A photovoltaic inverter adaptive maximum power point tracking system and method

Abstract

The present application relates to photovoltaic power generation control technical field, specifically to a kind of photovoltaic inverter adaptive maximum power point tracking system, including data acquisition module, for real-time acquisition photovoltaic array's output voltage and output current;Feature construction module is used to determine power variation rate and the variation of power variation rate, and the combination of two is determined as system state;Reward calculation module is used to calculate reward value based on the power increment of photovoltaic array and voltage disturbance;Rule adjustment module is used to update adaptive fuzzy rule base based on system state and reward value by reinforcement learning algorithm;Decision reasoning module is used to generate duty cycle adjustment amount;Control execution module is used to update the duty cycle of photovoltaic inverter based on duty cycle adjustment amount to track the maximum power point of photovoltaic array;The present application solves the problem of effectively improving the stability of output power, guarantees the stable and reliable operation of system, avoids the control failure caused by calculation error.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation control technology, specifically to an adaptive maximum power point tracking system and method for photovoltaic inverters. Background Technology

[0002] In photovoltaic power generation systems, maximum power point tracking control is crucial for improving energy conversion efficiency. Fuzzy control algorithms are widely used in MPPT technology due to their good robustness.

[0003] Currently, traditional fuzzy control MPPT algorithms for photovoltaic inverters mainly rely on a fixed fuzzy rule base. This rule base is pre-defined by technicians based on expert experience and existing experimental data, defining the control actions the system should take under specific input conditions. However, this method based on static empirical rules has exposed significant technical bottlenecks in practical applications. The operating environment of photovoltaic systems, especially light intensity and ambient temperature, is constantly and unpredictably changing. Fixed fuzzy rules cannot dynamically adapt to these complex and changing conditions, resulting in poor system adaptability and difficulty in continuously optimizing its control strategy. This, in turn, affects the tracking accuracy and response speed of the maximum power point, and may even cause power oscillations. Therefore, how to enable fuzzy control systems to break free from their dependence on fixed rules and endow them with the ability to learn online and self-optimize in order to adapt to complex and changing environments, thereby achieving efficient, accurate, and stable maximum power point tracking, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides an adaptive maximum power point tracking system for photovoltaic inverters. Specifically, the technical solution of this invention includes:

[0005] The data acquisition module is used to collect the output voltage and output current of the photovoltaic array in real time.

[0006] The feature construction module is used to determine the power change rate and the change in power change rate based on the output voltage and output current, and to determine the combination of the two as the system state.

[0007] The reward calculation module is used to calculate the reward value based on the power increment and voltage disturbance of the photovoltaic array;

[0008] The rule adjustment module is used to update the adaptive fuzzy rule base based on the system state and reward value using a reinforcement learning algorithm.

[0009] The decision reasoning module is used to perform fuzzy reasoning based on the rate of change of power and the change in the rate of change of power, and calls the updated adaptive fuzzy rule base to generate the duty cycle adjustment amount.

[0010] The control execution module is used to update the duty cycle of the photovoltaic inverter based on the duty cycle adjustment amount in order to track the maximum power point of the photovoltaic array.

[0011] Optionally, the feature construction module determines the rate of power change and the amount of variation in the rate of power change, including:

[0012] The instantaneous output power is calculated based on the current sampled values ​​of the output voltage and output current and the historical values ​​of the previous cycle.

[0013] Based on the instantaneous output power and output voltage of the current cycle and the previous cycle, the power change rate is determined. In the calculation of the power change rate, a preset zero disturbance prevention term is added to the denominator.

[0014] The change in the power change rate is determined based on the current power change rate compared to the previous cycle.

[0015] Optional, the determination of system state includes:

[0016] Call the preset membership function to map the power change rate and the change in the power change rate into fuzzy linguistic variables;

[0017] The fuzzy linguistic variables corresponding to the power change rate and the amount of change of the power change rate are combined into ordered pairs to uniquely determine the system state.

[0018] Optionally, the fuzzy linguistic variables are divided into seven fuzzy subsets, including negative large, negative medium, negative small, zero, positive small, positive medium, and positive large.

[0019] Optionally, the reward calculation module calculates the reward value, including:

[0020] Obtain the power increment and voltage disturbance of the photovoltaic array;

[0021] The power increment and voltage disturbance are weighted by a preset weighting coefficient and standardized by a preset rated power and rated voltage to generate a reward value.

