A multi-sensor fusion solar street lamp artificial intelligence self-adaptive control method and system

Abstract

The application discloses a kind of solar street lamp artificial intelligence self-adapting control method and system of multi-sensor fusion, belong to lighting control technical field, including obtaining local first meteorological sensing data, numerical weather prediction data and the current state of charge of energy storage device;Data is input into artificial intelligence prediction model, and the prediction result of future time period photovoltaic power generation power sequence and load power consumption power sequence is obtained;The charging control strategy and discharging control strategy that minimizes preset objective function are solved using model predictive control rolling optimization, wherein the charging control strategy dynamically adjusts charging cutoff voltage and charging current, and the discharging control strategy dynamically adjusts lighting brightness and color temperature;Control solar street lamp to execute the above strategy;The application realizes the control span from passive response to active prediction, can actively reserve energy before the arrival of continuous rain and other adverse weather, effectively prolong the night guarantee lighting length, improve energy resilience and utilization efficiency under complex meteorological conditions.

Description

Technical Field

[0001] This invention relates to the field of lighting control technology, and in particular to an artificial intelligence adaptive control method and system for solar streetlights that integrates multiple sensors. Background Technology

[0002] Solar streetlights, as a green and independent road lighting solution, have been widely used in urban secondary roads, rural roads, industrial parks, and remote areas. A typical solar streetlight system consists of photovoltaic modules, energy storage batteries, charge and discharge controllers, and LED lighting loads. Its operation is highly dependent on the effective capture of solar energy resources and the rational allocation of battery energy.

[0003] Currently, the charging and discharging control of solar streetlights generally adopts a passive response strategy based on real-time status feedback. The controller mainly tracks the maximum power point (MPPT) based on the open-circuit voltage of the photovoltaic module or the terminal voltage of the battery, and realizes charging control through traditional algorithms such as the perturbation observation method and the incremental conductance method. On the discharging side, fixed time-brightness curves (such as full power at night and reduced power at night) or simple battery voltage threshold dimming (such as reducing brightness when the voltage is lower than the set value) are mostly used to ensure basic lighting.

[0004] This passive-response control faces a core bottleneck in practical applications: a lack of foresight regarding future energy supply and load demand. This leads to inefficient battery energy management and severely inadequate resilience in providing lighting support under extreme weather conditions such as continuous rain. When faced with several consecutive days of rain, traditional controllers only passively reduce load power after the battery voltage has significantly decreased. By this time, the remaining power is often insufficient to support basic nighttime lighting, causing streetlights to turn off prematurely and directly impacting traffic and public safety. Furthermore, because future irradiance conditions cannot be predicted, the system cannot proactively adjust its charging strategy before severe weather arrives (such as appropriately increasing the charging cutoff voltage to store more energy) or dynamically optimize the discharge plan for the following days. It can only react passively after the fact, resulting in low energy utilization efficiency.

[0005] Therefore, how to endow solar streetlights with the ability to predict future energy conditions and proactively optimize charging and discharging strategies based on this prediction in order to fundamentally improve the system's resilience in lighting under continuous adverse weather conditions is a key technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-sensor fusion-based artificial intelligence adaptive control method and system for solar streetlights, in order to solve the problem that existing solar streetlight control methods lack the ability to predict future energy supply and load demand, resulting in insufficient lighting resilience and low energy utilization efficiency under severe weather conditions such as continuous rain.

[0007] To achieve the above objectives, this invention provides a multi-sensor fusion-based artificial intelligence adaptive control method for solar streetlights, comprising the following steps: S1. Obtain the first meteorological sensor data of the target street light location, the numerical weather forecast data for the future preset time period, and the current state of charge of the energy storage device; S2. Input the first meteorological sensor data, numerical weather forecast data and the current state of charge into the pre-trained artificial intelligence prediction model deployed in the edge computing unit to obtain the prediction results including the photovoltaic power generation power sequence and the load power consumption sequence for future periods; S3. Based on the prediction results, before the arrival of the future period, use model prediction control to solve the charging control strategy and discharging control strategy that minimize the preset objective function. S4. Control the solar streetlights to execute charging and discharging control strategies.

