Photovoltaic panel energy management system based on multi-mode perception and intelligent decision

Abstract

The invention discloses a photovoltaic panel energy management system based on intelligent control, and the system achieves the self-adaptive adjustment of the angle of a photovoltaic panel through the linkage of a photosensitive sensor array and a double-shaft steering engine, optimizes the output of a Buck circuit through the self-adaptive PID algorithm of an STM32 controller, and builds a data closed loop through a Hi3861 controller, a Huawei cloud Internet of Things platform and a Pantou large model. The system adopts a multi-sensor fusion technology to monitor illumination, temperature and humidity, voltage and current and other parameters in real time, and the data precision is improved through a Kalman filtering algorithm; based on an illumination prediction model and a fault early warning model trained by historical data, potential risks can be predicted in advance, and system parameters can be automatically adjusted. Actual measurement results show that the average daily generating capacity of the system is improved by 20%-25% compared with that of a traditional fixed installation mode, the cycle life of a battery is prolonged by 30% or above, the abnormal response time is shorter than 50 ms, and the system has the advantages of being high in intelligence, reliability and adaptability and can be widely applied to the fields of distributed photovoltaic power generation, intelligent home energy management and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of solar energy utilization, in particular to a photovoltaic panel energy management system integrating intelligent control, data communication and adaptive regulation functions. More specifically, the system realizes intelligent management of the whole link of photovoltaic energy collection, storage and application through multi-sensor fusion technology, edge computing and cloud computing collaborative architecture, and is suitable for distributed photovoltaic power generation systems, intelligent home energy management and industrial Internet of Things scenarios. BACKGROUND

[0002] In practical applications, traditional photovoltaic systems encounter some technical difficulties.

[0003] Firstly, there is a bottleneck in light energy conversion efficiency. Fixed photovoltaic panels have relatively low daily light utilization due to the influence of the change of the sun's azimuth and elevation angle with time. Moreover, local shadow, such as bird droppings, fallen leaves and the like, can cause "hot spot effect", which not only reduces power generation efficiency, but also can damage the panel. Secondly, the energy storage system management is relatively extensive. The lead-acid battery charge and discharge control has always relied on fixed thresholds, such as a charging upper limit of 14.4V, without considering the influence of temperature on the internal resistance of the battery. At the same time, due to the lack of real-time assessment of the battery health status, the actual service life of the battery is lower than the designed service life. Thirdly, the system is not highly intelligent. The existing photovoltaic monitoring system mainly focuses on data collection, and lacks deep analysis capability. Moreover, once an abnormal situation occurs, the response basically relies on manual intervention, and the fault handling delay time is relatively long. For example, a certain distributed photovoltaic power station failed to detect the inverter fault in time, resulting in a large loss of power.

[0004] Chinese Patent No. CN119882838A proposes an adaptive photovoltaic tracking control method and system, which collects and fuses multi-source data such as weather and light through multiple sensors, inputs the adaptive photovoltaic tracking control model after preprocessing and edge computing. The model is based on neural network construction, uses feature analysis for dimension reduction, and optimizes parameters through reward function, combines particle filtering algorithm to correct angle error in real time, and realizes automatic switching of tracking mode in different environments such as cloudy and sunny days. The disadvantage is that the function is limited to photovoltaic panel tracking control, and does not involve battery energy storage management; the data application only serves the tracking model, lacks cloud deep analysis and strategy closed loop; does not integrate a relatively complete edge computing and cloud computing collaborative architecture, and the system has insufficient scalability; the hardware security protection mechanism is missing, and the optimization target does not cover energy storage efficiency, equipment health status and other indicators, making it difficult to maximize the efficiency of the whole system throughout the life cycle.

