AI intelligent photovoltaic energy storage mobilization device

Abstract

The application provides an AI intelligent photovoltaic energy storage mobilization device, and relates to the technical field of automatic control of new energy power systems, and has the characteristics that the device comprises a photovoltaic power generation unit, an energy storage battery unit, a multi-dimensional data acquisition module, an ultra-short-term electricity price prediction module, an AI adaptive decision engine and a safe execution control module; the multi-dimensional data acquisition module acquires system operation data in real time at a sampling frequency of no higher than 1 minute; the ultra-short-term electricity price prediction module is used to obtain a predicted electricity price sequence within a preset time window after the current time; and the AI adaptive decision engine is in communication connection with the multi-dimensional data acquisition module and the ultra-short-term electricity price prediction module respectively. The application has the advantages that through the cooperation of the edge side AI adaptive decision engine and the ultra-short-term electricity price prediction module, 15-minute granularity rolling optimization scheduling is realized, the maximum full life cycle benefit is realized, the prediction accuracy is high, and the decision reliability is strong.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology for new energy power systems, and in particular to an AI intelligent photovoltaic energy storage dispatching device. Background Technology

[0002] With the large-scale grid connection of photovoltaic power generation and the rapid development of energy storage technology, photovoltaic-storage integrated systems have become an important form of distributed energy. Against the backdrop of the gradual advancement of the electricity spot market, electricity prices exhibit high-frequency fluctuations. How to intelligently schedule energy storage charging and discharging and photovoltaic output based on real-time electricity price signals to maximize system revenue is the core challenge currently facing photovoltaic-storage systems.

[0003] Traditional photovoltaic energy storage systems often employ fixed peak-valley arbitrage strategies or simple rule-based control, making them ill-suited to adapt to drastic electricity price fluctuations of 15 minutes or even shorter periods. Their coarse-grained decision-making fails to capture short-term electricity price spikes or negative price opportunities; they neglect the dynamic balance between battery lifespan degradation and economic returns, leading to battery losses; they lack robust handling of forecast uncertainties, making prediction bias prone to malfunctions; and the lack of a decoupled design for a safe execution layer means that AI commands directly driving the equipment pose safety risks.

[0004] Therefore, there is an urgent need for an intelligent dispatching device that can accurately respond to 15-minute electricity price fluctuations while balancing economy and security. Summary of the Invention

[0005] The purpose of this invention is to provide an AI-powered intelligent photovoltaic energy storage scheduling device that solves the problems of insufficient decision granularity, uncontrollable lifespan loss, and lack of safety redundancy in the existing technology. Through the collaboration of the edge-side AI adaptive decision engine and the ultra-short-term electricity price prediction module, it achieves rolling optimization scheduling with a 15-minute granularity.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0007] An AI-powered intelligent photovoltaic energy storage and control device includes a photovoltaic power generation unit, an energy storage battery unit, a multi-dimensional data acquisition module, an ultra-short-term electricity price prediction module, an AI adaptive decision engine, and a safety execution control module.

[0008] The photovoltaic power generation unit is used to convert solar energy into electrical energy. The energy storage battery unit is connected to the photovoltaic power generation unit and the power grid through a bidirectional converter, and is used to store or release electrical energy.

[0009] The multidimensional data acquisition module collects system operation data in real time at a sampling frequency of no more than 1 minute. The system operation data includes real-time photovoltaic power generation, energy storage battery state of charge (SOH / SOC), real-time load power consumption, real-time grid-connected node electricity price, and circuit breaker status signals.

