Energy management system and method and electronic equipment

Abstract

The invention relates to the technical field of Internet of Things, and provides an energy management system and method and electronic device.The system comprises a cloud resource platform and a local distributed system which are connected through a network, and the local distributed system comprises a core node and a plurality of computing nodes; the core node is used for splitting an inference operation task into a plurality of sub-tasks and distributing the plurality of sub-tasks to each calculation node; the computing node is used for acquiring task data required by the distributed subtasks and inputting the task data into the currently deployed energy management model to obtain an equipment control strategy output by the energy management model; wherein the energy management model is downloaded from the cloud resource platform by the core node and then deployed to each computing node. By executing the reasoning operation task on the local distributed system, the operation efficiency can be improved, the privacy disclosure risk can be reduced, and the model can be flexibly adjusted and updated by downloading the model from the cloud resource platform and deploying the model.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things, and in particular to an energy management system, method and electronic equipment. Background Art

[0002] As people's living standards continue to improve, the power consumption of household appliances has gradually become a major factor in household energy consumption, which also poses challenges to power infrastructure. In the field of smart homes, home energy management can optimize device operation and achieve reasonable energy distribution, thereby achieving the goals of improving energy efficiency, reducing energy consumption and living costs.

[0003] Currently, home energy management systems primarily utilize smart meters and IoT sensors to collect operational data from household appliances. This data is then uploaded to cloud servers for analysis and processing, generating optimized control strategies to regulate device operating status. However, this architecture, which relies on centralized cloud processing, has significant limitations. First, frequent data transmission exposes household electricity usage information (such as household electricity consumption patterns and device usage habits) to external networks, posing a security risk of privacy leakage. Second, the round-trip network transmission and cloud computing inevitably introduce communication delays, reducing system response efficiency. Furthermore, the system's normal operation is highly dependent on the stability of the network connection. A network outage or cloud service failure will directly render the energy management function ineffective. Summary of the Invention

[0004] The present invention provides an energy management system, method and electronic device to address the defects of existing energy management, such as privacy leakage risk, low system operation efficiency and poor stability.

[0005] The present invention provides an energy management system, comprising a cloud resource platform and a local distributed system connected via a network, wherein the local distributed system comprises a core node and a plurality of computing nodes, wherein the plurality of computing nodes are all communicatively connected to the core node; The core node is used to split the reasoning operation task into multiple subtasks and distribute the multiple subtasks to each computing node; The computing node is used to obtain task data required for the assigned subtask and input the task data into the currently deployed energy management model to obtain the device control strategy output by the energy management model, and the device control strategy is used to adjust the operating status of the electrical equipment; The energy management model is downloaded from the cloud resource platform by the core node and then deployed to each computing node.

[0006] According to an energy management system provided by the present invention, the local distributed system further comprises a networking node, wherein the networking node is communicatively connected to the core node and the cloud resource platform respectively; The networked node is configured to access the check update interface of the cloud resource platform to detect whether the energy management model on the cloud resource platform has been updated; The networked node is further configured to, upon detecting that the energy management model has been updated, download the updated energy management model based on the model download interface of the cloud resource platform and send the updated energy management model to the core node; The core node is used to deploy the updated energy management model to each computing node.

[0007] According to an energy management system provided by the present invention, the networked nodes establish point-to-point internal data communication with the core nodes, the networked nodes are communicated with the cloud resource platform through the external Internet, the computing nodes and the core nodes are connected through an internal communication bus, and no communication is established between the computing nodes and the networked nodes.

[0008] According to an energy management system provided by the present invention, the local distributed system further comprises a plurality of sensors, and the plurality of sensors are all connected to an internal communication bus; The multiple sensors are used to collect perception content data in the task data at a preset frequency, and transmit the perception content data to each computing node through the internal communication bus. The perception content data includes device operation information and / or environmental perception information of the electrical device.

[0009] According to an energy management system provided by the present invention, the task data further includes symbol content data, and the symbol content data includes external factors and / or user habits that affect the use of the electrical equipment; The user habits are recorded locally by the computing nodes and / or the core node, and the external factors are downloaded from the cloud resource platform by the networked nodes at a preset frequency and then transmitted to the computing nodes via the core node.

[0010] According to an energy management system provided by the present invention, the computing node is further configured to collect incremental training data when executing the assigned subtask, and transmit the incremental training data to the core node at a preset frequency; The core node is further configured to perform incremental training on a local energy management model based on the incremental training data to obtain an energy management model with fine-tuned parameters.

[0011] According to an energy management system provided by the present invention, the core node is further configured to record model training information when incremental training is performed on a local energy management model, the model training information including algorithm structure information, parameter update differences, and training metadata of the energy management model, the training metadata including at least one of the number of training rounds, batch size, number of inputs, and learning rate; The networked node is further configured to upload the model training information to the cloud resource platform at preset time intervals, so that the cloud resource platform can update and publish the energy management model on the cloud resource platform based on all received model training information.

[0012] According to an energy management system provided by the present invention, the local distributed system includes multiple systems, each of which is deployed in a user's home space, and the cloud resource platform includes a federated learning platform and an information resource platform; The federated learning platform is used to update the energy management model using a federated learning method based on the model training information uploaded by each local distributed system, and publish the updated energy management model when it meets preset publishing conditions for download and deployment by each local distributed system; The information resource platform is used to publish software update information and public information, and the public information includes external factors that affect the use of the electrical equipment.

[0013] The present invention also provides an energy management method, which is applied to a local distributed system, wherein the local distributed system includes a core node and multiple computing nodes, and the method includes: Based on the core node, the reasoning operation task is split into multiple subtasks, and the multiple subtasks are assigned to each computing node; Based on each computing node, task data required for the assigned subtask is obtained, and the task data is input into a currently deployed energy management model to obtain a device control strategy output by the energy management model, wherein the device control strategy is used to adjust the operating status of the electrical device; The energy management model is downloaded from the cloud resource platform by the core node and then deployed to each computing node.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements any of the above-mentioned energy management methods when executing the computer program.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which implements any of the above-mentioned energy management methods when executed by a processor.

[0016] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above energy management methods.