[0005] The Chinese patent with the publication number CN120016865A provides a multi-path parallel BUCK circuit and control method. The circuit obtains the output current of each branch through a current acquisition module, calculates the average current and determines whether it needs current sharing control. When needed, it calculates the adjustment error according to the current error, calculates the incremental value using a formula to adjust the branch. The circuit realizes multi-path parallel through shared capacitors, reducing the size. The control method improves control efficiency through current sampling and feedback. The difference between this patent and the present invention is that the present invention focuses on the whole link management of photovoltaic panel energy, covering dynamic tracking, energy storage management, intelligent decision-making, and cloud collaboration, while the patent mainly focuses on BUCK circuit current sharing control, lacks multi-modal perception, cloud big data analysis, and energy storage device state management, and is difficult to solve the core problems of low light utilization rate, extensive energy storage, and insufficient intelligence of traditional photovoltaic systems.

[0006] In terms of comprehensive performance of photovoltaic systems, traditional photovoltaic systems have certain room for improvement compared to the present invention. Traditional photovoltaic systems still have room for optimization in terms of light utilization rate, battery cycle life, fault response speed, daily average power generation, and intelligent management. Based on this, the present invention proposes a photovoltaic panel energy management system that integrates multi-modal perception and intelligent decision-making. SUMMARY

[0007] The present invention focuses on the difficulties of traditional photovoltaic systems, such as low light utilization rate, extensive energy storage management, and lack of intelligence, and aims to create a photovoltaic energy management system that integrates dynamic tracking, intelligent control, and data-driven technology. Its specific goals include achieving high-precision tracking of photovoltaic panels to the sun's orientation, reducing angle error, dynamically optimizing battery charging and discharging strategies based on environmental parameters, constructing a device fault prediction model to provide early warning of potential risks, and reducing system response delay through the cooperation of edge computing and cloud computing.

[0008] To achieve the above-mentioned purposes, the present invention provides the following technical solutions:

[0009] ① Photovoltaic panel dynamic tracking subsystem

[0010] Photosensitive sensor array: embed a 5x5 matrix layout of photosensitive sensors, model BH1750FVI, in the photovoltaic panel. The data of each sensor is transmitted to the STM32 through the I 2 C bus.

[0011] Dual-axis servo motor drive: horizontal axis uses MG996R servo motor, vertical axis uses SG90 servo motor.

[0012] Angle adjustment algorithm:

[0013]

[0014] Wherein, For the light-sensitive array gradient field, K p = 0.2, K i = 0.01, K d = 0.05.

[0015] ②Energy management subsystem

[0016] Buck circuit design: input voltage greater than 18V, output voltage 3.3-12V adjustable; the key parameters in this scheme involve power MOSFET, Schottky diode and output filter capacitor. Power MOSFET selects IRF540N, its on-resistance ≤77mΩ; Schottky diode adopts SS34, forward voltage drop ≤0.4V; output filter capacitor is 470μF / 35V, which belongs to low ESR type.

[0017] PID control algorithm optimization: incremental PID algorithm is adopted, and anti-integral saturation strategy is adopted:

[0018] Δu(k) = K p [e(k)-e(k-1)]+K i e(k)+K d [e(k)-2e(k-1)+e(k-2

[0019] Adaptive parameter setting: Where, K p0 = 0.8, α = 0.01 / ℃, β = 0.005 / ℃·s, T is the environmental temperature.

[0020] ③Data acquisition and communication subsystem

[0021] Multi-sensor fusion: temperature and humidity sensor DHT22; battery voltage and current monitoring INA219; data fusion algorithm: Where, is the fusion estimate, z k is the sensor measurement, K k is the Kalman gain.