[0010] The ultra-short-term electricity price prediction module is used to obtain the predicted electricity price sequence within a preset time window after the current moment. The predicted electricity price sequence consists of multiple consecutive time segments, each lasting 15 minutes, and each time segment corresponds to a predicted average electricity price. Preferably, the ultra-short-term electricity price prediction module includes a multi-source data interface unit, a feature engineering unit, and a hybrid prediction model unit. The multi-source data interface unit is used to obtain spot market clearing prices, time-of-use pricing policies, and ancillary service market prices in real time from the power trading center API, the power grid dispatch data network, or a third-party aggregator platform. The feature engineering unit is used to extract historical electricity price time-series features, meteorological correlation features, and load supply-demand ratio features. The hybrid prediction model unit adopts a weighted fusion architecture of LSTM neural network and XGBoost regression model, outputting electricity price prediction curves with a 15-minute resolution for the next 4 to 24 hours. The error feedback correction unit is used to compare the predicted value of the previous period with the actual value and update the model weight parameters in real time.

[0011] The AI ​​adaptive decision engine is communicatively connected to the multi-dimensional data acquisition module and the ultra-short-term electricity price prediction module, and is equipped with an edge computing processor to receive the system operation data and the predicted electricity price sequence. With a scheduling cycle of 15 minutes, and with the goal of maximizing the net revenue of the system's entire life cycle, it combines the energy storage battery life loss model to generate the optimal charging and discharging power command and photovoltaic output adjustment command for the current scheduling cycle.

[0012] The objective function of the AI ​​adaptive decision engine is:

[0013] in, and These represent the discharge and charging power during time period t, respectively. and For discharge and charge efficiency, Forecast electricity prices for time period t. For direct photovoltaic load power supply, Electricity price as a substitute for self-use electricity The cost of energy storage battery lifespan degradation based on rainflow counting method, The time limit is 15 minutes. Constraints include: upper and lower limits and rate of change constraints of energy storage SOC, charging and discharging power ramp-up constraints, photovoltaic inverter capacity constraints, grid connection point power factor constraints, and reverse current protection thresholds.

[0014] Preferably, the AI ​​adaptive decision engine further includes a strategy interpretation and confidence assessment unit, used to perform sensitivity analysis on the generated scheduling instructions; when the confidence of the predicted electricity price is lower than a preset threshold, it automatically switches to a conservative benchmark strategy, which only responds to the determined peak-valley price difference signal and shields the spot price fluctuation signal with higher uncertainty.

[0015] The safety execution control module is communicatively connected to the AI ​​adaptive decision engine, the bidirectional converter, and the photovoltaic inverter. It performs electrical safety checks before executing the optimal charge / discharge power command and the photovoltaic output adjustment command, and controls the charge / discharge state of the energy storage battery unit and the active power output of the photovoltaic power generation unit after the checks are passed. Preferably, the safety execution control module includes an instruction buffer queue, a hard limit protection unit, and a smooth transition unit. The instruction buffer queue temporarily stores scheduling commands issued by the AI ​​decision engine to prevent command jitter. The hard limit protection unit operates independently of the AI ​​decision engine; when it detects that the energy storage current or voltage exceeds the physical safety threshold, it forcibly takes over control and performs an emergency shutdown or power limiting operation. The smooth transition unit generates a power ramp curve during the instruction switching process between two adjacent 15-minute scheduling cycles to avoid sudden changes in the output power of the energy storage converter impacting the power grid.

[0016] The AI ​​adaptive decision engine triggers rolling optimization calculations and dynamically corrects the scheduling instructions for the next scheduling cycle before the end of each 15-minute scheduling cycle, based on the latest collected real-time electricity price and the updated predicted electricity price sequence.

[0017] Preferably, the device further includes a cloud-based collaborative training platform, whereby the edge computing processor periodically uploads anonymized operating data and decision effect labels to the cloud. The cloud uses massive samples to retrain the global model and sends the optimized model parameter increments to the edge side for online learning and updates.

[0018] Preferably, the device further includes a human-computer interaction terminal for visually displaying the current 15-minute band electricity price trend, the expected return curve of energy storage, and the AI ​​decision-making logic; the human-computer interaction terminal is also equipped with a manual intervention interface, and when the user inputs a forced charging and discharging command, the AI ​​adaptive decision engine incorporates the user command as a hard constraint into the next round of optimization.

