent algorithm to generate load prediction results. The input features of the load prediction model include historical load data, average daily temperature, percentage of residential users, and weekday identifiers. The training process of the load prediction model employs an adaptive learning rate, initially set to 0.001. When the validation set loss does not decrease for six consecutive rounds, the learning rate decays to 0.5 times its original value. The training batch size is set to 64, and the number of iterations is no less than 200 rounds. A Dropout layer is used during the training process to suppress overfitting, with a Dropout probability of 0.2.

[0023] For example, the load forecasting model aims to minimize the forecasting error by adjusting the model parameters using the steepest descent method. ,in Including parameters such as weights, biases, number of network layers, and learning rate, the loss function of this model is expressed as: in, Indicates the number of training samples. Indicates the actual load. Indicates preset time The load forecast results.

[0024] The control strategy execution layer 104 is used to initiate corresponding control strategies based on the load forecast results, in response to any region's load exceeding the line carrying capacity threshold. The feedback and iterative optimization layer 105 is used to feed back the control execution results to edge nodes and the cloud in real time, update the load database through the cloud, and the edge nodes correct the workload allocation strategy based on the feedback data, iteratively optimizing the load forecast model parameters to achieve a closed-loop connection across the entire link. The load forecast results also include mapping the topology and load distribution of the physical power grid to a digital twin of the power grid, inputting the characteristic parameters uploaded from the edge side, historical load data, and simulated operating condition data output by the digital twin into the load forecast model, and dynamically adjusting the attention mechanism and hidden layer dimension of the load forecast model using a Bayesian optimization algorithm to achieve load forecasting for a preset time period. The power grid digital twin is constructed using a hybrid modeling method, wherein the physical layer of the power grid digital twin uses Kirchhoff's laws to construct the power grid topology model, and the behavioral layer of the power grid digital twin uses a long short-term memory network to simulate the dynamic characteristics of the load. The regulation strategy is generated using a multi-objective optimization algorithm, with the dual objectives of minimizing regulation delay and minimizing grid energy consumption, and the resource allocation ratio is optimized through the steepest descent algorithm.

[0025] For example, using the minimization of control delay and the minimization of grid energy consumption as dual objectives can be used to verify whether the control delay and energy consumption have achieved the optimization goals: in, Indicates the region Maximum allowable control delay For the region Maximum permissible grid energy consumption Indicates an indicator function.

[0026] The control strategies include: Level 1 control pushes real-time electricity price signals to industrial user terminals through edge nodes to guide production during off-peak hours; Level 2 control remotely controls edge nodes of commercial buildings through the cloud to limit the short-term power of non-core loads such as air conditioning and lighting; Level 3 control uses smart switches to adopt graph game-related equilibrium strategies for interruptible residential loads to plan charging periods and power, and balance the load of charging stations.

[0027] The graph game-related equilibrium strategy constructs an undirected graph game model with the goal of minimizing charging delay, and solves the joint probability distribution through linear programming to achieve load balancing.

[0028] The smart grid load control system, integrated with the Internet of Things (IoT), adopts a five-layer architecture. It utilizes IoT technology to achieve real-time sensing, prediction, control, and optimization of the smart grid load in a closed loop. The IoT sensing layer 101 is used for data acquisition and transmission. On the user side, it includes smart meters for collecting data such as electricity consumption, voltage, and current. On the distribution network side, it includes line sensors for monitoring line power and temperature. On the transformer side, it includes monitoring terminals for collecting transformer operating status data. Data is aggregated to the IoT gateway and transmitted to edge nodes via hybrid communication networks such as 5G, LoRa, and fiber optics. The edge node load allocation layer 102 is used for dynamic resource allocation and task preprocessing, then calculates the load threshold for each edge node. This load threshold can be determined based on historical data and real-time traffic. If the load of a node exceeds the threshold, a pheromone update mechanism using a particle swarm optimization algorithm improved with a pheromone strategy or a fusion of ant colony algorithms is employed to optimize the virtual machine resource allocation ratio. A semidefinite relaxation algorithm is used to distribute terminal tasks to load-balanced edge nodes, avoiding single-point overload. The load forecasting layer 103 is used to build a dynamic load forecasting model based on LSTM. The input features of this model include historical load data, average daily temperature, the proportion of residential users, and weekday identifiers. The number of network layers, neurons, and learning rate are optimized in real time using the steepest descent algorithm or gradient descent method, and the output is the load forecast result after a preset time, such as 15 minutes. The regulation strategy execution layer 104 is used for real-time load regulation. When the load in any area exceeds the line's carrying capacity threshold, price signals can be used to guide users to adjust their electricity consumption behavior, or energy storage systems or renewable energy sources can be activated to supplement power supply. Alternatively, power rationing or rotational shutdowns can be implemented for non-critical loads (such as air conditioners and charging piles). The feedback and iterative optimization layer 105 is used for end-to-end closed-loop optimization. It can feed back the regulation execution results to edge nodes and the cloud in real time. The cloud updates the load database to provide new data for model training, and edge nodes optimize the load forecasting model parameters by correcting workload allocation strategies.

