A method for intelligent control of photovoltaic modules based on TinyML

Abstract

The application discloses a kind of photovoltaic module intelligent control method based on TinyML, specifically related to photovoltaic system control technical field, pre-training under model quantization control based on original collection data to TinyML framework, according to the spatial distribution of photovoltaic module array surface temperature, the generalization ability of model is evaluated by k-fold cross validation, and the model is optimized by hyperparameter search, for obtaining illumination intensity threshold and temperature threshold, MCU reads the environmental data when photovoltaic module is used, according to different environmental data, MCU adjusts photovoltaic module voltage through dynamic control strategy, and carries out fault diagnosis to photovoltaic module through the time series data of sensor and TinyML algorithm. Through adaptive MPPT algorithm, the energy conversion efficiency of photovoltaic system is significantly improved, and the stability and reliability of the system are improved according to the dynamic adjustment of control strategy.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic system control technology, and in particular to a smart control method for photovoltaic modules based on TinyML. Background Technology

[0002] With the increasing global demand for renewable energy, photovoltaic (PV) systems have been widely adopted due to their clean and renewable characteristics. One of the core components of a PV system is the PV module controller, whose main functions include maximum power point tracking (MPPT), inverter control, data monitoring, and remote management. Traditional PV module controllers typically employ fixed algorithms and hardware designs, which, while meeting basic control requirements, lack intelligence and adaptive capabilities.

[0003] Traditional photovoltaic (PV) module controllers suffer from numerous problems. Most traditional controllers employ fixed MPPT algorithms, such as Perturbation and Observation (P&O) and Incremental Conductivity (INC), which are prone to misjudgments under conditions of significant variations in light and temperature, leading to decreased system efficiency. Existing controllers cannot dynamically adjust control strategies based on environmental changes, resulting in unstable performance in complex environments. Traditional controllers also have weak data processing capabilities, hindering real-time data analysis and intelligent decision-making. Furthermore, traditional controllers consume significant power when handling complex tasks, impacting the overall energy efficiency of the system. All these issues affect the efficiency and safety of PV modules. Summary of the Invention

[0004] In view of the above-mentioned shortcomings of the prior art, the technical problem to be solved by the present invention is to propose a smart control method for photovoltaic modules based on TinyML, which is used to solve the problem that photovoltaic modules cannot be analyzed and controlled in real time in the prior art.

[0005] The technical solution adopted by this invention to solve its technical problem is a smart control method for photovoltaic modules based on TinyML, characterized by including the following steps:

[0006] S1. Based on the original collected data, the TinyML framework is pre-trained under model quantization control. According to the spatial distribution of surface temperature of photovoltaic module array, the generalization ability of the model is evaluated by k-fold cross-validation. The model is optimized by hyperparameter search to obtain the light intensity threshold and temperature threshold.

[0007] S2. Deploy the trained TinyML algorithm to the MCU. The MCU reads the environmental data of the photovoltaic module during use. When the environmental data is higher than the light intensity threshold, the MCU adjusts the photovoltaic module voltage using the perturbation-observation method; when the environmental data is lower than the light intensity threshold, the MCU adjusts the photovoltaic module voltage using the incremental conductance method; when the environmental data is higher than the temperature threshold, the MCU adjusts the photovoltaic module voltage using the TinyML algorithm; when the environmental data is lower than the temperature threshold, the MCU adjusts the photovoltaic module voltage using both the perturbation-observation method and the incremental conductance method simultaneously.

[0008] S3. When the MCU adjusts the voltage of the photovoltaic module, the MCU uses the timing data of the sensor and the TinyML algorithm to diagnose the fault of the photovoltaic module. When the photovoltaic module fails, the MCU activates the alarm system and starts the edge bypass.

[0009] Furthermore, before step S1, TinyML is pruned using TensorFlow to remove redundant weights and connections.

[0010] Furthermore, when the environmental data exceeds the temperature threshold, the voltage and current of the photovoltaic module are adjusted by TinyML to compensate for the temperature of the photovoltaic module.

[0011] Furthermore, when the environmental data is lower than the light intensity, the photovoltaic module is controlled by the MCU to switch to a low-power mode at the microampere level.

[0012] Furthermore, the low-power mode switches the capacitor to a miniature solid-state capacitor via the LDSW communication unit.

