Grid-connected inverter parameter optimization method based on deep learning and quantum inheritance

Abstract

The invention provides a grid-connected inverter parameter optimization method based on deep learning and quantum inheritance, relates to the field of power electronics and intelligent control, and solves the technical problems that the self-adaptive capability is poor due to the fact that linkage with a real-time operation state is failed, and the optimal comprehensive performance of a system is difficult to realize by taking a single index as an optimization target mostly. The method comprises the following steps: carrying out dimensionality reduction and feature extraction through a deep auto-encoder network to obtain a feature vector; a population containing N quantum chromosomes is initialized, and a possible solution is covered while the population is initialized by using the superposition characteristic of a quantum state. And inputting the feature vector and a real-time control parameter set of the inverter into a quantum genetic optimization algorithm model, and iteratively calculating a fitness value of a solution represented by each chromosome in the population. And issuing the optimal control parameter set corresponding to the optimal fitness value to an actual inverter control system for parameter adjustment to form closed-loop optimization. The method and the device are used in the parameter optimization process of the grid-connected inverter.

Description

Technical Field

[0001] This application relates to the field of power electronics and intelligent control, and in particular to a method and apparatus for optimizing grid-connected inverter parameters based on deep learning and quantum genetics. Background Technology

[0002] Grid-connected inverters are the core interface equipment between new energy power generation systems and the power grid, and their performance directly affects grid stability, power quality, and equipment lifespan. Several key parameters in the inverter control system (such as filter parameters, controller gain, and PWM strategy) need to be rationally configured according to operating conditions and load characteristics. Traditional optimization methods, such as genetic algorithms (GA), particle swarm optimization (PSO), and simulated annealing (SA), mostly use a single performance index (such as focusing only on steady-state error or harmonic distortion rate) as the optimization objective. They cannot comprehensively consider multiple dimensions of system dynamic performance (such as overshoot and settling time), steady-state performance (such as steady-state error and system efficiency), and harmonic distortion rate, easily leading to situations where one aspect is prioritized at the expense of others, making it difficult to achieve optimal overall system performance. Furthermore, when searching for optimal parameters, existing algorithms either easily get trapped in local optima due to limited search range or have slow convergence speed due to a lack of efficient search strategy guidance, failing to balance global applicability and convergence efficiency. Summary of the Invention

[0003] This application provides a method and apparatus for optimizing grid-connected inverter parameters based on deep learning and quantum genetics, which solves the technical problems of existing technologies failing to link with real-time operating status, resulting in poor adaptive capability, and often using a single index as the optimization target, making it difficult to achieve the optimal overall system performance.

[0004] To achieve the above objectives, this application adopts the following technical solution: Firstly, a parameter optimization method for grid-connected inverters based on deep learning and quantum genetics includes: acquiring the original multi-dimensional operating data sequence from the historical database of the grid-connected inverter; performing dimensionality reduction and feature extraction through a deep autoencoder network to obtain feature vectors representing the system's operating state; initializing a population containing N quantum chromosomes, each quantum chromosome consisting of a gate unit representing the control loop selection state and a parameter encoding unit representing the control loop parameter values, connected in series; utilizing the superposition property of quantum states to simultaneously cover possible solutions during population initialization; inputting the feature vectors and the inverter's real-time control parameter set into the quantum genetic optimization algorithm model; driving population evolution through probability amplitude-based quantum rotating gate operations; and iteratively calculating the fitness value of the solution represented by each chromosome in the population; the fitness value is calculated by a pre-trained fitness evaluation function and used to comprehensively evaluate the system's dynamic performance, steady-state performance, and harmonic distortion rate. The optimal fitness value in the current population is compared with the preset performance threshold. If the fitness value is better than the performance threshold, the optimized control parameter set corresponding to the optimal fitness value is sent to the actual inverter control system for parameter adjustment. The inverter output waveform data after parameter adjustment is fed back to the historical database as new raw data to form a closed-loop optimization.