[0022] Multi-level communication architecture: STM32→Hi3861: UART protocol, baud rate 115200bps, data frame format:

[0023] Header(2Bytes)+SensorData(32Byres)+CRC16(2Byres);

[0024] Hi3861→Huawei Cloud: MQTT protocol, QoS = 1, heartbeat packet interval 60s; application layer: custom JSON format, such as:

[0025]

[0026]

[0027] IV. Intelligent prediction and decision subsystem

[0028] Huawei cloud platform deployment: data storage: uses the time series database InfluxDB, stores frequency 1Hz, can save historical data for a long time; model training: disc big model based on Transformer architecture, the training data of the model is rich and diverse, and the data can run for a long time; fault simulation data, such as occlusion, short circuit, battery aging and other scenes; there are also meteorological data, including light, temperature, humidity, wind speed and other aspects.

[0029] Prediction algorithm: light intensity prediction model:

[0030] Where I is the light intensity, M is the meteorological data, w i / v j is the weight coefficient, n=24, m=6;

[0031] Fault warning model: P(fault) = σ(W·X+b)

[0032] Where X = [ΔT, ΔV, ΔI, corr(I meas , I pred )] is the feature vector, and σ is the Sigmoid function.

[0033] Compared with the prior art, the present application has the following beneficial effects:

[0034] 1. Improve light energy conversion efficiency and increase power generation. The 5x5 matrix light-sensitive sensor array detects the light difference of each area of the photovoltaic panel in real time, and the dual-axis steering drive device (angle adjustment accuracy ≤0.5°) realizes high-precision tracking of the photovoltaic panel to the sun, dynamically optimizing the light angle. The measured data shows that the daily average power generation of the system is increased by 20%-25% compared with the traditional fixed installation method, especially in the morning and evening periods when the light angle changes significantly, the power generation efficiency is improved more obviously.

[0035] 2. Optimize energy storage management and prolong battery life. The adaptive PID algorithm is used to dynamically adjust the output voltage or current of the Buck circuit, and the PID parameters (K p , K i , Kd) are corrected in real time according to the environmental temperature and light intensity, avoiding the overcharge and overdischarge problems caused by traditional fixed threshold control. At the same time, the multi-stage protection relay (main relay response time ≤10ms, branch relay ≤5ms) can quickly respond to local shadow, overcurrent and other abnormalities, reducing battery loss. The cycle life of the storage battery is prolonged by more than 30%.

[0036] 3. Intelligent fault early warning and rapid response. The fault early warning model based on the Pangu model (Transformer architecture) can predict potential equipment failures up to 48 hours in advance, with an accuracy of ≥95%; the system abnormal response time is <50ms, which can cut off the fault loop or adjust the operating parameters in a short time to avoid the expansion of the fault.

[0037] 4. Edge and cloud cooperation to realize intelligent management of the whole link. Relying on the cooperative architecture of STM32 edge computing and Huawei cloud platform, the data precision is improved through Kalman filtering algorithm to fuse multi-sensor data (temperature and humidity, voltage and current, etc.); combined with the LSTM light prediction model (prediction error ≤15%), the system operation strategy is optimized in advance to form a closed-loop control of "data collection-cloud analysis-instruction feedback".

[0038] 5. Low-power design, suitable for complex environments. With intelligent sleep mechanism, it automatically enters low-power state (power consumption ≤5mW) in insufficient light, low battery and other scenes, only retaining necessary wake-up modules to work, reducing energy loss during non-working period, especially suitable for remote areas or unattended scenes.

[0039] 6. High scalability and wide applicability. The system architecture is compatible with distributed photovoltaic power generation, smart home energy management, industrial Internet of Things and other scenes, and can flexibly adapt to photovoltaic panels and energy storage devices of different specifications through modular design. At the same time, the communication architecture based on MQTT protocol supports large-scale device access, providing reliable support for the intelligent upgrading of photovoltaic energy network.