[0013] Furthermore, the photovoltaic modules are arranged in a grid pattern. When the environmental data of one of the photovoltaic modules changes, the MPPT algorithm is used to select the appropriate module through TinyML.

[0014] Furthermore, the color changes of each voxel in different directions are captured by the spherical harmonic function, and the time-series data includes statistical features, frequency domain features, and time domain features.

[0015] Furthermore, the perturbation observation method continuously applies perturbations to the operating voltage of the photovoltaic module, and adjusts the photovoltaic module voltage accordingly based on the output power of the photovoltaic module under different voltages.

[0016] Furthermore, the incremental conductivity method monitors the conductivity changes of the photovoltaic module and obtains the maximum power point based on the voltage-current characteristic curve of the photovoltaic module, which is then used to adjust the voltage of the photovoltaic module.

[0017] Furthermore, the TinyML algorithm adjusts the photovoltaic module voltage based on the nonlinear relationship between the environmental data and the output power.

[0018] Compared with the prior art, the present invention has at least the following beneficial effects:

[0019] (1) The intelligent control method and system for photovoltaic modules based on TinyML described in this invention enables the controller to dynamically adjust the control strategy according to environmental changes, and significantly improves the energy conversion efficiency of the photovoltaic system through the adaptive MPPT algorithm. This, in turn, enhances the stability and reliability of the system.

[0020] (2) The low-power mode design extends the lifespan of the controller and reduces maintenance costs;

[0021] (3) The system adopts intelligent fault detection and diagnosis functions to effectively prevent and reduce system failures and improve system security. Attached Figure Description

[0022] Figure 1 This is a flowchart of a smart control method for photovoltaic modules based on TinyML.

[0023] Figure 2 This is a schematic diagram of a photovoltaic module intelligent control system based on TinyML. Detailed Implementation

[0024] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings. However, the present invention is not limited to these embodiments.

[0025] Example 1

[0026] To enable real-time data analysis and intelligent control of photovoltaic modules, such as Figure 1 As shown, the present invention discloses a smart control method for photovoltaic modules based on TinyML, characterized by comprising the following steps:

[0027] S1. Based on the original collected data, the TinyML framework is pre-trained under model quantization control. According to the spatial distribution of surface temperature of photovoltaic module array, the generalization ability of the model is evaluated by k-fold cross-validation. The model is optimized by hyperparameter search to obtain the light intensity threshold and temperature threshold.

[0028] S2. Deploy the trained TinyML algorithm to the MCU. The MCU reads the environmental data of the photovoltaic module during use. When the environmental data is higher than the light intensity threshold, the MCU adjusts the photovoltaic module voltage using the perturbation-observation method; when the environmental data is lower than the light intensity threshold, the MCU adjusts the photovoltaic module voltage using the incremental conductance method; when the environmental data is higher than the temperature threshold, the MCU adjusts the photovoltaic module voltage using the TinyML algorithm; when the environmental data is lower than the temperature threshold, the MCU adjusts the photovoltaic module voltage using both the perturbation-observation method and the incremental conductance method simultaneously.

[0029] S3. When the MCU adjusts the voltage of the photovoltaic module, the MCU uses the timing data of the sensor and the TinyML algorithm to diagnose the fault of the photovoltaic module. When the photovoltaic module fails, the MCU activates the alarm system and starts the edge bypass.

[0030] The microprocessor reads raw data such as current, voltage, temperature, and gravitational acceleration via an ADC (Analog-to-Digital Converter), as well as environmental parameter data such as wind speed and humidity. Low-pass filtering and mean filtering are used to remove noise from this data. Interpolation or averaging methods are used to fill in missing values ​​in the environmental and operational data collected by the sensors, ensuring the completeness and continuity of the sensor data used for analysis and control. Obviously erroneous measurements are identified and removed, thus cleaning the raw data. The cleaned data is scaled to the range [0,1] and converted to a distribution with a mean of 0 and a standard deviation of 1. After verification, cleaning, and normalization, the data is used as the input vector for the MPPT algorithm model, improving the algorithm's accuracy.