[0005] Based on the above technical solution, the grid-connected inverter parameter optimization method based on deep learning and quantum genetics provided in this application constructs a quantum genetic optimization algorithm model. This model allows the extracted system state feature vector, real-time control parameters, and the solution sequence decoded by the quantum population to be input into a pre-trained fitness evaluation function. Through three sub-functions—dynamic, steady-state, and harmonic—the system performance is comprehensively evaluated, and a comprehensive fitness value is output. This deeply couples the optimization process with the actual system state, making the optimization objective no longer a single indicator but the optimal overall system performance. Furthermore, the search process possesses both globality and fast convergence. This effectively addresses the problem that existing optimization algorithms, operating under fixed system models or conditions, fail to integrate with real-time operating states and exhibit poor adaptive capabilities.

[0006] In conjunction with the first aspect above, in one possible implementation, initializing a population containing N quantum chromosomes includes: each quantum chromosome consists of a gate unit containing m qubits and a parameter encoding unit containing n qubits connected in series, wherein the gate unit is used to select the control loop type of the inverter through the qubit state, the parameter encoding unit is used to encode the parameter value of the selected control loop through the qubit state, and each qubit is a superposition state of probability amplitudes. Where α and β are complex probability amplitudes, satisfying The probability amplitude of each qubit in each quantum chromosome. Initialize to Used to put qubits into a state of state and The probabilities of superposition states are equal.

[0007] In conjunction with the first aspect mentioned above, in one possible implementation, the quantum genetic optimization algorithm model includes: a three-layer evaluation network used to calculate the fitness value F of the solution represented by each quantum chromosome, comprising an input layer, a quantum solution mapping layer, and a fitness calculation layer, wherein the input layer is connected to the feature vector V. feat and the real-time control parameter set C of the inverter real The quantum solution mapping layer, in each iteration, modifies the population... Each quantum chromosome q in i The quantum observation operation collapses the sequence into a binary solution sequence B. i And the binary solution sequence B i The middle gate control unit is decoded into the control loop selection identifier G. i The binary solution sequence B i The parameter encoding unit is partially decoded into a parameter value set P. i Summary control loop selection identifier G i、 Parameter value set P i、 Feature vector V feat and real-time control parameter set C real Both outputs are fed to the fitness calculation layer. The neural network model is then trained under supervised conditions using historical data and corresponding system performance metrics to obtain the fitness evaluation function. The fitness evaluation function Embedded within the fitness computation layer, the fitness computation layer receives the output of the quantum solution mapping layer and feeds it into the pre-trained fitness evaluation function. In the calculation, the corresponding quantum chromosome q is obtained. i fitness value F i It is a comprehensive index evaluation value used to characterize the dynamic performance, steady-state performance and harmonic distortion rate of a system.

[0008] In conjunction with the first aspect mentioned above, in one possible implementation, the fitness evaluation function... This includes dynamic performance evaluation subfunctions, steady-state performance evaluation subfunctions, and harmonic distortion rate evaluation subfunctions. The dynamic performance evaluation subfunction selects the identifier G through an analog control loop. i and parameter value set P i The overshoot M of the system's step response is obtained from the system's dynamic response. p and the settling time t of the system step response s ,pass Calculate the dynamic performance index f d , of which M p,max and t s,max This is the preset maximum allowable overshoot and settling time.p and w s The preset dynamic performance weighting coefficients, and w p +w s =1. The steady-state performance evaluation subfunction selects the identifier G through the analog control loop. i and parameter value set P i The steady-state response of the system under the given conditions yields the steady-state error e. ss和 System efficiency η, through Calculate the steady-state performance index f s , where e ss,max and η max For the preset maximum allowable steady-state error and ideal efficiency, w e and w η The preset steady-state performance weighting coefficients are w. e +w η =1. The harmonic distortion rate evaluation subfunction selects the identifier G through an analog control loop. i and parameter value set P i The system output waveform was analyzed and Fourier analysis was performed to obtain the THD. Calculate the harmonic performance f of the integrated system h THD max The preset maximum permissible total harmonic distortion (THD). The dynamic performance index f... d Steady-state performance index f s and the overall system harmonic performance f h Weighted fusion yields the fitness value F i .