[0017] The energy management system, method, and electronic device provided by the present invention use core nodes to split complex inference calculation tasks into multiple subtasks and assign them to individual computing nodes for collaborative processing. This significantly improves the computational efficiency of energy management models and meets the high-real-time requirements of home energy management. The entire computation process is completed within a local distributed system, reducing data transmission and cloud latency. This allows for rapid processing of inference calculation tasks, timely generation of device control strategies, and efficient adjustment of the operating status of electrical devices. Furthermore, data processing within the local distributed system reduces the need to upload household electricity usage data to the cloud, reduces the risk of privacy leaks, and ensures the security of users' private information, such as their home energy usage. Furthermore, core nodes can retrieve the latest energy management models from a cloud resource platform and dynamically deploy them to computing nodes, helping to ensure algorithm optimization and updates and offering greater flexibility. Compared to traditional cloud-based management solutions, the distributed architecture of the present invention maintains energy management functions through local nodes even during network outages, offering greater stability, reliability, and fault tolerance. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or related technologies, the following is a brief introduction to the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is one of the structural diagrams of the energy management system provided by the present invention; Figure 2 This is the second structural diagram of the energy management system provided by the present invention; Figure 3 This is the third structural diagram of the energy management system provided by the present invention; Figure 4 This is the fourth structural diagram of the energy management system provided by the present invention; Figure 5 It is a structural diagram of the energy management model provided by the present invention; Figure 6 This is the fifth structural diagram of the energy management system provided by the present invention; Figure 7 This is a schematic diagram of the overall process of the energy management system provided by the present invention; Figure 8 It is a flow chart of the energy management method provided by the present invention; Figure 9 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0021] With the booming economy and the continuous improvement of people's living standards, household appliances are becoming increasingly popular and versatile. This has led to a continuous increase in the proportion of electricity consumed by household appliances in household energy consumption, gradually becoming the dominant factor in household energy consumption. This change not only puts pressure on household financial expenditures but also imposes stricter requirements and challenges on power infrastructure. However, current household energy utilization efficiency is generally low, and energy waste is relatively serious, resulting in unnecessary resource depletion and economic losses.

[0022] The rise of the smart home sector has brought new opportunities for home energy management. By precisely adjusting the operating status and timing of various electrical devices, home energy management optimizes energy distribution and usage, achieving multiple goals, including improving energy efficiency, reducing energy consumption, and lowering the cost of living. Existing smart home systems have taken a significant step forward in home energy management, providing preliminary solutions that have achieved certain energy conservation, carbon reduction, and cost reduction, playing a positive role in the rational use of household energy.

[0023] Currently, users prefer to leverage artificial intelligence (AI) methods for home energy management, hoping to leverage its powerful data processing and learning capabilities to further enhance the intelligence of their home energy management. However, existing AI-based home energy management systems and methods suffer from numerous drawbacks that are difficult to address. One approach relies on cloud-based computing. While this approach allows for flexible adjustments to AI algorithms to suit diverse home energy management needs, it also presents significant drawbacks. Because all data must be uploaded to the cloud for analysis, private information such as users' household electricity usage habits and device usage is at risk of being stolen or misused during network transmission, making privacy and security difficult to guarantee. Furthermore, data transmission latency makes it difficult to issue energy management commands in a timely manner, impacting the system's real-time and accuracy, especially in scenarios requiring rapid response.

[0024] Another approach places all computations on local devices. While this avoids privacy concerns associated with data transmission, limited local computing power and storage resources make algorithm updates and optimizations relatively difficult. Local devices struggle to adapt to evolving energy management needs and new technological developments. Furthermore, because data is limited to a single household and lacks broader sample support, the algorithm's training efficiency and generalization capabilities are limited, hindering sustained improvement in energy management effectiveness and performance.

[0025] These shortcomings make it difficult for existing solutions to balance privacy and security, real-time response, and algorithm adaptability, hindering the further development and widespread application of home energy management systems. To address this, the present invention provides an energy management system and method that can protect user data privacy while improving algorithm flexibility and computational efficiency, thereby overcoming these shortcomings.

[0026] Figure 1 This is one of the structural diagrams of the energy management system provided by the present invention, such as Figure 1 As shown, the system includes a cloud resource platform 110 and a local distributed system 120 connected via a network, wherein the local distributed system 120 includes a core node 121 and a plurality of computing nodes 122, and the plurality of computing nodes 122 are all in communication connection with the core node 121; The core node 121 is used to split the reasoning operation task into multiple subtasks and distribute the multiple subtasks to each computing node 122; The computing node 122 is used to obtain task data required for the assigned subtask and input the task data into the currently deployed energy management model to obtain the device control strategy output by the energy management model, and the device control strategy is used to adjust the operating status of the electrical equipment; The energy management model is downloaded from the cloud resource platform 110 by the core node 121 and then deployed to each computing node.

[0027] Specifically, the energy management system provided by embodiments of the present invention may include a cloud-based resource platform and a local distributed system, which interact via a network connection. The cloud-based resource platform is a resource aggregation and service provision system based on cloud computing technology. It can be deployed on a cloud server and provide online services to all users' local distributed systems. In the energy management system, the cloud-based resource platform primarily stores and manages energy management models, facilitating downloads and updates to core nodes.

[0028] A local distributed system is composed of multiple independent computing devices (nodes) located in different locations within a home, interconnected through a specific network architecture, and collaboratively completing specific tasks. In a home energy management system, the local distributed system serves as an execution terminal, deployed in each user's home space. It is responsible for managing and controlling the electrical devices within the home. It can adjust the operating status of electrical devices in real time based on the device control strategies output by the energy management model to achieve rational energy allocation and efficient utilization.

[0029] It can be understood that network connectivity refers to the process of connecting different devices or systems through physical media (such as optical fiber, network cables, wireless signals, etc.) and communication protocols (such as TCP / IP), enabling them to communicate and exchange data. In energy management systems, network connectivity enables information transmission and interaction between cloud resource platforms and local distributed systems. For example, local distributed systems and cloud resource platforms can be connected via home broadband networks (such as fiber broadband, ADSL, etc.) or mobile networks (such as 4G, 5G). Home routers serve as network access devices, connecting local distributed systems to the internet and establishing a communication link with the cloud resource platform. Through this connection, core nodes can download energy management models from the cloud resource platform and upload relevant data (such as model training data) to the cloud for storage and analysis.

[0030] Furthermore, the local distributed system can include a core node and multiple compute nodes. The core node plays a management, control, and coordination role. It is responsible for splitting the inference calculation task into multiple subtasks and rationally allocating these subtasks to each compute node based on the computing power and load of each compute node. At the same time, the core node is also responsible for downloading the energy management model from the cloud resource platform and deploying it to each compute node, ensuring that the entire local distributed system can perform calculations and control according to a unified model.

[0031] Here, inference computing refers to the process of using a trained model to process and analyze input data to produce corresponding output results. In energy management systems, inference computing tasks typically involve calculating control strategies for individual appliances, such as adjusting air conditioning temperature settings or turning off unnecessary lighting, based on the current operating status of household appliances, environmental data, external factors affecting appliance usage, and user habits.

[0032] Subtasks are relatively independent small tasks that are derived from the decomposition of an inference operation task. Each subtask has clear input data, operation logic, and expected output results. For example, one subtask might analyze the current operating data of the living room air conditioner and calculate the optimal temperature setting for the living room air conditioner based on an energy management model; another subtask might process the data of the bedroom air conditioner and provide recommended adjustments to its operating status.

[0033] Compute nodes are devices that perform specific computing tasks within a local distributed system. Each compute node is assigned subtasks by the core node. Upon receiving a subtask, the compute node obtains the required task data and inputs this data into the currently deployed energy management model. The model then calculates device control strategies and sends these strategies to the corresponding electrical devices to adjust their operating status.

[0034] It should be understood that communications connections exist between multiple compute nodes, as well as between each compute node and the core node. Here, a communications connection refers to a logical channel established between two or more devices for data transmission and interaction. It specifies the format, rate, protocol, and other rules for data transmission, ensuring accurate and reliable information exchange between devices. In a local distributed system, communications connections enable real-time transmission of task information, model data, control strategies, and more between core nodes and compute nodes, and between individual compute nodes.

