Energy scheduling method of household solar power supply system, electronic device, readable storage medium and computer program product

Abstract

This application relates to the field of solar power supply technology, and more particularly to energy dispatching methods, electronic devices, readable storage media, and computer program products for residential solar power supply systems. The method includes acquiring meteorological data and historical power generation data; generating solar power generation forecasts for a preset future time period based on a multimodal data fusion prediction model; performing strategic-level, tactical-level, and execution-level dispatching based on the solar power generation forecasts; generating dispatching instructions, wherein strategic-level dispatching executes long-term energy balance planning on a daily cycle, tactical-level dispatching executes short-term planning for charging and discharging power and load control on an hourly cycle, and execution-level dispatching executes real-time power adjustment on a minute cycle; and issuing dispatching instructions to inverters, energy storage systems, and controllable loads to control energy flow and power. This achieves cross-timescale collaborative dispatching energy management from long-term planning to real-time execution.

Description

Technical Field

[0001] This application relates to the field of solar power technology, and in particular to energy dispatching methods, electronic devices, readable storage media, and computer program products for residential solar power systems. Background Technology

[0002] A typical residential solar power system usually includes a photovoltaic array, energy storage batteries, an inverter, and household loads. Its operational goal is to maximize solar energy utilization, reduce dependence on the power grid, and ensure the safe and stable operation of the system while meeting electricity demand. However, solar power generation is significantly affected by weather conditions, exhibiting strong intermittency and volatility. Simply relying on real-time power balance is insufficient to achieve optimal scheduling; therefore, forward-looking decision-making based on power generation forecasts, energy storage status, and grid information is necessary.

[0003] Currently, most residential solar power systems on the market employ rule-based scheduling strategies for energy management. For example, they set charging and discharging thresholds for the battery's state of charge (SOC). When the photovoltaic power exceeds the load, the battery is charged; when it falls below the load, the battery discharges or draws power from the grid. For instance, Chinese patent CN121485195A discloses an automatic control method for battery management. This method first acquires the status information of each battery in the battery pack, the park load, and the current electricity price to determine if charging or discharging conditions are met. If the photovoltaic power is sufficient and the battery is not fully charged, photovoltaic charging is initiated, dynamically adjusting the charging power and stopping when the charging end standard is reached. If the discharging start condition is met, the power of the battery pack is dynamically allocated and adjusted in real time until the discharging end standard is met, achieving safe and efficient charging and discharging management. This scheme automatically determines whether to charge or discharge, prioritizing photovoltaic charging when conditions are met and stopping once the standard is reached.

[0004] While such strategies are simple in structure and easy to implement, their scheduling decisions rely solely on the current state, lacking the ability to predict future changes in power generation, demand, and prices. This results in an inability to reserve energy storage capacity in advance during periods of continuous rain or sudden load fluctuations. Furthermore, because the charging and discharging thresholds are fixed, they fail to dynamically adjust based on battery aging and the impact of charging and discharging rates on lifespan, easily leading to overcharging, over-discharging, or high-frequency deep charging and discharging, accelerating capacity decay. In other words, current scheduling methods are insufficient in handling multi-source information fusion and multi-timescale coordination.

[0005] Therefore, how to integrate weather forecasting, power generation forecasting, energy storage status, and load priority, and achieve cross-timescale collaborative scheduling of energy management from long-term planning to real-time execution, in order to solve the technical problems of poor system operation stability and large equipment lifespan loss caused by single scheduling strategies and insufficient foresight, is an urgent problem to be solved.

[0006] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0007] The main purpose of this application is to provide an energy dispatching method, electronic device, readable storage medium and computer program product for a household solar power system, which aims to solve the technical problems of poor system operation stability and large equipment life loss caused by single dispatching strategy and insufficient foresight.

[0008] To achieve the above objectives, this application proposes an energy dispatching method for a residential solar power system, the energy dispatching method for the residential solar power system comprising:

[0009] Acquire meteorological data and historical power generation data, and generate solar power generation forecasts for future preset time periods based on a multimodal data fusion prediction model;

[0010] Based on the solar power generation forecast results, strategic-level scheduling, tactical-level scheduling, and execution-level scheduling are executed to generate scheduling instructions. The strategic-level scheduling executes long-term energy balance planning on a daily cycle, the tactical-level scheduling executes short-term planning for charging and discharging power and load control on an hourly cycle, and the execution-level scheduling executes real-time power adjustment on a minute cycle.

[0011] The dispatching instructions are sent to the inverter, energy storage system and controllable load to control the flow and power of energy.

[0012] In one embodiment, the meteorological data includes time-series cloud images; the step of acquiring meteorological data and historical power generation data, and generating solar power generation prediction results for a future preset time period based on a multimodal data fusion prediction model includes:

[0013] The time-series cloud images are processed using a three-dimensional convolutional neural network to extract dynamic features of the clouds.

[0014] Based on the LSTM-Transformer hybrid model, combined with the historical power generation data, weather type coding, and cloud dynamic characteristics, the power generation prediction value is output for each time point.

[0015] In one embodiment, the time-series cloud image includes multiple consecutive frames of satellite cloud images, and the step of processing the time-series cloud image based on a three-dimensional convolutional neural network to extract dynamic features of the cloud layer includes:

[0016] Preprocessing operations are performed on the continuous multi-frame satellite cloud images to construct a time-series image dataset;

[0017] In the aforementioned time-series image dataset, spatial shape features and temporal movement features of clouds are extracted simultaneously using 3D convolution kernels.

[0018] Based on the attention mechanism, the feature weights of the spatial shape features and the movement features are dynamically adjusted in the temporal dimension at different time steps to output a dynamic feature vector of the cloud layer.