[0008] Preferably, the first meteorological sensor data includes ambient light intensity, photovoltaic module backsheet temperature, ambient humidity, liquid precipitation, and wind speed; the numerical weather forecast data includes cloud cover, precipitation probability, and temperature.

[0009] Preferably, in step S2, the artificial intelligence prediction model is a long short-term memory network model or a time-series fusion converter model, and the current time features are configured to be input into the artificial intelligence prediction model together with the first meteorological sensor data and numerical weather forecast data, so that the model outputs the photovoltaic power generation power sequence and the load power consumption sequence at the same time.

[0010] Preferably, before inputting the ambient light intensity into the artificial intelligence prediction model in step S2, the method further includes calculating the dust attenuation factor using the readings of a dust transmittance sensor on the surface of the photovoltaic module, according to the following formula: ; in, Dust attenuation factor, To calibrate the light transmittance under clean conditions, This represents the current measured transmittance. Dust attenuation factor The ambient light intensity is compensated and corrected, and the corrected ambient light intensity is used as the model input.

[0011] Preferably, the preset objective function in step S3 is: ; in, The objective function value, To predict the length of the time domain, for The necessary lighting electricity demand at all times, for The available electricity is based on the state of charge of the energy storage device and the predicted power generation. For reference to comfortable brightness, This represents the actual output brightness. , These are the weighting coefficients.

[0012] Preferably, the charging control strategy generated in step S3 includes: when the prediction result indicates that the number of consecutive rainy days in the future exceeds a set value, and the current state of charge is lower than a safety threshold, the charging cutoff voltage of the energy storage device is increased according to the following relationship: ; in, This is the charging cutoff voltage. The reference cutoff voltage, For reference, the state of charge threshold, The current state of charge, This is for adjusting the coefficient.

[0013] Preferably, the discharge control strategy generated in step S3 includes: when a power shortage is predicted, adjusting the lighting brightness during preset off-peak nighttime hours according to the following relationship: ; in, This represents the actual output brightness. For minimum safe brightness, For full brightness, The current state of charge, To avoid the lowest state of charge for deep discharge, The target is a high-charge state; and the lighting color temperature is switched from the first color temperature to a lower second color temperature.

[0014] Preferred options also include: S5. Obtain the actual photovoltaic power generation in real time, calculate the deviation between the actual power generation and the predicted power generation in step S2, and perform online incremental training on the artificial intelligence prediction model based on the deviation to update the model parameters.

[0015] An AI-driven adaptive control system for solar streetlights, incorporating multi-sensor fusion, includes: The weather sensor module is used to collect primary weather sensor data. The positioning and communication module is used to acquire geographic location information and numerical weather forecast data for a future preset time period; The battery management module is used to obtain the current state of charge of the energy storage device; The edge computing and control module is configured as follows: It receives the first meteorological sensor data collected by the meteorological sensor module, the numerical weather forecast data obtained by the positioning and communication module, and the current state of charge obtained by the battery management module; Run a pre-trained artificial intelligence prediction model to output the photovoltaic power generation sequence and load power consumption sequence for future periods; Model predictive control rolling optimization is used to minimize a preset objective function, generating charging control commands and discharging control commands. The preset objective function is constructed to minimize the weighted sum of squares of the gap between the necessary lighting power demand and the available power in the future time period, and the weighted sum of squares of the deviation between the reference comfortable brightness and the actual output brightness. The charge / discharge execution module is used to adjust the charging cut-off voltage and charging current according to the charging control command, and to adjust the brightness of the lighting load according to the discharging control command.

[0016] Therefore, the present invention employs the above-mentioned multi-sensor fusion-based artificial intelligence adaptive control method and system for solar streetlights, which has the following beneficial effects: (1) This invention integrates local multi-source meteorological sensor data with external numerical weather forecast data, and uses a pre-trained artificial intelligence prediction model to jointly predict the photovoltaic power generation and load power consumption in the future period. Combined with the model prediction control rolling optimization of the charging and discharging strategy, the system can actively adjust the charging cut-off voltage and dynamically plan the discharge plan before the arrival of severe weather such as continuous rain, effectively extending the nighttime lighting guarantee time under extreme weather conditions and reducing the risk of streetlights going out in advance.