[0040] The specific embodiments of the present application will be described in further detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0041] In the drawings:

[0042] Figure 1 is a system overall architecture diagram;

[0043] Figure 2 is a photosensitive sensor array layout diagram;

[0044] Figure 3 is a Buck circuit principle flow chart;

[0045] Figure 4 is a double-shaft servo control flow chart;

[0046] Figure 5 is a data communication protocol stack diagram;

[0047] Figure 6 is a Pangu model training and reasoning flow chart;

[0048] Figure 7 is a system exception handling timing diagram;

[0049] Figure 8 This is a comparison chart of the photovoltaic panel angle adjustment effect. DETAILED DESCRIPTION

[0050] The proposed photovoltaic panel energy management system, based on multimodal sensing and intelligent decision-making, aims to address the core issues of traditional photovoltaic systems: low light utilization, extensive energy storage management, and insufficient intelligence. Through a collaborative architecture combining multi-sensor fusion, edge computing, and cloud computing, the system achieves intelligent management of the entire photovoltaic energy collection, storage, and application chain. Its overall architecture comprises six major components: photovoltaic panels, sensors, controllers, actuators, communication modules, and a cloud platform. These components work closely together through the flow of data and commands, forming a highly coordinated management system.

[0051] 1. Overall system architecture and core components (parent system)

[0052] As attached Figure 1 and attached Figure 2 As shown, the photovoltaic panel serves as an energy source. Its surface is embedded with a 5×5 matrix array of photosensors (model BH1750FVI). Each sensor is spaced 10 cm apart, with its edge 5 cm from the panel edge, enabling real-time sensing of differences in light intensity across different areas. Embedded beneath the photovoltaic panel are dual-axis servos—an MG996R servo for the horizontal axis and an SG90 servo for the vertical axis—that adjust the photovoltaic panel's angle for precise sun tracking. The STM32 controller serves as the control center, receiving sensor data, calculating control signals using a PID algorithm, driving the servos and Buck circuit, and connecting to the Hi3861 communication module via a UART interface. The Hi3861 controller, acting as the communication hub, uploads local data to Huawei Cloud via the MQTT protocol. It also receives commands from the cloud and forwards them to the STM32 for execution. The Huawei Cloud platform and the Pangu Big Data model serve as cloud-based intelligence centers, leveraging historical data to train light prediction and fault warning models, providing optimization strategies for the local system. This creates a closed loop of "data collection, cloud analysis, and command feedback."

[0053] 2. Photovoltaic panel dynamic tracking subsystem

[0054] As attached Figure 2 , Attachment Figure 4 and attached Figure 8 As shown in the photovoltaic panel dynamic tracking subsystem, the photosensitive sensor array is evenly distributed on the photovoltaic panel surface in the form of a 5×5 matrix. Each sensor is connected to the photovoltaic panel through I 2C bus transmits data to STM32 controller, STM32 reads the light intensity value of each sensor at a frequency of 10 Hz, forms a light matrix containing 64 data points, which can cover the whole surface of the photovoltaic panel and detect the light intensity difference caused by local shadow. In terms of dual-axis servo control, the MG996R servo of the horizontal axis and the SG90 servo of the vertical axis are controlled by the PWM signal output by STM32. STM32 calculates the light gradient in horizontal and vertical directions according to the photosensitive array data, judges whether the angle needs to be adjusted, and if so, uses the PID algorithm to calculate the duty cycle of the PWM signal output to the servo, the formula is:

[0055]

[0056] where K p , K i , K d are 0.2, 0.01, 0.05 respectively. By adjusting these three parameters, the photovoltaic panel angle quickly and smoothly approaches the optimal light direction, and the angle feedback signal is received to form a closed-loop control, ensuring that the angle adjustment accuracy is ≤0.5°. In sunny environment test, it is found that the dynamic tracking system improves the daily power generation of the fixed installed photovoltaic panel by 22.3%, for example, during the period of 9:00-10:00, the dynamic tracking system generates 152Wh, which is 21.6% higher than the fixed installed 125Wh, and the angle error is only 0.3°. The test situation of each time period is shown in the following table:

[0057]