[0019] In one embodiment, the step of the strategic layer scheduling to execute a long-term energy balance plan on a daily cycle includes:

[0020] Based on the solar power generation prediction results and the baseline of household electricity consumption patterns, a power generation-electricity consumption-energy storage balance strategy is generated.

[0021] Based on dynamic electricity price information, a charging and discharging strategy is planned to supplement grid power storage during off-peak hours and prioritize the use of energy storage discharge during peak hours.

[0022] When performing energy storage discharge operations, reserve energy storage redundancy according to the preset emergency capacity.

[0023] In one embodiment, the step of the tactical layer scheduling to execute short-term planning for charge / discharge power and load control on an hourly basis includes:

[0024] The tactical layer scheduling is executed based on a reinforcement learning scheduling engine.

[0025] The state space of the reinforcement learning scheduling engine includes real-time power generation, battery state of charge, grid purchase price, and household load priority. The action space of the reinforcement learning scheduling engine includes battery charging and discharging power adjustment, interruptible load control, and grid interaction power limitation.

[0026] In one embodiment, the reward function of the reinforcement learning scheduling engine is a multi-objective weighted sum: R = w1·R_power generation + w2·R_energy storage + w3·R_grid + w4·R_user.

[0027] In one embodiment, the energy dispatching method for the household solar power system further includes:

[0028] When a power generation fluctuation event is detected, the fluctuation data is recorded;

[0029] A grayscale prediction model is established based on the fluctuation data;

[0030] The cloud movement speed of the meteorological data is transmitted to the grayscale prediction model to obtain the fluctuation compensation coefficient;

[0031] Based on the fluctuation compensation coefficient, a correction operation is performed on the solar power generation prediction results.

[0032] In addition, to achieve the above objectives, this application also proposes an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the energy dispatching method for a home solar power system as described above.

[0033] In addition, to achieve the above objectives, this application also proposes a readable storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the energy dispatching method for a household solar power system as described above.

[0034] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the energy dispatching method for a home solar power system as described above.

[0035] One or more technical solutions proposed in this application have at least the following technical effects:

[0036] By generating solar power generation forecasts for future preset time periods, residential solar power systems can anticipate the changing trends of photovoltaic output over a period of time. Through execution-level scheduling, real-time power adjustment is performed on a minute-by-minute basis, allowing for advance adjustment of the charging and discharging status of the energy storage system based on the forecast results. In other words, by combining forward-looking forecasts with minute-level real-time power adjustment, the impact of photovoltaic power fluctuations on the grid side is smoothed, the power change rate at the grid connection point is reduced, and thus the impact on the distribution network is reduced.

[0037] By using tactical-level scheduling to plan charging and discharging power and load control on an hourly basis, frequent deep charging and discharging of the energy storage system is avoided. Since residential solar power systems can predict power generation shortfalls or surpluses in the next few hours, the tactical level can rationally arrange charging and discharging depth and rate. In other words, by combining daily strategic planning with hourly charging and discharging strategy optimization, the operating range and charging / discharging rate of the energy storage system can be actively controlled, reducing the number of deep charging and discharging cycles and thus extending the actual cycle life of the battery.

[0038] By decomposing scheduling tasks into three time scales—strategic, tactical, and execution—the compromise between high-precision, long-term optimization and millisecond-level rapid response by a single scheduler is avoided. This layered design ensures that each layer focuses solely on the problem within its own time scale, and that information is exchanged between layers through prediction results and state variables, thus guaranteeing both global optimization capabilities and millisecond-level response. Attached Figure Description

[0039] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a flowchart illustrating the first embodiment of the energy dispatching method for a household solar power system according to this application.

[0042] Figure 2 This is a flowchart illustrating an embodiment of step S200 of the energy dispatching method for a household solar power system according to this application.

[0043] Figure 3 This is a flowchart illustrating the second embodiment of the energy dispatching method for a household solar power system according to this application.

[0044] Figure 4 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the energy dispatching method of the household solar power supply system in the embodiments of this application.

[0045] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0046] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0047] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0048] It should be noted that the execution subject of the energy dispatching method for a household solar power system in this application can be the energy dispatching system of the household solar power system, or a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions. This embodiment does not specifically limit it in this way. The following uses the energy dispatching system of the household solar power system as the execution subject as an example to describe this embodiment and the following embodiments.

[0049] Based on this, this application proposes an energy dispatching method for a household solar power supply system according to the first embodiment. Please refer to [link / reference]. Figure 1The energy dispatching method for the household solar power system includes steps S100 to S300:

[0050] Step S100: Obtain meteorological data and historical power generation data, and generate solar power generation prediction results for the future preset time period based on the multimodal data fusion prediction model.

[0051] It should be noted that the preset time period can be 24 hours, 48 ​​hours, or 72 hours in the future, and the time resolution of the solar power generation prediction results can be 5 minutes, 15 minutes, or 1 hour. In practical applications, the appropriate resolution can be selected based on household electricity demand and computing resources. A multimodal data fusion prediction model refers to a model that simultaneously utilizes two or more different types of data (such as meteorological image data, numerical weather prediction data, historical power data, etc.) for comprehensive prediction.

[0052] Optionally, meteorological data can be obtained from local micro-weather stations, providing hourly weather forecasts for the next 24 hours. Meteorological data may include light intensity (W / m²), temperature (°C), cloud cover (%, 0-100), and probability of precipitation (%). If no local weather station is available, data can be obtained from a public weather API. Historical power generation data includes the inverter's actual power generation over the past 30 days, with an hourly average, totaling 720 data points.