[0009] In conjunction with the first aspect mentioned above, in one possible implementation, the quantum rotation gate operation includes: obtaining the best individual q in the current population. best With the current individual q i Observe the binary value b of the j-th bit in the solution sequence after collapse. best,j and b i,j ,pass Calculate the fitness value F of the current best individual best With the current individual fitness value F i The difference F. Through Determine the quantum rotation gate pair q i Each quantum bit The rotation angle θj, where The rotation angle calculated for the j-th qubit in generation t. For the preset rotation angle lookup table, The preset adaptive adjustment coefficient η(t) is used, where the adaptive adjustment coefficient η(t) dynamically decreases as the iteration number t increases, and T maxη is the maximum number of iterations. max and η min These are the preset maximum and minimum coefficient values. (Using a quantum rotation gate matrix) Probability amplitude of updating qubit , where α j and β j For the current individual q i The probability amplitude of the j-th qubit. and This is the new probability amplitude after being updated by the quantum rotation gate.

[0010] In conjunction with the first aspect mentioned above, in one possible implementation, the construction of the preset rotation angle lookup table includes: based on the statistical results of the relationship between ΔF and binary bit changes in historical optimization data, preset rotation angle values ​​for different ΔF and bit differences, and summarizing them to form the preset rotation angle lookup table. The rotation angle value refers to the value that allows the population to converge to the global optimum.

[0011] In conjunction with the first aspect mentioned above, one possible implementation involves a deep autoencoder network performing dimensionality reduction and feature extraction. This includes constructing a deep autoencoder network comprising an input layer, multiple encoding layers, a bottleneck layer, multiple decoding layers, and an output layer. The input layer receives raw multi-dimensional operating data sequences obtained from a historical database of grid-connected inverters. The multiple encoding layers consist of at least two fully connected layers, each using a non-linear activation function for feature transformation, progressively reducing the data dimensionality. Finally, a low-dimensional feature vector is output at the bottleneck layer. The decoding layer is symmetrical to the encoding layer and is used to reconstruct data from the feature vector at the bottleneck layer. The data dimensionality is progressively restored through multiple fully connected layers and activation functions, and the reconstructed data is output at the output layer. The raw multi-dimensional operating data sequences include time-series data of voltage, current, and power parameters, with each data sample containing measurements from multiple time steps. The raw multi-dimensional operating data sequences from the historical database are used as the training dataset, with each data sample serving as both input and desired output. The network parameters are optimized using a backpropagation algorithm, and the deep autoencoder network is trained by minimizing the reconstruction loss function. After training, the decoder part is discarded, retaining only the encoder part as a feature extractor. The reconstruction loss function is... Where N is the number of training samples, x i It is the i-th original data sample, x i ′ is the mean squared error loss defined as the i-th data sample in the network reconstruction. It is the Euclidean norm.

[0012] In conjunction with the first aspect mentioned above, in one possible implementation, the actual inverter control system performs parameter adjustment, including: selecting identifier G according to the control loop. iDetermine the target control loop type, which may be a voltage control loop, current control loop, or power control loop. Set the parameter value set P... i The parameter configuration instructions are converted into recognizable parameter configuration commands for the target control loop according to preset mapping rules. These rules are based on a control loop parameter specification table defined in a historical database, which stores parameter addresses and scaling factors corresponding to different control loop types. The parameter configuration commands are then sent to the digital signal processor (DSP) of the actual inverter control system. The DSP updates the adjustable parameters in the control algorithm based on the configuration commands and calculates the duty cycle or frequency using a PID control law to generate a PWM control signal. This PWM control signal drives the inverter's power switching devices, adjusting the inverter's output waveform.

[0013] In conjunction with the first aspect mentioned above, one possible implementation involves forming a closed-loop optimization loop: using high-precision sensors to collect time-series data of voltage, current, and power at the inverter output. The collected data is then filtered and normalized by a data preprocessing unit to generate new raw multi-dimensional operating data sequences. These new raw data sequences are then fed back to a historical database in real time to update the training data of the deep autoencoder network, thus forming a closed-loop optimization cycle.