[0029] For example, the IoT sensing layer includes devices such as smart meters deployed on the user side, line sensors on the distribution network side, and transformer monitoring terminals, which collect load data and operating status information such as voltage, current, power, electricity consumption, and equipment temperature of each node in real time, and aggregate the data to the IoT gateway through wireless communication technology.

[0030] The edge node load distribution layer can also convert the massive amounts of data from the IoT sensing layer through an IoT gateway and then upload them to the cloud data center via communication networks such as 5G / fiber optics. The cloud data center cleans, verifies, and standardizes the data, removes outliers, and fills in missing data to form a structured load database, providing high-quality data support for subsequent analysis.

[0031] The load forecasting layer and control strategy execution layer construct a load forecasting and control model based on historical and real-time load data in a cloud database. The model's operation process is as follows: First, load impact factor data samples are collected. Through the IoT sensing layer, various factors affecting power grid load are continuously collected, including historical load data, user type, meteorological data (temperature, humidity, season), and date type (weekday, holiday), and stored in a historical database. Some key parameters are shown in Table 1 below: Table 1: Data on various factors affecting power grid load Then, a load forecasting model was constructed using the LSTM (Long Short-Term Memory) algorithm. Historical load data and influencing factors were used as inputs, with load data for the next 24 hours as the output, to train the model. During training, model performance was optimized by adjusting parameters such as the number of network layers, neurons, and learning rate, keeping the prediction error below 5%.

[0032] Based on load forecasting results and the real-time operating status of the power grid, the system automatically generates multi-level control strategies. For example, when it is predicted that the load in a certain area will exceed the line carrying capacity threshold, the corresponding control strategy is activated.

[0033] The control strategy execution layer can also have real-time monitoring and anomaly detection functions. By comparing the deviation between real-time load data and predicted data, when the deviation exceeds a set threshold, it is determined to be a potential anomaly, immediately triggering an alarm and locating the abnormal node, and combining historical fault data to assist in judging the fault type.

[0034] The feedback and iterative optimization layer is installed in intelligent control devices at each power consumption node, including smart circuit breakers, controllable load switches, and smart sockets. When the control strategy execution layer issues a control command, the feedback and iterative optimization layer receives the command via the Internet of Things (IoT) and automatically executes load adjustment operations, such as disconnecting unnecessary circuits and adjusting equipment operating power. Simultaneously, the execution results are fed back to the cloud, forming a closed-loop control process of "sensing-analysis-decision-execution-feedback," achieving dynamic balance and efficient utilization of the power grid load.

[0035] By collecting and intelligently analyzing grid load data in real time through IoT sensing modules, load fluctuations can be accurately predicted and power supply strategies dynamically adjusted, reducing equipment damage and maintenance costs caused by overload. Simultaneously, multi-level control strategies enable rational load allocation, improving energy efficiency, reducing unit energy consumption in grid operation, and saving operating costs for power companies. Furthermore, the integration of IoT technology with smart grids will drive the industrialization of related sensing devices, communication modules, and control systems, promoting technological maturity and reducing product costs, resulting in significant economic benefits for large-scale deployment.