[0031] The TinyML framework was pre-trained under model quantization control based on the raw collected data. The time-series data was processed, including statistical, frequency domain, and temporal features. Specifically, statistical measures such as mean, variance, maximum, and minimum were calculated, and the frequency components of these measures were extracted using Fast Fourier Transform. Features such as power, energy, and zero-crossing rate were also calculated. These statistical measures summarize the overall characteristics of the time-series data, helping the model better understand the data distribution. Frequency domain features help capture periodicity and oscillation patterns in the data, enabling the model to capture the time dependence and dynamic change patterns of the data.

[0032] Furthermore, the photovoltaic modules are arranged in a grid pattern. Based on the spatial distribution of surface temperature of the photovoltaic module array, and combined with data from multiple sensors, Kalman filtering, fuzzy logic, and other techniques are used to improve the accuracy and reliability of the data. A lightweight machine learning model is trained based on the extracted features, enabling it to run on a resource-constrained microcontroller. The model is first simplified. TensorFlow performs pruning operations on TinyML to remove redundant weights and connections. The TensorFlow Model Optimization Toolkit is used to prune the model, removing unimportant weights and connections to reduce the model size. Floating-point weights are converted to low-precision fixed-point numbers (such as 8-bit integers) to reduce model size and computational complexity, and unimportant weights or connections are removed, further compressing the model. In the simplified model, the data is divided into training and validation sets. The generalization ability of the model is evaluated by k-fold cross-validation. By splitting the data multiple times for training and validation, a more stable and reliable performance estimate is obtained. The model is optimized by hyperparameter search to find the best combination of hyperparameters to improve the model's performance. Based on the trained TinyML framework model, the light intensity threshold and temperature threshold are obtained.

[0033] Furthermore, the trained TinyML is deployed into the MCU. Through component integration and platform environmental sensors, the controller edge collects input voltage, current, and temperature data in real time, combined with environmental data such as light intensity collected by the system. When the environmental data of one of the photovoltaic modules changes, TinyML performs MPPT algorithm selection. When the environmental data is higher than the light intensity threshold, the MCU adjusts the photovoltaic module voltage using a perturbation-observation method, continuously applying perturbations to the photovoltaic module's operating voltage. Based on the output power of the photovoltaic module under different voltages, the voltage is adjusted accordingly to ensure the system can respond quickly and maintain efficient energy conversion under high light conditions. When the environmental data is lower than the light intensity threshold, the MCU adjusts the photovoltaic module voltage using an incremental conductance method. It monitors changes in the photovoltaic module's conductance, obtains the maximum power point based on the photovoltaic module's voltage-current characteristic curve, compares the obtained conductance with the current-voltage ratio, and adjusts the photovoltaic array's operating voltage accordingly until it finds and stabilizes at the maximum power point, thus adjusting the photovoltaic module voltage. To improve system stability and accuracy, ensuring accurate tracking of the maximum power point even under low light conditions; when environmental data exceeds the temperature threshold, the MCU adjusts the photovoltaic module voltage using the TinyML algorithm, adjusting the voltage based on the nonlinear relationship between environmental data and output power, effectively handling complex nonlinear relationships and ensuring efficient energy conversion even at high temperatures; when environmental data falls below the temperature threshold, the MCU simultaneously adjusts the photovoltaic module voltage using both the perturbation-observation method and the incremental conductance method, employing a hybrid MPPT strategy. This strategy combines the advantages of both algorithms, alternately executing the perturbation-observation and incremental conductance steps, determining the direction of the next adjustment based on the result of the previous step. Once the ambient temperature returns to normal, the MCU can switch back to a single MPPT method, either the perturbation-observation method or the incremental conductance method, simplifying the control logic to improve system stability and response speed under low-temperature conditions. Utilizing a trained TinyML algorithm model, the maximum power point tracking strategy is dynamically optimized based on real-time changes in light intensity and ambient temperature.