[0019] Preferably, the multi-dimensional data acquisition module further includes a power quality monitoring submodule, used to collect harmonic distortion rate and voltage sag events; when power quality is abnormal, the AI ​​adaptive decision engine automatically adjusts the weight of the power quality governance item in the objective function, prioritizing power supply reliability rather than economy.

[0020] This invention also provides an AI-powered intelligent photovoltaic energy storage dispatching method, comprising the following steps:

[0021] S1. Collect the real-time operating status of the system and the actual settlement price of the current 15-minute band at minute-level frequency;

[0022] S2. Based on multi-source data fusion, predict the electricity price sequence for multiple 15-minute bands in the future and calculate the prediction confidence level;

[0023] S3. Construct a comprehensive objective function that includes electricity price revenue, self-consumption savings revenue, and battery life loss, and solve for the optimal charging and discharging power and photovoltaic output setting value for the current 15-minute band under multiple electrical constraints.

[0024] S4. Perform security verification and smoothing on the solution results, and generate executable control instructions;

[0025] S5. Send instructions to the energy storage converter and photovoltaic inverter to execute the scheduling for this cycle;

[0026] S6. Five minutes before the end of this cycle, trigger rolling optimization using the latest measured data to generate instructions for the next cycle, achieving closed-loop adaptive control with a 15-minute granularity.

[0027] Preferably, the response strategy for the 15-minute band electricity price in step S3 specifically includes: when the predicted electricity price is on an upward trend and the current SOC has not reached the upper limit, starting charging in advance or reducing the discharge power to lock in the low-price charging window; when the predicted electricity price is on a downward trend and the current SOC has not reached the lower limit, delaying discharge or starting low-power floating charging to avoid high-price discharge losses; when photovoltaic output is excessive and the current 15-minute band electricity price is negative or lower than the energy storage cycle cost, actively limiting photovoltaic output or activating curtailment protection to avoid negative revenue operation; when the deviation between the real-time electricity price and the predicted electricity price exceeds a threshold, immediately triggering intraday rescheduling to correct the charging and discharging plan for the remaining period.

[0028] The beneficial effects of this invention are as follows:

[0029] 1. A high-resolution electricity price sequence for the next 424 hours is obtained through an ultra-short-term electricity price forecasting module, and then optimized in 15-minute cycles using an AI adaptive decision engine. This captures spot market price fluctuations, locks in electricity in advance during low-price windows, releases electricity precisely during high-price windows, and proactively avoids losses during periods of negative electricity prices, thereby improving net returns throughout the entire lifecycle.

[0030] 2. The objective function of the AI ​​decision engine explicitly integrates the energy storage battery life depreciation cost based on the rainflow counting method. When generating charge and discharge commands, the system automatically weighs current electricity revenue against long-term battery depreciation, avoiding excessive overdraft of battery life for short-term profits and improving the full life-cycle value of energy storage assets.

[0031] 3. Set up a decoupled architecture between the AI ​​decision-making layer and the safety execution layer, and ensure that the hard limit protection operates independently to ensure equipment safety under extreme operating conditions.

[0032] 4. Through cloud collaboration and error feedback mechanisms, the model continuously iterates to adapt to the electricity price characteristics of different regions and seasons.

[0033] 5. Supports seamless switching between manual intervention and automatic optimization to meet emergency dispatch needs in special scenarios. Attached Figure Description

[0034] Figure 1 This is an overall architecture diagram of an AI intelligent photovoltaic energy storage and regulation device according to the present invention.

[0035] Figure 2 This is a schematic diagram of the internal data flow of an AI adaptive decision engine.

[0036] Figure 3 Optimize the timing logic diagram for a 15-minute rolling timer.

[0037] Figure 4 Workflow diagram for the safe execution control module. Detailed Implementation

[0038] To make the content of this invention easier to understand, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Identical components are indicated by the same reference numerals.

