Household load intelligent scheduling device

Abstract

The invention discloses a household load intelligent scheduling device, which comprises a data acquisition module and a linkage control module, and is characterized in that the data acquisition module is used for acquiring a load data sets corresponding to a electric devices in a preset time period; the linkage control module is used for acquiring a target time sequence prediction model; determining a first predicted total load data set in the first future time period according to the a load data sets and a target time sequence prediction model; acquiring a system control strategy corresponding to the home load system; determining a first scheduling scheme corresponding to the first future time period according to a system control strategy and the first predicted total load data set; determining a first scheduling instruction corresponding to the first scheduling scheme; and sending the first scheduling instruction to a local control module. By adopting the embodiment of the invention, the requirements of real-time scheduling and control of families can be met.

Description

Technical Field

[0001] This application relates to the field of power control technology, and in particular to a smart household load dispatching device. Background Technology

[0002] With the development of urbanization, the refined management of urban residents' electricity load and the improvement of energy efficiency have become important aspects of smart grid construction.

[0003] Currently, most residential load dispatching systems use cloud-based prediction models for electricity consumption decisions. They rely on big data platforms to analyze historical electricity data and generate load forecast curves to achieve optimized power allocation. Although this type of architecture has powerful computing capabilities, the data transmission latency between the cloud and terminal devices often exceeds tens or even hundreds of milliseconds, making it difficult to meet the needs of real-time household dispatching and control.

[0004] Therefore, how to meet the needs of real-time scheduling and control in homes has become an urgent problem to be solved. Summary of the Invention

[0005] This application provides a smart household load scheduling device that can meet the needs of real-time household scheduling and control.

[0006] In a first aspect, embodiments of this application provide a smart household load scheduling device, the device comprising: a data acquisition module and a linkage control module, both the linkage control module and the data acquisition module being communicatively connected to a household load system, the household load system comprising a local control module and a electrical devices, where a is a positive integer; wherein: The data acquisition module is used to collect the load data set of each of the a electrical devices within a preset time period to obtain a load data set; The linkage control module is used to acquire a target time-series prediction model; determine a first predicted total load dataset within a first future time period based on the a load datasets and the target time-series prediction model; the start time of the first future time period is later than the end time of the preset time period; acquire the system control strategy corresponding to the household load system; determine a first scheduling scheme corresponding to the first future time period based on the system control strategy and the first predicted total load dataset; determine a first scheduling instruction corresponding to the first scheduling scheme; and send the first scheduling instruction to the local control module; the first scheduling instruction is used to instruct the local control module to schedule the load of the a electrical devices according to the first scheduling scheme.

[0007] Optionally, the first predicted total load dataset includes: multiple predicted total load values ​​and the prediction time corresponding to each total load value; the system control strategy includes a threshold control strategy, and in determining the first scheduling scheme based on the system control strategy and the first predicted total load dataset, the linkage control module is specifically used for: Determine the first load threshold corresponding to the household load system; The predicted load curve segment is determined based on the multiple predicted total load values ​​and the predicted time corresponding to each total load value; the horizontal axis of the predicted load curve is time, and the vertical axis is the load value. Determine the partial load curves in the predicted load curve segment that are greater than the first load threshold, resulting in b partial load curves; where b is a natural number. The first scheduling scheme is determined based on the b partial load curves, the a electrical devices, and the a load datasets.

[0008] Optionally, in determining the first scheduling scheme based on the b partial load curves, the a electrical devices, and the first predicted total load dataset, the linkage control module is specifically used for: Determine the time period corresponding to each of the b partial load curves to obtain b time periods; Based on the b partial load curves and the first load threshold, b load values ​​to be adjusted are determined; Determine the adjustable electrical devices among the a electrical devices to obtain c adjustable electrical devices; c is a positive integer less than or equal to a; Determine c priorities corresponding to the c adjustable electrical devices; each adjustable electrical device corresponds to one priority. Based on the c priorities, the b load values ​​to be adjusted, and the c adjustable electrical devices, determine b scheduling schemes corresponding to the b time periods; The first scheduling scheme is determined based on the b scheduling schemes.

[0009] Optionally, in determining the b scheduling schemes corresponding to the b time periods based on the c priorities, the b load values ​​to be adjusted, and the c adjustable electrical devices, the linkage control module is specifically used for: Obtain a first time period and its corresponding first load value to be adjusted; the first time period is any one of the b time periods. Based on the a load datasets, determine c first average load values ​​corresponding to the c adjustable electrical devices; each adjustable electrical device corresponds to one first average load value; Determine c minimum load values ​​corresponding to the c adjustable electrical devices; and a minimum load value corresponding to each adjustable electrical device; Based on the c first average load values ​​and the c minimum load values, determine c adjustable load values; Determine the total adjustable load value corresponding to the c adjustable load values; When the total adjustable load value is greater than or equal to the first adjustable load value, c adjustable load values ​​corresponding to the c adjustable electrical devices are determined according to the c priorities and the c adjustable load values, with each adjustable load value corresponding to one adjustable electrical device; and a scheduling scheme corresponding to the first time period is determined according to the c adjustable load values. When the total adjustable load value is less than the first load value to be adjusted, d first average load values ​​are determined based on the c priorities and the c first average load values; the sum of the d first average load values ​​is greater than the first load value to be adjusted; d is an integer less than or equal to c; the scheduling scheme corresponding to the first time period is determined based on the d first average load values ​​and the c adjustable electrical equipment.

[0010] Optionally, the system control strategy includes a sliding window strategy. In determining the first scheduling scheme based on the system control strategy and the first predicted total load dataset, the linkage control module is specifically used for: Get the preset window length; A first time window is determined based on the preset window length and the first future time period; the start time of the first time window is equal to the start time of the first future time period. The first average predicted load value corresponding to the first time window is determined based on the first predicted total load dataset. When the first average predicted load value is greater than the second load threshold, the adjustable electrical equipment among the a electrical devices is identified, resulting in c adjustable electrical devices; the first working period of the c adjustable electrical devices within the first time window is identified, resulting in c first working periods; the c first working periods are adjusted according to the second load threshold and the first preset rule to obtain c second working periods; and the first scheduling scheme is determined based on the c second working periods and the c adjustable electrical devices.

[0011] Optionally, when the system control strategy includes a rolling scheduling strategy, the linkage control module is further specifically used for: Obtain the target electricity consumption change of the household load system at the current moment; The a load datasets are updated based on the target electricity consumption changes to obtain a updated load datasets; Based on the a updated load datasets and the target time-series prediction model, a second predicted total load dataset is determined within a second future time period; the start time of the second future time period is later than the end time of the preset time period, and the duration of the second future time period is less than the duration of the first future time period. Based on the second predicted total load dataset and the second preset rule, determine the equipment operation schedule corresponding to the a electrical devices; A second scheduling scheme is determined based on the equipment operation schedule; Determine the second scheduling instruction corresponding to the second scheduling scheme; The second scheduling instruction is sent to the local control module.

[0012] Optionally, the device further includes: a cloud server, the linkage control module being connected to the cloud server, the cloud server having a preset teacher model and a preset student model deployed thereon, and a preset load database stored thereon; Regarding the acquisition of the target time-series prediction model, the linkage control module is specifically used for: A model request message is sent to the cloud server; after receiving the model request message, the cloud server uses the data in the preset load database to train the preset teacher model and the preset student model to obtain the target time series prediction model. Receive first response data from the cloud server in response to the model request information; the first response data includes the installation package of the target time series prediction model; The target time-series prediction model is installed into the linkage control module based on the first response data; Specifically, in the step of training the preset teacher model and the preset student model using data from the preset load database to obtain the target time-series prediction model, the cloud server is used for: Extract the training dataset from the preset load database; The preset teacher model is trained using the training dataset until the model converges, thus obtaining the target teacher model. The preset student model is distilled and trained using the training dataset and the target teacher model to obtain the target student model; The target time series prediction model is determined based on the target student model.

[0013] Optionally, in the step of using the training dataset and the target teacher model to distill and train the preset student model to obtain the target student model, the cloud server is specifically used for: Distillation training data is extracted from the preset load database to obtain the distillation training dataset; The distillation training dataset is input into the target teacher model and the preset student model respectively to obtain the teacher output and the student output. Determine the actual output results corresponding to the distillation training dataset; Based on the actual output results and the student output results, determine the hard label loss; Based on the teacher's output and the student's output, determine the soft label loss; The target loss is determined based on the hard label loss and the soft label loss; The parameters of the preset student model are optimized based on the target loss and the preset optimization algorithm to obtain the target student model.