[0014] Secondly, a grid-connected inverter parameter optimization device based on deep learning and quantum genetics is provided, comprising: a communication unit and a processing unit; the communication unit is used to acquire the original multi-dimensional operating data sequence from the historical database of the grid-connected inverter; the processing unit is used to perform dimensionality reduction and feature extraction through a deep autoencoder network to obtain a feature vector representing the system's operating state. A population containing N quantum chromosomes is initialized, each quantum chromosome consisting of a gating unit representing the control loop selection state and a parameter encoding unit representing the control loop parameter values, connected in series. The superposition property of quantum states is utilized to simultaneously cover possible solutions during population initialization. The feature vector and the real-time control parameter set of the inverter are input into the quantum genetic optimization algorithm model, and the population evolution is driven by a quantum rotating gate operation based on probability amplitude, iteratively calculating the fitness value of the solution represented by each chromosome in the population. The fitness value is calculated by a pre-trained fitness evaluation function and is used to comprehensively evaluate the system's dynamic performance, steady-state performance, and harmonic distortion rate. The optimal fitness value in the current population is compared with the preset performance threshold. If the fitness value is better than the performance threshold, the optimized control parameter set corresponding to the optimal fitness value is sent to the actual inverter control system for parameter adjustment. The inverter output waveform data after parameter adjustment is fed back to the historical database as new raw data to form a closed-loop optimization.

[0015] Thirdly, this application provides a grid-connected inverter parameter optimization device based on deep learning and quantum genetics, comprising: a processor and a storage medium; the storage medium includes instructions, and the processor is used to execute the instructions to implement the method described in the first aspect and any possible implementation thereof. This grid-connected inverter parameter optimization device based on deep learning and quantum genetics can be an electronic device or a chip within an electronic device.

[0016] Fourthly, this application provides a computer-readable storage medium storing instructions that, when executed on a grid-connected inverter parameter optimization device based on deep learning and quantum genetics, cause the device to perform the method described in the first aspect and any possible implementation thereof.

[0017] Fifthly, this application provides a computer program product containing instructions that, when the computer program product is run on the grid-connected inverter parameter optimization device based on deep learning and quantum genetics, causes the grid-connected inverter parameter optimization device based on deep learning and quantum genetics to perform the method described in the first aspect and any possible implementation thereof.

[0018] This application provides a method and apparatus for optimizing grid-connected inverter parameters based on deep learning and quantum genetics. By constructing a quantum genetic optimization algorithm model, the extracted system state feature vector, real-time control parameters, and the solution sequence decoded by the quantum population are input into a pre-trained fitness evaluation function. Through three sub-functions—dynamic, steady-state, and harmonic—the system performance is comprehensively evaluated, and a comprehensive fitness value is output. This deeply couples the optimization process with the actual system state, making the optimization objective no longer a single metric but the optimal overall system performance. Furthermore, the search process possesses both globality and fast convergence. This effectively addresses the shortcomings of existing optimization algorithms, which optimize under fixed system models or operating conditions, failing to integrate with real-time operating states and exhibiting poor adaptive capabilities.

[0019] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0020] Figure 1 A system architecture diagram of a grid-connected inverter parameter optimization system based on deep learning and quantum genetics provided in the embodiments of this application; Figure 2 A flowchart illustrating the grid-connected inverter parameter optimization method based on deep learning and quantum genetics provided in this application embodiment; Figure 3 A flowchart illustrating another grid-connected inverter parameter optimization method based on deep learning and quantum genetics provided for embodiments of this application; Figure 4 A flowchart illustrating another grid-connected inverter parameter optimization method based on deep learning and quantum genetics provided for embodiments of this application; Figure 5 A flowchart illustrating another grid-connected inverter parameter optimization method based on deep learning and quantum genetics provided for embodiments of this application; Figure 6 A flowchart illustrating another grid-connected inverter parameter optimization method based on deep learning and quantum genetics provided for embodiments of this application; Figure 7 A schematic diagram of the structure of a grid-connected inverter parameter optimization device based on deep learning and quantum genetics provided in an embodiment of this application; Detailed Implementation