[0039] Example 1: Application of an AI-powered intelligent photovoltaic energy storage and dispatching device in an industrial park microgrid scenario

[0040] An industrial park is equipped with a 2MW photovoltaic power generation system and a 1MW / 2MWh lithium iron phosphate energy storage system. The park's annual electricity consumption is approximately 8 million kWh, with a peak load of approximately 1.5MW. The park is connected to a 10kV distribution network and participates in provincial electricity spot market transactions.

[0041] Deploying the device described in this invention in the power distribution room of the industrial park specifically includes:

[0042] Photovoltaic power generation unit: consisting of a 2MW rooftop photovoltaic array and photovoltaic inverter, which has been built and put into operation.

[0043] Energy storage battery unit: It consists of a 1MW / 2MWh lithium iron phosphate battery cluster, a bidirectional converter (PCS) and a battery management system (BMS), and is connected to the photovoltaic system and the park's 380V bus via the AC side.

[0044] Multi-dimensional data acquisition module: Deploy data acquisition terminals and connect to photovoltaic inverters, energy storage BMS / PCS, park smart meters and grid connection point monitoring devices through communication protocols such as RS485 and ModbusTCP. Collect real-time photovoltaic power, energy storage SOC / SOH, load power, real-time electricity price of grid connection nodes and circuit breaker status at a sampling frequency of 1 minute.

[0045] The ultra-short-term electricity price forecasting module deploys an edge computing server to obtain real-time spot market clearing prices, time-of-use pricing policies, and ancillary service market prices through the power trading center's API interface. The feature engineering unit extracts historical electricity price time-series features (past 72-hour price series), meteorological correlation features (24-hour irradiance forecast, temperature forecast), and load supply-demand ratio features. The hybrid forecasting model unit employs a weighted fusion architecture of an LSTM neural network and an XGBoost regression model, outputting electricity price forecast curves with a 15-minute resolution for the next 24 hours. The error feedback correction unit compares the predicted and actual values ​​from the previous period every 15 minutes, updating the model weight parameters in real-time.

[0046] AI Adaptive Decision Engine: Deployed on an edge computing server, configured with an Intel Xeon processor and GPU accelerator card. The engine receives system operation data from the multi-dimensional data acquisition module and predicted electricity price sequences from the ultra-short-term electricity price prediction module. It optimizes the system with a 15-minute scheduling cycle, using maximizing the net revenue over the entire system's lifecycle as the objective function. In the objective function... Taken from the predicted electricity price series, The system uses rainflow counting to calculate the charge-discharge cycle depth of energy storage batteries in real time and converts it into lifetime loss cost. Constraints include: energy storage SOC operating range of 20%~90%, SOC change rate not exceeding 5% per minute, charge-discharge power ramp-up rate not exceeding 20% / minute of rated power, photovoltaic inverter capacity not exceeding 2MW, grid connection point power factor not lower than 0.95, and reverse current protection threshold set at 20kW.

[0047] The strategy interpretation and confidence assessment unit performs sensitivity analysis on the generated dispatch instructions and automatically switches to a conservative benchmark strategy when the confidence level of the predicted electricity price is below 70%.

[0048] Safety Execution Control Module: Deploys an independent PLC controller, communicating with the AI ​​decision engine, PCS, and photovoltaic inverter. An instruction buffer queue temporarily stores the three most recent scheduling instructions. A hard limit protection unit independently detects the DC-side voltage of the energy storage (threshold set to ±10% of the battery's rated voltage) and the charging / discharging current (threshold set to 120% of the rated current), forcibly shutting down the system within 50 milliseconds if limits are exceeded. A smooth transition unit generates a power ramp curve when switching instructions between adjacent scheduling cycles, with a ramp time set to 30 seconds.

[0049] Cloud-based collaborative training platform: Deployed on the public cloud, edge computing servers upload anonymized running data and decision effect labels daily. The cloud retrains the global model monthly based on accumulated samples and distributes the optimized model parameters incrementally to the edge side.

