MPPT (Maximum Power Point Tracking) control method and system based on adaptive fuzzy neural network

Abstract

The invention relates to the technical field of photovoltaic power generation system control, and provides an MPPT control method and system based on an adaptive fuzzy neural network, and the method comprises the steps: synchronously collecting voltage, current, illumination and temperature signals, and generating original sensing data; transmitting the generated original sensing data to a memory mapping area through a hardware channel; sensing data is obtained from the memory mapping area, calculation is executed, a power variation and an illumination variation are predicted, a fuzzy reasoning algorithm is operated based on the power variation and the illumination variation, and a power point decision scheme is obtained; a fuzzy rule is updated based on the predicted power variation and illumination variation, the updated fuzzy rule is fed back to the dual-core heterogeneous processor in real time, and fuzzy reasoning logic is optimized; and driving the four-phase interlaced power topological structure to work in an interlaced manner according to the power point decision scheme so as to realize maximum power point tracking. According to the invention, rapid tracking and high-precision stable control of the maximum power point of the photovoltaic array in a complex environment can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic power generation system control, and particularly relates to an MPPT control method and system based on an adaptive fuzzy neural network. BACKGROUND

[0002] Conventional MPPT algorithms (such as incremental conductance method and perturbation and observation method) have problems such as high steady-state oscillation rate (>1.5%), large dynamic response delay (>150 ms), and local shadow tracking failure rate >30% in a nonlinear scene. Moreover, the conventional MPPT scheme also has the following problems: Fuzzy control improves environmental adaptability, but its fixed rule base is difficult to cover the multi-peak IV curve scene, and the success rate under local shadow is less than 70%; Intelligent optimization algorithms (particle swarm, genetic algorithm) have high computational complexity and are difficult to run in real time on an embedded platform; Neural network control relies on a large amount of training data, and the model generalization ability is limited by hardware resources. SUMMARY

[0003] The present application aims to solve at least one of the technical problems in the background art, and provides an MPPT control method and system based on an adaptive fuzzy neural network.

[0004] To achieve the above-mentioned purpose, the present application provides an MPPT control method based on an adaptive fuzzy neural network, comprising: Synchronously collecting voltage, current, light, and temperature signals through a multi-modal sensing module to generate original sensing data; Transferring the original sensing data generated by the multi-modal sensing module to a memory mapping area through a hardware channel through a DMA zero-copy transmission unit; A dual-core heterogeneous processor acquires sensing data from the memory mapping area of the DMA zero-copy transmission unit, performs calculation, predicts power variation and light variation, runs a fuzzy reasoning algorithm based on the power variation and the light variation, and acquires a power point decision scheme; A dynamic rule base acquires the predicted power variation and the light variation, updates fuzzy rules based on the predicted power variation and the light variation of the dual-core heterogeneous processor, and feeds back the updated fuzzy rules to the dual-core heterogeneous processor in real time to optimize fuzzy reasoning logic; Driving a four-phase interleaved power topology structure to work in an interleaved manner according to the power point decision scheme to realize maximum power point tracking.

[0005] According to one aspect of the present application, the sampling rate of the multi-modal sensing module is ≥200 kHz.

[0006] According to one aspect of the present invention, the dual-core heterogeneous processor includes a Cortex-M4 core and a Cortex-M7 core; The Cortex-M4 core obtains sensor data from the memory-mapped area of ​​the DMA zero-copy transfer unit, performs calculations, and predicts changes in power and illumination. The Cortex-M7 core uses the power change and illumination change output by the Cortex-M4 core to run a fuzzy inference algorithm to obtain a power point decision scheme. Specifically, when the power change of the Cortex-M4 core is greater than 15W or the light change is greater than 1000Lux / s, the Cortex-M7 core is triggered to run the fuzzy inference algorithm.

[0007] According to one aspect of the present invention, the dynamically expanded rule base updates fuzzy rules based on power and illumination changes predicted by a dual-core heterogeneous processor, including: The dynamically expanding rule base continuously checks whether the power change of the Cortex-M4 core output is greater than 15W or the light change is greater than 1000Lux / s through an independent thread or interrupt service routine; When the power change is greater than 15W or the illumination change is greater than 1000Lux / s, the dynamic expansion rule base collects sensor data during periods of abrupt power or illumination changes and generates new fuzzy rules through a decision tree algorithm. Check whether the new fuzzy rule is redundant or conflicting with the existing fuzzy rules in the dynamically expanded rule base. If the new fuzzy rule has logical overlap with the existing fuzzy rule, merge them. If the new fuzzy rule conflicts with the existing fuzzy rule, overwrite the existing fuzzy rule with the new fuzzy rule.

[0008] According to one aspect of the present invention, the four-phase interleaved power topology includes four Boost circuits connected in parallel. Based on the power point decision scheme provided by the dual-core heterogeneous processor, the switching signals of each Boost circuit are staggered in time to achieve phase interleaving, thereby realizing maximum power point tracking.