[0022] Optionally, the rule adjustment module updates the adaptive fuzzy rule base, including:

[0023] Maintain the action value function, which is used to record the adjustment action value under each system state;

[0024] The action value function is updated based on the system state and reward value, combined with the preset learning rate and discount factor;

[0025] Based on the updated action value function, select the adjustment action and determine it as the rule adjustment amount;

[0026] Use rule adjustments to update the fuzzy rules corresponding to the system state.

[0027] Optionally, the selection of adjustment actions adopts... -greedy strategy.

[0028] Optional updates to fuzzy rules include:

[0029] The rule adjustment is applied to update the membership function center point of the consequent of the fuzzy rule that uniquely corresponds to the system state.

[0030] Optionally, the decision reasoning module generates the duty cycle adjustment, including:

[0031] The power change rate and its variation are fuzzified to generate fuzzy linguistic variables;

[0032] The updated adaptive fuzzy rule base is used to infer fuzzy linguistic variables and obtain fuzzy output.

[0033] The centroid method is used to defuzzify the fuzzy output in order to obtain the duty cycle adjustment amount.

[0034] Optionally, the control execution module updates the duty cycle of the photovoltaic inverter, including:

[0035] The duty cycle adjustment is added to the duty cycle of the previous control cycle to generate the duty cycle of the current control cycle.

[0036] An adaptive maximum power point tracking method for photovoltaic inverters includes:

[0037] The data acquisition process involves real-time acquisition of the photovoltaic array's output voltage and current.

[0038] The feature construction step involves determining the power change rate and the variation amount of the power change rate based on the output voltage and the output current, and determining the combination of the two as the system state.

[0039] The reward calculation steps involve calculating the reward value based on the power increment and voltage disturbance of the photovoltaic array.

[0040] The rule adjustment step involves updating the adaptive fuzzy rule base using a reinforcement learning algorithm based on the system state and the reward value.

[0041] The decision reasoning step involves using the power change rate and the amount of change in the power change rate as a basis, and calling the updated adaptive fuzzy rule base to perform fuzzy reasoning to generate the duty cycle adjustment amount.

[0042] The control execution steps update the duty cycle of the photovoltaic inverter based on the duty cycle adjustment amount to track the maximum power point of the photovoltaic array.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. This system introduces reinforcement learning algorithms, enabling the fuzzy rule base to be updated and self-optimized online in real time. It eliminates reliance on fixed expert experience and can automatically adapt to complex and changing working environments such as light intensity and temperature, as well as conditions like photovoltaic module aging. This significantly improves the system's environmental adaptability and autonomous optimization capabilities under all operating conditions. Simultaneously, by introducing reinforcement learning algorithms to optimize fuzzy rules in real time, it forms an adaptive closed-loop control, solving the problem of traditional fuzzy control relying on fixed expert experience and being unable to adapt to changing operating conditions. Compared to existing technologies, this invention achieves faster dynamic response speed, higher steady-state tracking accuracy, and stronger oscillation suppression capabilities, thereby achieving efficient, accurate, and stable maximum power point tracking under complex and changing environments such as light intensity and temperature.

[0045] 2. This system constructs an adaptive closed-loop control from observation, evaluation, decision-making to learning. Through a precise reward mechanism, it guides the control strategy to evolve towards the optimal goal. This not only accelerates the dynamic response speed to the maximum power point, but also improves the tracking accuracy of the system under stable working conditions, thereby maximizing energy conversion efficiency under various operating conditions.

[0046] 3. The reward function of this system integrates two evaluation dimensions: power increment and voltage disturbance. Through weighted and standardized processing, it guides the system to actively suppress voltage oscillations near the maximum power point while pursuing faster power increase. This effectively improves the stability of output power and ensures the stable and reliable operation of the system.

[0047] 4. By fuzzifying continuous state characteristics into discrete linguistic variables, this system enhances its tolerance to measurement noise and minor disturbances. At the same time, by adding a term to prevent zero disturbances in key calculations, the numerical stability of the algorithm is ensured. This makes the entire control system exhibit stronger robustness and reliability in actual hardware applications, avoiding control failures caused by calculation errors. Attached Figure Description

[0048] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0049] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0051] Example 1:

[0052] Please see Figure 1 An adaptive maximum power point tracking system for a photovoltaic inverter, comprising:

[0053] The data acquisition module is used to collect the output voltage and output current of the photovoltaic array in real time.