[0050] Human-computer interaction terminal: Deployed in the park's central control room, equipped with a 55-inch touch screen, visually displaying the current 15-minute band electricity price trend, the expected return curve of energy storage, and the AI ​​decision-making logic. It also has a manual intervention button, supporting users to input forced charging and discharging commands.

[0051] Operation process:

[0052] Taking a typical day as an example, the working process of the device is explained in detail:

[0053] At 08:00, the multi-dimensional data acquisition module collected the current system status: sunny weather, continuously rising irradiance, photovoltaic output of 1.2MW, energy storage SOC of 65%, and load power of 0.8MW, while also displaying the real-time electricity price. The ultra-short-term electricity price prediction module outputs a 24-hour electricity price prediction curve, showing that the electricity price is rising from 10:00 to 14:00 and begins to fall after 14:00.

[0054] From 08:00 to 08:15, the AI ​​adaptive decision engine uses the current state as the initial condition and solves the problem within a 15-minute optimization window. Given that the electricity price is on an upward trend and the SOC is 65%, which is below the upper limit of 90%, the engine decides to reduce the discharge power to 0.1MW, instead of the originally planned 0.3MW, to reserve energy in advance to lock in the low-price charging window. After the safety execution control module passes the verification, the instruction is sent to the PCS for execution.

[0055] At 09:45, the weather suddenly changed, and the irradiance dropped sharply, causing the photovoltaic output to plummet from 1.5MW to 0.6MW. The multi-dimensional data acquisition module captured this change in real time with a sampling frequency of 1 minute. Five minutes before the end of the current 15-minute cycle, the engine triggered rolling optimization. Based on the latest photovoltaic output and the updated electricity price forecast, it recalculated the optimal strategy for the next cycle: increasing the discharge power from 0.2MW to 0.5MW to compensate for the insufficient photovoltaic output and ensure power supply to the load.

[0056] At 12:00, the electricity price reached its daily peak of 0.68 yuan / kWh, and the energy storage SOC had been charged to 85% under the previous strategy. The engine decided to discharge at a power of 0.8MW between 12:00 and 12:15 to achieve arbitrage during the high electricity price period.

[0057] At 14:30, the real-time electricity price plummeted due to the surge in wind power generation in neighboring provinces, with a deviation rate of 36% from the predicted value, exceeding the threshold. The confidence assessment unit determined that the prediction confidence was below 70%, and the engine automatically switched to a conservative benchmark strategy, suspending all charging and discharging operations based on spot price signals and only responding to the determined peak-valley price difference.

[0058] At 16:00, the photovoltaic output dropped to 0.3MW, and the start-up of the production line's night shift caused the load power to rise to 1.1MW. The engine decided to discharge at 0.7MW to prioritize power supply to the load and reduce electricity purchased from the grid.

[0059] After a full day of operation, the system tallied the daily economic benefits: photovoltaic power generation was approximately 15,000 kWh, energy storage underwent 2.3 charge-discharge cycles with an equivalent cycle depth of 65%, peak-valley arbitrage profits increased, electricity cost savings from self-consumption of photovoltaic power increased, and battery lifespan depreciation costs decreased, resulting in increased net profit for the day. This represents a significant improvement in returns compared to traditional fixed-period charge-discharge strategies.

[0060] Example 2: Application of an AI-powered intelligent photovoltaic energy storage dispatching device in residential photovoltaic energy storage scenarios

[0061] A residential villa is equipped with a 5kW photovoltaic power generation system and a 10kWh residential energy storage system. The household's average daily electricity consumption is about 25kWh, and the user is connected to the residential time-of-use electricity price.

[0062] The device configuration is similar to that of Embodiment 1, but simplified according to the characteristics of residential scenarios: the multi-dimensional data acquisition module communicates with the smart meter, photovoltaic inverter and energy storage BMS via WiFi; the ultra-short-term electricity price prediction module obtains the next day's time-of-use electricity price from the grid company's public API (no spot market prediction required); the AI ​​adaptive decision engine is deployed in the home gateway (low-power ARM processor); the safety execution control module is integrated inside the energy storage unit; and the human-machine interaction terminal is a mobile APP.