[0009] According to one aspect of the present invention, the Cortex-M4 core obtains sensor data from the memory-mapped area of ​​the DMA zero-copy transfer unit, performs neural network calculations, and predicts power changes and illumination changes. The neural network parameter tuning layer is a 3-layer backpropagation structure, including: The input layer has 6 nodes: receiving the fuzzy output ΔP / Δt, ΔL / Δt, ΔT / Δt and the historical error mean; where ΔP is the power change, ΔL is the illumination change, Δt is the time change, and ΔT is the temperature change; The hidden layer has 12 nodes: the LeakyReLU activation function is used, with a decay coefficient α = 0.01; The output layer has three nodes: the output PWM duty cycle adjustment amount ΔD∈[-5%,+5%], the membership function parameter correction amount, and the learning rate adjustment instruction.

[0010] According to one aspect of the invention, the fuzzy inference algorithm employs an adjustable Gaussian membership function. ; Where x represents the variable input to the fuzzy inference algorithm, the center value c is generated by K-means clustering, and the variance σ is adaptively adjusted in the interval [0.1, 0.5].

[0011] To achieve the above objectives, the present invention also provides an MPPT control system based on an adaptive fuzzy neural network, comprising: The multimodal sensing module synchronously collects voltage, current, light, and temperature signals to generate raw sensing data. The DMA zero-copy transfer unit transmits the raw sensing data collected and generated by the multimodal sensing module to the memory-mapped area through a hardware channel; The dual-core heterogeneous processor acquires sensor data from the memory mapping area of ​​the DMA zero-copy transfer unit, performs calculations, predicts power changes and illumination changes, and runs a fuzzy inference algorithm based on the power changes and illumination changes to obtain a power point decision scheme. The rule base is dynamically expanded. Based on the power and illumination changes predicted by the dual-core heterogeneous processor, the fuzzy rules are updated and the updated fuzzy rules are fed back to the dual-core heterogeneous processor in real time to optimize the fuzzy inference logic. The four-phase interleaved power topology, based on the power point decision scheme given by the dual-core heterogeneous processor, operates in an interleaved manner to achieve maximum power point tracking.

[0012] To achieve the above objectives, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the MPPT control method based on an adaptive fuzzy neural network as described above.

[0013] To achieve the above objectives, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the MPPT control method based on an adaptive fuzzy neural network as described above.

[0014] According to the present invention, a collaborative optimization mechanism of fuzzy logic and neural networks is used to achieve rapid maximum power point tracking and high-precision stable control of photovoltaic arrays under complex environments. The present invention achieves breakthrough performance with tracking efficiency ≥99.2%, response time ≤80ms, and steady-state power oscillation amplitude <1.5% through a dynamic fuzzy rule expansion mechanism, a lightweight neural network architecture, and dual-core collaborative computing technology. This reduces energy consumption by 40% compared to traditional methods and is applicable to photovoltaic power plants, distributed energy storage, and new energy vehicles. Attached Figure Description

[0015] Figure 1 The flowchart illustrates an MPPT control method based on an adaptive fuzzy neural network according to an embodiment of the present invention. Detailed Implementation

[0016] The invention will now be discussed with reference to exemplary embodiments. It should be understood that the described embodiments are merely intended to enable those skilled in the art to better understand and thus implement the invention, and are not intended to imply any limitation on the scope of the invention.

[0017] As used herein, the term "comprising" and its variations are to be interpreted as open-ended terms meaning "including but not limited to". The term "based on" is to be interpreted as "at least partially based on". The terms "one embodiment" and "an embodiment" are to be interpreted as "at least one embodiment".

[0018] Figure 1 The flowchart schematically illustrates an MPPT control method based on an adaptive fuzzy neural network according to an embodiment of the present invention. Figure 1 As shown, in this embodiment, the MPPT control method based on an adaptive fuzzy neural network includes: The multimodal sensing module synchronously collects voltage, current, light, and temperature signals to generate raw sensing data. The raw sensing data acquired and generated by the multimodal sensing module is transmitted to the memory-mapped area via a hardware channel through the DMA zero-copy transfer unit. The dual-core heterogeneous processor obtains sensor data from the memory mapping area of ​​the DMA zero-copy transfer unit, performs calculations, predicts power changes and illumination changes, runs a fuzzy inference algorithm based on the power changes and illumination changes, and obtains a power point decision scheme (PWM duty cycle instruction). The rule base is dynamically expanded to obtain the predicted power change and illumination change, and the fuzzy rules are updated based on the power change and illumination change predicted by the dual-core heterogeneous processor. The updated fuzzy rules are fed back to the dual-core heterogeneous processor in real time to optimize the fuzzy inference logic. The four-phase interleaved power topology is driven to operate interleaved according to the power point decision scheme to achieve maximum power point tracking.

[0019] Furthermore, according to one embodiment of the present invention, the sampling rate of the multimodal sensing module is ≥200 kHz. In this embodiment, synchronous acquisition by the multimodal sensing module means that all signals are measured at the same point in time (reducing errors caused by asynchrony), which is suitable for dynamic systems (such as motor monitoring). This requires the module's hardware (sensors and ADC) to work together and the signal transmission path to be balanced.

[0020] In this implementation, the ADC sampling rate refers to the number of samples per second (sampling rate ≥ 200 kHz = 200,000 times / second) to ensure that signal details are not lost (according to Nyquist's theorem, the sampling rate must be greater than twice the highest frequency of the signal). For example, if acquiring a 100 kHz high-frequency signal (such as switching power supply ripple), a sampling rate of 200 kHz meets the minimum requirement (to avoid aliasing distortion).