[0021] In the description of this application, unless otherwise stated, "" means "or," for example, A / B can mean A or B. "And / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The words "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0022] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0023] To address the shortcomings of existing technologies that often use a single performance indicator (such as focusing only on steady-state error or harmonic distortion rate) as the optimization objective, which fails to comprehensively consider the system's dynamic performance (overshoot, settling time), steady-state performance (steady-state error, system efficiency), and harmonic distortion rate, it is crucial to address these shortcomings. This approach is prone to overlooking certain aspects and failing to achieve optimal overall system performance. Furthermore, existing algorithms are susceptible to getting stuck in local optima due to limitations in their search scope, or experiencing slow convergence due to a lack of strategy guidance, making it difficult to balance global applicability with convergence efficiency. Moreover, optimizations are often based on fixed system models or specific operating conditions, failing to integrate with the real-time operating status of the inverter. This makes them unable to adapt to fluctuations in operating conditions (such as load changes and grid voltage fluctuations), exhibiting poor adaptive capabilities and difficulty in timely parameter adjustments to maintain optimal performance, thus failing to meet the optimization needs of complex and variable operating environments. Furthermore, feature extraction relies on manual selection based on expert experience or linear dimensionality reduction methods, which are difficult to adapt to the highly nonlinear data of power electronic systems, affecting the accuracy of optimization input. This application provides a grid-connected inverter parameter optimization method based on deep learning and quantum genetics. This method constructs a quantum genetic optimization algorithm model, which inputs the extracted system state feature vector, real-time control parameters, and the solution sequence decoded by the quantum population into a pre-trained fitness evaluation function. Through three sub-functions—dynamic, steady-state, and harmonic—the system performance is comprehensively evaluated, and a comprehensive fitness value is output. This deeply couples the optimization process with the actual system state, making the optimization objective no longer a single indicator but the optimal overall system performance. The search process also possesses both globality and convergence speed. This effectively solves the problem that existing optimization algorithms, when optimized under fixed system models or operating conditions, fail to link with real-time operating states and have poor adaptive capabilities.

[0024] like Figure 1As shown in the embodiments of this application, the grid-connected inverter parameter optimization method based on deep learning and quantum genetics includes: Step 101: Obtain the original multi-dimensional operation data sequence from the historical database of the grid-connected inverter, and perform dimensionality reduction and feature extraction through a deep autoencoder network to obtain the feature vector representing the system's operating state.

[0025] The grid-connected inverter historical database refers to a database storing historical operating data of grid-connected inverters, including time-series data of parameters such as voltage, current, and power. The original multi-dimensional operating data sequence refers to multi-dimensional time-series data obtained from the historical database, with each data sample including measurements from multiple time points. A deep autoencoder network is a neural network structure composed of an encoder and a decoder. The encoder compresses high-dimensional data into low-dimensional features, and the decoder reconstructs the data from the features, learning effective features by minimizing the reconstruction error during training. Dimensionality reduction refers to the process of simplifying data representation by reducing data dimensionality while retaining important information. Feature extraction refers to the process of extracting feature vectors representing the key characteristics of the data from the original data. Feature vectors are low-dimensional vectors obtained after dimensionality reduction, capable of representing key features of the system's operating state.

[0026] In some implementations, the original multi-dimensional operating data sequence, including time series data of voltage, current and power parameters, is first obtained from the historical database of grid-connected inverters. Then, this original data is input into a deep autoencoder network to reduce the dimensionality and extract features from the original data, thereby obtaining a feature vector that characterizes the operating state of the system.

[0027] Step 102: Initialize a population containing N quantum chromosomes. Each quantum chromosome consists of a gating unit representing the selection state of the control loop and a parameter encoding unit representing the parameter values ​​of the control loop, connected in series. The superposition property of quantum states is utilized to simultaneously cover all possible solutions during population initialization.

[0028] Initialization refers to setting the initial state of the population. The population is a set of quantum chromosomes. A quantum chromosome is a data structure used in quantum genetic algorithms, consisting of gated units and parameter encoding units connected in series. A gated unit contains m qubits and is used to select the inverter's control loop type based on the qubit states, where the control loop selection state indicates which control loop is being selected. A parameter encoding unit contains n qubits and is used to encode the parameter values ​​of the selected control loop based on the qubit states. Series connection means that the gated units and parameter encoding units are linked together to form a quantum chromosome. The superposition property of quantum states means that a qubit can simultaneously be in a superposition state. and The state is determined, and the probability is expressed using probability amplitude. A possible solution refers to all possible combinations of control loop choices and parameter values.