[0063] Typical daily operation:

[0064] The period from 00:00 to 06:00 is the off-peak electricity period. The engine predicts that the weather will be sunny the next day (with sufficient photovoltaic output). The decision is made to charge the energy storage from 20% SOC to 90% with a power of 3.3kW during the off-peak electricity period to lock in low-priced electricity.

[0065] From 8:00 AM to 4:00 PM is the peak solar power generation period, with solar power output prioritizing household loads and surplus electricity stored in energy storage. The system monitors the State of Charge (SOC) and load power in real time and dynamically adjusts the charging power. By 1:00 PM, the SOC has reached 95%, and there is still surplus solar power output. However, due to the low grid connection price for solar power for users, the economics are poor. Therefore, the system activates the "anti-reverse current" protection, actively limiting the output of the solar inverter to prevent backflow of power into the grid.

[0066] The peak electricity period is from 18:00 to 22:00. The energy storage discharges at a power of 3kW to cover the peak electricity consumption of families during dinner and at night, avoiding the high price of electricity purchased online.

[0067] Full-day economic calculation: Photovoltaic power generation is approximately 28 kWh, and energy storage for peak charging and discharging is approximately 8 kWh. Peak-valley arbitrage profits increase, and electricity cost savings from self-consumption of photovoltaic power increase. Daily electricity expenses are lower than without an energy storage system. Lifetime attrition costs are reduced, and the overall lifecycle economics are significantly better than a simple two-charge, two-discharge strategy without lifespan awareness.

[0068] Example 3: Application of an AI-powered intelligent photovoltaic energy storage dispatching device in a transformer substation energy storage scenario

[0069] A rural transformer substation is equipped with a 300kW photovoltaic system (multi-household distributed access), a 200kW / 400kWh energy storage system, and a 400kVA transformer, covering approximately 80 households. The main problems faced by the substation are: backfeeding during peak photovoltaic power generation at midday, causing transformer overload, and low voltage during peak evening load periods.

[0070] The device is deployed on the side of the distribution box in the transformer substation. The key configuration differences are as follows: the multi-dimensional data acquisition module additionally collects the three-phase voltage, current and power factor of the low-voltage side of the transformer in the transformer substation; in addition to electricity price prediction, the ultra-short-term electricity price prediction module also predicts the total photovoltaic output of the transformer substation based on meteorological data; the objective function of the AI ​​adaptive decision engine adds transformer overload penalty terms and voltage limit violation penalty terms; and the power quality monitoring submodule collects harmonic distortion rate and voltage sag events in real time.

[0071] Operational Results: After the device was put into operation, the number of midday overload events of the transformers in the distribution area decreased from an average of 2.3 times per day to 0.1 times, and the voltage qualification rate improved in the evening. The energy storage system averaged 2.1 charge-discharge cycles per day, and the average monthly peak-valley arbitrage profit increased. When a grid fault occurs, the safety execution control module quickly completes a seamless switch from grid connection to off-grid, ensuring uninterrupted power supply to important loads such as the health center and communication base stations in the distribution area.

[0072] Example 4: Application of an AI-powered intelligent photovoltaic energy storage dispatching device in a virtual power plant aggregation scenario

[0073] In this embodiment, multiple sets of the device described in this invention are deployed on different user sides (multiple industrial parks, commercial buildings, and charging stations), and uniformly scheduled through a cloud aggregation platform. In addition to performing local optimization, the AI ​​adaptive decision engine on each edge side also receives aggregated scheduling instructions (such as demand response events and auxiliary service calls) issued by the cloud virtual power plant platform.

[0074] When the power grid issues a demand response event, the cloud platform decomposes the response power target to each edge node. Each node's AI adaptive decision engine incorporates the response command as a new constraint into the objective function, optimizing the power and time periods for participation in the response while ensuring local power supply. After aggregation, users can obtain additional benefits by participating in ancillary power services.