[0034] Furthermore, the system monitors the real-time operating status of photovoltaic modules, including parameters such as voltage, current, and temperature. Through wireless communication technologies such as LDSW or LoRa, the data is transmitted to the cloud for analysis and management. By utilizing sensor timing data, the TinyML algorithm, and machine learning data analysis capabilities on the cloud platform, the controller can perform fault detection and diagnosis tasks. When the MCU adjusts the photovoltaic module voltage, it uses sensor timing data and the TinyML algorithm to diagnose faults in the photovoltaic modules. These fault detection technologies are based on comprehensive analysis of current, voltage, temperature, power, and environmental timing data, combined with machine learning capabilities. Specific thresholds and operating trend analyses are set for different types of modules. When a photovoltaic module malfunctions, depending on the fault type, the system activates an alarm via the MCU and automatically initiates an edge bypass or disconnect control mechanism. The microprocessor cuts off the power supply, preventing damage through relays or protection circuits. Simultaneously, the cloud platform provides diagnostic suggestions and guidance services such as cleaning, repair, and replacement, assisting users in quickly resolving problems, effectively preventing and reducing system failures, and improving system security.

[0035] Furthermore, to ensure that the photovoltaic modules always operate in optimal condition, TinyML employs cutting-edge technologies such as fine-tuning step size, integrating temperature compensation mechanisms, and real-time data fusion to accurately predict the voltage value at the maximum power point. This allows for temperature compensation by adjusting the voltage and current of the photovoltaic modules. By using low-power microprocessors, sensors, and communication units, and optimizing hardware design, overall energy consumption is effectively reduced. In low-light or nighttime environments, where ambient light intensity is lower than the controller's operating threshold, the MCU automatically switches the photovoltaic modules to a microampere-level low-power mode. Through the ultra-low-power technology of the LDSW communication unit, requiring only miniature solid-state capacitors, continuous monitoring operation for over 48 hours is guaranteed, extending the controller's lifespan and reducing maintenance costs.

[0036] Example 2

[0037] In a second aspect, the present invention provides a smart control system for photovoltaic modules based on TinyML, such as... Figure 2 As shown, the system performs intelligent control operations on the corresponding photovoltaic modules according to the modules.

[0038] The system includes:

[0039] The MCU is used to acquire data, which involves reading data from various sensors and processing the data. Data processing includes executing the MPPT algorithm and fault detection, and controlling the output to adjust the voltage and current of the MPPT controller.

[0040] Sensors are used to collect various data, including: current sensors for measuring the output current of photovoltaic modules; voltage sensors for measuring the input voltage of modules and the output voltage of the bus; temperature sensors for measuring the surface temperature of modules and the ambient temperature; acceleration sensors for measuring the installation tilt angle and vibration of modules; and system environment sensors for measuring environmental parameters such as wind speed, air pressure, humidity, and ambient temperature.

[0041] The communication module, which can be either LDSW narrowband trunking or PLC power line carrier, is used for internal system communication to achieve wireless data transmission.

[0042] The power management module is used to supply power, ensuring the stable operation of the MCU and sensors, while also managing the charging and discharging process of the micro battery.

[0043] The storage module has an internal FLASH memory used to store sensor data, log files, configuration parameters, etc., and the storage capacity can be increased by expanding the EEPROM.

[0044] The microprocessor chosen is the ARM Cortex-M4 microprocessor, which features low power consumption and high performance. The microprocessor, sensors, voltage regulation inverter circuit, communication module, etc. are integrated into a compact package and installed in the junction box of the photovoltaic module, thereby realizing real-time data monitoring and remote management functions, allowing users to understand the system status at any time and improve operation and maintenance efficiency.

[0045] In summary, the present invention provides a TinyML-based intelligent control method for photovoltaic modules. This method pre-trains the TinyML framework under model quantization control based on raw collected data. According to the spatial distribution of surface temperature on the photovoltaic module array, the generalization ability of the model is evaluated through k-fold cross-validation, and the model is optimized through hyperparameter search to obtain light intensity and temperature thresholds. The trained TinyML is then deployed to an MCU. The MCU reads environmental data during photovoltaic module operation. When the environmental data is higher than the light intensity threshold, the MCU adjusts the photovoltaic module voltage using a perturbation-observation method; when the environmental data is lower than the light intensity threshold, the MCU adjusts the photovoltaic module voltage using an incremental conductance method; when the environmental data is higher than the temperature threshold, the MCU adjusts the photovoltaic module voltage using the TinyML algorithm; when the environmental data is lower than the temperature threshold, the MCU adjusts the photovoltaic module voltage simultaneously using both the perturbation-observation method and the incremental conductance method. While adjusting the photovoltaic module voltage, the MCU performs fault diagnosis on the photovoltaic module using time-series data from sensors and the TinyML algorithm. When a fault occurs in the photovoltaic module, the MCU activates an alarm system and initiates edge bypass. The adaptive MPPT algorithm significantly improves the energy conversion efficiency of the photovoltaic system. The controller can dynamically adjust the control strategy according to environmental changes, improving the stability and reliability of the system. In low light or nighttime environments, it automatically uses a low-power mode, extending the life of the controller and reducing maintenance costs. Operators can monitor data in real time and manage remotely, allowing users to understand the system status at any time and improving operation and maintenance efficiency. It can also perform intelligent fault detection and diagnosis, effectively preventing and reducing system failures and improving system security.