[0021] A high sampling rate can also improve synchronization: all channels (voltage, current, etc.) are sampled at the same rate to ensure data point time alignment (microsecond-level delay).

[0022] Furthermore, according to one embodiment of the present invention, the DMA zero-copy transfer unit moves data directly from the source (sensor) to the target memory area (0x2400_0000) without passing through the CPU's internal cache or temporary buffer. This reduces additional steps in data transfer (e.g., the overhead of data transfer in CPU registers), thereby optimizing latency.

[0023] Because DMA is managed by a hardware controller (rather than the CPU core executing instructions), the CPU only needs to initialize the DMA settings (such as configuring the source and destination addresses) and can then perform other tasks in parallel, with CPU utilization ≤5%. The brief overhead of initializing DMA (triggering interrupts or setting registers) accounts for a tiny percentage of the entire transfer cycle.

[0024] The memory-mapped area (0x2400_0000) is typically associated with a specific peripheral or reserved memory (in ARM Cortex-M series chips, it represents the virtual address of the physical device). This allows the system to directly access this area for efficient read and write operations (sensor data can be processed by the algorithm module immediately after being written).

[0025] In this implementation, zero-copy combined with a dedicated memory mapping area reduces CPU load by approximately 80-95% (compared to traditional methods), especially in high-bandwidth applications (audio or video transmission). When the sensor outputs 1MB of data per second, DMA can transfer data at near the hardware limit while CPU utilization remains stable at a low level.

[0026] Furthermore, according to one embodiment of the present invention, the dual-core heterogeneous processor includes a Cortex-M4 core and a Cortex-M7 core; The Cortex-M4 core obtains sensor data from the memory-mapped area of ​​the DMA zero-copy transfer unit, performs calculations, and predicts changes in power and illumination. The Cortex-M7 core uses the power change and illumination change output by the Cortex-M4 core to run a fuzzy inference algorithm to obtain a power point decision scheme. Specifically, when the power change of the Cortex-M4 core is greater than 15W or the light change is greater than 1000Lux / s, the Cortex-M7 core is triggered to run the fuzzy inference algorithm.

[0027] In this embodiment, inter-core collaboration is achieved through the HSEM semaphore synchronization task, which utilizes 768KB of shared RAM to exchange intermediate results (e.g., the M4-identified △P>15W event triggers M7 inference).

[0028] Furthermore, according to one embodiment of the present invention, the dynamically expanded rule base updates fuzzy rules based on the power change and illumination change predicted by the dual-core heterogeneous processor, including: The dynamically expanding rule base continuously checks whether the power change of the Cortex-M4 core output is greater than 15W or the light change is greater than 1000Lux / s through an independent thread or interrupt service routine; When the power change is greater than 15W or the illumination change is greater than 1000Lux / s, the dynamic expansion rule base collects sensor data during periods of abrupt power or illumination changes and generates new fuzzy rules through a decision tree algorithm. Check whether the new fuzzy rule is redundant or conflicting with the existing fuzzy rules in the dynamically expanded rule base. If the new fuzzy rule has logical overlap with the existing fuzzy rule, merge them. If the new fuzzy rule conflicts with the existing fuzzy rule, overwrite the existing fuzzy rule with the new fuzzy rule.

[0029] Redundancy refers to the logical overlap between new fuzzy rules and existing fuzzy rules in the rule base in terms of input conditions and output conclusions. In other words, the control logic of the two is essentially the same or can be completely covered by existing rules.

[0030] Redundancy includes: Input redundancy: The fuzzy input interval (range of error and error rate of change) of the new fuzzy rule is a subset of the input interval of the existing fuzzy rule, and the output conclusions are completely consistent.

[0031] Output redundancy: The input conditions of the new fuzzy rule are exactly the same as those of the existing fuzzy rule, but the output conclusions are the same (the membership function or control strategy is completely consistent).

[0032] A conflict occurs when a new fuzzy rule and an existing fuzzy rule produce different output conclusions under the same input conditions, making it impossible for the system to clearly define which rule to execute.

[0033] Conflicts include: Direct conflict: The new fuzzy rule has exactly the same input conditions as the existing fuzzy rule, but the output conclusions are contradictory (e.g., one rule outputs an increase in the duty cycle, while another rule outputs a decrease in the duty cycle).

[0034] Indirect conflict: Input conditions partially overlap (not a full inclusion relationship), but the output conclusions in the overlapping areas are inconsistent (e.g., the membership assignment of overlapping intervals leads to opposite control strategies).

[0035] In this embodiment, the dynamically expandable rule base uses a dynamically scalable storage structure (database table, in-memory hash table, or linked list) to store the initial rules. Each rule contains: Input variables (environmental parameters, device status); Output actions (adjust power, adjust lighting); Membership function (defining the threshold for fuzzy logic); In this embodiment, during initialization, at least 50 expert experience rules are preloaded into the dynamically expanded rule base (IF illumination > 500 Lux THEN reduce power by 10%), and persistent storage is achieved through a database or configuration file.