[0054] The feature construction module is used to determine the power change rate and the change in power change rate based on the output voltage and output current, and to determine the combination of the two as the system state.

[0055] The reward calculation module is used to calculate the reward value based on the power increment and voltage disturbance of the photovoltaic array;

[0056] The rule adjustment module is used to update the adaptive fuzzy rule base based on the system state and reward value using a reinforcement learning algorithm.

[0057] The decision reasoning module is used to perform fuzzy reasoning based on the rate of change of power and the change in the rate of change of power, and calls the updated adaptive fuzzy rule base to generate the duty cycle adjustment amount.

[0058] The control execution module is used to update the duty cycle of the photovoltaic inverter based on the duty cycle adjustment amount in order to track the maximum power point of the photovoltaic array.

[0059] This embodiment provides an adaptive maximum power point tracking system for photovoltaic inverters. The system aims to solve the technical problem that the traditional fuzzy control MPPT algorithm has poor adaptability and cannot continuously optimize the control strategy under complex and ever-changing light and temperature conditions due to the fixed fuzzy rule base. The system constitutes an adaptive closed-loop control system. Its purpose is to give the fuzzy controller the ability to learn online and optimize itself by introducing a reinforcement learning mechanism, thereby improving the steady-state tracking accuracy of the maximum power point and suppressing power oscillations while ensuring fast dynamic response.

[0060] In a specific implementation scenario, the system is integrated into the digital signal processor (DSP) of a photovoltaic grid-connected inverter. The overall system process begins with real-time data acquisition, followed by precise construction of state features, and then enters a decision-making and optimization loop driven by reinforcement learning, ultimately generating high-precision control commands to adjust the operating point of the photovoltaic array.

[0061] The system includes:

[0062] The data acquisition module aims to provide real-time, high-precision raw input data for the entire control system. In this embodiment, this module utilizes high-precision voltage and current sensors built into the photovoltaic inverter. Sampling period... This refers to the data acquisition time interval, which defines the frequency of system state updates. It is derived from a fixed value preset based on the inverter hardware performance and control dynamic response requirements. This module uses a sampling period... The output voltage of the photovoltaic array at time k is collected synchronously. With output current ;

[0063] The feature construction module aims to extract key state features from raw voltage and current data that can accurately characterize the dynamic behavior of the system, providing a decision-making basis for subsequent fuzzy inference and reinforcement learning. In this embodiment, this module is based on the collected data... and Two core characteristic quantities were calculated: power change rate. and the change in the rate of change of power The combination of these two characteristic quantities uniquely determines the system state at time k. ;

[0064] The reward calculation module aims to construct a quantitative evaluation index to assess the quality of the system's previous control action, thereby providing a clear learning signal for the reinforcement learning agent. In this embodiment, the module calculates a scalar reward value based on the power increment and voltage perturbation of the photovoltaic array. ;

[0065] The rule adjustment module aims to achieve adaptive evolution of fuzzy control rules. In this embodiment, the module employs the Q-Learning reinforcement learning algorithm, based on the current system state. Compared with the newly calculated reward value An adaptive fuzzy rule base is updated online; the update here is relative to the current state. Fine-tune specific rules that are directly related;

[0066] The decision-making reasoning module aims to generate an optimal control adjustment command based on the current system state and using the latest, optimized fuzzy rule base. In this embodiment, this module receives the power change rate. and the change in the rate of change of power As input, the system calls an adaptive fuzzy rule library, which is updated in real time by the rule adjustment module, to perform fuzzy inference and ultimately generate a precise duty cycle adjustment value. ;

[0067] The control execution module aims to physically apply the control commands generated by the decision reasoning module to the photovoltaic inverter, thereby achieving actual adjustment of the photovoltaic array's operating point. In this embodiment, the module adjusts the duty cycle. Duty cycle compared to the previous cycle The duty cycle of the current period is obtained by accumulating the results. This drives the DC / DC converter in the inverter;

[0068] This embodiment constructs a closed loop of observation-evaluation-decision-execution-learning through the collaborative work of the above modules. It solves the problem that the traditional fuzzy control MPPT method relies on static empirical rules and cannot adapt to changes in operating conditions. By introducing reinforcement learning, the system can continuously optimize the fuzzy rule base during operation, enabling the MPPT control strategy to automatically adapt to different light intensities, temperature changes, and even photovoltaic module aging, ultimately achieving fast, accurate, and stable tracking of the maximum power point under all operating conditions.