[0075] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An AI-powered intelligent photovoltaic energy storage and dispatching device, characterized in that, It includes a photovoltaic power generation unit, an energy storage battery unit, a multi-dimensional data acquisition module, an ultra-short-term electricity price forecasting module, an AI adaptive decision engine, and a safety execution control module; The photovoltaic power generation unit is used to convert solar energy into electrical energy; The energy storage battery unit is connected to the photovoltaic power generation unit and the power grid via a bidirectional converter, and is used to store or release electrical energy. The multi-dimensional data acquisition module collects system operation data in real time at a sampling frequency of no more than 1 minute. The system operation data includes real-time photovoltaic power generation, energy storage battery state of charge (SOH / SOC), real-time load power consumption, real-time grid-connected node electricity price, and circuit breaker status signals. The ultra-short-term electricity price prediction module is used to obtain the predicted electricity price sequence within a preset time window after the current moment. The predicted electricity price sequence consists of multiple consecutive time segments, each time segment lasting 15 minutes, and each time segment corresponds to a predicted average electricity price. The AI ​​adaptive decision engine is communicatively connected to the multi-dimensional data acquisition module and the ultra-short-term electricity price prediction module, and is equipped with an edge computing processor to receive the system operation data and the predicted electricity price sequence. With a scheduling cycle of 15 minutes, and with the goal of maximizing the net revenue of the system's entire life cycle, it combines the energy storage battery life loss model to generate the optimal charging and discharging power command and photovoltaic output adjustment command for the current scheduling cycle. The safety execution control module is communicatively connected to the AI ​​adaptive decision engine, the bidirectional converter, and the photovoltaic inverter, respectively. It is used to perform electrical safety verification before executing the optimal charging and discharging power command and the photovoltaic output adjustment command, and to control the charging and discharging state of the energy storage battery unit and the active power output of the photovoltaic power generation unit after the verification is passed. The AI ​​adaptive decision engine triggers rolling optimization calculations and dynamically corrects the scheduling instructions for the next scheduling cycle before the end of each 15-minute scheduling cycle, based on the latest collected real-time electricity price and the updated predicted electricity price sequence.

2. The AI ​​intelligent photovoltaic energy storage dispatching device according to claim 1, characterized in that, The ultra-short-term electricity price prediction module includes a multi-source data interface unit, a feature engineering unit, and a hybrid prediction model unit. The multi-source data interface unit is used to obtain spot market clearing prices, time-of-use pricing policies and ancillary service market prices in real time from the power trading center API, the power grid dispatch data network or third-party aggregator platforms. The feature engineering unit is used to extract historical electricity price time-series features, meteorological correlation features, and load supply-demand ratio features; The hybrid prediction model unit adopts a weighted fusion architecture of LSTM neural network and XGBoost regression model, and outputs the electricity price prediction curve with a resolution of 15 minutes for the next 4 to 24 hours. The error feedback correction unit is used to compare the predicted value with the actual value in the previous period and update the model weight parameters in real time.

3. The AI ​​intelligent photovoltaic energy storage dispatching device according to claim 1, characterized in that, The objective function of the AI ​​adaptive decision engine is: in, and These represent the discharge and charging power during time period t, respectively. and For discharge and charge efficiency, Forecast electricity prices for time period t. For direct photovoltaic load power supply, Electricity price as a substitute for self-use electricity The cost of energy storage battery lifespan degradation based on rainflow counting method, It lasts for 15 minutes; The constraints include: upper and lower limits and rate of change of energy storage SOC, charging and discharging power ramp-up rate constraints, photovoltaic inverter capacity constraints, grid connection point power factor constraints, and anti-reverse current protection thresholds.

4. The AI ​​intelligent photovoltaic energy storage dispatching device according to claim 1, characterized in that, The AI ​​adaptive decision engine also includes a strategy interpretation and confidence assessment unit, which is used to perform sensitivity analysis on the generated scheduling instructions. When the confidence of the predicted electricity price is lower than a preset threshold, it automatically switches to a conservative benchmark strategy. The conservative benchmark strategy only responds to the determined peak-valley price difference signal and shields the spot price fluctuation signal with higher uncertainty.