[0014] Optionally, in determining the target time-series prediction model based on the target student model, the linkage control module is specifically used for: Determine the required performance parameters corresponding to the target student model; Obtain the target performance parameters corresponding to the linkage control module; When the target performance parameter is greater than or equal to the required performance parameter, the target student model is determined as the target time series prediction model; When the target performance parameter is less than the required performance parameter, the target student model is quantized to obtain the target time series prediction model.

[0015] Optionally, after the local control module schedules the load of the a electrical devices according to the first scheduling scheme, the linkage control module is further specifically used for: Obtain the actual load dataset for the first future time period; The target error is determined based on the actual load dataset and the first predicted total load dataset; When the target error is greater than a preset error, a model update parameter package is obtained; the parameters of the target time series prediction model are updated according to the model update parameter package.

[0016] Implementing this application will have the following beneficial effects: As can be seen, the intelligent household load scheduling device described in this application, on the one hand, communicates directly with the household load system through the data acquisition module, without any transmission links across the cloud or wide area network. The data acquisition delay depends only on the local communication protocol, which greatly reduces the transmission delay. On the other hand, by deploying the target time series prediction model in the linkage control module, the model makes predictions locally based on the collected data, and then directly completes decision analysis, scheduling scheme generation, and command issuance locally without relying on cloud relay, forming a local closed loop of "prediction-decision-execution", further compressing the delay of the entire process and ensuring real-time scheduling response, thereby meeting the needs of real-time household scheduling and control. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the background art, the accompanying drawings used in the embodiments of this application or the background art will be described below.

[0018] Figure 1 This is a schematic diagram of the structure of a household load system provided in an embodiment of this application; Figure 2 This is an application scenario diagram of a home load intelligent scheduling device provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a smart home load scheduling device provided in an embodiment of this application; Figure 4 This is a schematic diagram of another intelligent household load scheduling device provided in an embodiment of this application; Figure 5 This is a flowchart of a method for determining a target time series prediction model provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of another intelligent household load scheduling device provided in the embodiments of this application; Figure 7 This is a flowchart of a household load intelligent scheduling method provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0019] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0020] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0021] It should be understood that the term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this document indicates that the preceding and following related objects are in an "or" relationship. In the embodiments of this application, "multiple" refers to two or more.

[0022] In the embodiments of this application, "at least one item" or its similar expression refers to any combination of these items, including any combination of a single item or a plurality of items. "One or more" means one or more, while "multiple" means two or more. For example, "at least one item" of a, b, or c can represent the following seven cases: a, b, c; a and b; a and c; b and c; a, b, and c. Each of a, b, and c can be an element or a set containing one or more elements.

[0023] In this application embodiment, "connection" refers to various connection methods such as direct connection or indirect connection to realize communication between devices. This application embodiment does not limit this in any way.

[0024] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0025] The electronic device described in this application embodiment can be a home load intelligent scheduling device.

[0026] The following describes the relevant content, concepts, meanings, technical issues, technical solutions, and beneficial effects involved in the embodiments of this application.

[0027] First, let me explain some of the technical terms used in this application: The load of electrical equipment refers to the instantaneous power consumed by electrical equipment during operation (in watts or kilowatts), which is a core indicator reflecting the power consumption intensity of the equipment.

[0028] Distillation training is a machine learning technique that aims to improve the efficiency and performance of a model by transferring knowledge from a large pre-trained model (teacher model) to a smaller model (student model).

[0029] Model quantization: A model compression and deployment optimization technique that reduces the model's storage size and computational complexity by converting the model's parameters (e.g., weights, biases) from a high-precision data format (e.g., 32-bit floating-point numbers float32) to a low-precision data format (e.g., 8-bit integers int8) within an acceptable range of precision loss.

[0030] Zigbee is a low-power, short-range, low-latency wireless communication protocol suitable for local data transmission between IoT terminal devices. It supports simultaneous communication of multiple devices, ensuring both real-time scheduling response speed and adapting to the low-power requirements of home scenarios.

[0031] Power line carrier (PLC) protocol, also known as power line communication, is a communication technology protocol that uses existing power lines as the communication medium to transmit both electrical energy and data signals. Its core principle is to use modulation and demodulation technology to superimpose high-frequency data signals onto the power frequency lines, achieving "data transmission over power lines" and enabling communication between devices without additional wiring.

[0032] DiffLoad network: This is a complex time-series forecasting network designed for electricity load forecasting, with ≥120,000 parameters. By learning from large-scale historical load data (including individual device load sequences, timestamps, holidays, and other features), this network possesses high-precision forecasting capabilities to capture complex trends such as load peaks and power surges.

[0033] Single-layer gated recurrent unit (GRU) network: This is a simplified structure of the gated recurrent unit (GRU) (containing only one hidden layer) and belongs to the category of time series prediction neural networks. Its core feature is that it controls the flow of information in time series data through update gates and reset gates. It has a simple structure, low computational cost, and short inference latency.

[0034] Please see Figure 1 , Figure 1 This is a schematic diagram of a household load system provided in an embodiment of this application; it can be seen that the household load system (hereinafter referred to as the system) may include a local control module and a electrical appliances, etc., which are not limited here; wherein: The local control module, as the core control hub of the home load system, can communicate with a electrical devices (e.g., via Zigbee, Wi-Fi, or power line carrier protocols), and send control commands to the a electrical devices to drive them to adjust their operating status, such as cutting off the power supply to a device or adjusting its power output.

[0035] Each electrical device (a_t) acts as the direct load generator and scheduling target, serving as the "execution terminal" for load optimization. When not scheduled, it operates normally according to user demand, generating the corresponding load. Simultaneously, it feeds back its load data (e.g., real-time power, operating status) to the local control module, providing the initial basis for load forecasting and scheduling decisions. During scheduling, it adjusts its operating status according to instructions from the local control module to achieve peak shaving and valley filling or load smoothing. For example, for delay-start devices (e.g., washing machines, dishwashers) among the a_t_t devices, a "delayed start" instruction is executed to avoid peak load periods.

[0036] Please see Figure 2 , Figure 2 This is an application scenario diagram of a household load intelligent dispatching device provided in an embodiment of this application; it can be seen that the household load intelligent dispatching device (hereinafter referred to as the device) is communicatively connected to the linkage control module in the household load system, and the linkage control module is communicatively connected to a electrical devices, wherein the a electrical devices include: the first electrical device, the second electrical device, ..., the a-th electrical device; wherein: The household load intelligent dispatching device is the core decision-making unit of the entire system. It integrates functions such as data acquisition, model prediction, strategy generation, and command issuance. It is responsible for analyzing electricity consumption data, predicting load trends, and generating dispatching commands.

[0037] Home load system: is a collection and control system of electrical equipment in a home, including a local control module and a electrical equipment.

[0038] Local control module: It is the intermediate control layer of the home load system, and plays the role of "command relay and execution". It can receive scheduling commands from the device and then convert the commands into specific control operations for each electrical device (such as delayed start, power limit, etc.).

[0039] A single electrical device is the generator and scheduling object of the load, including all electrical devices in the home. It can adjust its operating status according to the instructions of the local control module to realize intelligent load scheduling.

[0040] Please see Figure 3 , Figure 3This is a schematic diagram of a household load intelligent scheduling device provided in an embodiment of this application; it can be seen that the device may include a data acquisition module and a linkage control module. Both the linkage control module and the data acquisition module are communicatively connected to the household load system. The household load system includes a local control module and a electrical devices, where a is a positive integer; wherein: The data acquisition module is used to collect the load data set of each of the a electrical devices within a preset time period to obtain a load data set; In this embodiment of the application, the linkage control module may include one of the following: a single-chip microcomputer, a microcontroller, a programmable logic controller, etc., which are not limited here; the preset time period can be preset in advance or defaulted, for example, 8:00-12:00 on a certain day.

[0041] It should be explained that in practical applications, the local control module can be a smart meter; the aforementioned device can be installed inside the smart meter.