[0036] This system significantly improves the stability and reliability of the power grid by constructing a closed-loop system of "sensing-transmission-analysis-control," reducing power outages caused by uneven load distribution and ensuring normal power supply for industrial production, commercial operations, and residential life. In response to extreme weather or peak electricity demand, the system can quickly respond and implement orderly control measures, avoiding the social impact of large-scale power outages. Simultaneously, by guiding users to use electricity rationally, it helps cultivate energy-saving awareness, promotes the optimization of the social energy structure and sustainable development, and is of great significance for building a safe, efficient, and green smart grid system.

[0037] Compared with existing technologies, this invention provides a smart grid load control system integrated with the Internet of Things (IoT), comprising: an IoT sensing layer, an edge node load allocation layer, a load prediction layer, a control strategy execution layer, and a feedback and iterative optimization layer, all connected by communication. The IoT sensing layer includes smart meters deployed on the user side of the smart grid, line sensors deployed on the distribution network side, and monitoring terminals deployed on the transformer side. The smart meters, line sensors, and monitoring terminals are used to collect real-time data from each user node, aggregate the real-time data to an IoT gateway, and transmit the real-time data through a hybrid communication network. The edge node load allocation layer receives the real-time data and transmits it to edge computing nodes. It calculates the load threshold for each edge node using a balanced initialization strategy, and in response to any edge node's load exceeding the threshold, adjusts the load of the current edge node using a particle swarm optimization algorithm improved with a pheromone strategy. The resource allocation ratio of virtual machines within the edge nodes is determined, and then the semidefinite relaxation algorithm is used to distribute terminal tasks to load-balanced edge nodes, and the terminal task data is preprocessed. The load prediction layer is used to construct an LSTM long short-term memory network load prediction model that integrates real-time edge data. With load data at a preset time as the output, the model is trained by dynamically adjusting the number of network layers, neurons, and learning rate using the steepest descent algorithm to generate load prediction results. The regulation strategy execution layer is used to initiate corresponding regulation strategies based on the load prediction results, in response to any area load exceeding the line carrying capacity threshold. The feedback and iterative optimization layer is used to feed back the regulation execution results to the edge nodes and the cloud in real time. The load database is updated through the cloud, and the edge nodes correct the workload allocation strategy based on the feedback data, iteratively optimizing the load prediction model parameters to achieve a closed-loop end-to-end system. With IoT technology as the core support, the real-time collection and transmission of load data of each node in the power grid is achieved by deploying IoT sensing devices. Combined with the load prediction model, this solves the problems of response lag, data silos, and single strategies in traditional power grid load regulation. While ensuring power supply stability, it significantly improves energy utilization efficiency and meets the requirements of smart grids for efficient, economical, and reliable operation.

[0038] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0039] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0040] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0041] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0042] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0043] If the aforementioned integrated units are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0044] The embodiments of the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A smart grid load control system integrating the Internet of Things, characterized in that, The system includes: an IoT sensing layer with communication connectivity, an edge node load distribution layer, a load prediction layer, a control strategy execution layer, and a feedback and iterative optimization layer. The IoT sensing layer includes smart meters deployed on the user side of the smart grid, line sensors deployed on the distribution network side, and monitoring terminals deployed on the transformer side. The smart meters, line sensors, and monitoring terminals are used to collect real-time data from each user node, aggregate the real-time data to the IoT gateway, and transmit the real-time data through a hybrid communication network. The real-time data includes node voltage, current, power, electricity consumption, equipment temperature, and user electricity consumption pattern data. The user electricity consumption pattern data includes user type identification and electricity consumption time distribution. The edge node load distribution layer is used to receive the real-time data and transmit it to the edge computing nodes. It uses a balanced initialization strategy to calculate the load threshold of each edge node. In response to any edge node's load exceeding the threshold, it adjusts the resource allocation ratio of the virtual machines in the current edge node using a particle swarm algorithm improved by the pheromone strategy. Then, it uses a semidefinite relaxation algorithm to distribute the terminal tasks to the load-balanced edge nodes and completes the preprocessing of the terminal task data. The load prediction layer is used to construct an LSTM (Long Short-Term Memory) network load prediction model that integrates real-time edge data. It takes load data at a preset time as output and trains the load prediction model by dynamically adjusting the number of network layers, neurons, and learning rate through the steepest descent algorithm to generate load prediction results. The input features of the load prediction model include historical load data, average daily temperature, percentage of residential users, and weekday identifiers. The control strategy execution layer is used to activate the corresponding control strategy based on the load forecast results, in response to any regional load exceeding the line carrying capacity threshold. The feedback and iterative optimization layer is used to feed back the control execution results to the edge nodes and the cloud in real time. The load database is updated through the cloud, and the edge nodes correct the workload allocation strategy based on the feedback data and iteratively optimize the load prediction model parameters to achieve a closed loop across the entire link.