[0069] Example 2:

[0070] The feature construction module determines the rate of power change and the amount of variation in the rate of power change, including:

[0071] The instantaneous output power is calculated based on the current sampled values ​​of the output voltage and output current and the historical values ​​of the previous cycle.

[0072] Based on the instantaneous output power and output voltage of the current cycle and the previous cycle, the power change rate is determined. In the calculation of the power change rate, a preset zero disturbance prevention term is added to the denominator.

[0073] The change in the power change rate is determined based on the current power change rate compared to the previous cycle.

[0074] Based on Example 1, this embodiment limits the specific implementation of the feature construction module; its purpose is to ensure that the constructed state feature quantity can accurately and robustly reflect the local geometric features of the PV characteristic curve, so as to provide high-quality input for subsequent control decisions.

[0075] This module calculates the instantaneous output power; to obtain the system's output capability at the current moment, this embodiment introduces the instantaneous output power. The calculation method is as follows:

[0076]

[0077] in, Represents at discrete time points The instantaneous output power, measured in watts (W), is derived from the current period sampled value provided by the data acquisition module. and Calculated; Represents at discrete time points The output voltage, measured in volts (V), is sourced in real time by a sensor. Represents at discrete time points The output current, measured in amperes (A), is obtained in real time from a sensor; this calculation logic follows the basic physical definition of electric power, and the result is compared with historical values. Used together in subsequent calculations;

[0078] To characterize the local slope of the PV curve at the current operating point, this embodiment introduces the power change rate. The calculation method is as follows:

[0079]

[0080] in, It is the discretized derivative of power with respect to voltage, with the physical dimension of current A, and characterizes the instantaneous conductance increment of the system; its source is calculated from the power and voltage values ​​of the current and previous cycles; the core function of this parameter is to determine whether the current operating point is to the left, right or near the maximum power point. These are the instantaneous output power of the current cycle and the previous cycle, respectively; These are the output voltages for the current cycle and the previous cycle, respectively. This is a zero-disturbance prevention term, its function is to prevent the voltage value from being within two consecutive sampling periods. and The calculation error of a denominator of zero due to perfect equality is avoided to ensure the numerical stability of the algorithm. This is a common practice in numerical computation, and in this embodiment, it is preset to a very small positive constant that matches the resolution of the system voltage sampling hardware, for example... Its dimension is consistent with the voltage term in the denominator, which is volt (V);

[0081] To further determine the dynamic trend of the operating point approaching the maximum power point, i.e., the local curvature information of the PV curve, this embodiment introduces the variation in the rate of power change. The calculation method is as follows:

[0082]

[0083] in, yes The time difference, with its physical dimension in amperes (A), reflects the changing trend of the slope of the PV curve, i.e., the change in the conductance increment within each sampling period; its source is the power change rate of the current and previous periods. The calculated parameter helps the system determine whether it is converging toward the maximum power point or diverging, thus enabling more precise adjustments.

[0084] Through the specific calculation method described above, this embodiment can more accurately capture the dynamic characteristics of the PV curve; the corrected The calculation formula will exclude zero disturbance terms. Placing it in the denominator resolves the issue of inconsistent dimensions, significantly enhances the algorithm's robustness in the steady-state operating region, and avoids computational errors caused by hardware sampling precision limitations; simultaneously, by constructing and The system not only knows the current operating point's location, but also its movement trend, providing sufficient and necessary state information for achieving fast and low-oscillation MPPT control.

[0085] Example 3:

[0086] Determining the system state includes:

[0087] Call the preset membership function to map the power change rate and the change in the power change rate into fuzzy linguistic variables;

[0088] The fuzzy linguistic variables corresponding to the power change rate and the change in power change rate are combined into ordered pairs to uniquely determine the system state.

[0089] Fuzzy linguistic variables are divided into seven fuzzy subsets, including negative large, negative medium, negative small, zero, positive small, positive medium, and positive large.