[0042] In a specific embodiment, a corresponding metering unit can be deployed at each electrical device (or its circuit branch): for example, a smart socket (connected to mobile devices such as washing machines and humidifiers), a rail-mounted power monitor (installed in the circuit of fixed equipment such as air conditioners and charging piles), or an appliance-embedded metering module (a power monitoring chip integrated into high-end home appliances). The metering unit can be interconnected with the data acquisition module via a local communication protocol (e.g., Zigbee) to form a multi-source acquisition network of "main incoming line + branch devices," ensuring that each electrical device has a corresponding metering node. Then, the data acquisition module initiates continuous sampling for each of the a electrical devices based on the device type and power consumption characteristics. Sampling objects: For each device, collect its core electrical parameters during operation, such as active power (directly reflecting the load size), current, voltage, etc., without limitation.

[0043] Sampling frequency: The sampling frequency is automatically adjusted according to the power stability of the equipment (for example, the sampling frequency of intermittently operating equipment such as washing machines and dishwashers is set to once per second to capture load changes at the moment of start-up and shutdown; the sampling frequency of low-power stable equipment such as routers is set to once per minute to reduce redundant data), ensuring that the collected data can reflect the load characteristics without occupying too many storage resources.

[0044] Device identification binding: Each sampled data carries a unique "device ID" (e.g., "washing machine_001", "air conditioner_002"), ensuring that subsequent data processing can accurately associate it with the corresponding electrical equipment, laying the foundation for distinguishing the a datasets.

[0045] Furthermore, the acquired electrical signals (voltage and current are analog quantities) can be converted into digital signals through the built-in A / D conversion circuit of the data acquisition module. Analog signals (e.g., continuous waveforms of current) are converted into discrete digital values ​​(e.g., 16-bit or 32-bit integers) by an analog-to-digital converter, facilitating subsequent digital circuit processing and storage.

[0046] The converted digital signal can contain key information such as "device ID, sampling timestamp, active power value, current value, and voltage value", forming the original time-series data record for each device. For example, a data record can be "washing machine_001, 19:00:00, 0.8kW, 3.5A, 220V".

[0047] Before writing the raw time-series data records to the local cache, the data acquisition module can perform data preprocessing to eliminate noise and anomalies and standardize the format, as follows: Noise reduction: By using moving average filtering (smoothing high-frequency noise) and threshold filtering (removing instantaneous abnormal values ​​caused by poor contact, such as power sudden changes to 0 or data far exceeding the rated value), the load data of each device is ensured to truly reflect the actual operating status.

[0048] Standardization processing: Normalize parameters such as active power and current to a preset range (e.g., [0,1]) to eliminate the impact of power level differences between different devices (e.g., 1.5kW for air conditioner and 0.1kW for humidifier) ​​on subsequent model inputs, while preserving the relative trend of change.

[0049] Timing stamp calibration: Based on a unified system time (e.g., the real-time clock of the local control module), calibrate the timestamp of each device's data to ensure that the timing data of a device are aligned in the time dimension (e.g., the load of each device at the same moment can be superimposed to calculate the total load), and avoid time misalignment caused by communication delays.

[0050] The preprocessed clean data is written to the local cache of the data acquisition module and stored according to the device ID, thus obtaining a load datasets. For example, the dataset of washing machine_001 can be: [(t1,P1),(t2,P2),...,(tn,Pn)] (t is the timestamp, P is the active power).

[0051] It should be explained that in practical applications, the above-mentioned device supports flexible configuration of the acquisition mode. When the system has low requirements for prediction accuracy (for example, it only needs to determine whether the total load in the future exceeds the safety threshold, without implementing precise scheduling of individual devices), the data acquisition module can collect only the total load data of all electrical devices in the household, without collecting the independent load data of individual electrical devices. This simplifies the acquisition process and significantly reduces resource consumption.

[0052] The linkage control module is used to acquire a target time-series prediction model; determine a first predicted total load dataset within a first future time period based on the a load datasets and the target time-series prediction model; the start time of the first future time period is later than the end time of the preset time period; acquire the system control strategy corresponding to the household load system; determine a first scheduling scheme corresponding to the first future time period based on the system control strategy and the first predicted total load dataset; determine a first scheduling instruction corresponding to the first scheduling scheme; and send the first scheduling instruction to the local control module; the first scheduling instruction is used to instruct the local control module to schedule the load of the a electrical devices according to the first scheduling scheme.

[0053] In this embodiment of the application, the first future time period can be preset or defaulted in advance, for example, the next 30 minutes.

[0054] In a specific embodiment, the linkage control module can acquire the target time series prediction model; then, based on a load datasets and the target time series prediction model, it can determine the first predicted total load dataset within the first future time period. Specifically, the a load datasets can be input into the target time series prediction model, and the total load data within the first future time period can be predicted through the target time series prediction model, thereby obtaining the first predicted total load dataset.

[0055] Next, the system control strategy corresponding to the household load system can be obtained. Specifically, historical load data of the household load system can be obtained and analyzed to obtain system power consumption characteristic indicators. These indicators may include at least one of the following: load fluctuation rate, peak load frequency, maximum load value, proportion of adjustable equipment, regularity of power consumption periods, etc., without limitation. The system control strategy is determined based on the system power consumption characteristic indicators. For example, a pre-stored mapping relationship between preset power consumption characteristic indicators and control strategies can be used to determine the system control strategy corresponding to the system power consumption characteristic indicators. The system control strategy may include one of the following: threshold control strategy, sliding window strategy, rolling scheduling strategy, etc., without limitation.

[0056] Then, based on the system control strategy and the first predicted total load dataset, a first scheduling scheme for a first future time period can be generated; the first scheduling instruction corresponding to the first scheduling scheme can be determined. Specifically, a pre-stored mapping relationship between the scheduling scheme and the scheduling instruction can be used to determine the first scheduling instruction corresponding to the first scheduling scheme; then, the first scheduling instruction can be sent to the local control module, and the local control module executes the first scheduling scheme according to the first scheduling instruction.

[0057] Optionally, the first predicted total load dataset includes: multiple predicted total load values ​​and the predicted time corresponding to each total load value; the system control strategy includes a threshold control strategy, and in determining the first scheduling scheme based on the system control strategy and the first predicted total load dataset, the linkage control module is specifically used to perform the following steps: A1. Determine the first load threshold corresponding to the household load system; A2. Determine the predicted load curve segment based on the multiple predicted total load values ​​and the predicted time corresponding to each total load value; the horizontal axis of the predicted load curve is time, and the vertical axis is the load value. A3. Determine the partial load curves in the predicted load curve segment that are greater than the first load threshold, resulting in b partial load curves; b is a natural number. A4. Determine the first scheduling scheme based on the b partial load curves, the a electrical devices, and the a load datasets.

[0058] In this embodiment of the application, the first load threshold refers to the maximum power consumption limit that the circuit can safely carry.

[0059] In a specific embodiment, the maximum load capacity (hardware limit) of the household load system circuit can be obtained. Specifically, the specifications of the incoming line and circuit breaker of the household load system can be scanned, and the rated current of the circuit can be determined based on the specifications. For example, "C40" represents a rated current of 40 amps. Then, the rated power, which is the maximum load capacity of the circuit, can be obtained by multiplying the rated current by the household voltage (usually 220 volts). Then, calculations can be performed based on the maximum load capacity of the circuit, as follows: First load threshold = maximum circuit load × (1 - preset safety redundancy coefficient); The preset safety redundancy coefficient can be preset in advance or set by default. For example, the preset safety redundancy coefficient can be 20%. According to the above formula, the first load threshold can be obtained. It should be noted that the first load threshold can also be set manually by the user. Next, multiple predicted total load values ​​and their corresponding predicted times can be combined to obtain multiple coordinate points. These coordinate points are then fitted using a preset curve fitting method (e.g., polynomial fitting) to obtain predicted load curve segments. Then, partial load curves exceeding a first load threshold within the predicted load curve segments can be identified, resulting in b partial load curves. Specifically, a first straight line parallel to the x-axis can be drawn based on the first load threshold. This first straight line and the predicted load curve segments are placed in the same coordinate system. The predicted load curve segments are then cut using this first straight line, dividing the target current curve into upper and lower parts. The portion above this first straight line is the partial load curve exceeding the first load threshold, thus obtaining b partial load curves. Finally, a first scheduling scheme can be determined based on the b partial load curves, a electrical devices, and a load datasets.

[0060] In this way, by determining the first load threshold and defining the circuit's safe carrying capacity boundary, the risk of tripping and overload caused by overload is avoided from the source, ensuring the safety of household electricity and the power grid. In addition, by predicting the load curve segment and extracting the curve of the over-threshold part, the time period and load intensity that need to be dispatched in the future can be accurately located, avoiding indiscriminate dispatching throughout the time period and reducing resource waste.