[0017] (2) On the discharge control side, when a power supply gap is predicted, the present invention smoothly adjusts the lighting brightness according to the current state of charge and switches the lighting color temperature from a higher color temperature to a lower color temperature, thereby reducing system power consumption and minimizing the impact of sudden changes in brightness on visual comfort.

[0018] (3) The present invention compensates and corrects the ambient light intensity by using a dust attenuation factor, and uses the deviation between the actual power generation and the predicted power generation to conduct online incremental training of the prediction model. It can gradually adapt to the local microclimate characteristics of the installation site and the dust accumulation state of the photovoltaic modules, and maintain the prediction accuracy and the effectiveness of the control strategy in long-term operation.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] Figure 1 This is a flowchart of an artificial intelligence adaptive control method for solar streetlights based on multi-sensor fusion, according to the present invention. Figure 2This is a diagram of the architecture of an AI adaptive control system for solar streetlights based on multi-sensor fusion, as described in this invention. Detailed Implementation

[0021] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely illustrates selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0022] Example like Figure 1 As shown, Finally, it should be noted that the present invention provides a multi-sensor fusion-based artificial intelligence adaptive control method for solar streetlights, comprising the following steps: S1. Obtain the first meteorological sensor data of the target street light location, the numerical weather forecast data for the future preset time period, and the current state of charge of the energy storage device.

[0023] The primary meteorological sensor data includes ambient illuminance, photovoltaic module backsheet temperature, ambient humidity, liquid precipitation, and wind speed; numerical weather prediction data includes cloud cover, precipitation probability, and temperature. Ambient illuminance is collected by an omnidirectional illuminance sensor installed on the same plane as the photovoltaic module; photovoltaic module backsheet temperature is collected by a thermistor or thermocouple attached to the photovoltaic module backsheet; ambient humidity and liquid precipitation are collected by an integrated temperature and humidity sensor and a piezoelectric rain gauge; and wind speed is collected by an ultrasonic anemometer. The current state of charge of the energy storage device is estimated in real time by the battery management module using coulomb counting combined with open-circuit voltage correction. Numerical weather prediction data is periodically retrieved by the positioning and communication module from an external meteorological data service interface via a cellular network based on the latitude and longitude of the target streetlight. The preset time period is selected from the next 24 to 72 hours, and the specific duration can be flexibly configured according to the storage and computing resources of the streetlight controller. Current time characteristics include the current time and month, used to assist the model in capturing the periodic usage patterns of the lighting load.

[0024] In a preferred embodiment, before inputting ambient illuminance into the artificial intelligence prediction model, the method further includes calculating a dust attenuation factor using readings from a dust transmittance sensor on the surface of the photovoltaic module, according to the following formula: ; in, Dust attenuation factor, To calibrate the light transmittance under clean conditions, This represents the current measured transmittance.

[0025] The dust transmittance sensor consists of a set of miniature reference photodiodes and a replaceable transparent glass plate covering them. The transmittance is calculated by periodically measuring the short-circuit current of the reference photodiodes and comparing it with the factory calibration value. A dust attenuation factor η is used to compensate for and correct the original ambient illuminance value, specifically by dividing the ambient illuminance by... The corrected ambient illuminance is obtained and then used as the model input. This correction step can effectively reduce the illuminance measurement deviation caused by dust accumulation on the surface of photovoltaic modules, making the power generation prediction closer to the actual light-receiving state of the modules.

[0026] S2. Input the first meteorological sensor data, numerical weather forecast data and the current state of charge into the pre-trained artificial intelligence prediction model deployed in the edge computing unit to obtain the prediction results including the photovoltaic power generation power sequence and the load power consumption sequence for future periods.