[0035] Before executing inference operations, the core node can first download the energy management model from the cloud resource platform and deploy it to each compute node. The model download and deployment process can be specifically implemented through the following steps: First, the core node establishes a communication connection with the cloud resource platform via the network. The core node sends a request to the cloud resource platform, explicitly specifying the identifier of the energy management model to be downloaded (such as the model name, version number, etc.). After receiving the request, the cloud resource platform sends the corresponding energy management model file to the core node. After receiving the model file, the core node parses it and reads the model's structural information, parameter information, etc. The core node then deploys the parsed model to each compute node, for example, by copying the model file to the compute node's designated storage location, or by remotely calling the compute node's interface to transfer the model data to the compute node's memory, and initializes the model's runtime environment on the compute node.

[0036] After the energy model is locally deployed, it can be used to perform inference operations. The core node first analyzes the inference operation task, understanding factors such as the data input size, the model's computational complexity, and the performance characteristics of the compute nodes. Based on the results of this task analysis, the core node can partition the input data. For example, household appliances can be grouped by area (such as living room, bedroom) or type (such as air conditioner, refrigerator), with each group corresponding to a portion of the input data. This way, the data for each group can serve as input for a subtask. Furthermore, if the energy management model's computational process can be broken down into multiple relatively independent steps, the core node can also decompose the model's computation into multiple substeps, each corresponding to a subtask. For example, if the model first performs data preprocessing, then feature extraction, and finally policy generation, these three steps can be divided into three subtasks.

[0037] After splitting the inference computation tasks, the core node can further evaluate the performance of each compute node, including factors such as computing power (such as the number of CPU cores and memory size), current load (such as the number of tasks being executed and CPU utilization), and network bandwidth. This evaluation allows the core node to understand the processing power and availability of each compute node. Based on the node evaluation results, the core node formulates a task allocation strategy. This task allocation strategy can be round-robin (assigning subtasks to each compute node in sequence), load balancing (preferentially assigning subtasks to compute nodes with lower current loads), or computing power-based allocation (assigning more computationally intensive subtasks to compute nodes with greater computing power).

[0038] The core node assigns subtasks to the compute nodes according to the established task allocation strategy. This allocation process may include sending task descriptions (including the subtask's input data address, computational logic, and so on) to the compute nodes over the network. After receiving the subtasks, the compute nodes begin executing the corresponding computations.

[0039] Specifically, after receiving a subtask, each computing node first obtains the task data required for that subtask. Here, task data required for a subtask refers to the data required to input into the energy management model in order to complete the subtask. For example, task data may include device operating data, environmental perception data, external factors affecting electrical device usage (such as weather forecasts, extreme weather warnings, power supply status, etc.), and user habits. This data can be obtained through sensor collection, access to public information, local records, and other methods.

[0040] After obtaining the task data required for the subtask currently being processed, the computing node can input this data into the currently deployed energy management model, thereby obtaining the device control strategy output by the model. Here, the currently deployed energy management model refers to the algorithm model that the core node has recently downloaded from the cloud resource platform and deployed to each computing node for performing home energy management reasoning operations. It should be understood that due to the continuous development of home energy management needs and technologies, the core node will regularly download updated energy management models from the cloud resource platform and redeploy them to the computing nodes to ensure that the model can always adapt to new situations and requirements.

[0041] It's important to note that a device control strategy refers to a series of tasks that change the operating state of an electrical device, such as "Set the bedroom air conditioner to 26°C dehumidification" or "Schedule the washing machine to start at 3:00 AM." A control strategy here is an event instruction, not a control signal. The energy management model only generates event instructions, while the control signals for the specific electrical devices required to implement these event instructions are handled by subsequent modules based on the specific device model and interface. For example, the control strategy "Set the bedroom air conditioner to 26°C dehumidification" is actually an API (Application Programming Interface) call, transmitted to the IoT remote controller of a certain brand of air conditioner. The remote controller then converts it into a specific control signal for that brand of air conditioner, controlling the mode and temperature of the bedroom air conditioner.

[0042] The system provided by the embodiments of the present invention uses core nodes to split complex inference computing tasks into multiple subtasks and assign them to individual computing nodes for collaborative processing. This significantly improves the computational efficiency of energy management models and meets the high-real-time requirements of home energy management. The entire computation process is completed within the local distributed system, reducing data transmission and cloud latency. This allows for rapid processing of inference computing tasks, timely generation of device control strategies, and efficient adjustment of the operating status of electrical devices. Furthermore, data processing within the local distributed system reduces the need to upload household electricity usage data to the cloud, minimizing the risk of privacy leaks and safeguarding the security of users' private information, such as their home energy usage. Furthermore, core nodes can retrieve the latest energy management models from the cloud resource platform and dynamically deploy them to computing nodes, helping to ensure algorithm optimization and updates and providing greater flexibility. Compared to traditional solutions that rely solely on the cloud for management, the distributed architecture of the present invention maintains energy management functions through local nodes even during network outages, offering greater stability, reliability, and fault tolerance.

[0043] Based on the above embodiments, Figure 2 This is the second structural diagram of the energy management system provided by the present invention, such as Figure 2As shown, the local distributed system 120 further includes a networking node 123, and the networking node 123 is communicatively connected to the core node 121 and the cloud resource platform 110 respectively; The network node 123 is used to access the check update interface of the cloud resource platform 110 to detect whether the energy management model on the cloud resource platform 110 has been updated; The network node 123 is further configured to, upon detecting that the energy management model has been updated, download the updated energy management model based on the model download interface of the cloud resource platform 110 and send the updated energy management model to the core node 121; The core node 121 is used to deploy the updated energy management model to each computing node 122 .

[0044] It's important to note that a network node is a specific node within a local distributed system responsible for communicating and interacting with the cloud resource platform. Acting as a bridge between the local system and the external cloud, it handles data transmission, information exchange, and key operations (such as model update detection and downloading), ensuring that the local distributed system can access the latest information and resources from the cloud resource platform.

[0045] Specifically, the cloud resource platform provides an update check interface that returns information such as whether a model update is available and the update type. Networked nodes can access this update check interface at a set interval to obtain updated information about the energy management model. This interval can be configured based on actual needs. For example, the access frequency can be set to once a day, once a week, or dynamically adjusted based on specific business logic.

[0046] At a set time interval, a networked node can initiate an access request to the cloud resource platform's check-update interface. This request will contain necessary information, such as the networked node's identity and authentication information, so that the cloud resource platform can recognize and process the request. The cloud resource platform's check-update interface is a specific network interface provided by the cloud resource platform that allows external systems (such as networked nodes in a local distributed system) to query the cloud resource platform for updates to specific resources (such as energy management models) by sending specific requests, as well as information about the type of update.