[0061] Optionally, in determining the first scheduling scheme based on the b partial load curves, the a electrical devices, and the first predicted total load dataset, the linkage control module is specifically used to perform the following steps: B1. Determine the time period corresponding to each of the b partial load curves to obtain b time periods; B2. Based on the b partial load curves and the first load threshold, determine b load values ​​to be adjusted; B3. Determine the adjustable electrical devices among the a electrical devices to obtain c adjustable electrical devices; c is a positive integer less than or equal to a; B4. Determine the c priorities corresponding to the c adjustable electrical devices; each adjustable electrical device corresponds to one priority. B5. Based on the c priorities, the b load values ​​to be adjusted, and the c adjustable electrical devices, determine the b scheduling schemes corresponding to the b time periods; B6. Determine the first scheduling scheme based on the b scheduling schemes.

[0062] In this embodiment of the application, the time period corresponding to each of the b partial load curves is determined, resulting in b time periods. Specifically, for each partial load curve, its start time and end time can be obtained, and its corresponding time period is determined based on the start time and end time. For example, assuming that the start time of a certain partial load curve is ta and the end time is tb, then its corresponding time period is [ta, tb]. Then, the maximum load value of each of the b partial load curves can be extracted to obtain b maximum load values. The first load threshold is subtracted from these b maximum load values ​​to obtain b load values ​​to be adjusted.

[0063] Next, we can identify the adjustable electrical devices among the 'a' electrical devices, resulting in 'c' adjustable electrical devices. Specifically, we can collect the device attributes of the 'a' electrical devices to obtain 'a' device attributes, as follows: Automatic identification: Smart devices (e.g., appliances connected to smart sockets) actively report device attributes (e.g., "Device type: washing machine, supports delayed start: yes") through communication protocols. User configuration: Non-smart devices (e.g., old-fashioned water heaters) are configured by the user by manually entering device properties (e.g., "pauseable: yes, power: 500W"). Historical data retrospective analysis: By analyzing the load dataset of the data acquisition module, the operating patterns of the equipment are analyzed, thereby obtaining the equipment attributes. For example, if a certain piece of equipment starts at a fixed time every day and the start time can fluctuate, it can be inferred that it is a device that can be delayed.

[0064] Then, based on the a device attributes, we can determine all electrical devices among the a electrical devices that meet at least one of the following conditions: delayed start, pause or restart, and adjustable power, thus obtaining c adjustable electrical devices. For example, if the device attributes of a washing machine are "delayed start, restart", then it is determined to be an adjustable electrical device. As another example, if the device attribute of a smart door lock is "cannot be closed", then it is determined not to be an adjustable electrical device.

[0065] Furthermore, c priorities corresponding to c adjustable electrical devices can be determined. For example, a pre-stored mapping relationship between adjustable electrical devices and priorities can be used to determine the c priorities corresponding to the c adjustable electrical devices. Then, based on the c priorities, b load values ​​to be adjusted, and c adjustable electrical devices, b scheduling schemes corresponding to b time periods can be determined. Finally, a first scheduling scheme can be determined based on the b scheduling schemes. Specifically, conflict checks can be performed on these b scheduling schemes to check whether any devices are repeatedly scheduled (e.g., the same device is scheduled for delay in two overlapping time periods). If a conflict exists, then: Prioritize scheduling for high-priority time periods (e.g., periods when the load exceeds the threshold more severely); For low-priority time periods, replace them with other low-priority devices that do not conflict; In addition, if the scheduling objects or methods of multiple time periods are the same (for example, all of them need to delay the same washing machine), they are merged into a unified instruction (for example, "delay the start time of washing machine 1 in both time periods 1 and 3") to reduce instruction redundancy; after integrating and verifying the b schemes, the scheduling equipment, adjustment amount, operation type (delay or power reduction, etc.) and execution time of each time period are clarified to form the final executable first scheduling scheme.

[0066] In this way, by first identifying the time period exceeding the threshold and the quantity to be adjusted, and then focusing on the adjustable equipment, the system avoids indiscriminate scheduling of all equipment throughout the day and improves resource utilization efficiency. In addition, by matching and adjusting resources in different time periods and superimposing global conflict verification, the system ensures that scheduling instructions are consistent with the actual adjustment capabilities of the equipment and guarantees the feasibility of the solution.

[0067] Optionally, in determining the b scheduling schemes corresponding to the b time periods based on the c priorities, the b load values ​​to be adjusted, and the c adjustable electrical devices, the linkage control module is specifically used to perform the following steps: C1. Obtain the first time period and its corresponding first load value to be adjusted; the first time period is any one of the b time periods; C2. Based on the a load datasets, determine c first average load values ​​corresponding to the c adjustable electrical devices; each adjustable electrical device corresponds to one first average load value; C3. Determine the c minimum load values ​​corresponding to the c adjustable electrical devices; one minimum load value corresponding to each adjustable electrical device; C4. Determine c adjustable load values ​​based on the c first average load values ​​and the c minimum load values; C5. Determine the total adjustable load value corresponding to the c adjustable load values; C6. When the total adjustable load value is greater than or equal to the first adjustable load value, determine c adjustable load values ​​corresponding to the c adjustable electrical devices according to the c priorities and the c adjustable load values, with each adjustable load value corresponding to one adjustable electrical device; determine the scheduling scheme corresponding to the first time period according to the c adjustable load values; C7. When the total adjustable load value is less than the first load value to be adjusted, determine d first average load values ​​based on the c priorities and the c first average load values; the sum of the d first average load values ​​is greater than the first load value to be adjusted; d is an integer less than or equal to c; determine the scheduling scheme corresponding to the first time period based on the d first average load values ​​and the c adjustable electrical devices.

[0068] In this embodiment, a first time period and its corresponding first load value to be adjusted are obtained. Then, based on a load datasets, c first average load values ​​corresponding to c adjustable electrical devices are determined. Specifically, c load datasets corresponding to c adjustable electrical devices can be found from the a load datasets, and the average value of each load dataset in the c load datasets is calculated to obtain c first average load values. Then, c minimum load values ​​corresponding to c adjustable electrical devices can be determined. Specifically, the instruction manuals of c adjustable electrical devices are obtained, and the minimum power of the device is queried from the instruction manuals to obtain c minimum power, which are also c minimum load values.

[0069] It should be explained that the minimum load value refers to the minimum power consumption that the equipment can achieve through scheduling without damaging the hardware or affecting the basic functions.

[0070] Next, we can subtract the corresponding lowest load value from the c lowest load values ​​from the c first average load values ​​to obtain c adjustable load values. Then, we can add these c adjustable load values ​​to obtain the total adjustable load value. When the total adjustable load value is greater than or equal to the first load value to be adjusted, we can determine the c load values ​​to be adjusted corresponding to the c adjustable electrical equipment based on the c priorities and the c adjustable load values. Specifically, we can sort the c adjustable electrical equipment from low to high priority (i.e., from high to low scheduling priority) according to the c priorities to obtain a sorted equipment list, for example, equipment 1, equipment 2, ..., equipment c (corresponding to adjustable load values ​​K1, K2, ..., Kc). We initialize the allocated adjustment amount sum to 0, and the remaining adjustment amount rem = T, where T represents the first load value to be adjusted. We then iterate through the equipment according to the sorted equipment list. For device i, take the smaller value between its adjustable load value Ki and the remaining adjustable amount rem as the load value to be adjusted for the device; where i is greater than 0 and is an integer less than or equal to c. Update sum = sum + the load value to be adjusted for device i, and rem = rem - the load value to be adjusted for device i; If rem=0, stop traversing and set the load values ​​to be adjusted for the remaining untraversed devices to 0; finally, c load values ​​to be adjusted are obtained (non-zero values ​​only appear in the first n devices, where n is a positive integer less than or equal to c, and the final sum=T).

[0071] Then, a scheduling scheme corresponding to the first time period can be determined based on c load values ​​to be adjusted. Specifically, values ​​greater than 0 can be selected from the c load values ​​to obtain n load values ​​to be adjusted, and the n adjustable electrical devices corresponding to these n load values ​​can be determined. Based on these n load values, n operation types corresponding to the n adjustable electrical devices can be determined. Based on the n operation types (e.g., delayed start, reduced power, suspended operation, etc.), the n load values ​​to be adjusted, and the n adjustable electrical devices, a scheduling scheme corresponding to the first time period can be generated. For example, a certain scheduling scheme can be: Washing machine (W01): Operation type: delayed start; Execution details: Originally scheduled to start at 18:40, adjusted to start at 19:00 (after the first time period ends), and will not run from 18:30 to 19:00, contributing 1.2kW of regulation.