[0090] This embodiment further optimizes the process of determining the system state based on Embodiment 2; its purpose is to optimize the continuous state characteristic quantities. , It is transformed into a discrete, linguistic state description, which can directly correspond to the discrete fuzzy rule base and the state space of reinforcement learning, thus solving the problem of transformation from continuous perception to discrete decision-making.

[0091] This process calls a preset membership function to determine the power change rate. Change in the rate of change of power Mapping to fuzzy linguistic variables; membership function This refers to a system that takes precise input values. Mapped to The membership function describes the degree to which an input value belongs to a certain fuzzy set; in this embodiment, the membership function is initially a Gaussian function, with its center point... With width The initial settings are made by technicians based on their experience with the PV curves of typical photovoltaic arrays;

[0092] The fuzzy linguistic variables corresponding to the power change rate and the change in power change rate are combined into ordered pairs to uniquely determine the system state. Each continuous input is divided into seven fuzzy subsets, which constitute the values ​​of the fuzzy linguistic variables. These seven fuzzy subsets are: {Negative Large NB, Negative Medium NM, Negative Small NS, Zero ZE, Positive Small PS, Positive Medium PM, Positive Large PB}.

[0093] System status Depend on and It is composed of the corresponding fuzzy subsets; if Belongs to Zhengxiao PS, If a system belongs to the negative smaller NS, then the current system state is... It is then uniquely determined as (PS, NS); since each variable has 7 fuzzy subsets, the total state space... It includes There are discrete states, and each state corresponds one-to-one with a fuzzy rule of the form IFeisPSANDΔeisNSTHEN...;

[0094] By using this fuzzy state definition method, this embodiment achieves effective dimensionality reduction and discretization of the system state. This not only enhances the system's tolerance to measurement noise and small disturbances, but more importantly, it constructs a finite and definite discrete state space. This is a prerequisite for applying tabular reinforcement learning algorithms such as Q-Learning, and lays the foundation for adaptive adjustment of the rule base.

[0095] Example 4:

[0096] The reward calculation module calculates the reward value, including:

[0097] Obtain the power increment and voltage disturbance of the photovoltaic array;

[0098] The power increment and voltage disturbance are weighted by a preset weighting coefficient and standardized by a preset rated power and rated voltage to generate a reward value.

[0099] This embodiment, based on embodiment 1, specifies the specific implementation of the reward calculation module; its purpose is to design a reward function that can accurately guide the reinforcement learning agent to learn toward the composite goal of quickly and stably tracking the maximum power point;

[0100] The core of this module is the reward value. The calculation, it Always The actions taken at any given moment are evaluated, and the calculation method is as follows:

[0101]

[0102] in, : Indicates that the system is in After the action is executed, at any time The environmental reward obtained at any given time is a dimensionless scalar value, which is calculated in real time by this module;

[0103] The power increment depends on the current and previous cycle power values, which are derived from the calculation results of the feature construction module.

[0104] The voltage disturbance depends on the current and previous cycle voltage values, which are obtained from the data acquisition module.

[0105] The weighting coefficient is a preset dimensionless hyperparameter used to balance the relationship between tracking speed and steady-state performance. It is derived from offline simulation based on typical illumination change curves to obtain the desired system response characteristics.

[0106] Rated power and rated voltage are used to standardize power increments and voltage disturbances with different physical dimensions, converting them into dimensionless values ​​so that meaningful weighted summation can be performed; their source is the photovoltaic array design specifications.

[0107] The reward function designed in this embodiment has clear physical meaning and engineering orientation; it integrates the core objective of maximizing power with the key constraint of suppressing oscillations into the same optimization function; through standardization, the universality of the reward is ensured; by introducing weight coefficients, the flexibility to adjust the system between speed and stability is provided; this reward signal can effectively guide the reinforcement learning algorithm to learn an excellent control strategy that can both respond quickly to environmental changes and maintain low oscillations near the maximum power point.

[0108] Example 5:

[0109] The rule adjustment module updates the adaptive fuzzy rule base, including:

[0110] Maintain the action value function, which is used to record the adjustment action value under each system state;

[0111] The action value function is updated based on the system state and reward value, combined with the preset learning rate and discount factor;

[0112] Based on the updated action value function, select the adjustment action and determine it as the rule adjustment amount;

[0113] Use rule adjustments to update the fuzzy rules corresponding to the system state.

[0114] The selection of adjustment actions adopts -greedy strategy.