[0047] When the networked node obtains information that the model has been updated by accessing the check update interface, it indicates that the cloud resource platform has optimized, improved or corrected the management model. At this time, the networked node can send a download request to the model download interface of the cloud resource platform to obtain the current updated energy management model. Here, the cloud resource platform also provides a model download interface, which can provide an updated energy management model, including both algorithm structure and model parameters. When only the model parameter information is updated, the model download interface returns the parameters and content tags; when the algorithm structure is updated, the interface returns the complete energy management model and content tags, and also includes guidance and prompt information for distributed inference deployment.

[0048] After downloading the updated energy management model, the networked nodes will send it to the core node, which will then deploy the updated energy management model (i.e., the latest version of the model currently downloaded) to each local compute node. It should be understood that before deploying the model, the core node will first check the status of each compute node, including its operational status (whether it is online and operating properly), sufficient storage space, and current load. This ensures that the compute nodes can properly receive and deploy the new model.

[0049] It is understandable that the updated energy management model can be obtained by collecting more actual operation data and retraining the model. Its accuracy and efficiency in energy management have been improved, and it can more accurately generate equipment control strategies based on home environment and user needs.

[0050] Based on any of the above embodiments, Figure 3 This is the third structural diagram of the energy management system provided by the present invention, such as Figure 3 As shown, the networking node 123 establishes point-to-point internal data communication with the core node 121, the networking node 123 is communicated with the cloud resource platform 110 through the external Internet 130, the computing nodes 122 and the core node 121 are connected via an internal communication bus 124, and no communication is established between the computing nodes 122 and the networking node 123.

[0051] Specifically, to further ensure user privacy and security, embodiments of the present invention isolate internal and external communications within a local distributed system. Specifically, the local distributed system also includes an internal communication bus, to which all compute nodes and core nodes are connected. Each compute node only communicates internally with core nodes and other compute nodes via the bus. Networked nodes only establish point-to-point internal data communications with core nodes, and only networked nodes communicate with the cloud resource platform via the external internet. Individual compute nodes do not communicate with other networked nodes.

[0052] Here, an internal communication bus refers to a hardware or software mechanism used to transmit and communicate data between compute nodes and core nodes within a local distributed system. It provides a standardized communication interface, enabling nodes to easily exchange data and collaborate. The internal communication bus in a local distributed system can be implemented using a network, with compute nodes and core nodes connected to the bus via wireless networks (Wi-Fi) or Ethernet. The internal communication bus is based on the local area network (LAN) of the distributed system. The LAN and wide area network (WAN) use different link layer and physical layer devices to provide physical isolation within the network.

[0053] In a local distributed system, only networked nodes establish a WAN connection, exchanging data with the cloud resource platform via the internet. Networked nodes exchange data only with core nodes. Networked nodes and core nodes can be hosted by different physical devices, communicating via dedicated connections. Alternatively, networked nodes and core nodes can be hosted by the same physical device, using separate network ports for connecting to the LAN and WAN. In this case, networked nodes and core nodes communicate only within the device itself.

[0054] In an embodiment of the present invention, networked nodes communicate with core nodes through point-to-point communication. This direct communication method reduces the intermediate links in data transmission, reduces data transmission delays and packet loss rates, and improves the efficiency and reliability of data transmission. As the interface between the local distributed system and the external Internet, networked nodes are responsible for communicating with the cloud resource platform. By isolating networked nodes from other computing nodes, system security and isolation can be enhanced. In addition, computing nodes focus on processing local tasks and executing control strategies, and do not participate in communications with external networks, ensuring the security and stability of internal system data.

[0055] Based on any of the above embodiments, Figure 4 This is the fourth structural diagram of the energy management system provided by the present invention, such as Figure 4 As shown, the local distributed system 120 further includes a plurality of sensors 125 , and the plurality of sensors 125 are all connected to the internal communication bus 124 ; The multiple sensors 125 are used to collect perception content data in the task data at a preset frequency, and transmit the perception content data to each computing node 122 through the internal communication bus 124. The perception content data includes device operation information and / or environmental perception information of the electrical device.

[0056] Specifically, task data can include sensory content data, which is collected at a preset frequency by multiple locally deployed sensors. Here, sensory content data refers to physical quantities or status codes directly collected by sensors and devices that reflect real-time status. This data includes both device operating information and / or environmental perception information. By monitoring device status and environmental parameters in real time, sensory content data provides dynamic data support for energy management, enabling immediate response and proactive planning.

[0057] Specifically, device operation information refers to data directly collected through built-in sensors or interfaces in electrical devices, such as power, temperature, operating status code, mode status code, and reservation requests. Environmental perception information refers to information collected by sensors deployed in the home environment (such as temperature and humidity sensors, biometric sensors, and cameras), such as temperature and humidity, light intensity, and the behavior of family members.

[0058] Based on any of the above embodiments, the task data further includes symbol content data, and the symbol content data includes external factors and / or user habits that affect the use of the electrical device; The user habits are recorded locally by the computing nodes 122 and / or the core node 121 , and the external factors are downloaded from the cloud resource platform 110 by the networking node 123 at a preset frequency and then transmitted to the computing nodes 122 via the core node 121 .

[0059] Specifically, symbolic content data refers to indirect observational data related to electrical equipment usage, expressed in symbolic form. This data includes information such as external factors and / or user habits. Symbolic content data uses abstract symbols to describe macroeconomic conditions and user behavior patterns, providing logical rules and historical experience for energy management.

[0060] Here, external factors refer to public information obtained from external sources, such as weather forecasts, extreme weather warnings, power supply status, grid load status, and energy prices. This information can be obtained by networked nodes accessing cloud resource platforms at a preset frequency. After obtaining relevant data on external factors, networked nodes can convert it into symbol sequences to ensure a uniform data format. External factors are public, timely, and macroscopic, and therefore constitute public data outside the scope of user privacy.

[0061] User habits refer to information gathered from historical user behavior records, such as sleep and rest schedules, activity ranges, and device associations. These can be recorded as symbol sequences through local logs or user configurations, forming a personalized dataset. User habits are private, long-term, and personalized, making them core data within the user privacy domain.

[0062] Based on any of the above embodiments, after obtaining the symbolic content data and the perception content data of the task data, these data may be input into the energy management model to obtain the device control strategy output by the model. Figure 5 This is a schematic diagram of the structure of the energy management model provided by the present invention. Figure 5 As shown, the energy management model can be a distributed AI (Artificial Intelligence) model that uses distributed multi-task learning to handle different control tasks. Figure 5 The distributed multi-task AI shown in the figure is the energy management model provided by an embodiment of the present invention. This model can be deployed on each computing node of a local distributed system. The model input mainly includes symbolic content data and perception content data, and outputs a control strategy for household appliances. The distributed AI model includes a shared layer for feature extraction and several dedicated task heads. The shared layer is executed collaboratively by all computing nodes, while the task heads are executed individually by specific computing nodes. The number of task heads can be expanded on demand.