[0072] When the total adjustable load value is less than the first load value to be adjusted, d first average load values ​​are determined based on c priorities and c first average load values. Specifically, equipment can be selected from low to high priority, and their average load values ​​can be accumulated until the sum exceeds the first load value to be adjusted, thus obtaining d first average load values. Finally, d adjustable electrical devices corresponding to the d first average load values ​​among the c adjustable electrical devices can be determined. Combined with the preset operation type (partial time-period shutdown operation within the first time period), the shutdown duration is allocated according to the priority of the d adjustable electrical devices to ensure that the total adjustment amount meets the first load value to be adjusted, thus obtaining the scheduling scheme corresponding to the first time period.

[0073] In this way, by adapting to two scenarios, namely "sufficient or insufficient adjustable total load", the adjustment amount is accurately allocated or low-priority equipment is added to supplement the load, ensuring that the first load to be adjusted meets the standard and avoiding the risk of overload. In addition, the adjustment potential is calculated based on data such as the actual average load and minimum load of the equipment, avoiding ineffective scheduling that is out of the equipment's capacity and ensuring the implementation of the solution.

[0074] Optionally, the system control strategy includes a sliding window strategy. In determining the first scheduling scheme based on the system control strategy and the first predicted total load dataset, the linkage control module is specifically used to perform the following steps: D1. Get the preset window length; D2. Determine a first time window based on the preset window length and the first future time period; the start time of the first time window is equal to the start time of the first future time period. D3. Determine the first average predicted load value corresponding to the first time window based on the first predicted total load dataset; D4. When the first average predicted load value is greater than the second load threshold, determine the adjustable electrical equipment among the a electrical devices to obtain c adjustable electrical devices; determine the first working period of the c adjustable electrical devices in the first time window to obtain c first working periods; adjust the c first working periods according to the second load threshold and the first preset rule to obtain c second working periods; determine the first scheduling scheme according to the c second working periods and the c adjustable electrical devices.

[0075] In this embodiment, the preset window length, the second load threshold, and the first preset rule can all be preset in advance or defaulted.

[0076] In a specific embodiment, a preset window length is obtained; the first future time period can be processed according to the preset window length to obtain a first time window. For example, assuming the preset window length is 10 minutes and the first future time period is [tc, td], then the first time window is [tc, tc+10]. Next, load data in the first time window can be obtained from the first predicted total load dataset to obtain at least one load data. The average value of the at least one load data is calculated, which is the first average predicted load value.

[0077] If the first average predicted load value is not greater than the second load threshold, it indicates that the system is in normal working condition and no scheduling is required.

[0078] When the first average predicted load value is greater than the second load threshold, it indicates that the system has an overload risk. We can identify a adjustable electrical devices out of a number of electrical devices, resulting in c adjustable electrical devices. Then, we can determine the first working period of these c adjustable electrical devices within a first time window, resulting in c first working periods. Specifically, for each adjustable electrical device, we can obtain its corresponding original operating plan, resulting in multiple original operating plans. For example, a washing machine is scheduled to run from 8:30-9:00 and 19:00-19:30. Then, we can determine the overlapping intervals between these multiple original operating plans and the first time window, resulting in c overlapping intervals, which are also c first working periods.

[0079] Then, c first working periods can be adjusted according to the second load threshold and the first preset rule to obtain c second working periods. Specifically, c priorities corresponding to c adjustable electrical devices can be obtained, and the c first working periods can be adjusted according to the c priorities and the first preset rule to ensure that the total load within the window is not greater than the second load threshold, thereby obtaining c second working periods. The first preset rule can be: 1. Priority is adjusted from low to high; 2. Prioritize adjustments through "staggered peak hours (staggering overlapping time periods)" and "shortening runtime". If these measures cannot be met, "delay to outside the window".

[0080] For example, assuming the first time window is 10:00-10:10 (a total of 10 minutes, the total load within this window needs to be controlled to be ≤ the second load threshold of 2.5kW); c=3, the specific work plans for c adjustable electrical devices are as follows:

[0081] It can be seen that only the total load during the period from 10:02 to 10:05 is 3.0kW > 2.5kW, requiring adjustment. Adjustment is performed according to the first preset rule. First, the device with the lowest priority (washing machine, priority 1) is adjusted, attempting "off-peak" adjustment: the washing machine's first working period (10:02-10:06) is shifted to after the air conditioner ends (after 10:05), avoiding the overlapping period of 10:02-10:05. The adjusted temporary working period for the washing machine is 10:05-10:09 (still within the window, duration unchanged). After adjustment, there are no periods within the first time window exceeding the second load threshold. Therefore, the c second working periods can be: 10:05-10:09 for the washing machine, 10:00-10:05 for the air conditioner, and 10:07-10:10 for the electric water heater.

[0082] Finally, based on c second working periods and c adjustable electrical devices, a first scheduling scheme is determined. Specifically, for each second working period and its corresponding adjustable electrical device, a corresponding running time adjustment instruction can be generated, resulting in c adjustment instructions. For example, an adjustment instruction could be "Device ID: Washing Machine, Delay Flag: True, Time Offset: 00:03} / / Delay the start-up of the washing machine by 3 minutes (i.e., delay the start-up time of the washing machine from 10:02 to 10:05)". These c adjustment instructions constitute the first scheduling scheme.

[0083] In this way, by focusing on the adjustable equipment and adjusting the working period of the equipment according to the first preset rule, the operation of the core equipment is not affected, and the load is controlled within the safe threshold.

[0084] Optionally, when the system control strategy includes a rolling scheduling strategy, the linkage control module is further specifically used for: E1. Obtain the target electricity consumption change of the household load system at the current moment; E2. Update the a load datasets according to the target electricity consumption changes to obtain a updated load datasets; E3. Based on the a updated load datasets and the target time-series prediction model, determine the second predicted total load dataset within the second future time period; the start time of the second future time period is later than the end time of the preset time period, and the duration of the second future time period is less than the duration of the first future time period. E4. Based on the second predicted total load dataset and the second preset rule, determine the equipment operation schedule corresponding to the a electrical devices; E5. Determine the second scheduling scheme based on the equipment operation schedule; E6. Determine the second scheduling instruction corresponding to the second scheduling scheme; E7. Send the second scheduling instruction to the local control module.

[0085] In this embodiment of the application, the second preset rule can be preset in advance or defaulted.

[0086] In a specific embodiment, the power consumption change of the system at the current moment can be collected by the data acquisition module to obtain the target power consumption change. For example, the target power consumption change could be the sudden offline status of the air conditioner or the alarm of the washing machine malfunction, etc., which is not limited here. Then, the load data of each device corresponding to the target power consumption change can be collected by the data acquisition module to obtain at least one load data. Adding this at least one load data to a load dataset can obtain a updated load dataset.

[0087] Next, the updated load datasets can be input into the target time series prediction model to predict the total load data for the second future time period (ultra-short period, e.g., 5 minutes), thus obtaining the second predicted total load dataset. Then, based on the second predicted total load dataset and the second preset rule, an optimal start-up and shutdown schedule (i.e., equipment operation schedule) can be dynamically calculated and updated for the aforementioned a electrical devices to cope with rapid load fluctuations.

[0088] It should be explained that the second preset rule can be the same as or different from the first preset rule.

[0089] Then, the equipment operation schedule is used as the second scheduling scheme; next, the second scheduling instruction corresponding to the second scheduling scheme can be determined. For example, a pre-stored mapping relationship between the preset scheduling scheme and the scheduling instruction can be used to determine the second scheduling instruction corresponding to the second scheduling scheme. Finally, the second scheduling instruction can be sent to the local control module so that the local control model can execute the second scheduling scheme.

[0090] In this way, by updating load data based on real-time electricity consumption changes and combining ultra-short-term (short-term long-term) time-series forecasting, the prediction deviation caused by load fluctuations is significantly reduced. In addition, by directly generating equipment operation schedules and scheduling instructions, the instructions go directly to the local control module, reducing intermediate delays and ensuring that the scheduling plan is implemented quickly, taking into account both load safety and user experience.