2. The system according to claim 1, characterized in that, The hybrid communication network has a transmission rate of no less than 300kbps, a communication bandwidth of no less than 200Mbps, and a real-time data acquisition cycle of no more than 1s.

3. The system according to claim 2, characterized in that, The particle swarm optimization algorithm with improved pheromone strategy has a pheromone evaporation factor ranging from 0.05 to 0.45, a pheromone intensity ranging from 0.4 to 1.4, a learning factor ranging from 1.4 to 2.0, and an inertia weight that decreases linearly from 0.5 to 0.2 with the number of iterations.

4. The system according to claim 3, characterized in that, The training process of the load prediction model adopts an adaptive learning rate. The initial learning rate is set to 0.

001. When the validation set loss does not decrease for 6 consecutive rounds, the learning rate is reduced to 0.5 times the original value. The training batch size is set to 64, and the number of iterations is no less than 200 rounds. The load prediction model uses a Dropout layer to suppress overfitting during training, with a Dropout probability of 0.

2.

5. The system according to claim 4, characterized in that, The preprocessing of the terminal task data adopts the sequence probability proportional test theory to calculate the probability that the terminal task data is accurately learned, establishes a mapping relationship between the data sample size and the privacy leakage risk, and when the privacy leakage risk exceeds a preset threshold, the terminal task data is diverted to the encrypted edge node, and the privacy protection in the terminal task data processing process is achieved through homomorphic encryption technology.

6. The system according to claim 5, characterized in that, The load forecasting results also include constructing a digital twin of the power grid to map the topology and load distribution of the physical power grid, inputting the feature parameters uploaded from the edge side, historical load data and simulated operating condition data output by the digital twin into the load forecasting model, and dynamically adjusting the attention mechanism and hidden layer dimension of the load forecasting model through a Bayesian optimization algorithm to achieve load forecasting for a preset time period.

7. The system according to claim 6, characterized in that, The power grid digital twin is constructed using a hybrid modeling approach. The physical layer of the power grid digital twin uses Kirchhoff's laws to construct the power grid topology model, while the behavioral layer of the power grid digital twin uses a long short-term memory network to simulate the dynamic characteristics of the load.

8. The system according to claim 7, characterized in that, The regulation strategy is generated using a multi-objective optimization algorithm, with the dual objectives of minimizing regulation delay and minimizing grid energy consumption, and the resource allocation ratio is optimized through the steepest descent algorithm.

9. The system according to claim 8, characterized in that, The control strategy includes: Level 1 control pushes real-time electricity price signals to industrial user terminals through edge nodes to guide production during off-peak hours; Level 2 control remotely controls edge nodes of commercial buildings through the cloud to limit the short-term power of non-core loads such as air conditioning and lighting; Level 3 control uses smart switches to adopt graph game-related equilibrium strategies for interruptible residential loads to plan charging periods and power, and balance the load of charging stations.

10. The system according to claim 9, characterized in that, The graph game-related equilibrium strategy constructs an undirected graph game model with the goal of minimizing charging delay, and solves the joint probability distribution through linear programming to achieve load balancing.

Images