[0063] The input symbolic content data consists of two parts: one is announcement information from external information sources (i.e., information about external factors obtained from cloud resource platforms through networked nodes), such as weather forecasts, extreme weather warnings, power supply status, grid load status, energy prices, etc.; the other is locally recorded user habits, such as sleep schedules, activity ranges, and device associations. The input perceptual content data is information obtained through sensors and electrical devices, such as sensor data, device operating status, family member behavior status, reservation requirements, and mode settings for large electrical devices. The announcement information in the symbolic content data is obtained through external communication and is public information. The user habits and perceptual content in the symbolic content data are generated and obtained locally and are considered user privacy.

[0064] The output device control strategy is a series of tasks that change the operating state of electrical devices, such as "Set the bedroom air conditioner to 26°C for dehumidification" or "Schedule the washing machine to start at 11:00 PM." The control strategy here is an event instruction, not a control signal. The distributed AI model only generates the control strategy. The control signals for the specific electrical devices required to implement these event instructions are processed by other subsequent modules based on the specific device models and interfaces.

[0065] The shared layer extracts common features and semantic information from the input. These extracted features and semantics are represented as data within the neural network. The shared layer consists of three submodules: the symbolic content processing module, the perceptual content processing module, and the fusion processing module. The symbolic content processing module specifically processes the input symbolic content data, while the perceptual content processing module specifically processes the input perceptual content data. These two modules initially extract features and semantics of the corresponding content, which are then fed into the fusion processing module. The fusion processing module combines the preliminary features and semantics generated by these two modules, processing them into features and semantics that can be used by the task head. The symbolic content processing module uses an attention mechanism to extract semantic information from the symbolic content data, while the perceptual content processing module uses a convolutional neural network (CNN) to extract features and a long short-term memory network (LSTM) to handle temporal dependencies. The fusion processing module uses a fully connected layer to fuse the two features and semantics.

[0066] The task head is the output layer for a specific task. The number of task heads is scalable. Each task head reads and further processes the features and semantics extracted by the shared layer to generate specific outputs for a specific task. Task heads use neural network methods to process features and semantics. Depending on the task, different neural network methods can be used. Possible neural network methods include deep neural networks (DNNs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), long short-term memory networks (LSTMs), and attention mechanisms. Different tasks use different task heads, configured according to the specific task. For example, the memory head is specifically designed to record user habits, converting recognized user habits into symbol sequences. The control head is the task head used to generate control decisions for electrical equipment. It is divided into two categories: real-time tasks and planning tasks. Tasks are divided according to electrical equipment and regions, and the electrical equipment controlled by different tasks do not overlap. The real-time control head predicts the demand status of electrical equipment and generates immediate task events by comparing the actual status of the electrical equipment with the demand status. The planning control head generates a timed sequence of task events based on the input appointment demand.

[0067] Based on any of the above embodiments, the computing node 122 is further configured to collect incremental training data when executing the assigned subtask, and transmit the incremental training data to the core node 121 according to a preset frequency; The core node 121 is further configured to perform incremental training on the local energy management model based on the incremental training data to obtain an energy management model with fine-tuned parameters.

[0068] It should be noted that incremental training data refers to data, such as gradient information, that is gradually collected during model training as new data or new tasks are generated. This data is used to update and optimize the existing model. Compute nodes collect this data for incremental training each time a distributed inference task is executed. Core nodes collect this accumulated incremental training data from compute nodes at a set frequency and perform incremental training on their local energy management model, fine-tuning the local model parameters. After fine-tuning, the core node generates a locally updated energy management model.

[0069] Specifically, the core nodes and computing nodes can pre-agreed on the frequency of collecting incremental training data, such as once every 10 minutes, once every hour, etc. This frequency can be adjusted based on the system's computing resources, data volume, and the real-time requirements of model training.

[0070] Compute nodes continuously collect and store incremental training data while executing subtasks. When the preset collection time is reached, the compute nodes organize and package the collected incremental training data. For example, they may merge multiple gradients or organize them into a specific format for easier transmission.

[0071] At the collection time, the core node sends a request to each compute node to collect incremental training data, based on the set frequency. This request may include the compute node's identification information and data format requirements. After receiving the request from the core node, the compute node sends the prepared incremental training data to the core node via the internal communication bus. The core node receives and stores this data for subsequent incremental model training.

[0072] Specifically, after receiving incremental training data from various compute nodes, the core node first integrates this data. For example, it aggregates the gradient information collected by different compute nodes to generate a global gradient, which reflects the optimization direction of the model across the entire local distributed system. The core node then uses this integrated incremental training data (e.g., global gradient information) to update the parameters of the local energy management model. It should be understood that incremental training is an iterative process. After completing a parameter update, the core node continues to collect subsequent incremental training data and repeats the aforementioned data integration and parameter update steps to continuously optimize the model.

[0073] It's understood that a fine-tuned energy management model refers to a model that has undergone incremental training, where the original model's parameters have been adjusted and optimized. Through incremental training, the model automatically adjusts its internal parameters based on new data and task requirements, making it more accurate and efficient in handling energy management tasks. For example, the model can better adapt to the operating characteristics of different electrical devices, environmental changes, and personalized user needs, thereby generating more reasonable device control strategies and achieving more efficient energy management.

[0074] In this embodiment of the present invention, the environment in which the local distributed system operates has unique characteristics, such as a specific combination of electrical devices and user usage habits. Local model fine-tuning allows the energy management model to be fine-tuned to user behavior, better adapting to these local characteristics, improving the model's performance and accuracy in the local environment, and achieving a high degree of personalization.

[0075] Based on any of the above embodiments, the core node 121 is further configured to record model training information when performing incremental training on the local energy management model, the model training information including algorithm structure information, parameter update differences, and training metadata of the energy management model, the training metadata including at least one of the number of training rounds, batch size, number of inputs, and learning rate; The networked node 123 is further configured to upload the model training information to the cloud resource platform 110 at a preset time interval, so that the cloud resource platform 110 updates and publishes the energy management model on the cloud resource platform 110 based on all received model training information.

[0076] It's important to note that in local distributed systems, core nodes record model training information each time a model is fine-tuned. Networked nodes then upload this information to the cloud resource platform at set intervals. Once the cloud resource platform collects the training information uploaded by all users' local distributed systems, it can update the energy management model based on this information.

[0077] Specifically, model training information refers to data generated during model training that reflects the state, process, and results of model training. For example, model training information may include algorithm structure information, parameter update deltas, training metadata, and so on. Here, algorithm structure information describes the algorithm architecture used by the energy management model, including the number of layers in the model, the number of neurons in each layer, the type of activation function, and the network topology. For example, for a neural network-based energy management model, the algorithm structure information would clearly indicate the structure of the input layer, hidden layer, and output layer, as well as how the layers are connected.

[0078] The parameter update delta is the difference between the new and old parameter values ​​during each parameter update during incremental model training. It reflects the changes in model parameters during training. By analyzing the parameter update delta, we can understand the model's sensitivity to different data and the direction and magnitude of parameter adjustments.

[0079] Training metadata refers to auxiliary information about the model training process, including the number of training epochs, batch size, number of inputs, and learning rate. The number of training epochs refers to the number of times the model traverses the entire training dataset during training. The batch size refers to the number of samples used to calculate the gradient during each parameter update. The number of inputs refers to the total number of samples used in model training. The learning rate is a hyperparameter that controls the step size for model parameter updates.