[0053] A multiple linear regression model is constructed as a multimodal data fusion prediction model, for example, P_solar = a0 + a1×light intensity + a2×temperature + a3×cloud cover + a4×rainfall probability + a5×P_history, where P_history is the actual power generation at the same time the previous day; P_solar is the power generation (kW) at the prediction time. Using historical data from the past 30 days, coefficients a0 to a5 are fitted using the least squares method. Then, the hourly weather forecast values ​​for the next 24 hours and the historical power at the same time the previous day are input to calculate the hourly power generation prediction values ​​for the next 24 hours.

[0054] In this way, basic scheduling needs can be met while taking into account computational efficiency, providing lightweight and highly responsive decision support for home energy management.

[0055] Step S200: Based on the solar power generation prediction results, perform strategic-level scheduling, tactical-level scheduling, and execution-level scheduling to generate scheduling instructions. The strategic-level scheduling performs long-term energy balance planning on a daily basis, the tactical-level scheduling performs short-term planning for charging and discharging power and load control on an hourly basis, and the execution-level scheduling performs real-time power adjustment on a minute-by-minute basis.

[0056] In this embodiment, when performing strategic-level scheduling, the inputs are the hourly power generation forecast value P_solar[h] for the next 24 hours, the typical household electricity consumption baseline P_load_base[h] (obtained by averaging the electricity consumption during the same period over the past week), and the peak-valley electricity price period for the next day. The net electricity consumption ΔE[h] = P_solar[h] - P_load_base[h] is calculated hourly. If ΔE[h] > 0 and the current battery SOC < 90%, it means there is a surplus in power generation and the battery still has charging potential, so the decision is to charge the battery for that hour. The charging power can be set to min(ΔE[h], 5kW). If ΔE[h] < 0 and the current battery SOC > 20%, it means insufficient power generation and sufficient battery power, so the decision is to supplement the battery by discharging for that hour, with the discharge power set to min(-ΔE[h], 5kW). If insufficient power generation and battery SOC ≤ 20%, the decision is to draw power from the grid.

[0057] When executing tactical-level scheduling, the strategic decisions mentioned above are adjusted based on real-time status in 15-minute cycles. Real-time generation power, battery SOC, current electricity price, and current total load power are read. The real-time generation power is compared with the predicted generation power. If the actual generation power is more than 20% lower than the predicted generation power, and SOC > 30%, battery discharge power is increased, and electricity purchase is reduced. If SOC > 85% and there is still a photovoltaic surplus, interruptible loads (such as water heaters and electric vehicle charging) are activated. If the grid electricity price is during peak hours and SOC > 30%, priority is given to discharging, and the purchased electricity power is limited to no more than 2kW. In other words, when executing tactical-level scheduling, the target values ​​for battery charging and discharging power, the switching commands for interruptible loads, and the upper limit of grid interaction power are output.

[0058] During execution-level scheduling, the actual power is tracked to the target value issued by the tactical layer using a PID controller with a 1-minute cycle. Specifically, the current battery charging / discharging power P_actual is read, and the error e = P_target - P_actual is calculated, where P_target is the charging / discharging power target value issued by the tactical layer. After proportional, integral, and derivative triple correction, a PWM duty cycle signal is output to drive the inverter, and it is updated every 50ms to bring P_actual to convergence.

[0059] To further ensure the safety of charging and discharging, the battery voltage is monitored in real time. If the battery voltage exceeds 4.25V / cell or falls below 2.5V / cell, charging and discharging are stopped immediately. The inverter temperature is also monitored in real time. If the inverter temperature exceeds 85℃, the inverter is derated by 50%.

[0060] In step S300, the scheduling command is sent to the inverter, energy storage system and controllable load to control the flow and power of energy.

[0061] It should be noted that inverters are used to control the direction and magnitude of power flow between photovoltaic arrays, batteries, the grid, and household loads, supporting bidirectional operation of rectification (AC to DC) and inversion (DC to AC). Energy storage systems are energy storage units that include battery packs and battery management systems, used to perform charging or discharging according to instructions and to provide feedback on states such as SOC, voltage, and current. Controllable loads refer to electrical devices that can receive switching or power adjustment commands through a communication interface, such as smart sockets, inverter air conditioners, and electric water heaters, responding to dispatch commands to connect, disconnect, or adjust operating power, participating in demand-side response.

[0062] Dispatch commands include at least power setting commands, mode switching commands, and switching / regulation commands. The power setting command is sent to the inverter and energy storage system to set their target output or absorbed power value; a positive value indicates battery discharge to the load, and a negative value indicates battery charging from the photovoltaic / grid. The mode switching command is sent to the inverter to control its connection status with the public grid. The switching / regulation command is sent to the controllable load to control its start / stop or power regulation.

[0063] Understandably, by adjusting the on-time and duty cycle of the power switches in the inverter, the power flow path can be changed. For example, controlling the inverter to operate in grid-connected mode converts photovoltaic DC power into AC power to supply the load; controlling the inverter's internal DC / DC converter to step down the voltage and store excess photovoltaic energy into the battery; controlling the battery DC power to be converted into AC power via the inverter to supply the load; and rectifying the AC power back into DC power to charge the battery.

[0064] In the technical solution provided in this embodiment, by generating solar power generation predictions for a preset future time period, the residential solar power supply system can anticipate the changing trend of photovoltaic output over a future period. Through execution-level scheduling, real-time power adjustment is performed on a minute-by-minute basis to adjust the charging and discharging state of the energy storage system in advance based on the prediction results. For example, five minutes before a predicted power drop, the battery is switched from charging mode to discharging mode or its discharge power is increased. In other words, by combining forward-looking prediction with minute-level real-time power adjustment, the impact of photovoltaic power fluctuations on the grid side is smoothed, the power change rate at the grid connection point is reduced, and thus the impact on the distribution network is minimized.