[0046] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0047] Furthermore, in this invention, descriptions involving terms such as "first," "second," and "a" are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0048] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0049] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

Claims

1. A smart control method for photovoltaic modules based on TinyML, characterized in that, Including the following steps: S1. Based on the original collected data, the TinyML framework is pre-trained under model quantization control. According to the spatial distribution of surface temperature of photovoltaic module array, the generalization ability of the model is evaluated by k-fold cross-validation. The model is optimized by hyperparameter search to obtain the light intensity threshold and temperature threshold. S2. Deploy the trained TinyML algorithm to the MCU. The MCU reads the environmental data of the photovoltaic module during use. When the environmental data is higher than the light intensity threshold, the MCU adjusts the photovoltaic module voltage using the perturbation-observation method; when the environmental data is lower than the light intensity threshold, the MCU adjusts the photovoltaic module voltage using the incremental conductance method; when the environmental data is higher than the temperature threshold, the MCU adjusts the photovoltaic module voltage using the TinyML algorithm; when the environmental data is lower than the temperature threshold, the MCU adjusts the photovoltaic module voltage using both the perturbation-observation method and the incremental conductance method simultaneously. S3. When the MCU adjusts the voltage of the photovoltaic module, the MCU uses the timing data of the sensor and the TinyML algorithm to diagnose the fault of the photovoltaic module. When the photovoltaic module fails, the MCU activates the alarm system and starts the edge bypass.

2. The intelligent control method for photovoltaic modules based on TinyML according to claim 1, characterized in that, Before step S1, TinyML is pruned using TensorFlow to remove redundant weights and connections.

3. The intelligent control method for photovoltaic modules based on TinyML according to claim 1, characterized in that, When the environmental data exceeds the temperature threshold, the voltage and current of the photovoltaic module are adjusted by TinyML to compensate for the temperature of the photovoltaic module.

4. The intelligent control method for photovoltaic modules based on TinyML according to claim 1, characterized in that, When the environmental data is lower than the light intensity, the photovoltaic module is controlled by the MCU to switch to a low-power mode at the microampere level.

5. The intelligent control method for photovoltaic modules based on TinyML according to claim 4, characterized in that, The low-power mode switches the capacitor to a miniature solid-state capacitor via the LDSW communication unit.

6. The intelligent control method for photovoltaic modules based on TinyML according to claim 1, characterized in that, The photovoltaic modules are arranged in a grid pattern. When the environmental data of one of the photovoltaic modules changes, the MPPT algorithm is used to select the appropriate module using TinyML.

7. The intelligent control method for photovoltaic modules based on TinyML according to claim 1, characterized in that, The time-series data includes statistical features, frequency domain features, and time domain features.

8. The intelligent control method for photovoltaic modules based on TinyML according to claim 1, characterized in that, The perturbation observation method continuously applies perturbations to the operating voltage of the photovoltaic module, and adjusts the photovoltaic module voltage accordingly based on the output power of the photovoltaic module under different voltages.

9. The intelligent control method for photovoltaic modules based on TinyML according to claim 1, characterized in that, The incremental conductivity method monitors the conductivity changes of the photovoltaic module and obtains the maximum power point based on the voltage-current characteristic curve of the photovoltaic module, which is then used to adjust the voltage of the photovoltaic module.

10. The intelligent control method for photovoltaic modules based on TinyML according to claim 1, characterized in that, The TinyML algorithm adjusts the photovoltaic module voltage based on the nonlinear relationship between the environmental data and the output power.

Images