[0091] Optional, please refer to Figure 4 , Figure 4 This is a schematic diagram of another intelligent household load scheduling device provided in this application embodiment. As can be seen, in addition to a data acquisition module and a linkage control module, the device also includes a cloud server. The linkage control module communicates with the cloud server. The cloud server deploys preset teacher models and preset student models, and stores a preset load database. Regarding the acquisition of the target time-series prediction model, the linkage control module is specifically used to perform the following steps: F1. Send a model request message to the cloud server; after receiving the model request message, the cloud server uses the data in the preset load database to train the preset teacher model and the preset student model to obtain the target time series prediction model. F2. Receive the first response data from the cloud server in response to the model request information; the first response data includes the installation package of the target time series prediction model; F3. Install the target time-series prediction model into the linkage control module based on the first response data.

[0092] In this embodiment, the linkage control module can first detect whether the target time series prediction model is deployed in the local database. If it is, the target time series prediction model can be obtained directly from the local database. If it is not, the linkage control module can send model request information to the cloud server.

[0093] After receiving the model request information, the cloud server calls its own preset load database and loads pre-configured preset teacher and student models. It then trains these two models using data from the preset load database to obtain the target time-series prediction model. Next, the target time-series prediction model is packaged into first response data, which is then sent to the linkage control module. This first response data may include the installation package of the target time-series prediction model, as well as model configuration files (containing prediction duration, load parameter thresholds, etc.) and integrity verification files, etc., without limitation.

[0094] After receiving the first response data, the linkage control module first verifies that the installation package is not damaged through the aforementioned integrity verification file. After the verification is successful, the target time series prediction model installation package contained in the first response data is installed locally, and the model initialization verification is triggered (input test load data and confirm that the prediction output meets the accuracy requirements). After the initialization verification is successful, the local deployment of the target time series prediction model is completed.

[0095] Thus, by using massive cloud data and powerful computing to support dual-model training, the target time series prediction model achieves both high accuracy and lightweight design, solving the problems of insufficient local data and computing power. In addition, by installing the target time series prediction model into the linkage control module, the network latency of relying on real-time cloud prediction is avoided, thereby enabling faster prediction response and adapting to the needs of rapid load fluctuation scheduling.

[0096] Specifically, in the step of training the preset teacher model and the preset student model using data from the preset load database to obtain the target time-series prediction model, the cloud server is used for: F11. Extract the training dataset from the preset load database; F12. Train the preset teacher model using the training dataset until the model converges to obtain the target teacher model; F13. Using the training dataset and the target teacher model, the preset student model is distilled to obtain the target student model; F14. Determine the target time series prediction model based on the target student model.

[0097] In this embodiment, the preset teacher model can be a DiffLoad network; the preset student model can be a single-layer GRU network.

[0098] In a specific embodiment, a first preset number (e.g., 10,000) of load data can be randomly extracted from a preset load database to form a training dataset. Then, the preset teacher model can be trained using the training dataset to learn the load change patterns (e.g., the correlation between equipment start-up and shutdown, time periods and future loads). The training is iterated until the prediction error stabilizes at a preset low level (i.e., the model converges), and finally a target teacher model that can accurately predict the load is obtained.

[0099] Furthermore, the pre-set student model can be distilled and trained using the training dataset and the target teacher model to obtain the target student model; finally, the target time series prediction model can be determined based on the target student model.

[0100] In some embodiments, through knowledge distillation technology, the target student model can fully inherit the core load prediction capabilities of the teacher model while maintaining extreme lightweightness (e.g., approximately 3900 parameters and a size of only 4KB). A single forward inference operation of the target student model takes only 50ms, and its computational efficiency is suitable for short-term load prediction needs at the minute or hour level, enabling rapid response to real-time prediction scenarios. The target student model can generate point-by-point load estimates for the next 30 minutes to 2 hours, used to accurately identify load peaks, power surges, or abnormal fluctuations; the prediction results can be transmitted to the linkage control module via an internal interface, providing core data support for the generation of scheduling strategies.

[0101] In this way, by training through distillation, student models can inherit the accuracy of teacher models while achieving lightweight design (fewer parameters and lower computing power requirements), thus solving the pain point of insufficient local computing power.

[0102] Optionally, in the step of using the training dataset and the target teacher model to distill and train the preset student model to obtain the target student model, the cloud server is specifically used to perform the following steps: G1. Extract distillation training data from the preset load database to obtain the distillation training dataset; G2. Input the distillation training dataset into the target teacher model and the preset student model respectively to obtain the teacher output and the student output. G3. Determine the actual output result corresponding to the distillation training dataset; G4. Determine the hard label loss based on the actual output and the student output. G5. Determine the soft label loss based on the teacher's output and the student's output; G6. Determine the target loss based on the hard label loss and the soft label loss; G7. Optimize the parameters of the preset student model according to the target loss and the preset optimization algorithm to obtain the target student model.

[0103] In this embodiment of the application, the preset optimization algorithm can be preset in advance or defaulted.

[0104] In a specific embodiment, a second preset number (e.g., 5000) of load data is randomly extracted from a preset load database to obtain a distillation training dataset. Next, the distillation training dataset can be input into a target teacher model to obtain teacher output results. Alternatively, the distillation training dataset can be input into a preset student model to obtain student output results. Furthermore, the actual output results corresponding to the distillation training dataset can be determined. Specifically, for each input data point in the distillation training dataset, the actual output result is the "actual total load value for the corresponding prediction period." This can be achieved by searching the preset load database for the "actual total load value at the end of the prediction period" based on the "timestamp" of each data point in the distillation training dataset. For example, assuming the input data timestamp is 19:00:00, and the load is predicted for the next 5 minutes, the actual total load value at 19:05:00 is extracted. Thus, the actual output result can be obtained. Alternatively, the distillation training dataset can be manually labeled to obtain the actual output result.

[0105] Then, based on the actual output and student output, the hard-label loss is determined. Specifically, since load forecasting is a continuous value regression task, the mean squared error can be chosen as the first loss function. The actual output and student output are substituted into the first loss function to calculate the hard-label loss. Next, the teacher output and student output can also be substituted into the first loss function to calculate the soft-label loss. Further, based on the hard-label loss and soft-label loss, the target loss is determined. For example, the average of the hard-label loss and soft-label loss can be calculated, which is the target loss. Alternatively, a preset weighting coefficient can be obtained. The preset weighting coefficient can be preset in advance or defaulted, and is calculated according to the following formula: Target loss = Hard label loss × Preset weight coefficient + Soft label loss (1 - Preset weight coefficient); Based on the above formula, the target loss can be obtained. Finally, the parameters of the preset student model are optimized according to the target loss and the preset optimization algorithm to obtain the target student model. The preset optimization algorithm may include one of the following: Adaptive Moment Estimation Optimization Algorithm (Adam Algorithm), Root Mean Square Propagation Optimization Algorithm (RMSprop Algorithm), etc., which are not limited here.

[0106] In this way, by binding the hard-labeled loss to the real results, the student model is ensured to fit the actual load pattern and maintain the bottom line of accuracy; the soft-labeled loss allows the student model to learn the high-precision knowledge of the teacher model, realize "experience transfer", and improve the prediction ability; the target loss balances the two, avoids overfitting or deviation from reality, and combines with optimization algorithms to efficiently optimize parameters, so that the final target student model retains its lightweight nature (adapting to local computing power) and approaches the accuracy of the teacher model, perfectly supporting ultra-short-term load prediction.

[0107] Optional, please refer to Figure 5 , Figure 5 This is a flowchart illustrating a method for determining a target time series prediction model according to an embodiment of this application. In the step of determining the target time series prediction model based on the target student model, the linkage control module is specifically used to execute... Figure 5 The steps described: H1. Determine the required performance parameters corresponding to the target student model; H2. Obtain the target performance parameters corresponding to the linkage control module; H3. When the target performance parameter is greater than or equal to the required performance parameter, the target student model is determined as the target time series prediction model; H4. When the target performance parameter is less than the required performance parameter, the target student model is quantized to obtain the target time series prediction model.

[0108] In the embodiments of this application, each performance parameter may include one of the following: storage capacity, computing power level, data processing rate, etc., which are not limited here.