[0058] III. Energy management subsystem

[0059] As shown in the accompanying Figure 3 , the Buck circuit in the energy management subsystem is designed to have an input voltage greater than 18V and an output voltage adjustable between 3.3-12V. The key devices include IRF540N power MOSFET with on-resistance ≤77mΩ, SS34 Schottky diode with forward voltage drop ≤0.4V, and 470μF / 35V low ESR output filter capacitor. Its working principle is that when the MOSFET is turned on, the photovoltaic panel current flows through the MOSFET to the inductor and the battery, and the inductor stores energy; when the MOSFET is turned off, the inductor releases energy, the current is maintained through the freewheeling diode to the battery, and the output voltage is controlled by adjusting the duty cycle of the PWM signal. Traditional photovoltaic systems use fixed threshold to control battery charging and discharging, without considering temperature influence. This system introduces adaptive PID algorithm, which dynamically adjusts PID parameters according to environmental temperature and light intensity, and the proportional coefficient formula is:

[0060] where K p0= 0.8, a = 0.01 / ℃, β = 0.005 / ℃·s, when the temperature rises, K p Automatic increase, accelerate the charging response speed, avoid overcharge. Set the main relay and branch relay in the battery discharge circuit, the main relay response time ≤10ms, branch relay response time ≤5ms, when the photovoltaic panel local temperature difference > 15℃, charging current > 5A for more than 30 seconds, respectively, execute the operation of cutting off the main relay, corresponding area branch relay or reducing the PWM duty cycle, etc., to protect the safety of the battery.

[0061] Four, data acquisition and communication subsystem

[0062] As shown in the attached Figure 1 and attached Figure 5 The data acquisition and communication subsystem integrates multiple sensor data such as temperature and humidity sensor DHT22 and battery voltage and current monitor INA219, and uses Kalman filtering algorithm for data fusion, the formula is:

[0063] Among them is the fusion estimate, z k is the sensor measurement value, through recursive calculation to effectively suppress noise and improve data accuracy. In the communication architecture, STM32 and Hi3861 communicate through UART protocol, baud rate 115200bps, data frame format is Header(2 bytes) + SensorData(32 bytes) + CRC16(2 bytes); Hi3861 uploads data to Huawei Cloud through MQTT protocol, QoS = 1, heartbeat packet interval 60 seconds, application layer uses custom JSON format, for example:

[0064]

[0065] After Huawei Cloud receives the data, it cleanses the abnormal values and stores the valid data into InfluxDB time series database, and pushes it to the function workflow for real-time analysis.

[0066] Five, intelligent prediction and decision subsystem

[0067] As shown in the attached Figure 1 and attached Figure 6As shown, the intelligent prediction and decision subsystem is based on the Huawei Cloud platform deployment, and the data storage uses the InfluxDB time series database to store historical data at a frequency of 1 Hz. Model training is performed using the Pangu large model, and the light prediction model is based on the LSTM network architecture, with input features such as light intensity, temperature, humidity, and timestamp for the previous 24 hours, and output of light intensity prediction values for the next 12 hours, with a prediction error of ≤15%. The fault warning model uses the Transformer architecture, with input system operating parameters and output device failure probability, with potential faults warned up to 48 hours in advance, with an accuracy of ≥95%. The Python example code for the function workflow is as follows:

[0068]

[0069] The Huawei Cloud generates instructions that are transmitted to the Hi3861 via the MQTT protocol and then to the STM32 for execution, forming a "prediction-decision-execution" closed loop.