[0115] The updating of fuzzy rules includes:

[0116] The rule adjustment is applied to update the membership function center point of the consequent of the fuzzy rule that uniquely corresponds to the system state.

[0117] This embodiment, based on embodiment 1, elaborates on the core workflow of the rule adjustment module; its purpose is to establish a clear, reinforcement learning-driven mechanism for online optimization of fuzzy rules.

[0118] This module maintains an action value function. Action value function Used to assess a specific system state Next, perform a certain adjustment action. The expected value of the long-term cumulative reward it can bring; its function is to provide a numerical basis for the agent to select the optimal action in each state; in this embodiment, the function is based on a... The system is stored in the form of a Q-table, where 49 corresponds to 49 discrete system states and N corresponds to the number of discrete actions; when the system is first started, all values ​​in the Q-table are initialized to zero.

[0119] In each control cycle, the module updates the action value function based on the system state and reward value; this update follows the core rules of the Q-Learning algorithm, and its mathematical form is as follows:

[0120]

[0121] in, The current state, current action, and next state are derived from the system's real-time operational trajectory.

[0122] : Execute action Then, the instant reward value is provided by the reward calculation module;

[0123] Learning rate Its function is to determine the extent to which newly acquired reward information covers the old value estimate; its source is an empirical setting based on control theory, for example... ;

[0124] Discount factor Its function is to define the degree of importance attached to future rewards; its source is also experience-based settings, such as... ;

[0125] Next state The set of all possible actions;

[0126] The module selects and adjusts actions based on the updated action value function; this embodiment adopts... -Greedy strategy for action selection; The greedy strategy is a strategy that trades off between exploration and exploitation; its logic is: to... The probability is used to select the optimal action with the highest Q value in the current state, while... A small probability is used to randomly select an action; in this embodiment... It can be preset to a small constant, for example This is to ensure that the known optimal strategy is used most of the time, while retaining a certain amount of exploration capability;

[0127] Selected action It is determined as a rule adjustment amount; in this embodiment, the action space Defined as a discrete set of adjustment values, for example ,in It is a preset basic adjustment step size; its size can be determined according to the adjustment accuracy of the inverter duty cycle and the desired response speed, for example, it can be set to 0.005;

[0128] By utilizing rule adjustment variables, the fuzzy rules corresponding to the system state are updated; this update process is precise and unique; the rule adjustment variables are then applied to update the current system state. The center point of the membership function of the consequent of the uniquely corresponding fuzzy rule Its update logic is as follows:

[0129]

[0130] :exist Time, and state The center point of the output membership function of the corresponding fuzzy rule; this center point directly determines the size of the control output when the rule is activated;

[0131] For all other states The corresponding rule parameters remain unchanged;

[0132] Through this complete update mechanism, this embodiment constructs a clear path from environmental feedback to rule evolution; the application of the Q-Learning algorithm enables the system to learn from experience. The introduction of the -greedy strategy ensures comprehensive learning; while adjusting the learned amount By directly and specifically applying the consequent center point of a particular fuzzy rule, precise, local, and online optimization of the fuzzy rule base can be achieved.

[0133] Example 6:

[0134] The decision reasoning module generates the duty cycle adjustment, including:

[0135] The power change rate and its variation are fuzzified to generate fuzzy linguistic variables;

[0136] The updated adaptive fuzzy rule base is used to infer fuzzy linguistic variables and obtain fuzzy output.

[0137] The centroid method is used to defuzzify the fuzzy output in order to obtain the duty cycle adjustment amount.

[0138] Based on Example 1, this embodiment defines the specific process for the decision reasoning module to generate the duty cycle adjustment amount; its purpose is to describe how to use a fuzzy rule base optimized in real time through reinforcement learning to transform system state features into a precise and executable control command.