[0036] In this embodiment, sudden environmental changes are monitored in real time, and sensors are deployed to collect data in real time: △P (Power Change Rate): Calculated per second, current power - power of the previous second; △L (rate of change of illumination): (current illumination value - illumination value of the previous second) / time interval; Continuous checks via a separate thread or interrupt service routine (ISR): if (△P > 15W) or (△L > 1000 Lux / s): activate_rule_expansion() # Starts the process of adding a rule; In this implementation, the logic for generating new rules is as follows: Collect data streams during periods of abrupt change and generate new rules using a decision tree algorithm; For example, "IF Illumination increases suddenly (ΔL>1000) AND time = daytime THEN turn off auxiliary lighting"; Manual assistance can be used to extend the system, triggering alerts and prompting administrators to review the newly added rules, thus ensuring system security.

[0037] In this implementation, the new rules are appended and stored in a structured format (JSON / XML), while a fast retrieval index is built based on a hash index of environment parameters.

[0038] In this implementation, when adding a new rule, redundancy or conflict with existing rules is checked: If the new rules and old rules have overlapping logic, the membership functions are merged. If a conflict occurs and the opposite operation is performed, the old rule is overridden based on the confidence weight.

[0039] Capacity control: The rule base is dynamically expanded with a maximum of 500 rules, triggering an elimination mechanism: deleting infrequently used rules or merging similar rules.

[0040] Furthermore, according to one embodiment of the present invention, the four-phase interleaved power topology includes four Boost circuits (four-phase Boost circuits), which are connected in parallel. Based on the power point decision scheme given by the dual-core heterogeneous processor, the switching signals of each Boost circuit are staggered in time to achieve phase interleaving, thereby realizing maximum power point tracking.

[0041] In this embodiment, the four-phase Boost circuit operates in an interleaved manner (each phase offset by 90°), combined with GaN switches (high frequency, low loss), a 22μH inductor (energy storage), and a 470μF capacitor (filtering) to achieve efficient power conversion. This reduces input current ripple (through phase interleaving), improves efficiency (>95%), and allows for rapid response to processor instructions, adjusting the output voltage (within a 100V range) to track the maximum power point.

[0042] Furthermore, according to one embodiment of the present invention, the Cortex-M4 core obtains sensor data from the memory-mapped area of ​​the DMA zero-copy transfer unit, performs neural network calculations, and predicts the power change and the illumination change. The neural network parameter tuning layer is a 3-layer backpropagation structure, including: The input layer has 6 nodes: receiving the fuzzy output ΔP / Δt, ΔL / Δt, ΔT / Δt and the historical error mean; where ΔP is the power change, ΔL is the illumination change, Δt is the time change, and ΔT is the temperature change; The hidden layer has 12 nodes: the LeakyReLU activation function is used, with a decay coefficient α = 0.01; The output layer has three nodes: the output PWM duty cycle adjustment amount ΔD∈[-5%,+5%], the membership function parameter correction amount, and the learning rate adjustment instruction.

[0043] The necessity of the three-layer backpropagation structure in this embodiment: a. Input layer (6 nodes): Integrates multi-dimensional dynamic information The input includes fuzzified rates of change of physical quantities (ΔP / Δt, ΔL / Δt, ΔT / Δt) and historical error mean, covering the system's real-time dynamic characteristics (rate of change of power / rate of change of illumination / rate of change of temperature) and historical control performance (error mean). This design enables the network to capture both short-term fluctuations (Δt-related variables) and long-term trends (historical errors) of the system, providing a comprehensive basis for parameter adjustment decisions.

[0044] b. Hidden layer (12 nodes + LeakyReLU): Enhances nonlinear fitting capability The 12 nodes balance model complexity and generalization ability, avoiding overfitting; the LeakyReLU activation function (α=0.01) solves the "neuron death" problem of traditional ReLU, ensuring the effective transmission of negative input signals and improving the network's fitting accuracy for complex nonlinear relationships, making it particularly suitable for uncertainty modeling of dynamic systems.

[0045] c. Output layer (3 nodes): Multi-objective collaborative optimization Simultaneously output three types of key parameters: Control command (PWM duty cycle adjustment ΔD): directly acts on the actuator to achieve real-time closed-loop control; Model parameter correction (membership function parameters): Optimize the core components of the fuzzy system (membership function shape) to improve the adaptability of the fuzzy model; Learning rate adjustment command: Dynamically control the training step size of the neural network itself to avoid getting stuck in local optima or oscillations, and achieve adaptation at the meta-learning level.

[0046] In this implementation, the role of the backpropagation algorithm is: By optimizing network weights through gradient descent, system performance indicators (error, stability) are transformed into differentiable loss functions, enabling quantitative and automated parameter tuning. This differs from traditional heuristic parameter tuning methods, improving optimization efficiency and accuracy.

[0047] Traditional control systems (such as PID controllers) rely on fixed parameters or manual parameter tuning, making it difficult to cope with environmental changes (such as changes in light intensity or temperature). This implementation uses a neural network to analyze dynamic data in real time, adjust the control output (ΔD) online, and optimize the membership function parameters of the fuzzy model, enabling the system to have "self-evolution" capabilities and adapt to nonlinear and time-varying scenarios.

[0048] The learning rate adjustment instruction in the output layer further realizes the meta-learning mechanism: when the system error is large, the learning rate is increased (fast convergence), and when the error is small, the learning rate is decreased (fine adjustment), avoiding the limitations of the traditional fixed learning rate.