[0065] By employing tactical-level scheduling for short-term planning of charging and discharging power and load control on an hourly basis, frequent deep charging and discharging of the energy storage system is avoided. Since residential solar power systems can predict power generation shortfalls or surpluses in the next few hours, the tactical level can rationally arrange charging and discharging depth and rate. For example, if continuous cloudy or rainy weather is predicted for the next 6 hours, the tactical level will limit the discharge depth, such as ensuring the State of Charge (SOC) does not fall below 30%, rather than passively stopping discharge when the SOC falls below a threshold. In other words, by combining daily strategic planning with hourly charging and discharging strategy optimization, the operating range and charging / discharging rate of the energy storage system can be actively controlled, reducing the number of deep charging and discharging cycles and thus extending the actual cycle life of the battery.

[0066] By decomposing scheduling tasks into three time scales—strategic, tactical, and execution—the compromise between high-precision, long-term optimization and millisecond-level rapid response by a single scheduler is avoided. This layered design ensures that each layer focuses solely on the problem within its own time scale, and that information is exchanged between layers through prediction results and state variables, thus guaranteeing both global optimization capabilities and millisecond-level response.

[0067] In one feasible implementation, the meteorological data includes time-series cloud images, and step S100 may include steps S110 to S120:

[0068] Step S110: Process the time-series cloud image based on a three-dimensional convolutional neural network to extract dynamic features of the cloud layer;

[0069] Step S120: Based on the LSTM-Transformer hybrid model, and combining the historical power generation data, weather type coding, and cloud dynamic characteristics, output the power generation prediction value for each time point.

[0070] Optionally, the time-series cloud image includes multiple consecutive frames of satellite cloud images, and step S110 may further include steps S111 to S113:

[0071] Step S111: Perform preprocessing operations on the continuous multi-frame satellite cloud images to construct a time-series image dataset;

[0072] Step S112: In the time-series image dataset, extract the spatial shape features and temporal movement features of the cloud layer simultaneously using a 3D convolution kernel.

[0073] Step S113: Based on the attention mechanism, dynamically adjust the feature weights of the spatial shape features and the movement features at different time steps in the temporal dimension, and output the cloud dynamic feature vector.

[0074] It should be noted that infrared cloud images and water vapor cloud images over the target area can be obtained from publicly available meteorological satellites. Infrared cloud images reflect cloud top temperature, while water vapor cloud images reflect atmospheric water vapor content. Time-series data from the past 6 hours to the next 18 hours (24 hours in total) are used to create the aforementioned continuous multi-frame satellite cloud images. Preprocessing operations include cropping, alignment, normalization, and dataset construction. If one frame is taken every 15 minutes, the total number of frames is 96.

[0075] Specifically, the cropping operation includes cropping a square region centered on the latitude and longitude of the home location, with sides of 256 pixels, and removing irrelevant ocean and land boundaries. In the alignment operation, since satellites may have slight shifts at different times, phase correlation is used to register adjacent frames, ensuring that the same geographical location corresponds to the same pixel position in different frames. In the normalization operation, the brightness temperature value (in Kelvin) or reflectance (0-100%) of each pixel is linearly mapped to the [0,1] interval. For example, the infrared brightness temperature range is 180K-320K, and the mapping formula is (value-180) / (320-180). In the dataset construction operation, the 96 preprocessed images are stacked chronologically to form a four-dimensional tensor of size (96, 256, 256, C), where C is the number of channels, for example, two channels for infrared and water vapor, C=2. This tensor is the time-series image dataset.

[0076] 3D convolutional kernels slide simultaneously along both spatial (height × width) and temporal dimensions. A 3×3×3 kernel can be used, where the first 3 represents the temporal depth (one convolution covers three consecutive frames), and the last two 3s represent the spatial size (3×3 pixels). In the first 3D convolutional layer, with an input size of (96, 256, 256, 2), 32 3×3×3 kernels are used with a stride of 1 and "same" padding, resulting in an output size of (96, 256, 256, 32). In the second 3D convolutional layer, 64 3×3×3 kernels are used with a stride of 2 (downsampling), resulting in an output size of (48, 128, 128, 64). In the third 3D convolutional layer, 128 3×3×3 kernels are used with a stride of 2, resulting in an output size of (24, 64, 64, 128). After three convolutional layers, a feature map with dimensions (24, 64, 64, 128) is obtained, where 24 is the compressed time step number, 64×64 is the spatial size, and 128 is the number of feature channels. Each location in this feature map simultaneously encodes the cloud motion information of that spatiotemporal region.

[0077] Understandably, in a 96-frame (24-hour) time series, not all time steps are equally important for power generation prediction. For example, cloud images at night (20:00-6:00 the next day) are irrelevant to power generation; while the period of rapid cloud movement during the day (e.g., 10:00-15:00) has the greatest impact on power generation. The attention mechanism allows the model to automatically focus on key time periods. Specifically, the output feature map is denoted as X, with dimensions (T, H, W, C), where T=24, H=64, W=64, and C=128. Then, for each time step t, global average pooling is performed on the spatial dimension H×W to obtain a C-dimensional vector. This is repeated for all T time steps to obtain a (T, C) matrix, which is the global feature description for each time step. This matrix is ​​then input into a two-layer fully connected network. The first layer reduces the dimension of C=128 to 16 and uses ReLU activation; the second layer restores the dimension of 16 to 128 and uses Sigmoid activation. The output is a weight matrix of (T, C), where each element is between 0 and 1. Then, the original feature map X is reweighted by multiplying each channel by the weight corresponding to the time step. The weighted feature map is then subjected to global average pooling (simultaneously compressing the temporal and spatial dimensions) to obtain a vector of length 128. This vector is the final extracted cloud dynamic feature vector, which encodes the key patterns of cloud movement affecting sunlight occlusion over the next 24 hours.