[0139] This process follows a fuzzy control inference flow:

[0140] Blurring: The module receives the precise input value of the power change rate from the feature construction module. and the change in the rate of change of power Convert these two precise input values ​​into fuzzy linguistic variables and determine their membership degrees on seven fuzzy subsets;

[0141] Fuzzy Inference: The module performs inference on fuzzy linguistic variables based on the updated adaptive fuzzy rule base to obtain fuzzy output; the system iterates through all rules in the rule base; for each rule, its trigger strength is determined by... and The membership degree of the premise fuzzy set is obtained through fuzzy AND operation; this trigger strength is used to cut the fuzzy set of the rule consequent to obtain a weighted fuzzy output; the weighted fuzzy outputs generated by all triggered rules are finally merged into a total fuzzy output;

[0142] Defuzzification: Since the overall fuzzy output is still a fuzzy set, it needs to be converted into a precise numerical value. This embodiment uses the centroid method to defuzzify the fuzzy output to obtain the duty cycle adjustment amount. The centroid method is a method that calculates the abscissa of the geometric centroid of the total fuzzy output graphic and uses it as the final accurate output.

[0143] This embodiment ensures the logical consistency of control decisions through a standardized fuzzy inference process. The adaptive fuzzy rule base here is optimized in real time by the rule adjustment module, which means that the knowledge on which the inference is based is dynamically updated and continuously improved. The centroid method is used for defuzzification, which can comprehensively consider the contributions of all activated rules, so that the final generated duty cycle adjustment amount is... Smooth and precise, improving the stability of control.

[0144] Example 7:

[0145] The control execution module updates the duty cycle of the photovoltaic inverter, including:

[0146] The duty cycle adjustment is added to the duty cycle of the previous control cycle to generate the duty cycle of the current control cycle.

[0147] Based on Example 1, this embodiment specifies the method by which the control execution module updates the duty cycle of the photovoltaic inverter; its purpose is to convert the discrete adjustment quantity generated in the previous step into a continuous control signal for the inverter hardware, thereby completing the entire control closed loop.

[0148] This module is implemented by adding the duty cycle adjustment amount to the duty cycle of the previous control cycle to generate the duty cycle of the current control cycle. The mathematical derivation of its update logic is as follows:

[0149]

[0150] Current control cycle The final duty cycle of the DC / DC converter to be executed is calculated by this module;

[0151] Previous control cycle Duty cycle;

[0152] : Duty cycle adjustment amount generated by the decision reasoning module;

[0153] This update process is essentially a discrete-time integrator; the calculated new duty cycle It will be directly written into the hardware register in the inverter responsible for generating the PWM pulse width modulation signal, thereby changing the on-time of the power switch in real time and thus adjusting the operating point of the photovoltaic array;

[0154] This cumulative update method ensures the continuity and gradualness of duty cycle changes, avoiding excessive stress on power electronic devices that may result from sudden changes in control commands, thus guaranteeing the system's hardware safety. Through this cumulative operation, the advanced decisions generated by the intelligent algorithm are... This effectively translates into physical control of the underlying hardware. This completes the closed loop of the entire adaptive MPPT control.

[0155] Example 8:

[0156] An adaptive maximum power point tracking method for photovoltaic inverters includes:

[0157] The data acquisition process involves real-time acquisition of the photovoltaic array's output voltage and current.

[0158] The feature construction step is to determine the power change rate and the variation of the power change rate based on the output voltage and output current, and to determine the combination of the two as the system state.

[0159] The reward calculation steps involve calculating the reward value based on the power increment and voltage disturbance of the photovoltaic array.

[0160] The rule adjustment process involves updating the adaptive fuzzy rule base using a reinforcement learning algorithm, based on the system state and reward value.

[0161] The decision-making reasoning steps are based on the power change rate and the change in the power change rate, and the updated adaptive fuzzy rule base is called to perform fuzzy reasoning to generate the duty cycle adjustment amount.

[0162] The control execution steps update the duty cycle of the photovoltaic inverter based on the duty cycle adjustment to track the maximum power point of the photovoltaic array.