[0049] In this embodiment, the rate of change of physical quantities and historical errors integrated by the input layer are combined with: Causal relationship (ΔP / Δt reflects the correlation between system input and output); Dynamic characteristics (ΔL / Δt, ΔT / Δt characterize load and thermal response rate); Empirical knowledge (historical error mean reflects long-term control experience). This design allows the neural network not only to fit the data, but also to utilize domain knowledge (prior information of the fuzzy system) to improve robustness.

[0050] Furthermore, according to one embodiment of the present invention, the fuzzy inference algorithm employs an adjustable Gaussian membership function: ; Where x represents the variable input to the fuzzy inference algorithm, the center value c is generated by K-means clustering, and the variance σ is adaptively adjusted in the interval [0.1, 0.5].

[0051] In MPPT (Maximum Power Point Tracking) control scenarios, the variables input into the fuzzy inference algorithm typically refer to key physical quantities directly related to the operating state of the photovoltaic system, such as: The output voltage, current, or power of a photovoltaic cell; The amount of change in environmental parameters (light intensity, temperature); Input signals such as power difference and voltage / current deviation are used for fuzzy control.

[0052] Its core function is to convert the input variable x into the membership degree of a fuzzy set through a Gaussian membership function, so as to facilitate subsequent fuzzy inference and adaptive control, ultimately achieving maximum power point tracking of the photovoltaic system. The dynamic adjustment of the center value c and variance σ (by K-means clustering and adaptive adjustment mechanism) is used to optimize the applicability of the fuzzy rules and the control accuracy.

[0053] According to the above-described scheme of this invention, the present invention achieves rapid tracking and high-precision stable control of the maximum power point of a photovoltaic array under complex environments through a collaborative optimization mechanism of fuzzy logic and neural networks. This invention achieves breakthrough performance with tracking efficiency ≥99.2%, response time ≤80ms, and steady-state power oscillation amplitude <1.5% through a dynamic expansion mechanism of fuzzy rules, a lightweight neural network architecture, and dual-core collaborative computing technology. It reduces energy consumption by 40% compared to traditional methods and is applicable to photovoltaic power plants, distributed energy storage, and new energy vehicles.

[0054] Furthermore, to achieve the above objectives, the present invention also provides an MPPT control system based on an adaptive fuzzy neural network, comprising: The multimodal sensing module synchronously collects voltage, current, light, and temperature signals to generate raw sensing data. The DMA zero-copy transfer unit transmits the raw sensing data collected and generated by the multimodal sensing module to the memory-mapped area through a hardware channel; The dual-core heterogeneous processor acquires sensor data from the memory mapping area of ​​the DMA zero-copy transfer unit, performs calculations, predicts power changes and illumination changes, and runs a fuzzy inference algorithm based on the power changes and illumination changes to obtain a power point decision scheme. The rule base is dynamically expanded. Based on the power and illumination changes predicted by the dual-core heterogeneous processor, the fuzzy rules are updated and the updated fuzzy rules are fed back to the dual-core heterogeneous processor in real time to optimize the fuzzy inference logic. The four-phase interleaved power topology, based on the power point decision scheme given by the dual-core heterogeneous processor, operates in an interleaved manner to achieve maximum power point tracking.

[0055] Furthermore, according to one embodiment of the present invention, the sampling rate of the multimodal sensing module is ≥200 kHz. In this embodiment, synchronous acquisition by the multimodal sensing module means that all signals are measured at the same point in time (reducing errors caused by asynchrony), which is suitable for dynamic systems (such as motor monitoring). This requires the module's hardware (sensors and ADC) to work together and the signal transmission path to be balanced.

[0056] In this implementation, the ADC sampling rate refers to the number of samples per second (sampling rate ≥ 200 kHz = 200,000 times / second) to ensure that signal details are not lost (according to Nyquist's theorem, the sampling rate must be greater than twice the highest frequency of the signal). For example, if acquiring a 100 kHz high-frequency signal (such as switching power supply ripple), a sampling rate of 200 kHz meets the minimum requirement (to avoid aliasing distortion).

[0057] A high sampling rate can also improve synchronization: all channels (voltage, current, etc.) are sampled at the same rate to ensure data point time alignment (microsecond-level delay).

[0058] Furthermore, according to one embodiment of the present invention, the DMA zero-copy transfer unit moves data directly from the source (sensor) to the target memory area (0x2400_0000) without passing through the CPU's internal cache or temporary buffer. This reduces additional steps in data transfer (e.g., the overhead of data transfer in CPU registers), thereby optimizing latency.

[0059] Because DMA is managed by a hardware controller (rather than the CPU core executing instructions), the CPU only needs to initialize the DMA settings (such as configuring the source and destination addresses) and can then perform other tasks in parallel, with CPU utilization ≤5%. The brief overhead of initializing DMA (triggering interrupts or setting registers) accounts for a tiny percentage of the entire transfer cycle.

[0060] The memory-mapped area (0x2400_0000) is typically associated with a specific peripheral or reserved memory (in ARM Cortex-M series chips, it represents the virtual address of the physical device). This allows the system to directly access this area for efficient read and write operations (sensor data can be processed by the algorithm module immediately after being written).