[0070] Six, System Abnormality Handling

[0071] As shown in the accompanying Figure 7 As shown, the system abnormality handling process is as follows: at time t0, the photosensitive sensor detects a sudden drop in light intensity in a certain area, and calculates that the local temperature difference ΔT > 15℃, triggering an abnormality signal; at t0+80ms, the STM32 detects the abnormality signal and immediately adjusts the angle of the photovoltaic panel and reduces the charging current by 30%; at t0+200ms, the Hi3861 sends an alarm message to the Huawei Cloud via the MQTT protocol; at t0+350ms, the Huawei Cloud analyzes and issues instructions to shut off the corresponding branch relay; at t0+400ms, the STM32 controls the branch relay to be disconnected via the GPIO signal; at t0+435ms, the relay completes the disconnection action and the system enters a protection state. Other abnormal scenarios such as high-temperature environments (temperature ≥ 55℃) cause the STM32 to reduce the charging voltage, and if the temperature continues to rise, it triggers a hibernation mode; and overcharging of the battery (voltage > 14.5V) causes the main relay to be disconnected within 0.05 seconds and an alarm message to be sent.

[0072] Seven, Hardware and Software Implementation Details

[0073] In terms of hardware design, the main control board is centered around the STM32 minimum system. An 8MHz external crystal oscillator and a 32.768kHz RTC crystal oscillator are used. The input 5-12V voltage is converted to 3.3V by an AMS1117 voltage stabilizing chip. A separate 5V switching power supply is used to power the steering engine. The interfaces include a 12-bit ADC channel, a TIM3 / TIM4 timer output PWM signal interface, and a USART interface. The Buck circuit uses a four-layer PCB design. The top layer is a signal layer. The inner layer 1 is a 12V power plane. The inner layer 2 is a ground plane. The bottom layer is a power layer. The control power loop area is less than 1cm 2 , and the current sampling resistor uses a four-terminal Kelvin connection. The software architecture uses the FreeRTOS real-time operating system to create tasks such as FaultDetectionTask (priority 5), ControlTask (priority 4), CommunicationTask (priority 2), SensorTask (priority 3), and PowerManagementTask (priority 1), as shown in the following table:

[0074]

[0075] The C language pseudo code of SensorTask is as follows:

[0076]

[0077]

[0078] Eight, performance test and verification

[0079] In the light tracking performance test, as shown in Figure 7 and Figure 8 , the dynamic tracking system improves the power generation of the fixed installed photovoltaic panel by 18.9%-25.0% under sunny weather, with an angle error of 0.2-0.4°. After optimization, the average error decreases from 1.2° to 0.3°. The Buck circuit has a steady-state output voltage of 12.01V±0.05V, a ripple of <50mV, and a recovery time of <200μs after a load mutation. The efficiency reaches 95.2%-96.2%. The efficiency test results are shown in the following table:

[0080]

[0081] In terms of abnormal response speed, the detection time of local shading, high temperature, and battery overcharge is 120ms, 80ms, and 50ms, respectively. The response action recovery time is 20-500ms. The long-term running stability test runs continuously for 30 days, with a daily power generation fluctuation of <5%, a system failure rate of 0, a battery temperature of less than 50℃, and a voltage maintained between 12.5-13.8V.

[0082]

[0083] Nine, low-power management

[0084] When the light intensity < 100 Lux and lasts for 30 minutes, a serious fault is detected and cannot be automatically recovered, the battery power < 20% and no external charging input, the system enters the sleep state, the power consumption ≤ 5mW, only the RTC clock and the wake-up detection circuit work. The wake-up methods include photosensitive sensor threshold triggering, RTC timing wake-up and external interrupt wake-up.

[0085] Ten, summary

[0086] The system solves the core problems of traditional photovoltaic systems through multi-modal perception, intelligent decision-making, edge computing and cloud computing coordination. The actual measurement data shows that the daily average power generation is increased by 20%-25%, the battery life is extended by more than 30%, and the abnormal response is less than 50ms. It provides an efficient and reliable energy management solution for distributed photovoltaic power generation, smart home and other scenarios. In the future, it can further expand the function, promote the development of solar energy utilization to a more intelligent and more efficient direction.