[0078] In this way, the attention mechanism increases the model's prediction weight for periods of rapid cloud movement and reduces the weight for nighttime, thereby reducing power generation prediction errors.

[0079] The cloud dynamic feature vector, historical power generation data, and weather type encoding are input into the LSTM-Transformer hybrid model. The cloud dynamic feature vector has 128 dimensions; the historical power generation data includes 15-minute power generation sequences over the past 24 hours, totaling 96 time points, normalized to the [0,1] interval; the weather type encoding uses one-hot encoding of weather forecast types (such as sunny, partly cloudy, cloudy, overcast, rainy, thunderstorms, etc.) to obtain a 6-dimensional vector. Additionally, temperature, humidity, and wind speed can be added, resulting in a total of 9 dimensions.

[0080] In the LSTM-Transformer hybrid model, the LSTM layer inputs a sequence of historical power generation data from 96 time points into a two-layer LSTM network, with 128 hidden units per layer. The LSTM captures the long-term and short-term dependencies in the power generation data, such as whether there are similar patterns in the same time period over the past few days. It outputs an intermediate hidden state sequence of dimensions (96, 128). Subsequently, the hidden states (128-dimensional) of each time step output by the LSTM are concatenated with cloud dynamic features (128-dimensional) and the weather code of the current time step (9-dimensional) to obtain a feature matrix of dimensions (96, 265). This matrix integrates historical trends, cloud dynamics, and weather conditions.

[0081] After receiving the feature matrix, the Transformer encoder performs temporal modeling with a feedforward network through a four-layer multi-head self-attention mechanism (4 heads), with each layer's output maintaining 265 dimensions. The self-attention mechanism enables the model to associate dependencies between different time steps, such as the impact of the previous hour's power generation data on the next hour. The Transformer encoder outputs a feature sequence of size (96, 256). Subsequently, this feature sequence is input into a fully connected layer (input 256, output 1) to obtain power generation predictions for 96 time points (15 minutes for the next 24 hours). If it is necessary to predict the next 72 hours, the predicted 24-hour data can be used as historical input to generate the subsequent 48 hours recursively.

[0082] The model was trained using historical data from the past 90 days (actual photovoltaic power, corresponding satellite cloud images, and weather forecasts). The loss function was mean absolute error plus L2 regularization. The optimizer was Adam with a learning rate of 0.001 and a batch size of 32. Training was performed for approximately 200 epochs using either TensorFlow or PyTorch.

[0083] Thus, compared to the standalone LSTM model, this embodiment reduces the average prediction error over 72 hours by using the LSTM-Transformer hybrid model; compared to the standalone Transformer model, the LSTM preprocessing enables the model to converge stably even with a small amount of data.

[0084] As an alternative implementation method, refer to Figure 2 In step S200, the strategic layer scheduling execution of long-term energy balancing planning on a daily cycle includes steps S211 to S213:

[0085] Step S211: Based on the solar power generation prediction results and the baseline of household electricity consumption patterns, generate a power generation-consumption-energy storage balance strategy;

[0086] Step S212: Based on dynamic electricity price information, plan a charging and discharging strategy that supplements grid power storage during off-peak hours and prioritizes energy storage discharge during peak hours.

[0087] Step S213: When performing energy storage discharge operation, reserve energy storage redundancy according to the preset emergency capacity.

[0088] In this embodiment, a day is divided into three periods: daytime (7:00–18:00), nighttime (18:00–23:00), and late night (23:00–7:00). The total power generation and total power consumption for each period are calculated. If the total daytime power generation is greater than the sum of the total daytime power consumption and the rechargeable energy storage capacity (currently available spare capacity), a power generation-consumption-storage balance strategy is generated to prioritize charging the battery during the day. If the total daytime power generation is less than the total daytime power consumption, a power generation-consumption-storage balance strategy is generated to supply power through both photovoltaics and batteries during the day, with any shortfall being supplied by purchasing electricity from the grid; if the battery SOC is below 40%, electricity is purchased from the grid for charging. During the nighttime and late night periods, if the battery SOC is greater than the emergency capacity (e.g., 20%), a power generation-consumption-storage balance strategy is generated to prioritize discharging the battery for power supply; otherwise, electricity is purchased from the grid for power supply. Generate recommended charge and discharge power values ​​for 24 hours, for example, 2kW for charging from 8:00 to 12:00 and 1kW for discharging from 12:00 to 16:00. Store the recommended values ​​in tabular form for tactical reference.

[0089] Electricity price information can be obtained through a user-configured preset peak-valley timetable to determine the current time period. For example, off-peak electricity: 22:00-06:00; peak electricity: 09:00-12:00, 17:00-21:00; flat electricity: the rest of the time. For example, during off-peak hours, a mandatory charging / discharging strategy is adopted. Regardless of the current photovoltaic power generation status, as long as the battery SOC is below 80%, electricity is purchased from the grid at the maximum allowable power for charging, while battery discharge is prohibited. During peak hours, a mandatory charging / discharging strategy is adopted, prioritizing battery discharge to supply the load. The battery SOC is allowed to discharge to the emergency capacity lower limit (e.g., 15%). Electricity is only purchased from the grid when the battery discharge power is insufficient and photovoltaic power is also lacking, and the purchased power is limited to no more than 2kW, i.e., prioritizing energy storage discharge. During flat electricity hours, a flexible charging / discharging strategy is adopted, automatically determining charging and discharging based on real-time power generation and load conditions, but prioritizing maintaining the battery SOC between 40% and 60% to prepare for peak hours. In this way, by enforcing time-based rules, the system can respond to peak-valley electricity pricing and reduce electricity costs.