[0061] In this implementation, zero-copy combined with a dedicated memory mapping area reduces CPU load by approximately 80-95% (compared to traditional methods), especially in high-bandwidth applications (audio or video transmission). When the sensor outputs 1MB of data per second, DMA can transfer data at near the hardware limit while CPU utilization remains stable at a low level.

[0062] Furthermore, according to one embodiment of the present invention, the dual-core heterogeneous processor includes a Cortex-M4 core and a Cortex-M7 core; The Cortex-M4 core obtains sensor data from the memory-mapped area of ​​the DMA zero-copy transfer unit, performs calculations, and predicts changes in power and illumination. The Cortex-M7 core uses the power change and illumination change output by the Cortex-M4 core to run a fuzzy inference algorithm to obtain a power point decision scheme. Specifically, when the power change of the Cortex-M4 core is greater than 15W or the light change is greater than 1000Lux / s, the Cortex-M7 core is triggered to run the fuzzy inference algorithm.

[0063] In this embodiment, inter-core collaboration is achieved through the HSEM semaphore synchronization task, which utilizes 768KB of shared RAM to exchange intermediate results (e.g., the M4-identified △P>15W event triggers M7 inference).

[0064] Furthermore, according to one embodiment of the present invention, the dynamically expanded rule base updates fuzzy rules based on the power change and illumination change predicted by the dual-core heterogeneous processor, including: The dynamically expanding rule base continuously checks whether the power change of the Cortex-M4 core output is greater than 15W or the light change is greater than 1000Lux / s through an independent thread or interrupt service routine; When the power change is greater than 15W or the illumination change is greater than 1000Lux / s, the dynamic expansion rule base collects sensor data during periods of abrupt power or illumination changes and generates new fuzzy rules through a decision tree algorithm. Check whether the new fuzzy rule is redundant or conflicting with the existing fuzzy rules in the dynamically expanded rule base. If the new fuzzy rule has logical overlap with the existing fuzzy rule, merge them. If the new fuzzy rule conflicts with the existing fuzzy rule, overwrite the existing fuzzy rule with the new fuzzy rule.

[0065] In this embodiment, the dynamically expandable rule base uses a dynamically scalable storage structure (database table, in-memory hash table, or linked list) to store the initial rules. Each rule contains: Input variables (environmental parameters, device status); Output actions (adjust power, adjust lighting); Membership function (defining the threshold for fuzzy logic); In this embodiment, during initialization, at least 50 expert experience rules are preloaded into the dynamically expanded rule base (IF illumination > 500 Lux THEN reduce power by 10%), and persistent storage is achieved through a database or configuration file.

[0066] In this embodiment, sudden environmental changes are monitored in real time, and sensors are deployed to collect data in real time: △P (Power Change Rate): Calculated per second, current power - power of the previous second; △L (rate of change of illumination): (current illumination value - illumination value of the previous second) / time interval; Continuous checks via a separate thread or interrupt service routine (ISR): if (△P > 15W) or (△L > 1000 Lux / s): activate_rule_expansion() # Starts the process of adding a rule; In this implementation, the logic for generating new rules is as follows: Collect data streams during periods of abrupt change and generate new rules using a decision tree algorithm; For example, "IF Illumination increases suddenly (ΔL>1000) AND time = daytime THEN turn off auxiliary lighting"; Manual assistance can be used to extend the system, triggering alerts and prompting administrators to review the newly added rules, thus ensuring system security.

[0067] In this implementation, the new rules are appended and stored in a structured format (JSON / XML), while a fast retrieval index is built based on a hash index of environment parameters.

[0068] In this implementation, when adding a new rule, redundancy or conflict with existing rules is checked: If the new rules and old rules have overlapping logic, the membership functions are merged. If a conflict occurs and the opposite operation is performed, the old rule is overridden based on the confidence weight.

[0069] Capacity control: The rule base is dynamically expanded with a maximum of 500 rules, triggering an elimination mechanism: deleting infrequently used rules or merging similar rules.

[0070] Furthermore, according to one embodiment of the present invention, the four-phase interleaved power topology includes four Boost circuits (four-phase Boost circuits), which are connected in parallel. Based on the power point decision scheme given by the dual-core heterogeneous processor, the switching signals of each Boost circuit are staggered in time to achieve phase interleaving, thereby realizing maximum power point tracking.

[0071] In this embodiment, the four-phase Boost circuit operates in an interleaved manner (each phase offset by 90°), combined with GaN switches (high frequency, low loss), a 22μH inductor (energy storage), and a 470μF capacitor (filtering) to achieve efficient power conversion. This reduces input current ripple (through phase interleaving), improves efficiency (>95%), and allows for rapid response to processor instructions, adjusting the output voltage (within a 100V range) to track the maximum power point.