[0027] The artificial intelligence prediction model is either a long short-term memory network model or a time-series fusion transformer model. Current time features are configured to be input into the AI ​​prediction model along with primary meteorological sensor data and numerical weather prediction data, enabling the model to output both photovoltaic power generation and load power consumption sequences. This model is pre-trained on a cloud server before deployment. Training data comes from historical meteorological sensor data, corresponding actual photovoltaic power generation sequences, and actual load power consumption sequences collected over a long period from photovoltaic power plants or already deployed solar streetlights located near the target streetlights and with similar climate conditions. During training, historical meteorological sensor data, numerical weather prediction data, and the current state of charge are used as input features, while future photovoltaic power generation and load power consumption sequences are used as labels. Backpropagation and gradient descent algorithms are used to optimize the model parameters until the prediction error on the validation set converges to a preset range. After training, the model is compressed and quantized, converted to a format suitable for edge computing units, and then loaded into the flash memory of the streetlight controller via remote firmware upgrade or local burning, becoming the pre-trained AI prediction model. The edge computing unit uses a microcontroller with an embedded neural processing unit or a low-power embedded processor, which can perform model inference periodically at minute intervals while the streetlights are running normally, and output the photovoltaic power generation sequence and load power consumption sequence at a time step of 15 minutes or 30 minutes in the next 24 hours.

[0028] S3. Based on the prediction results, before the arrival of the future period, use model prediction control to solve the charging control strategy and discharging control strategy that minimize the preset objective function.

[0029] The preset objective function is: ; in, The objective function value, To predict the length of the time domain, for The necessary lighting electricity demand at all times, for The available electricity is based on the state of charge of the energy storage device and the predicted power generation. For reference to comfortable brightness, This represents the actual output brightness. , These are the weighting coefficients.

[0030] The specific process of rolling optimization is as follows: At the beginning of each control cycle, starting from the current moment, using the power generation sequence and load power consumption sequence for the next N time steps output by the artificial intelligence prediction model, and under the constraint of the current state of charge, the control variables such as the charging cutoff voltage and lighting brightness within the N time steps are optimized to find a set of control sequences that minimize the objective function J. However, only the control command of the first time step in this sequence is executed. In the next control cycle, data is collected again, predictions are updated, and optimization is performed again, and so on, in a cyclical manner. The weight coefficients w1 and w2 can be configured according to the priority strategy of the application scenario. If safety is prioritized, w1 is increased to ensure that the system prioritizes the power supply for necessary lighting. If comfort is prioritized, w2 is increased to ensure that the system maintains a lighting level as close as possible to the reference comfortable brightness.

[0031] The generated charging control strategy includes: when the forecast indicates that the number of consecutive rainy days in the future exceeds a set value, and the current state of charge is below a safe threshold, the charging cutoff voltage of the energy storage device is increased according to the following relationship: ; in, This is the charging cutoff voltage. The reference cutoff voltage, For reference, the state of charge threshold, The current state of charge, This is for adjusting the coefficient.

[0032] It is usually set to the standard charging cut-off voltage at room temperature recommended by the battery manufacturer. It can be configured to a value between 50% and 60%. Through experimental calibration, the charging cut-off voltage increases by several millivolts to tens of millivolts for every percentage point decrease in the state of charge, thus allowing for an appropriate increase in charge within a safe range. When the forecast does not indicate continuous rain or the current state of charge is already above the safe threshold, the charging cut-off voltage is maintained at [value missing]. Keep it unchanged and charge in the normal way to avoid the impact of long-term overcharging on battery life.

[0033] The generated discharge control strategy includes adjusting lighting brightness according to the following relationship during preset off-peak nighttime hours when a power shortage is predicted: ; in, This represents the actual output brightness. For minimum safe brightness, For full brightness, The current state of charge, To avoid the lowest state of charge for deep discharge, The system aims for a high charge state and switches the lighting color temperature from the primary color temperature to a lower secondary color temperature. The primary color temperature is typically neutral white or cool white light of 4000K to 5000K, while the secondary color temperature is warm yellow light of 2200K to 3000K. Lower color temperature lighting consumes less power and causes less interference with the human eye's dark adaptation while maintaining the same visual experience. The preset off-peak hours at night generally refer to 11 PM to 5 AM the next day, during which traffic flow and pedestrian activity are significantly reduced. Appropriately reducing the brightness can still meet the basic security and pedestrian safety needs. The brightness adjustment process is not a step-like jump, but a smooth transition over several seconds to tens of seconds to avoid discomfort to the human eye caused by sudden brightness changes. When sunlight conditions improve and the charge state returns to the normal range, the system automatically restores the brightness and color temperature to the normal operating state.