[0109] In a specific embodiment, the required performance parameters corresponding to the target student model can be determined. For example, the required performance parameters can be storage capacity. The constituent files of the target student model can be obtained, such as the size of the core model file, the size of the supporting files, and the reserved space for the runtime cache. The file sizes of these constituent files are calculated to obtain multiple file sizes. These multiple file sizes are added together to obtain the required performance parameters. For example, assuming the core model file size is 400kb, the supporting file size is 30kb, and the reserved space for the runtime cache is 50kb, then the required performance parameters = 400 (kb) + 30 (kb) + 50 (kb) = 480 (kb). Next, the target performance parameters corresponding to the linkage control module can be obtained. Similarly, the target performance parameters can also be storage capacity. The available storage capacity of the linkage control module can be determined, which is also the target performance parameter.

[0110] When the target performance parameter is greater than or equal to the required performance parameter, the target student model can be directly used as the target time series prediction model.

[0111] When the target performance parameters are less than the required performance parameters, the target student model can be quantized. Without significantly losing prediction accuracy, the storage and computing power consumption of the model can be reduced by compressing the number of parameter bits and simplifying the calculation logic. Finally, the target time series prediction model adapted to the local deployment of the linkage control module is obtained. Specifically, since quantization is a conventional technique, it will not be elaborated here.

[0112] In this way, by comparing performance parameters, the compatibility between the model and the linkage control module can be clarified, avoiding operational failures caused by blind deployment. When the two are compatible, the target student model can be reused directly, saving quantization overhead while retaining the original prediction accuracy. When they are mismatched, resource consumption is compressed through quantization, allowing the model to adapt to the module's capabilities and avoiding deployment failures due to hardware limitations.

[0113] Optionally, after the local control module schedules the load of the a electrical devices according to the first scheduling scheme, the linkage control module is further specifically used for: I1. Obtain the actual load dataset for the first future time period; I2. Determine the target error based on the actual load dataset and the first predicted total load dataset; I3. When the target error is greater than the preset error, obtain the model update parameter package; update the parameters of the target time series prediction model according to the model update parameter package.

[0114] In this embodiment of the application, the preset error can be preset in advance or defaulted.

[0115] In a specific embodiment, during the first future time period, the actual load dataset of the system can be collected by the data acquisition module; then, based on the actual load dataset and the first predicted total load dataset, the target error is determined. Specifically, the data in the actual load dataset and the first predicted total load dataset can be substituted into the mean absolute error calculation formula to calculate the mean absolute error, which is also the target error.

[0116] When the target error is greater than the preset error, the model update parameter package can be obtained from the cloud server. The model update parameter package is a data package that only contains the parameter differences between the old and new models. Specifically, it can include: difference parameter ΔW (the difference value between the old and new model parameters, which is completely consistent with the parameter dimension and data type of the target time series prediction model).

[0117] Then, the parameters of the target time series prediction model can be updated according to the model update parameter package. For example, assuming that a certain parameter of the target time series prediction model is W1old, the corresponding ΔW1 can be extracted from the model update parameter package, and W1old can be updated according to the preset update formula. The preset update formula is as follows: W1new = W1old + ΔW1; W1new represents the updated parameters; in this way, the parameters of the target time series prediction model can be updated.

[0118] If the target error is not greater than the preset error, it means that the prediction accuracy of the target time series prediction model is within the normal range and there is no need to update its parameters.

[0119] In this way, updates are triggered only when the error exceeds the limit, reducing unnecessary calculations and resource consumption. In addition, the model is updated by using a lightweight parameter package (i.e., the model update parameter package), which balances update speed and accuracy improvement, ensuring that the target time series prediction model continuously adapts to load changes and provides a reliable prediction basis for subsequent scheduling.

[0120] As can be seen, the intelligent household load scheduling device described in this application, on the one hand, communicates directly with the household load system through the data acquisition module, without any transmission links across the cloud or wide area network. The data acquisition delay depends only on the local communication protocol, which greatly reduces the transmission delay. On the other hand, by deploying the target time series prediction model in the linkage control module, the model makes predictions locally based on the collected data, and then directly completes decision analysis, scheduling scheme generation, and command issuance locally without relying on cloud relay, forming a local closed loop of "prediction-decision-execution", further compressing the delay of the entire process and ensuring real-time scheduling response, thereby meeting the needs of real-time household scheduling and control.

[0121] Please see Figure 6 , Figure 6 This is a schematic diagram of another intelligent household load dispatching device provided in this application embodiment; it can be seen that, in addition to a data acquisition module, a linkage control module, and a cloud server, the intelligent household load dispatching device may also include a user interaction module, which has the following functions: 1. Data reception and visualization The user interaction module can receive various types of data uploaded by the linkage control module or the local control module, including historical power consumption data, load forecast results, equipment control logs, and abnormal alarm records. These data are presented intuitively on the interactive interface in the form of charts, trend curves, or structured text for users to view. Examples include actual power consumption curves for the past 24 hours or 7 days, future load trend charts predicted by the system, operating status and control records of each connected device, monthly or quarterly power consumption reports, and energy-saving assessment reports.

[0122] 2. Proactive alerts for abnormal situations When the device detects abnormal power consumption, it can push notifications to users via a mobile application. Abnormal scenarios include: abnormal device operating power, continuous high energy consumption, failure to execute operations according to preset scheduling strategies, or deviations between actual and predicted power consumption exceeding a set threshold. The purpose of the notification is to guide users to promptly check the device status or adjust control strategies.

[0123] 3. Interact with users Users can participate in the device's operational decisions through the user interaction module. Specific operations include: manually switching connected devices on and off, setting or adjusting the operating priorities of each device, configuring peak-shaving power consumption periods, and setting personalized parameters such as the maximum allowable power threshold for each device. During subsequent load scheduling, the device will strictly refer to the parameters set by the user to ensure that automatic control behavior is executed within the user-authorized range.

[0124] Please see Figure 7 , Figure 7 This is a flowchart of a smart home load scheduling method provided in an embodiment of this application, applied to a home load system. The home load system includes a local control module and a electrical devices, where a is a positive integer. The smart home load scheduling method includes: S1. Collect the load data set of each of the a electrical devices within a preset time period to obtain a load data sets; S2. Obtain the target time-series prediction model; based on the a load datasets and the target time-series prediction model, determine the first predicted total load dataset within the first future time period; the start time of the first future time period is later than the end time of the preset time period; obtain the system control strategy corresponding to the household load system; based on the system control strategy and the first predicted total load dataset, determine the first scheduling scheme corresponding to the first future time period; determine the first scheduling instruction corresponding to the first scheduling scheme; send the first scheduling instruction to the local control module; the first scheduling instruction is used to instruct the local control module to schedule the load of the a electrical devices according to the first scheduling scheme.

[0125] Please see Figure 8 , Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include a processor, a memory, a communication interface, and one or more programs. The processor, memory, and communication interface can be interconnected via a bus. The one or more programs are stored in the memory and configured to be executed by the processor. In this embodiment, the programs include instructions for executing the home load intelligent scheduling method provided in the above embodiment.

[0126] 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 this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. 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 this application.

[0127] 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.

[0128] In the several embodiments provided in this application, 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.

[0129] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

[0130] The steps of the methods or algorithms described in the embodiments of this application can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in RAM, flash memory, ROM, EPROM, electrically erasable programmable read-only memory (EEPROM), registers, hard disk, portable hard disk, read-only optical disk (CD-ROM), or any other form of storage medium well known in the art. An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and storage medium can reside in an ASIC. Furthermore, the ASIC can reside in a terminal device or management device. Alternatively, the processor and storage medium can exist as discrete components in the terminal device or management device.

[0131] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in the embodiments of this application can be implemented, in whole or in part, by software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, in the form of a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated.

[0132] The aforementioned computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media.

[0133] The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0134] The modules / units included in the various devices and products described in the above embodiments can be software modules / units, hardware modules / units, or a combination of both. For example, for devices and products applied to or integrated into a chip, all modules / units can be implemented using hardware methods such as circuits, or at least some modules / units can be implemented using software programs that run on a processor integrated within the chip, while the remaining (if any) modules / units can be implemented using hardware methods such as circuits. For devices and products applied to or integrated into a chip module, all modules / units can be implemented using hardware methods such as circuits. Different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or different components of the chip module, or at least some modules / units can be implemented using hardware methods such as circuits. The implementation is achieved through a software program that runs on the processor integrated within the chip module. The remaining modules / units (if any) can be implemented using hardware methods such as circuits. For various devices and products applied to or integrated into terminal equipment, each of their modules / units can be implemented using hardware methods such as circuits. Different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or different components within the terminal equipment. Alternatively, at least some modules / units can be implemented through a software program that runs on the processor integrated within the terminal equipment, while the remaining modules / units (if any) can be implemented using hardware methods such as circuits.