[0072] Furthermore, according to one embodiment of the present invention, the Cortex-M4 core obtains sensor data from the memory-mapped area of ​​the DMA zero-copy transfer unit, performs neural network calculations, and predicts the power change and the illumination change. The neural network parameter tuning layer is a 3-layer backpropagation structure, including: The input layer has 6 nodes: receiving the fuzzy output ΔP / Δt, ΔL / Δt, ΔT / Δt and the historical error mean; where ΔP is the power change, ΔL is the illumination change, Δt is the time change, and ΔT is the temperature change; The hidden layer has 12 nodes: the LeakyReLU activation function is used, with a decay coefficient α = 0.01; The output layer has three nodes: the output PWM duty cycle adjustment amount ΔD∈[-5%,+5%], the membership function parameter correction amount, and the learning rate adjustment instruction.

[0073] The necessity of the three-layer backpropagation structure in this embodiment: a. Input layer (6 nodes): Integrates multi-dimensional dynamic information The input includes fuzzified rates of change of physical quantities (ΔP / Δt, ΔL / Δt, ΔT / Δt) and historical error mean, covering the system's real-time dynamic characteristics (rate of change of power / rate of change of illumination / rate of change of temperature) and historical control performance (error mean). This design enables the network to capture both short-term fluctuations (Δt-related variables) and long-term trends (historical errors) of the system, providing a comprehensive basis for parameter adjustment decisions.

[0074] b. Hidden layer (12 nodes + LeakyReLU): Enhances nonlinear fitting capability The 12 nodes balance model complexity and generalization ability, avoiding overfitting; the LeakyReLU activation function (α=0.01) solves the "neuron death" problem of traditional ReLU, ensuring the effective transmission of negative input signals and improving the network's fitting accuracy for complex nonlinear relationships, making it particularly suitable for uncertainty modeling of dynamic systems.

[0075] c. Output layer (3 nodes): Multi-objective collaborative optimization Simultaneously output three types of key parameters: Control command (PWM duty cycle adjustment ΔD): directly acts on the actuator to achieve real-time closed-loop control; Model parameter correction (membership function parameters): Optimize the core components of the fuzzy system (membership function shape) to improve the adaptability of the fuzzy model; Learning rate adjustment command: Dynamically control the training step size of the neural network itself to avoid getting stuck in local optima or oscillations, and achieve adaptation at the meta-learning level.

[0076] In this implementation, the role of the backpropagation algorithm is: By optimizing network weights through gradient descent, system performance indicators (error, stability) are transformed into differentiable loss functions, enabling quantitative and automated parameter tuning. This differs from traditional heuristic parameter tuning methods, improving optimization efficiency and accuracy.

[0077] Traditional control systems (such as PID controllers) rely on fixed parameters or manual parameter tuning, making it difficult to cope with environmental changes (such as load fluctuations and temperature drift). This implementation method uses a neural network to analyze dynamic data in real time, adjust the control output (ΔD) online, and optimize the membership function parameters of the fuzzy model, enabling the system to have "self-evolution" capabilities and adapt to nonlinear and time-varying scenarios.

[0078] The learning rate adjustment instruction in the output layer further realizes the meta-learning mechanism: when the system error is large, the learning rate is increased (fast convergence), and when the error is small, the learning rate is decreased (fine adjustment), avoiding the limitations of the traditional fixed learning rate.

[0079] In this embodiment, the rate of change of physical quantities and historical errors integrated by the input layer are combined with: Causal relationship (ΔP / Δt reflects the correlation between system input and output); Dynamic characteristics (ΔL / Δt, ΔT / Δt characterize load and thermal response rate); Empirical knowledge (historical error mean reflects long-term control experience). This design allows the neural network not only to fit the data, but also to utilize domain knowledge (prior information of the fuzzy system) to improve robustness.

[0080] Furthermore, according to one embodiment of the present invention, the fuzzy inference algorithm employs an adjustable Gaussian membership function. ; Where x represents the variable input to the fuzzy inference algorithm, the center value c is generated by K-means clustering, and the variance σ is adaptively adjusted in the interval [0.1, 0.5].

[0081] According to the above-described scheme of this invention, the present invention achieves rapid tracking and high-precision stable control of the maximum power point of a photovoltaic array under complex environments through a collaborative optimization mechanism of fuzzy logic and neural networks. This invention achieves breakthrough performance with tracking efficiency ≥99.2%, response time ≤80ms, and steady-state power oscillation amplitude <1.5% through a dynamic expansion mechanism of fuzzy rules, a lightweight neural network architecture, and dual-core collaborative computing technology. It reduces energy consumption by 40% compared to traditional methods and is applicable to photovoltaic power plants, distributed energy storage, and new energy vehicles.

[0082] Furthermore, to achieve the above objectives, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the MPPT control method based on an adaptive fuzzy neural network as described above.

[0083] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the MPPT control method based on an adaptive fuzzy neural network as described above.

[0084] Those skilled in the art will recognize that the modules and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented using electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0085] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described apparatus and equipment can be referred to the corresponding process in the foregoing method implementation, and will not be repeated here.

[0086] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0087] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the objectives of the embodiments of the present invention, depending on actual needs.

[0088] In addition, the functional modules in the embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0089] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the sending / receiving methods of various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0090] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in this application.