[0090] Optionally, in tactical-level scheduling, before each discharge decision, it is confirmed that if the decrease in battery power corresponding to the SOC minus the discharge power exceeds the preset emergency capacity, then discharge is allowed; otherwise, discharge is prohibited, and a forced switch to grid power purchase is initiated. In this way, by sacrificing 15% of available capacity, it ensures that households can still maintain critical load operation using the reserved battery power during grid outages. Furthermore, to prevent the battery from over-discharging and entering a dormant state, a mandatory discharge lower limit of 10% is set; when the SOC is less than or equal to 5%, the discharge function is locked.

[0091] In an optional implementation, the step of executing short-term planning for charging / discharging power and load control on an hourly cycle in the tactical layer scheduling includes executing the tactical layer scheduling based on a reinforcement learning scheduling engine; wherein the state space of the reinforcement learning scheduling engine includes real-time generation power, battery state of charge, grid purchase price, and household load priority, and the action space of the reinforcement learning scheduling engine includes battery charging / discharging power adjustment, interruptible load control, and grid interaction power limiting. Optionally, the reward function of the reinforcement learning scheduling engine is a multi-objective weighted sum form: R = w1·R_generation + w2·R_energy storage + w3·R_grid + w4·R_user.

[0092] In this embodiment, for a state space containing real-time generation power, SOC, electricity price, and load priority, four types of data are collected every 15 minutes, normalized, and input into the neural network. This enables the scheduling engine to perceive the overall environment and provide an information basis for optimal decision-making. The reward function adopts a multi-objective weighted sum of generation, energy storage, grid, and user objectives. Four sub-rewards are designed and their weights are optimized, transforming the multi-objective optimization problem into a single scalar reward, making reinforcement learning solvable, and the relative importance of each objective is configurable. Compared with rule-based scheduling, the reinforcement learning scheduling engine has stronger adaptive capabilities, eliminates the need for manually designing complex rule tables, and can be retrained according to environmental changes (such as seasons and electricity price policies).

[0093] Optionally, the load priority follows a user-preset three-tier priority system. High-priority loads, including refrigerators, lighting, and communication equipment, require continuous and uninterrupted power supply. Medium-priority loads, such as air conditioners, water heaters, and washing machines, can be delayed or briefly interrupted. Low-priority loads, such as electric vehicle charging and entertainment equipment, can be flexibly interrupted or have their power adjusted. Understandably, the state space of the reinforcement learning scheduling engine is a multi-dimensional vector. Four variables—generation power, battery state of charge, grid electricity price, and household load priority—are concatenated into a 6-dimensional vector. This state vector is collected every 15 minutes, synchronized with the tactical layer scheduling cycle, and serves as the input to the reinforcement learning model.

[0094] For example, the action space can include two continuous actions, such as battery charging / discharging power and grid interaction power limiting; and three discrete actions, such as air conditioner operating mode switching, water heater start / stop control, and electric vehicle charging switch. The reinforcement learning model outputs a set of action values ​​every 15 minutes and sends them to the execution layer. The execution layer uses a PID controller to track the actual power to the target value and executes load switching.

[0095] In the reward function, the generation efficiency reward R_generation encourages the priority use of solar power to reduce electricity purchases from the grid. The lower the proportion of electricity purchased, the higher the reward. Its value ranges from 0 to 1, with 1 for complete self-sufficiency and 0 for complete grid dependence. The energy storage health reward R_energy storage helps avoid battery overcharging, over-discharging, and rapid charging and discharging, extending battery life. The grid stability reward R_grid helps reduce sudden changes in grid power interaction, smoothing the load curve. Its value ranges from -1 to 0, with 0 for no change and -1 for the largest sudden change. The user comfort reward R_user prioritizes power supply to high-priority loads. Its value ranges from 0 to 1, approaching 1 when all high-priority loads are powered and 0 when all are interrupted. In scenarios emphasizing generation efficiency and energy storage lifespan, weights w1=0.4, w2=0.3, w3=0.2, and w4=0.1 can be obtained through Bayesian optimization offline tuning.

[0096] Thus, by setting a reward function, for each action performed (charging / discharging, load switching, limiting grid power), the environment calculates and weights the four sub-rewards based on the new state. The goal is to maximize the cumulative discounted reward, thereby learning to achieve a balance among generation, energy storage, the grid, and users.

[0097] Reference Figure 3 Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description and will not be repeated hereafter. Based on this, the energy dispatching method of the household solar power supply system may further include steps S410 to S440:

[0098] Step S410: When a power generation fluctuation event is detected, record the fluctuation data;

[0099] Step S420: Establish a grayscale prediction model based on the fluctuation data;

[0100] Step S430: Transmit the cloud movement speed of the meteorological data to the grayscale prediction model to obtain the fluctuation compensation coefficient;

[0101] Step S440: Perform a correction operation on the solar power generation prediction result based on the fluctuation compensation coefficient.

[0102] In this embodiment, the actual power generation of the photovoltaic array, P_actual(t), is monitored in real time. The sampling period can be 1 second, and the average value is taken over 1 minute. The power standard deviation σ_5min over the past 5 minutes is calculated. If σ_5min > 0.15 × P_nominal and the duration exceeds 3 minutes, it is considered a power generation fluctuation event. P_nominal is the rated power of the photovoltaic array (e.g., 5kW). Understandably, in cloudy weather, the repeated obstruction of sunlight by cloud edges causes rapid power fluctuations on a second to minute scale, resulting in a significantly increased standard deviation. A 15% threshold can exclude minor fluctuations under normal weather conditions, such as changes in irradiance caused by wind speed.