[0135] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the embodiments of this application. It should be understood that the above descriptions are merely specific embodiments of the embodiments of this application and are not intended to limit the protection scope of the embodiments of this application. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solutions of the embodiments of this application should be included within the protection scope of the embodiments of this application.

Claims

1. A household load intelligent dispatching device, characterized in that, The device includes: a data acquisition module and a linkage control module. Both the linkage control module and the data acquisition module are communicatively connected to a home load system. The home load system includes a local control module and *a* electrical devices, where *a* is a positive integer. The data acquisition module is used to collect the load data set of each of the a electrical devices within a preset time period to obtain a load data set; The linkage control module is used to acquire a target time-series prediction model; determine a first predicted total load dataset within a first future time period based on the a load datasets and the target time-series prediction model; the start time of the first future time period is later than the end time of the preset time period; acquire the system control strategy corresponding to the household load system; determine a first scheduling scheme corresponding to the first future time period based on the system control strategy and the first predicted total load dataset; determine a first scheduling instruction corresponding to the first scheduling scheme; and send the first scheduling instruction to the local control module; the first scheduling instruction is used to instruct the local control module to schedule the load of the a electrical devices according to the first scheduling scheme.

2. The apparatus as claimed in claim 1, characterized in that, The first predicted total load dataset includes: multiple predicted total load values ​​and the prediction time corresponding to each total load value; The system control strategy includes a threshold control strategy. Regarding determining the first scheduling scheme based on the system control strategy and the first predicted total load dataset, the linkage control module is specifically used for: Determine the first load threshold corresponding to the household load system; The predicted load curve segment is determined based on the multiple predicted total load values ​​and the predicted time corresponding to each total load value; the horizontal axis of the predicted load curve is time, and the vertical axis is the load value. Determine the partial load curves in the predicted load curve segment that are greater than the first load threshold, resulting in b partial load curves; where b is a natural number. The first scheduling scheme is determined based on the b partial load curves, the a electrical devices, and the a load datasets.

3. The apparatus as described in claim 2, characterized in that, In determining the first scheduling scheme based on the b partial load curves, the a electrical devices, and the first predicted total load dataset, the linkage control module is specifically used for: Determine the time period corresponding to each of the b partial load curves to obtain b time periods; Based on the b partial load curves and the first load threshold, b load values ​​to be adjusted are determined; Determine the adjustable electrical devices among the a electrical devices to obtain c adjustable electrical devices; c is a positive integer less than or equal to a; Determine the c priorities corresponding to the c adjustable electrical devices; Each adjustable electrical device corresponds to a priority level; Based on the c priorities, the b load values ​​to be adjusted, and the c adjustable electrical devices, determine b scheduling schemes corresponding to the b time periods; The first scheduling scheme is determined based on the b scheduling schemes.

4. The apparatus as described in claim 3, characterized in that, In determining the b scheduling schemes corresponding to the b time periods based on the c priorities, the b load values ​​to be adjusted, and the c adjustable electrical devices, the linkage control module is specifically used for: Obtain a first time period and its corresponding first load value to be adjusted; the first time period is any one of the b time periods. Based on the a load datasets, determine the c first average load values ​​corresponding to the c adjustable electrical devices; Each adjustable electrical device corresponds to a first average load value; Determine the c minimum load values ​​corresponding to the c adjustable electrical devices; A minimum load value corresponding to each adjustable electrical device; Based on the c first average load values ​​and the c minimum load values, determine c adjustable load values; Determine the total adjustable load value corresponding to the c adjustable load values; When the total adjustable load value is greater than or equal to the first adjustable load value, c adjustable load values ​​corresponding to the c adjustable electrical devices are determined according to the c priorities and the c adjustable load values, with each adjustable load value corresponding to one adjustable electrical device; and a scheduling scheme corresponding to the first time period is determined according to the c adjustable load values. When the adjustable total load value is less than the first load value to be adjusted, d first average load values ​​are determined based on the c priorities and the c first average load values; the sum of the d first average load values ​​is greater than the first load value to be adjusted. d is an integer less than or equal to c; based on the d first average load values ​​and the c adjustable electrical devices, the scheduling scheme corresponding to the first time period is determined.

5. The apparatus as claimed in claim 1, characterized in that, The system control strategy includes a sliding window strategy. Regarding determining the first scheduling scheme based on the system control strategy and the first predicted total load dataset, the linkage control module is specifically used for: Get the preset window length; A first time window is determined based on the preset window length and the first future time period; the start time of the first time window is equal to the start time of the first future time period. The first average predicted load value corresponding to the first time window is determined based on the first predicted total load dataset. When the first average predicted load value is greater than the second load threshold, the adjustable electrical equipment among the a electrical devices is identified to obtain c adjustable electrical equipment; the first working period of the c adjustable electrical equipment in the first time window is identified to obtain c first working periods; The c first working periods are adjusted according to the second load threshold and the first preset rule to obtain c second working periods; the first scheduling scheme is determined according to the c second working periods and the c adjustable electrical devices.

6. The apparatus as claimed in claim 1, characterized in that, When the system control strategy includes a rolling scheduling strategy, the linkage control module is further specifically used for: Obtain the target electricity consumption change of the household load system at the current moment; The a load datasets are updated based on the target electricity consumption changes to obtain a updated load datasets; Based on the a updated load datasets and the target time-series prediction model, a second predicted total load dataset is determined within a second future time period; the start time of the second future time period is later than the end time of the preset time period, and the duration of the second future time period is less than the duration of the first future time period. Based on the second predicted total load dataset and the second preset rule, determine the equipment operation schedule corresponding to the a electrical devices; A second scheduling scheme is determined based on the equipment operation schedule; Determine the second scheduling instruction corresponding to the second scheduling scheme; The second scheduling instruction is sent to the local control module.

7. The apparatus according to any one of claims 1-6, characterized in that, The device further includes: a cloud server, and the linkage control module communicates with the cloud server; the cloud server is equipped with a preset teacher model and a preset student model, and stores a preset workload database; Regarding the acquisition of the target time-series prediction model, the linkage control module is specifically used for: A model request message is sent to the cloud server; after receiving the model request message, the cloud server uses the data in the preset load database to train the preset teacher model and the preset student model to obtain the target time series prediction model. Receive first response data from the cloud server in response to the model request information; the first response data includes the installation package of the target time series prediction model; The target time-series prediction model is installed into the linkage control module based on the first response data; Specifically, in the step of training the preset teacher model and the preset student model using data from the preset load database to obtain the target time-series prediction model, the cloud server is used for: Extract the training dataset from the preset load database; The preset teacher model is trained using the training dataset until the model converges, thus obtaining the target teacher model. The preset student model is distilled and trained using the training dataset and the target teacher model to obtain the target student model; The target time series prediction model is determined based on the target student model.

8. The apparatus as claimed in claim 7, characterized in that, In the process of distilling a preset student model using the training dataset and the target teacher model to obtain a target student model, the cloud server is specifically used for: Distillation training data is extracted from the preset load database to obtain the distillation training dataset; The distillation training dataset is input into the target teacher model and the preset student model respectively to obtain the teacher output and the student output. Determine the actual output results corresponding to the distillation training dataset; Based on the actual output results and the student output results, determine the hard label loss; Based on the teacher's output and the student's output, determine the soft label loss; The target loss is determined based on the hard label loss and the soft label loss; The parameters of the preset student model are optimized based on the target loss and the preset optimization algorithm to obtain the target student model.

9. The apparatus as claimed in claim 7, characterized in that, In the process of determining the target time-series prediction model based on the target student model, the linkage control module is specifically used for: Determine the required performance parameters corresponding to the target student model; Obtain the target performance parameters corresponding to the linkage control module; When the target performance parameter is greater than or equal to the required performance parameter, the target student model is determined as the target time series prediction model; When the target performance parameter is less than the required performance parameter, the target student model is quantized to obtain the target time series prediction model.

10. The apparatus according to any one of claims 1-3, characterized in that, After the local control module schedules the load of the a electrical devices according to the first scheduling scheme, the linkage control module is further specifically used for: Obtain the actual load dataset for the first future time period; The target error is determined based on the actual load dataset and the first predicted total load dataset; When the target error is greater than the preset error, obtain the model update parameter package; The target time series prediction model is updated with parameters based on the model update parameter package.

Images