[0034] S4. Control the solar streetlights to execute charging and discharging control strategies. Charging control commands are sent from the edge computing and control module to the charging / discharging execution module via a communication bus. The charging / discharging execution module includes a DC-DC converter circuit and a metal-oxide-semiconductor field-effect transistor (MOSFET) drive circuit. Based on the charging control commands, it adjusts the duty cycle of the pulse width modulation signal in real time to control the charging current, and controls the upper limit of battery charging by adjusting the set value of the charging cutoff voltage. Discharging control commands are also sent to the charging / discharging execution module. The constant current drive circuit in the charging / discharging execution module adjusts the current output to the LED array according to the brightness command and selects LED channels with different color temperatures to achieve color temperature switching.

[0035] S5. Real-time acquisition of actual photovoltaic power generation, calculation of the deviation between actual power generation and predicted power generation in step S2, and online incremental training of the artificial intelligence prediction model based on this deviation to update model parameters. Actual photovoltaic power generation is calculated by multiplying the output voltage and current of the photovoltaic modules collected in real-time by the battery management module. Deviation calculation is performed after each control cycle, aligning the actual power generation sequence and predicted power generation sequence of the past prediction cycle by time, and calculating the root mean square error. When the cumulative deviation exceeds a preset threshold or the preset number of online training cycles has elapsed, the first meteorological sensor data, numerical weather prediction data, state of charge, and actual power generation collected during this period are used as new training samples. The model parameters are updated once or multiple times with a small learning rate gradient descent within the edge computing unit, without changing the model structure, only fine-tuning the weight parameters. The online-trained model automatically replaces the original model parameters for prediction inference in the next cycle. Through this online incremental learning mechanism, the model can gradually adapt to the local microclimate characteristics of the installation site, such as local irradiance changes caused by surrounding building shading and seasonal dust deposition patterns, continuously improving prediction accuracy over long-term operation.

[0036] like Figure 2 As shown, a multi-sensor fusion solar street light artificial intelligence adaptive control system includes a weather sensing module, a positioning and communication module, a battery management module, an edge computing and control module, and a charging and discharging execution module.

[0037] The meteorological sensing module is used to collect primary meteorological sensing data, including at least an omnidirectional illuminance sensor, a photovoltaic module backsheet temperature sensor, an ambient temperature and humidity sensor, a piezoelectric rain gauge, and an ultrasonic wind speed sensor. Each sensor is connected to the edge computing and control module via an integrated circuit bus or a serial peripheral interface. The positioning and communication module includes a Global Navigation Satellite System (GNSS) positioning module and a cellular mobile communication module. These modules are used to acquire the latitude, longitude, altitude, and time information of the target streetlights and connect to an external meteorological data service interface via a cellular network to periodically acquire numerical weather forecast data for a preset future time period. The battery management module includes a voltage detection circuit, a current detection circuit, and a battery management chip. The battery management chip estimates the current state of charge of the energy storage device in real time using a coulomb counting method combined with open-circuit voltage correction, while simultaneously monitoring the battery surface temperature. The state of charge and temperature data are then transmitted to the edge computing and control module via the integrated circuit bus.