[0103] Fluctuation data includes, but is not limited to, the event start time t0; the actual power generation P_actual[t] every 15 minutes during the event (t = t0, t0+15min, t0+30min, …); and the baseline clear-day predicted power P_clear[t] at the corresponding time. The fluctuation compensation coefficient k[t] = P_actual[t] / P_clear[t]. k[t] reflects the degree of attenuation of the current actual power generation relative to clear-day conditions. For example, when the cloud cover is completely obscured, k[t] is close to 0; when there are no clouds, k[t] is close to 1; when the cloud edge enhancement effect causes the power to briefly exceed the clear-day value, k may be greater than 1. In addition, the system caches the k value sequence of the most recent 4 time points: (k_1, k_2, k_3, k_4), corresponding to the data of the past hour.

[0104] Optionally, a gray-scale prediction model based on GM(1,1) (a first-order univariate gray model) is adopted. Specifically, in the gray-scale prediction model, the original sequence k^{(0)} = (k_1, k_2, k_3, k_4) is first constructed from the four most recent fluctuation compensation coefficients, for example (0.62, 0.55, 0.48, 0.43) (indicating continuous cloud cover and a gradual decrease in power generation). Subsequently, the cumulative sequence k^{(1)} is calculated, where k^{(1)}_i = Σ_{j=1}^{i} k^{(0)}_j, for example (0.62, 1.17, 1.65, 2.08). Next, the GM(1,1) model assumes that the cumulative sequence satisfies the whitening differential equation, dk^{(1)} / dt + a·k^{(1)} = b, where a is the development coefficient and b is the gray action. Construct a matrix B and a vector Y, where each row of B is [-0.5(k^{(1)}_i + k^{(1)}_{i-1}), 1], and each element of Y is k^{(0)}_i (i=2,3,4). Solve for [a, b]^T = (B^TB)^{-1} B^TY. Then, using the solution of the differential equation k^{(1)}_{pred}(t+1) = (k^{(0)}_1 - b / a)·e^{-a·t} + b / a, and taking t=4 (next step), predict the accumulated value at the next time step. Finally, calculate the predicted value of the compensation coefficient k_pred = k^{(1)}_{pred}(5) - k^{(1)}_4.

[0105] For example, assuming the original sequence k^{(0)} = (0.62, 0.55, 0.48, 0.43), a = 0.15, b = 0.68, then k_pred = 0.38. This means that if the current compensation coefficient is 0.43, it is predicted to further decrease to 0.38 after 15 minutes, indicating that the cloud cover will continue to thicken.

[0106] It should be noted that the cloud movement speed can be obtained by simultaneously extracting the optical flow field of adjacent cloud images during the above-mentioned 3D convolutional neural network processing. The average moving speed vector v = (v_x, v_y) of the clouds above the target area is calculated, with the unit being pixels / frame. The actual speed is then obtained based on the conversion relationship between image resolution and actual geographical distance. This calculation is performed every 15 minutes.

[0107] Cloud movement speed is used as an auxiliary parameter to adjust the output of the grayscale prediction model. Specifically, the development coefficient is dynamically adjusted in GM(1,1). Since the faster the cloud movement speed, the faster the rate of change in power generation fluctuations, i.e., the larger the absolute value of 'a', an empirical formula is established: a = a_base + α·v, where v is the cloud speed and α is the fitting coefficient. Thus, when the cloud moves rapidly, the predicted compensation coefficient changes more drastically, better reflecting reality. In this way, by introducing cloud movement speed, the prediction accuracy of the grayscale prediction model is further improved. Whether under rapidly moving fractal cumulus clouds or slowly moving stratus clouds, the prediction error can be reduced through correction.

[0108] The gray-scale prediction model output k_pred is weighted and fused with the compensation coefficient k_current at the current time to prevent abrupt changes. This can be expressed by the formula: k_final = β·k_pred + (1-β)·k_current, where β is the smoothing coefficient.

[0109] The corrected power generation forecast is P_corrected(t) = P_original(t) × k_final, where P_original(t) is the original solar power generation forecast generated based on historical and meteorological data. The corrected value replaces the original forecast and is input into the three-layer scheduling architecture to generate better scheduling instructions.

[0110] For example, suppose the original prediction model outputs a power generation of 2.5 kW for the next 15 minutes. The current fluctuation compensation coefficient k_current = 0.48, indicating that the actual power generation is only 48% of that on a sunny day. The gray-scale prediction model outputs k_pred = 0.38, further decreasing the prediction. Taking β = 0.7, then k_final = 0.7 × 0.38 + 0.3 × 0.48 = 0.41. The corrected predicted power is 2.5 × 0.41 = 1.025 kW. Subsequently, based on this lower predicted value, the battery discharge power can be increased in advance or interruptible loads can be reduced to avoid purchasing electricity from the grid due to insufficient power generation later.

[0111] In the technical solution provided in this embodiment, by detecting power generation fluctuation events, weather conditions requiring correction, such as cloudy or rainy weather, are identified, avoiding unnecessary corrections under normal weather conditions. By transmitting cloud movement speed, the prediction model can respond to the cloud movement rate, further improving the correction accuracy. Compensation coefficients are obtained through a grayscale prediction model to correct the prediction, and the corrected prediction values ​​are integrated into the three-layer dispatching system, improving dispatching decisions under cloudy weather conditions and reducing grid power purchases and battery over-discharge.

[0112] This application provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the energy dispatching method of the home solar power supply system in the above embodiments.

[0113] The following is for reference. Figure 4 The diagram illustrates a structural schematic of an electronic device suitable for implementing embodiments of this application. The electronic devices in these embodiments may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 4 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0114] like Figure 4 As shown, the electronic device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the electronic device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. The communication device 1009 allows the electronic device to communicate wirelessly or wiredly with other devices to exchange data. Although the diagrams show electronic devices with various systems, it should be understood that it is not required to implement or have all of the systems shown. More or fewer systems may be implemented alternatively.