5. The AI ​​intelligent photovoltaic energy storage dispatching device according to claim 1, characterized in that, The safe execution control module includes an instruction buffer queue, a hard limit protection unit, and a smooth transition unit; The instruction buffer queue is used to temporarily store scheduling instructions issued by the AI ​​decision engine to prevent instruction jitter. The hard limit protection unit operates independently of the AI ​​decision engine. When it detects that the energy storage current or voltage exceeds the physical safety threshold, it forcibly takes over the control and performs an emergency shutdown or power limiting operation. The smooth transition unit is used to generate a power ramp curve during the instruction switching process between two adjacent 15-minute scheduling cycles, so as to avoid the sudden change in the output power of the energy storage converter from impacting the power grid.

6. The AI ​​intelligent photovoltaic energy storage dispatching device according to claim 1, characterized in that, The device also includes a cloud-based collaborative training platform. The edge computing processor periodically uploads de-identified running data and decision effect labels to the cloud. The cloud uses massive samples to retrain the global model and sends the optimized model parameters incrementally to the edge side for online learning and updates.

7. The AI ​​intelligent photovoltaic energy storage dispatching device according to claim 1, characterized in that, The device also includes a human-computer interaction terminal, which is used to visualize the current 15-minute band electricity price trend, the expected return curve of energy storage, and the AI ​​decision-making logic. The human-computer interaction terminal is also equipped with a manual intervention interface. When the user inputs a forced charging and discharging command, the AI ​​adaptive decision engine will incorporate the user command as a hard constraint into the next round of optimization.

8. The AI ​​intelligent photovoltaic energy storage dispatching device according to claim 1, characterized in that, The multidimensional data acquisition module also includes a power quality monitoring submodule, which is used to collect harmonic distortion rate and voltage sag events. When the power quality is abnormal, the AI ​​adaptive decision engine automatically adjusts the weight of the power quality governance item in the objective function, prioritizing power supply reliability rather than economy.

9. An AI-powered intelligent photovoltaic energy storage dispatching method, implemented based on the device described in any one of claims 18, characterized in that, Includes the following steps: S1. Collect the real-time operating status of the system and the actual settlement price of the current 15-minute band at minute-level frequency; S2. Based on multi-source data fusion, predict the electricity price sequence for multiple 15-minute bands in the future and calculate the prediction confidence level; S3. Construct a comprehensive objective function that includes electricity price revenue, self-consumption savings revenue, and battery life loss, and solve for the optimal charging and discharging power and photovoltaic output setting value for the current 15-minute band under multiple electrical constraints. S4. Perform security verification and smoothing on the solution results, and generate executable control instructions; S5. Send instructions to the energy storage converter and photovoltaic inverter to execute the scheduling for this cycle; S6. Five minutes before the end of this cycle, trigger rolling optimization using the latest measured data to generate instructions for the next cycle, achieving closed-loop adaptive control with a 15-minute granularity.

10. The AI-powered intelligent photovoltaic energy storage dispatching method according to claim 9, characterized in that, The response strategy for the 15-minute band electricity price in step S3 specifically includes: When the predicted electricity price is on an upward trend and the current SOC has not reached its limit, start charging in advance or reduce the discharge power to lock in a low-price charging window. When the predicted electricity price is on a downward trend and the current SOC has not reached the lower limit, delay discharge or start low-power floating charging to avoid losses from high-price discharge. When photovoltaic output is excessive and the current 15-minute band electricity price is negative or lower than the energy storage cycle cost, photovoltaic output is actively limited or curtailment protection is activated to avoid negative revenue operation; When the deviation between the real-time electricity price and the predicted electricity price exceeds a threshold, intraday rescheduling is immediately triggered to correct the charging and discharging plans for the remaining time period.

Images