[0080] Specifically, the networked nodes and the cloud resource platform pre-agreed on a time interval for uploading model training information, such as once every 24 hours or once a week. This time interval can be appropriately set based on factors such as system requirements, data volume, and network bandwidth. When the preset upload time arrives, the core node will provide the recorded model training information to the networked nodes. The networked nodes will encapsulate the model training information according to the communication protocol and data format specified by the cloud resource platform. The networked nodes then establish a network connection with the cloud resource platform via the external internet. Once the connection is established, the networked nodes send the encapsulated model training information data packets to the upload interface designated by the cloud resource platform (i.e., the training information upload interface). After receiving the data, the cloud resource platform will return a response message, notifying the networked nodes whether the data upload was successful.

[0081] The cloud resource platform receives, collects, and integrates model training information from multiple local distributed systems. Because different local distributed systems may have different data characteristics and application scenarios, the cloud resource platform needs to comprehensively analyze this information to obtain a more comprehensive picture of model training.

[0082] It's understandable that cloud resource platforms can employ model fusion technology to integrate model training information from different local distributed systems. For example, the cloud model's structure and parameters can be optimized based on the algorithmic structure information and parameter update differences of each local model. Methods such as weighted averaging and model ensembles can be used to combine the strengths of different local models to generate a more robust cloud model. Furthermore, based on the integrated training metadata (such as the number of training rounds, batch size, number of inputs, and learning rate), cloud model parameters can be further optimized. For example, the model performance of different local systems under different training parameters can be analyzed, and the optimal training parameter combination can be selected for retraining or parameter fine-tuning of the cloud model.

[0083] After the model update is complete, the cloud resource platform can evaluate and validate the updated cloud model using a dedicated test dataset. Evaluation metrics may include accuracy, recall, and mean squared error to ensure that the updated model has improved performance and meets the needs of real-world applications. If the updated cloud model passes evaluation and validation, the cloud resource platform will publish it for download and deployment across local distributed systems.

[0084] In an embodiment of the present invention, different local distributed systems may face different data and task scenarios. By uploading the model training information of each local distributed system to the cloud for integration and updating, knowledge sharing and collaborative optimization can be achieved. The cloud model can absorb the advantages and experience of each local system to improve the performance and generalization ability of the overall model. In addition, unified management and maintenance of the model on the cloud resource platform can reduce the maintenance cost and difficulty of the system. The cloud can provide a unified model version control, update and deployment mechanism to ensure that each system using the model can obtain the latest and optimal model version in a timely manner.

[0085] Based on any of the above embodiments, Figure 6 This is the fifth structural diagram of the energy management system provided by the present invention, such as Figure 6 As shown, the local distributed system 120 includes multiple systems, each of which is deployed in a user's home space. The cloud resource platform 110 includes a federated learning platform 111 and an information resource platform 112. The federated learning platform 111 is configured to update the energy management model using a federated learning method based on the model training information uploaded by each local distributed system 120, and publish the updated energy management model when it meets preset publishing conditions for download and deployment by each local distributed system 120; The information resource platform 112 is used to publish software update information and public information, and the public information includes external factors that affect the use of the electrical device.

[0086] It should be noted that the energy management system provided in this embodiment of the present invention is divided into two parts: a local distributed system and a cloud resource platform, which exchange data via the internet. This embodiment of the present invention utilizes a federated learning approach to build the system. The federated learning platform in the cloud resource platform serves as the central server for federated learning, and the local distributed system serves as the execution terminal. All users' local distributed systems execute the same AI model (i.e., the energy management model). The specific home energy management AI algorithm is executed on the local distributed system, while the cloud resource platform provides privacy-secure incremental training and the necessary public data resources.

[0087] Specifically, the cloud resource platform is deployed on cloud servers and provides online services for all users' local distributed systems. It primarily consists of a federated learning platform and an information resource platform. Network interactions between the cloud resource platform and local distributed systems are initiated by the local distributed systems, with the cloud resource platform providing only responses. Online services primarily include federated learning of AI models and the publication of other information and data.

[0088] The federated learning platform receives non-private model training information, such as model parameter update differences and training metadata, from all users' local distributed systems. Using federated learning methods, it aggregates these parameter update differences from different users to update the energy management model. The federated learning platform periodically releases new model parameters for users to download and use. It should be understood that the federated learning platform provides a framework for collaborative model training across multiple local distributed systems. In energy management scenarios, each local distributed system (deployed in a user's home space) can independently train a model and upload key information (such as model training data) from the training process to the federated learning platform. The federated learning platform is responsible for integrating this information and using federated learning methods to update and optimize the global energy management model, ensuring improved model performance and generalization without sharing local raw data.

[0089] It is understandable that the federated learning platform is deployed in a cloud computing cluster and provides three network service interfaces: the check update interface, the model download interface, and the training information upload interface. The check update interface returns information about whether an update exists and the type of update. The model download interface provides the AI ​​model, including the algorithm structure and parameters. When only the parameter information is updated, the interface only returns the parameters and content tags; when the algorithm structure is updated, the interface returns the complete AI model and content tags, and also includes guidance and prompt information for distributed inference deployment. The training information upload interface receives training data sent by the user, including the algorithm structure information of the AI ​​model used by the user, parameter update differences, local fine-tuning metadata, etc. Local fine-tuning metadata includes the number of training rounds, batch size, number of inputs, learning rate, etc.

[0090] The information resource platform is used to provide other public data, such as software updates and public information (i.e., external factors affecting the use of electrical equipment). This data can serve as input for energy management models. Specifically, the information resource platform includes an information publishing platform, which provides public information for user-deployed energy management models. This information includes weather forecasts and warnings, grid load trend statistics, and electricity price curves. This information is expressed as a series of data using agreed-upon symbols.

[0091] The local distributed system is deployed in the user's home and is used to execute AI algorithms for digital home energy management in real time. The local distributed system draws input from sensors, private records, and public data, performing distributed inference locally and in real time. The local distributed system logically consists of three types of nodes: core nodes, compute nodes, and networking nodes. The local distributed system also includes an internal communication bus, to which the core node, all compute nodes, and all sensors are connected. In the local distributed system, internal and external communications are isolated. Computing nodes only establish internal data communication with core nodes and other compute nodes via the bus. Networking nodes only establish point-to-point internal data communication with core nodes. Only networking nodes establish external internet connections. Computing nodes do not communicate with networking nodes.

[0092] The functions of core nodes include control and computing, playing a role in both AI model and data transmission. In terms of AI models, core nodes control the distributed task deployment of the algorithm, calculate the updated differences of AI model parameters, perform local fine-tuning of the AI ​​model, and record metadata for local fine-tuning. In terms of data transmission, core nodes control the transmission of bus data and network data. Here, distributed task deployment refers to the allocation of AI algorithm computing tasks to each node and the planning of data interaction between each node. Bus data transmission refers to the flow of data between each node on the bus, including the deployment of AI models, the execution of distributed AI algorithms, the backtransmission of gradients, and data interaction during updates. Network data transmission refers to the flow of data between local distributed systems and cloud resource platforms via the Internet, including downloading new AI models from the federated learning platform, uploading model training information to the federated learning platform, and downloading data from the information resource platform.