[0091] It should be understood that the sequence number of each step in the invention and its embodiments does not absolutely imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

Claims

1. An MPPT control method based on an adaptive fuzzy neural network, characterized in that, include: The multimodal sensing module synchronously collects voltage, current, light, and temperature signals to generate raw sensing data. The raw sensing data acquired and generated by the multimodal sensing module is transmitted to the memory-mapped area via a hardware channel through the DMA zero-copy transfer unit. The dual-core heterogeneous processor obtains sensor data from the memory mapping area of ​​the DMA zero-copy transfer unit, performs calculations, predicts power changes and illumination changes, runs a fuzzy inference algorithm based on power changes and illumination changes, and obtains a power point decision scheme. The rule base is dynamically expanded to obtain the predicted power change and illumination change, and the fuzzy rules are updated based on the power change and illumination change predicted by the dual-core heterogeneous processor. The updated fuzzy rules are fed back to the dual-core heterogeneous processor in real time to optimize the fuzzy inference logic. The four-phase interleaved power topology is driven to operate interleaved according to the power point decision scheme to achieve maximum power point tracking.

2. The MPPT control method based on an adaptive fuzzy neural network according to claim 1, characterized in that, The sampling rate of the multimodal sensing module is ≥200 kHz.

3. The MPPT control method based on an adaptive fuzzy neural network according to claim 1, characterized in that, The dual-core heterogeneous processor includes a Cortex-M4 core and a Cortex-M7 core; The Cortex-M4 core obtains sensor data from the memory-mapped area of ​​the DMA zero-copy transfer unit, performs calculations, and predicts changes in power and illumination. The Cortex-M7 core uses the power change and illumination change output by the Cortex-M4 core to run a fuzzy inference algorithm to obtain a power point decision scheme. Specifically, when the power change of the Cortex-M4 core is greater than 15W or the light change is greater than 1000Lux / s, the Cortex-M7 core is triggered to run the fuzzy inference algorithm.

4. The MPPT control method based on an adaptive fuzzy neural network according to claim 3, characterized in that, The dynamically expanded rule base updates fuzzy rules based on power and illumination changes predicted by the dual-core heterogeneous processor, including: The dynamically expanding rule base continuously checks whether the power change of the Cortex-M4 core output is greater than 15W or the light change is greater than 1000Lux / s through an independent thread or interrupt service routine; When the power change is greater than 15W or the illumination change is greater than 1000Lux / s, the dynamic expansion rule base collects sensor data during periods of abrupt power or illumination changes and generates new fuzzy rules through a decision tree algorithm. Check whether the new fuzzy rule is redundant or conflicting with the existing fuzzy rules in the dynamically expanded rule base. If the new fuzzy rule has logical overlap with the existing fuzzy rule, merge them. If the new fuzzy rule conflicts with the existing fuzzy rule, overwrite the existing fuzzy rule with the new fuzzy rule.

5. The MPPT control method based on an adaptive fuzzy neural network according to claim 1, characterized in that, The four-phase interleaved power topology includes four Boost circuits connected in parallel. Based on the power point decision scheme provided by the dual-core heterogeneous processor, the switching signals of each Boost circuit are staggered to achieve phase interleaving, thereby realizing maximum power point tracking.

6. The MPPT control method based on an adaptive fuzzy neural network according to claim 3, characterized in that, The Cortex-M4 core obtains sensor data from the memory mapping area of ​​the DMA zero-copy transfer unit, performs neural network calculations, and predicts power changes and illumination changes. The neural network parameter tuning layer is a 3-layer backpropagation structure, including: The input layer has 6 nodes: receiving the fuzzy output ΔP / Δt, ΔL / Δt, ΔT / Δt and the historical error mean; where ΔP is the power change, ΔL is the illumination change, Δt is the time change, and ΔT is the temperature change; The hidden layer has 12 nodes: the LeakyReLU activation function is used, with a decay coefficient α = 0.01; The output layer has three nodes: the output PWM duty cycle adjustment amount ΔD∈[-5%,+5%], the membership function parameter correction amount, and the learning rate adjustment instruction.

7. The MPPT control method based on an adaptive fuzzy neural network according to any one of claims 1-6, characterized in that, The fuzzy inference algorithm employs an adjustable Gaussian membership function: ； Where x represents the variable input to the fuzzy inference algorithm, the center value c is generated by K-means clustering, and the variance σ is adaptively adjusted in the interval [0.1, 0.5].

8. An MPPT control system based on an adaptive fuzzy neural network, characterized in that, include: The multimodal sensing module synchronously collects voltage, current, light, and temperature signals to generate raw sensing data. The DMA zero-copy transfer unit transmits the raw sensing data collected and generated by the multimodal sensing module to the memory-mapped area through a hardware channel; The dual-core heterogeneous processor acquires sensor data from the memory mapping area of ​​the DMA zero-copy transfer unit, performs calculations, predicts power changes and illumination changes, and runs a fuzzy inference algorithm based on the power changes and illumination changes to obtain a power point decision scheme. The rule base is dynamically expanded. Based on the power and illumination changes predicted by the dual-core heterogeneous processor, the fuzzy rules are updated and the updated fuzzy rules are fed back to the dual-core heterogeneous processor in real time to optimize the fuzzy inference logic. The four-phase interleaved power topology, based on the power point decision scheme given by the dual-core heterogeneous processor, operates in an interleaved manner to achieve maximum power point tracking.

9. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the MPPT control method based on an adaptive fuzzy neural network as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the MPPT control method based on an adaptive fuzzy neural network as described in any one of claims 1-7.

Images