Claims

1. A photovoltaic panel energy management system based on multimodal perception and intelligent decision-making, comprising photovoltaic panels, batteries, Buck circuits, and controllers, characterized by: ① The photovoltaic panel surface is evenly distributed with a photosensitive sensor array in a 5×5 matrix layout, with each sensor spaced 10 cm apart, to detect the difference in light intensity in different areas of the photovoltaic panel in real time; ② A dual-axis servo drive device is set under the photovoltaic panel, including a horizontal rotation axis and a pitch axis, which is controlled by the PWM signal output by the STM32 controller, and the angle adjustment accuracy is ≤0.5°; ③ The Buck circuit is controlled by the STM32 controller through the adaptive PID algorithm to adjust the output voltage / current. The proportional coefficient K in the algorithm is p , integral coefficient K i and differential coefficient K d Dynamic adjustment based on ambient temperature T and light intensity I, the calculation formula is: Among them, K p0 =0.8-1.2,K i0 =0.01-0.03, K d0 =0.04-0.06, α=0.005-0.015 / ℃, β=0.001-0.005 / ℃·s, γ=0.0001-0.0005 / Lux, δ=0.00001-0.00005 / Lux·s, ∈=0.01-0.05; ④) The STM32 controller communicates with the Hi3861 controller via UART. The Hi3861 uploads data to the Huawei Cloud IoT platform through the MQTT protocol and receives adjustment instructions generated by the Pangu model. The communication data frame contains timestamps, sensor raw data, calculation parameters, and device status information.

2. The system according to claim 1, wherein: The battery discharge circuit is equipped with a multi-stage protection relay, including a main relay and a branch relay. The main relay response time is ≤10ms, and the branch relay response time is ≤5ms. The STM32 controller controls the relay on and off through the GPIO signal. The control logic is based on the following conditions: ①When the battery voltage V b When the voltage is >14.5V or <10.5V, the main relay is cut off; ② When the local temperature difference of the photovoltaic panel ΔT>15℃, the corresponding regional branch relay is cut off; ③When the charging current I c >5A for more than 30 seconds, reduce the PWM duty cycle until I c ≤4.5A.

3. The system according to claim 1, wherein: The STM32 controller integrates a multi-sensor fusion module, including: ①DHT22 temperature and humidity sensor, measuring range -40-80℃ (temperature), 0-99.9%RH (humidity), with accuracy of ±0.5℃ and ±2%RH respectively; ② The voltage and current monitoring modules are INA219 and ACS712, with a measurement range of 0-26V / 0-5A, a resolution of 16 bits, and a sampling rate of ≥100Hz; ③ Using the Kalman filter algorithm to fuse multi-sensor data, the state estimation equation is: in, is the state estimate, F k is the state transfer matrix, K k is the Kalman gain, z k is the measured value, H k is the observation matrix.

4. The system according to claim 1, wherein: The Pangu model trains a light prediction model and an equipment failure warning model based on historical data, where: ① The light prediction model uses an LSTM network architecture, with inputs such as light intensity, temperature, humidity, and timestamp from the previous 24 hours, and outputs a predicted light intensity value for the next 12 hours, with a prediction error of ≤15%; ② The fault warning model uses a Transformer architecture, with system operating parameters (voltage, current, temperature, angle, etc.) as input and device failure probability as output. The warning lead time is ≥48 hours; ③ Huawei Cloud Platform generates control instructions based on the prediction results, including the optimal photovoltaic panel angle, PID parameter correction value, charge and discharge threshold adjustment, etc., and sends them to the Hi3861 controller through the MQTT protocol.

5. The system according to claim 1, wherein: The system also includes a low power management module, which enters a dormant state when one of the following conditions is met: ① Light intensity <100 Lux and duration ≥30 minutes; ②The system detects a serious fault and cannot recover automatically; ③The battery power is less than 20% and there is no external charging input; ④ In sleep mode, the system power consumption is ≤5mW, and only the RTC clock and wake-up detection circuit are kept working. The wake-up methods include light sensor threshold trigger, RTC timer wake-up or external interrupt wake-up.

Images