[0093] Compute nodes are used for distributed, real-time computation of energy management AI algorithms. The specific computational tasks performed by each compute node are deployed and dispatched by the core node. These nodes acquire the required data from sensors and other compute nodes via a bus, generating control signals for home appliances and systems used for home energy management. The core node fully controls the operation of the compute nodes, receiving computational tasks from them, executing them periodically, and transmitting the resulting fine-tuning data back to the core node.

[0094] Networking nodes are the only way for local distributed systems to connect to the public internet and are responsible for completing all network interaction tasks. These tasks include data transmission, identity verification, and data encryption and decryption. Networking nodes are fully controlled by core nodes, and all internet connections for local distributed systems are initiated by networking nodes.

[0095] It is understood that the internal communication bus in a local distributed system can be implemented using a network, with each distributed node and sensor connected to the internal communication bus via a wireless network or Ethernet. The internal communication bus is based on the local area network (LAN) established by the distributed system. The LAN and wide area network (WAN) use different link layer and physical layer devices to achieve physical isolation of the internal network.

[0096] In a local distributed system, only networked nodes establish a wide area network connection, exchanging data with online resource platforms via the internet. Networked nodes exchange data only with core nodes. Networked nodes and core nodes are hosted on separate physical devices, communicating via dedicated connections.

[0097] In a local distributed system, sensors, compute nodes, and core nodes can be hosted by the same or different physical devices. Sensors and compute nodes can be hosted by the same physical device. Core nodes and compute nodes can be hosted by the same physical device. One physical device can host multiple compute nodes.

[0098] Furthermore, in a locally distributed system, the LAN (local area network) of the internal communication bus can use the same link layer and physical layer equipment as the WAN. Network segmentation technologies such as virtual LANs and link layer access control can be used to isolate data between the LAN and WAN. Networking nodes and core nodes can also reside on the same physical device, using separate network ports to connect to the LAN and WAN. In this case, networking nodes and core nodes communicate only within the device.

[0099] Figure 7 This is a schematic diagram of the overall process of the energy management system provided by the present invention. Figure 7 As shown, inference is deployed and executed through the core nodes and computing nodes of the local distributed system. For training, the local AI model is updated through two channels: cloud-based federated learning and local fine-tuning.

[0100] First, let's explain the distributed inference process. After obtaining an updated AI model, the core node deploys distributed inference tasks to each distributed node based on their computing performance. After deployment, the compute nodes execute distributed inference at a set frequency. During distributed inference, networked nodes download the required resource data from the cloud resource platform at a set frequency, and sensors capture data at a set frequency as input for the AI ​​model's distributed inference.

[0101] Next, we'll explain the local fine-tuning process. Each time an AI model executes distributed inference, the compute nodes collect data for incremental training, such as gradient information. The core nodes collect accumulated training data from the compute nodes at a set frequency and perform incremental training on the local AI model, fine-tuning the local model parameters. After fine-tuning, the core nodes generate a locally updated AI model.

[0102] Finally, the online federated learning process is described. First, an initial AI model is established for use, and this pre-trained distributed AI model is released through the federated learning platform. The local distributed system downloads the latest AI model from the federated learning platform at a set frequency and then completes deployment. Within the local distributed system, each time fine-tuning is performed, the core node records parameter differences and training metadata such as the number of training rounds and data volume. At set intervals, the networked nodes upload the training information, including parameter differences and training metadata, recorded by the core nodes, to the federated learning platform. The federated learning platform collects the training information uploaded by all users' local distributed systems, aggregates the data using federated learning methods, and updates the AI ​​model. The federated learning platform evaluates the effectiveness of the updated new model and selectively releases the new model.

[0103] The system provided by the embodiments of the present invention can serve as the software and hardware carrier of a home energy management system, and can be fine-tuned according to user behavior habits, with a high degree of personalization; artificial intelligence algorithms and models can be updated instantly through the Internet, with strong flexibility; the artificial intelligence model can be trained by integrating usage data of all users, with high incremental training efficiency; tasks are executed locally on the user, with high operating efficiency; and there is no need to exchange private data on the public network, effectively protecting user privacy and security.

[0104] The energy management method provided by the present invention is introduced below. The energy management method described below and the energy management system described above can be referenced to each other.

[0105] Based on any of the above embodiments, Figure 8 It is a flow chart of the energy management method provided by the present invention, such as Figure 8 As shown, the method is applied to a local distributed system, wherein the local distributed system includes a core node and multiple computing nodes, and the method includes: Step 810: Split the reasoning operation task into multiple subtasks based on the core node, and assign the multiple subtasks to each computing node; Step 820: Based on each computing node, task data required for the assigned subtask is obtained, and the task data is input into the currently deployed energy management model to obtain a device control strategy output by the energy management model, wherein the device control strategy is used to adjust the operating status of the electrical device; The energy management model is downloaded from the cloud resource platform by the core node and then deployed to each computing node.

[0106] Specifically, before executing an inference task, the core node can download the energy management model from the cloud resource platform and deploy it to each compute node. Once the energy model is locally deployed, the model can be used to execute the inference task. The core node first analyzes the inference task, understanding factors such as the data input size, the model computation complexity, and the compute node performance characteristics. Based on the task analysis results, the core node can partition the input data. For example, household appliances can be grouped by area (e.g., living room, bedroom) or type (e.g., air conditioner, refrigerator), with each group corresponding to a portion of the input data. In this way, the data for each group can serve as input for a subtask. Furthermore, if the energy management model computation process can be broken down into multiple relatively independent steps, the core node can also decompose the model computation into multiple substeps, each corresponding to a subtask. For example, if the model first performs data preprocessing, then feature extraction, and finally policy generation, these three steps can be divided into three subtasks.

[0107] After splitting the inference computation task, the core node can further evaluate the performance of each compute node. This evaluation allows the core node to understand the processing power and availability of each compute node. Based on the node evaluation results, the core node can formulate a task allocation strategy and assign subtasks to the compute nodes according to the formulated task allocation strategy. After receiving the subtask, the compute node begins executing the corresponding computation task.

[0108] Specifically, after receiving a subtask, each compute node first obtains the task data required for that subtask. Once the compute node has obtained the task data required for the subtask being processed, it then inputs this data into the currently deployed energy management model, which then generates the device control strategy output by the model. A device control strategy here refers to a series of tasks that change the operating state of electrical devices, such as "Set the bedroom air conditioner to 26°C for dehumidification" or "Schedule the washing machine to start at 3:00 AM." The control strategy here is an event instruction, not a control signal. The energy management model only generates event instructions, while the control signals for the specific electrical devices required to implement these event instructions are processed by subsequent modules based on the specific device models and interfaces.