[0038] The edge computing and control module is the core control unit of the system. It employs a microcontroller with an embedded neural processing unit or a low-power embedded processor and is configured to: receive first meteorological sensor data collected by the meteorological sensor module, numerical weather forecast data acquired by the positioning and communication module, and the current state of charge acquired by the battery management module; run a pre-trained artificial intelligence prediction model to output the photovoltaic power generation sequence and load power consumption sequence for future periods; and use model predictive control rolling optimization to minimize a preset objective function, generating charging control commands and discharging control commands. The preset objective function is constructed to minimize the weighted sum of squares of the gap between necessary lighting power demand and available power supply in the future period, and the weighted sum of squares of the deviation between the reference comfortable brightness and the actual output brightness. The edge computing and control module is also responsible for scheduling and executing online incremental training, utilizing idle system computing resources during daytime photovoltaic power generation periods to fine-tune and update model parameters.

[0039] The charge / discharge execution module adjusts the charging cutoff voltage and charging current according to charging control commands, and adjusts the brightness of the lighting load according to discharging control commands. The module includes a synchronous buck or buck-boost DC-DC converter circuit, a metal-oxide-semiconductor field-effect transistor (MOSFET) power switch, a pulse-width modulation (PWM) drive circuit, and a dual-channel constant-current LED driver circuit. During charging, the DC-DC converter circuit converts the unstable voltage output from the photovoltaic module into a stable voltage suitable for battery charging, and controls the charging current by adjusting the PWM duty cycle; after reaching the charging cutoff voltage, it switches to float charging or stops charging. During discharging, the dual-channel constant-current LED driver circuit controls the current ratio of the cool white LED channel and the warm yellow LED channel respectively according to brightness and color temperature commands, achieving smooth brightness adjustment and seamless color temperature switching.

[0040] In a preferred embodiment, the weather sensing module, battery management module, edge computing and control module, and charge / discharge execution module are integrated onto the same printed circuit board and installed inside the streetlight pole or lamp head cavity. The antenna portion of the positioning and communication module extends to the outside of the pole or the top of the lamp head to ensure the reception quality of satellite and cellular signals. The entire system is jointly powered by photovoltaic modules and energy storage devices. The edge computing and control module and communication module adopt a low-power design, entering a sleep state during periods of no target activity at night, and waking up at set intervals to perform prediction and control calculations, further reducing the system's own power consumption.

[0041] Therefore, this invention employs a multi-sensor fusion-based artificial intelligence adaptive control method and system for solar streetlights. By fusing local multi-source meteorological sensor data with external numerical weather forecast data, it utilizes a pre-trained artificial intelligence prediction model to jointly predict future power generation and consumption. Based on this, it employs model predictive control to continuously optimize the charging and discharging strategy, achieving a technological leap from passive response to proactive prediction. Especially before the onset of severe weather such as continuous rain, the system can increase the charging cutoff voltage in advance to store more energy, and smoothly reduce brightness and switch to low-power color temperature during off-peak nighttime periods based on the state of charge, thereby maximizing the nighttime lighting duration while ensuring minimum safe lighting. The introduction of a dust attenuation factor and the coordination of an online incremental training mechanism enable the model to continuously adapt to changes in the local environment and equipment status, maintaining prediction accuracy and control strategy effectiveness during long-term operation, and overall improving the lighting resilience and energy utilization efficiency of solar streetlights under complex weather conditions.

[0042] The embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and such modifications or equivalent substitutions should not cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A multi-sensor fusion-based artificial intelligence adaptive control method for solar streetlights, characterized in that, Includes the following steps: S1. Obtain the first meteorological sensor data of the target street light location, the numerical weather forecast data for the future preset time period, and the current state of charge of the energy storage device; S2. Input the first meteorological sensor data, numerical weather forecast data and the current state of charge into the pre-trained artificial intelligence prediction model deployed in the edge computing unit to obtain the prediction results including the photovoltaic power generation power sequence and the load power consumption sequence for future periods; S3. Based on the prediction results, before the arrival of the future period, use model prediction control to solve the charging control strategy and discharging control strategy that minimize the preset objective function. S4. Control the solar streetlights to execute charging and discharging control strategies.

2. The multi-sensor fusion-based artificial intelligence adaptive control method for solar streetlights according to claim 1, characterized in that: The first meteorological sensor data includes ambient light intensity, photovoltaic module backsheet temperature, ambient humidity, liquid precipitation, and wind speed; the numerical weather prediction data includes cloud cover, probability of precipitation, and temperature.