[0163] It should be noted that the above 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 be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A photovoltaic inverter adaptive maximum power point tracking system, characterized in that, include: The data acquisition module is used to collect the output voltage and output current of the photovoltaic array in real time. The feature construction module is used to determine the power change rate and the change in power change rate based on the output voltage and output current, and to determine the combination of the two as the system state. The reward calculation module is used to calculate the reward value based on the power increment and voltage disturbance of the photovoltaic array; The rule adjustment module is used to update the adaptive fuzzy rule base based on the system state and reward value using a reinforcement learning algorithm. The decision reasoning module is used to perform fuzzy reasoning based on the rate of change of power and the change in the rate of change of power, and calls the updated adaptive fuzzy rule base to generate the duty cycle adjustment amount. The control execution module is used to update the duty cycle of the photovoltaic inverter based on the duty cycle adjustment amount in order to track the maximum power point of the photovoltaic array; The feature construction module determines the power change rate and the amount of variation in the power change rate, including: The instantaneous output power is calculated based on the current sampled values ​​of the output voltage and output current and the historical values ​​of the previous cycle. Based on the instantaneous output power and output voltage of the current cycle and the previous cycle, the power change rate is determined. In the calculation of the power change rate, a preset zero disturbance prevention term is added to the denominator. Determine the power change rate based on the current power change rate compared to the previous cycle. The amount of change; Power change rate The calculation method is as follows: in, It is the discretized derivative of power with respect to voltage, with the physical dimension of current A, and characterizes the instantaneous conductance increment of the system; These are the instantaneous output power of the current cycle and the previous cycle, respectively; These are the output voltages for the current cycle and the previous cycle, respectively. It is a term that prevents zero disturbances; The rule adjustment module updates the adaptive fuzzy rule base, including: Maintain the action value function, which is used to record the adjustment action value under each system state; The action value function is updated based on the system state and reward value, combined with the preset learning rate and discount factor; Based on the updated action value function, select the adjustment action and determine it as the rule adjustment amount; Update the fuzzy rules corresponding to the system state by adjusting the rule amount; The selection of the adjustment action adopts -greedy strategy; The update of the fuzzy rules includes: Apply rule adjustments to updates and the current system state. The center point of the membership function of the consequent of the uniquely corresponding fuzzy rule Its update logic is as follows: :exist Time, and state The center point of the output membership function of the corresponding fuzzy rule; this center point directly determines the size of the control output when the fuzzy rule is activated; For all other states .

2. The photovoltaic inverter adaptive maximum power point tracking system according to claim 1, characterized in that, The determination of the system state includes: Call the preset membership function to map the power change rate and the change in the power change rate into fuzzy linguistic variables; The fuzzy linguistic variables corresponding to the power change rate and the amount of change of the power change rate are combined into ordered pairs to uniquely determine the system state.

3. The photovoltaic inverter adaptive maximum power point tracking system according to claim 2, characterized in that, The fuzzy linguistic variables are divided into seven fuzzy subsets, including negative large, negative medium, negative small, zero, positive small, positive medium, and positive large.

4. The photovoltaic inverter adaptive maximum power point tracking system according to claim 1, characterized in that, The reward calculation module calculates the reward value, including: Obtain the power increment and voltage disturbance of the photovoltaic array; The power increment and voltage disturbance are weighted by a preset weighting coefficient and standardized by a preset rated power and rated voltage to generate a reward value.

5. The photovoltaic inverter adaptive maximum power point tracking system according to claim 4, characterized in that, The decision reasoning module generates the duty cycle adjustment amount, including: The power change rate and its variation are fuzzified to generate fuzzy linguistic variables; The updated adaptive fuzzy rule base is used to infer fuzzy linguistic variables and obtain fuzzy output. The centroid method is used to defuzzify the fuzzy output in order to obtain the duty cycle adjustment amount.

6. The photovoltaic inverter adaptive maximum power point tracking system according to claim 1, characterized in that, The control execution module updates the duty cycle of the photovoltaic inverter, including: The duty cycle adjustment is added to the duty cycle of the previous control cycle to generate the duty cycle of the current control cycle.

7. A photovoltaic inverter adaptive maximum power point tracking method, applied to the photovoltaic inverter adaptive maximum power point tracking system described in claim 1, characterized in that, include: The data acquisition process involves real-time acquisition of the photovoltaic array's output voltage and current. The feature construction step involves determining the power change rate and the variation amount of the power change rate based on the output voltage and the output current, and determining the combination of the two as the system state. The reward calculation steps involve calculating the reward value based on the power increment and voltage disturbance of the photovoltaic array. The rule adjustment step involves updating the adaptive fuzzy rule base using a reinforcement learning algorithm based on the system state and the reward value. The decision reasoning step involves using the power change rate and the amount of change in the power change rate as a basis, and calling the updated adaptive fuzzy rule base to perform fuzzy reasoning to generate the duty cycle adjustment amount. The control execution steps update the duty cycle of the photovoltaic inverter based on the duty cycle adjustment amount to track the maximum power point of the photovoltaic array.

Images