[0115] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0116] The electronic device provided in this application employs the energy dispatching method for a household solar power supply system as described in the above embodiments, which can solve the technical problems of poor system operation stability and high equipment lifespan loss caused by a single dispatching strategy and insufficient foresight. Compared with the prior art, the beneficial effects of the electronic device provided in this application are the same as those of the energy dispatching method for a household solar power supply system provided in the above embodiments, and other technical features of the electronic device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0117] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0118] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0119] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the energy dispatching method of the home solar power supply system in the above embodiments.

[0120] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0121] The aforementioned computer-readable storage medium may be included in an electronic device or may exist independently without being assembled into an electronic device.

[0122] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by an electronic device, the electronic device causes the following: to acquire meteorological data and historical power generation data; to generate a solar power generation forecast for a future preset time period based on a multimodal data fusion prediction model; to execute strategic-level scheduling, tactical-level scheduling, and execution-level scheduling based on the solar power generation forecast; to generate scheduling instructions, wherein the strategic-level scheduling performs long-term energy balance planning on a daily cycle, the tactical-level scheduling performs short-term planning for charging and discharging power and load control on an hourly cycle, and the execution-level scheduling performs real-time power adjustment on a minute cycle; and to issue the scheduling instructions to inverters, energy storage systems, and controllable loads to control the flow and power of energy.

[0123] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0124] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0125] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0126] The readable storage medium provided in this application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., a computer program) for executing the energy dispatching method of the above-described household solar power supply system. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the energy dispatching method of the household solar power supply system provided in the above embodiments, and will not be repeated here.

[0127] This application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the energy dispatching method for a residential solar power system as described above. Compared to the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the energy dispatching method for a residential solar power system provided in the above embodiments, and will not be repeated here.

[0128] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent scope of this application.

Claims

1. A method for energy scheduling of a domestic solar powered system, characterized in that, The energy dispatching method for the household solar power system includes: Acquire meteorological data and historical power generation data, and generate solar power generation forecasts for future preset time periods based on a multimodal data fusion prediction model; Based on the solar power generation forecast results, strategic-level scheduling, tactical-level scheduling, and execution-level scheduling are executed to generate scheduling instructions. The strategic-level scheduling executes long-term energy balance planning on a daily cycle, the tactical-level scheduling executes short-term planning for charging and discharging power and load control on an hourly cycle, and the execution-level scheduling executes real-time power adjustment on a minute cycle. The dispatching instructions are sent to the inverter, energy storage system and controllable load to control the flow and power of energy.

2. The energy dispatching method of the home solar power system according to claim 1, wherein, The meteorological data includes time-series cloud images; the steps of acquiring meteorological data and historical power generation data, and generating solar power generation prediction results for a future preset period based on a multimodal data fusion prediction model, include: The time-series cloud images are processed using a three-dimensional convolutional neural network to extract dynamic features of the clouds. Based on the LSTM-Transformer hybrid model, combined with the historical power generation data, weather type coding, and cloud dynamic characteristics, the power generation prediction value is output for each time point.

3. The energy dispatching method of the home solar power system according to claim 2, wherein, The time-series cloud image includes multiple consecutive frames of satellite cloud images. The step of processing the time-series cloud image based on a three-dimensional convolutional neural network to extract dynamic features of the cloud layer includes: Preprocessing operations are performed on the continuous multi-frame satellite cloud images to construct a time-series image dataset; In the aforementioned time-series image dataset, spatial shape features and temporal movement features of clouds are extracted simultaneously using 3D convolution kernels. Based on the attention mechanism, the feature weights of the spatial shape features and the movement features are dynamically adjusted in the temporal dimension at different time steps to output a dynamic feature vector of the cloud layer.

4. The energy dispatching method of claim 1, wherein, The steps involved in the strategic-level scheduling execution of long-term energy balance planning on a daily cycle include: Based on the solar power generation prediction results and the baseline of household electricity consumption patterns, a power generation-electricity consumption-energy storage balance strategy is generated. Based on dynamic electricity price information, a charging and discharging strategy is planned to supplement grid power storage during off-peak hours and prioritize the use of energy storage discharge during peak hours. When performing energy storage discharge operations, reserve energy storage redundancy according to the preset emergency capacity.

5. The method of claim 1, wherein, The steps of the tactical layer scheduling execution for short-term planning of charging / discharging power and load control on an hourly basis include: The tactical layer scheduling is executed based on a reinforcement learning scheduling engine. The state space of the reinforcement learning scheduling engine includes real-time power generation, battery state of charge, grid purchase price, and household load priority. The action space of the reinforcement learning scheduling engine includes battery charging and discharging power adjustment, interruptible load control, and grid interaction power limitation.

6. The energy dispatching method of the home solar power system according to claim 5, wherein, The reward function of the reinforcement learning scheduling engine is a multi-objective weighted sum: R = w1·R_power generation + w2·R_energy storage + w3·R_grid + w4·R_user.

7. The energy dispatching method for a household solar power supply system as described in claim 1, characterized in that, The energy dispatching method for the household solar power system also includes: When a power generation fluctuation event is detected, the fluctuation data is recorded; A grayscale prediction model is established based on the fluctuation data; The cloud movement speed of the meteorological data is transmitted to the grayscale prediction model to obtain the fluctuation compensation coefficient; Based on the fluctuation compensation coefficient, a correction operation is performed on the solar power generation prediction results.

8. An electronic device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the energy dispatching method for a residential solar power system as described in any one of claims 1 to 7.

9. A readable storage medium, characterized in that, The readable storage medium is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the energy dispatching method for a household solar power supply system as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the energy dispatching method for a residential solar power system as described in any one of claims 1 to 7.

Images