3. The multi-sensor fusion-based artificial intelligence adaptive control method for solar streetlights according to claim 1, characterized in that: In step S2, the artificial intelligence prediction model is a long short-term memory network model or a time-series fusion converter model, and the current time features are configured to be input into the artificial intelligence prediction model together with the first meteorological sensor data and numerical weather forecast data, so that the model outputs the photovoltaic power generation power sequence and the load power consumption sequence at the same time.

4. The multi-sensor fusion-based artificial intelligence adaptive control method for solar streetlights according to claim 2, characterized in that, Before inputting the ambient light intensity into the artificial intelligence prediction model in step S2, the method further includes calculating the dust attenuation factor using the readings of the dust transmittance sensor on the surface of the photovoltaic module, according to the following formula: ； in, Dust attenuation factor, To calibrate the light transmittance under clean conditions, This represents the current measured transmittance. Dust attenuation factor The ambient light intensity is compensated and corrected, and the corrected ambient light intensity is used as the model input.

5. The multi-sensor fusion-based artificial intelligence adaptive control method for solar streetlights according to claim 1, characterized in that, The preset objective function in step S3 is: ； in, The objective function value, To predict the length of the time domain, for The necessary lighting electricity demand at all times, for The available electricity is based on the state of charge of the energy storage device and the predicted power generation. For reference to comfortable brightness, This refers to the actual output brightness. , These are the weighting coefficients.

6. The multi-sensor fusion-based artificial intelligence adaptive control method for solar streetlights according to claim 1, characterized in that, The charging control strategy generated in step S3 includes: when the prediction result indicates that the number of consecutive rainy days in the future exceeds a set value, and the current state of charge is lower than a safe threshold, the charging cutoff voltage of the energy storage device is increased according to the following relationship: ； in, This is the charging cutoff voltage. The reference cutoff voltage, For reference, the state of charge threshold, The current state of charge, This is for adjusting the coefficient.

7. The multi-sensor fusion-based artificial intelligence adaptive control method for solar streetlights according to claim 1, characterized in that, The discharge control strategy generated in step S3 includes: when a power shortage is predicted, adjusting the lighting brightness during preset off-peak nighttime hours according to the following relationship: ； in, This refers to the actual output brightness. For minimum safe brightness, For full brightness, The current state of charge, To avoid the lowest state of charge for deep discharge, The target is a high-charge state; and the lighting color temperature is switched from the first color temperature to a lower second color temperature.

8. The multi-sensor fusion-based artificial intelligence adaptive control method for solar streetlights according to claim 1, characterized in that, Also includes: S5. Obtain the actual photovoltaic power generation in real time, calculate the deviation between the actual power generation and the predicted power generation in step S2, and perform online incremental training on the artificial intelligence prediction model based on the deviation to update the model parameters.

9. A multi-sensor fusion-based artificial intelligence adaptive control system for solar streetlights, used to implement the multi-sensor fusion-based artificial intelligence adaptive control method for solar streetlights as described in any one of claims 1-8, characterized in that, include: The weather sensor module is used to collect primary weather sensor data. The positioning and communication module is used to acquire geographic location information and numerical weather forecast data for a future preset time period; The battery management module is used to obtain the current state of charge of the energy storage device; The edge computing and control module is configured as follows: It receives the first meteorological sensor data collected by the meteorological sensor module, the numerical weather forecast data obtained by the positioning and communication module, and the current state of charge obtained by the battery management module; Run a pre-trained artificial intelligence prediction model to output the photovoltaic power generation sequence and load power consumption sequence for future periods; Model predictive control rolling optimization is used to minimize a preset objective function, generating charging control commands and discharging control commands. The preset objective function is constructed to minimize the weighted sum of squares of the gap between the necessary lighting power demand and the available power in the future time period, and the weighted sum of squares of the deviation between the reference comfortable brightness and the actual output brightness. The charge / discharge execution module is used to adjust the charging cut-off voltage and charging current according to the charging control command, and to adjust the brightness of the lighting load according to the discharging control command.

Images