[0109] The method provided by the embodiments of the present invention uses core nodes to split complex inference calculation tasks into multiple subtasks and assign them to individual computing nodes for collaborative processing. This significantly improves the computational efficiency of energy management models and meets the high-real-time requirements of home energy management. The entire calculation process is completed within the local distributed system, reducing data transmission and cloud latency. This allows for rapid processing of inference calculation tasks and timely generation of device control strategies, thereby efficiently adjusting the operating status of electrical devices. Furthermore, data processing within the local distributed system reduces the need to upload household electricity usage data to the cloud, reduces the risk of privacy leaks, and ensures the security of users' private information, such as their home energy usage. Furthermore, core nodes can retrieve the latest energy management models from the cloud resource platform and dynamically deploy them to computing nodes, helping to ensure algorithm optimization and updates and providing greater flexibility. Compared to traditional solutions that rely solely on the cloud for management, the distributed architecture of the present invention can maintain energy management functions through local nodes even during network outages, offering greater stability, reliability, and fault tolerance.

[0110] It should be noted that other embodiments or specific implementations of the energy management method of the present invention applied to a local distributed system can refer to the above-mentioned system embodiments and will not be described in detail here.

[0111] Figure 9 An example of a physical structure diagram of an electronic device is shown below. Figure 9 As shown, the electronic device may include: a processor 910, a communication interface 920, a memory 930 and a communication bus 940, wherein the processor 910, the communication interface 920, and the memory 930 communicate with each other via the communication bus 940. The processor 910 may call the logic instructions in the memory 930 to execute the energy management method, which is applied to a local distributed system, wherein the local distributed system includes a core node and multiple computing nodes. The method includes: based on the core node, splitting the reasoning operation task into multiple subtasks, and assigning the multiple subtasks to each computing node; based on each computing node, obtaining the task data required for the assigned subtask, and inputting the task data into the currently deployed energy management model to obtain the device control strategy output by the energy management model, wherein the device control strategy is used to adjust the operating status of the electrical device; wherein the energy management model is downloaded by the core node from the cloud resource platform and deployed to each computing node.

[0112] Furthermore, the logic instructions in the aforementioned memory 930 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the relevant art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0113] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the energy management method provided by the above-mentioned methods. The method is applied to a local distributed system, and the local distributed system includes a core node and multiple computing nodes. The method includes: based on the core node, splitting the reasoning operation task into multiple subtasks, and assigning the multiple subtasks to each computing node; based on each computing node, obtaining the task data required for the assigned subtask, and inputting the task data into the currently deployed energy management model to obtain the device control strategy output by the energy management model, and the device control strategy is used to adjust the operating status of the electrical equipment; wherein, the energy management model is downloaded by the core node from the cloud resource platform and deployed to each computing node.

[0114] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the energy management method provided by the above-mentioned methods. The method is applied to a local distributed system, and the local distributed system includes a core node and multiple computing nodes. The method includes: based on the core node, splitting the reasoning operation task into multiple subtasks, and assigning the multiple subtasks to each computing node; based on the each computing node, obtaining the task data required for the assigned subtask, and inputting the task data into the currently deployed energy management model to obtain the device control strategy output by the energy management model, and the device control strategy is used to adjust the operating status of the electrical equipment; wherein, the energy management model is downloaded by the core node from the cloud resource platform and deployed to the each computing node.

[0115] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0116] Through the description of the above embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the relevant technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An energy management system, characterized in that: It includes a cloud resource platform and a local distributed system connected via a network, wherein the local distributed system includes a core node and a plurality of computing nodes, and the plurality of computing nodes are all in communication connection with the core node; The core node is used to split the reasoning operation task into multiple subtasks and distribute the multiple subtasks to each computing node; The computing node is used to obtain task data required for the assigned subtask and input the task data into the currently deployed energy management model to obtain the device control strategy output by the energy management model, and the device control strategy is used to adjust the operating status of the electrical equipment; The energy management model is downloaded from the cloud resource platform by the core node and then deployed to each computing node.

2. The energy management system according to claim 1, characterized in that: The local distributed system further comprises a network node, wherein the network node is respectively communicatively connected to the core node and the cloud resource platform; The networked node is configured to access the check update interface of the cloud resource platform to detect whether the energy management model on the cloud resource platform has been updated; The networked node is further configured to, upon detecting that the energy management model has been updated, download the updated energy management model based on the model download interface of the cloud resource platform and send the updated energy management model to the core node; The core node is used to deploy the updated energy management model to each computing node.

3. The energy management system according to claim 2, characterized in that: The networking nodes establish point-to-point internal data communication with the core nodes, the networking nodes are communicated with the cloud resource platform via the external Internet, the computing nodes and the core nodes are connected via an internal communication bus, and no communication is established between the computing nodes and the networking nodes.

4. The energy management system according to claim 1, characterized in that: The local distributed system further comprises a plurality of sensors, wherein the plurality of sensors are all connected to an internal communication bus; The multiple sensors are used to collect perception content data in the task data at a preset frequency, and transmit the perception content data to each computing node through the internal communication bus. The perception content data includes device operation information and / or environmental perception information of the electrical device.

5. The energy management system according to claim 4, characterized in that: The task data also includes symbol content data, and the symbol content data includes external factors and / or user habits that affect the use of the electrical device; The user habits are recorded locally by the computing nodes and / or the core node, and the external factors are downloaded from the cloud resource platform by the networked nodes at a preset frequency and then transmitted to the computing nodes via the core node.

6. The energy management system according to any one of claims 1 to 5, characterized in that: The computing node is further configured to collect incremental training data when executing the assigned subtask, and transmit the incremental training data to the core node at a preset frequency; The core node is further configured to perform incremental training on a local energy management model based on the incremental training data to obtain an energy management model with fine-tuned parameters.

7. The energy management system according to any one of claims 1 to 5, characterized in that: The core node is further configured to record model training information when incrementally training the local energy management model, the model training information including algorithm structure information, parameter update differences, and training metadata of the energy management model, the training metadata including at least one of the number of training rounds, batch size, number of inputs, and learning rate; The networked node is further configured to upload the model training information to the cloud resource platform at preset time intervals, so that the cloud resource platform can update and publish the energy management model on the cloud resource platform based on all received model training information.

8. The energy management system according to any one of claims 1 to 5, characterized in that: The local distributed system includes multiple systems, each of which is deployed in a user's home space, and the cloud resource platform includes a federated learning platform and an information resource platform; The federated learning platform is used to update the energy management model using a federated learning method based on the model training information uploaded by each local distributed system, and publish the updated energy management model when it meets preset publishing conditions for download and deployment by each local distributed system; The information resource platform is used to publish software update information and public information, and the public information includes external factors that affect the use of the electrical equipment.

9. An energy management method, characterized in that: The method is applied to a local distributed system, the local distributed system including a core node and multiple computing nodes, and the method includes: Based on the core node, the reasoning operation task is split into multiple subtasks, and the multiple subtasks are assigned to each computing node; Based on each computing node, task data required for the assigned subtask is obtained, and the task data is input into a currently deployed energy management model to obtain a device control strategy output by the energy management model, wherein the device control strategy is used to adjust the operating status of the electrical device; The energy management model is downloaded from the cloud resource platform by the core node and then deployed to each computing node.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the energy management method according to claim 9 is implemented.

Images