[0029] In some implementations, a population containing N quantum chromosomes is first initialized. Then, each quantum chromosome is constructed by cascading a gating unit and a parameter encoding unit. The gating unit contains m qubits representing the control loop selection state, and the parameter encoding unit contains n qubits representing the control loop parameter values. Next, the probability amplitude of each qubit is initialized, such that each qubit is in a state... and The superposition of quantum states has equal probability; finally, by utilizing the superposition property of quantum states, the population is initialized to cover all possible solutions at the same time, that is, all combinations of control loop selection and parameter values.

[0030] Step 103: Input the eigenvectors and the real-time control parameter set of the inverter into the quantum genetic optimization algorithm model. Drive population evolution through quantum rotating gate operations based on probability amplitude, and iteratively calculate the fitness value of the solution represented by each chromosome in the population. The fitness value is calculated by the pre-trained fitness evaluation function and is used to comprehensively evaluate the dynamic performance, steady-state performance, and harmonic distortion rate of the system.

[0031] The real-time control parameter set of the inverter refers to the set of adjustable parameters currently used by the inverter control system. The quantum genetic optimization algorithm model is a three-layer evaluation network used to calculate the fitness value of the solution represented by each quantum chromosome, including an input layer, a quantum solution mapping layer, and a fitness calculation layer. Population evolution refers to the process of iteratively updating quantum chromosomes to drive the population towards better solutions. The fitness value is a numerical value calculated by a pre-trained fitness evaluation function, used to comprehensively evaluate the quality of solutions. The pre-trained fitness evaluation function is a function obtained by supervised training of the neural network model using historical data and corresponding system performance indicators. System dynamic performance refers to indicators such as overshoot and settling time of the system's step response. System steady-state performance refers to indicators such as steady-state error and efficiency. Harmonic distortion rate refers to the total harmonic distortion rate of the system's output waveform.

[0032] In some implementations, the feature vector and the real-time control parameter set of the inverter are input to the input layer of the quantum genetic optimization algorithm model. Then, in the quantum solution mapping layer, each quantum chromosome in the population is collapsed into a binary solution sequence through quantum observation operations, and the gating unit part of the sequence is decoded into a control loop selection identifier, and the parameter encoding unit part is decoded into a parameter value set. Next, the control loop selection identifier, parameter value set, feature vector, and real-time control parameter set are output to the fitness calculation layer. In the fitness calculation layer, the fitness value of the corresponding quantum chromosome is calculated through a pre-trained fitness evaluation function, which evaluates the system's dynamic performance, steady-state performance, and harmonic distortion rate. Finally, the population evolution is driven by a quantum rotation gate operation based on probability amplitude, and the fitness value of the solution represented by each chromosome in the population is calculated iteratively.

[0033] Step 104: Compare the optimal fitness value in the current population with the preset performance threshold. If the fitness value is better than the performance threshold, the optimized control parameter set corresponding to the optimal fitness value is sent to the actual inverter control system for parameter adjustment. The inverter output waveform data after parameter adjustment is fed back to the historical database as new raw data to form a closed-loop optimization.

[0034] In this context, the optimal fitness value in the current population refers to the highest fitness value among all quantum chromosomes in the current iteration, calculated using the fitness evaluation function. The preset performance threshold is a pre-defined performance standard value used to judge whether the optimization result is acceptable. The optimized control parameter set refers to the solution represented by the quantum chromosome corresponding to the optimal fitness value, including the decoded control loop selection identifier and parameter value set. The actual inverter control system refers to the hardware system that controls the operation of the grid-connected inverter, typically including a digital signal processor (DSP). Parameter adjustment refers to updating the adjustable parameters in the actual inverter control system based on the optimized control parameter set. The inverter output waveform data after parameter adjustment refers to the time-series data of the voltage, current, and other waveforms at the inverter's output after operating with the new parameter set. The new raw data refers to the pre-processed inverter output waveform data. Feedback refers to sending the new data back to the starting point. The historical database refers to a database storing the historical operating data of the grid-connected inverter.

[0035] In some implementations, the optimal fitness value in the current population is compared with a preset performance threshold. If the fitness value is determined to be better than the performance threshold, the optimized control parameter set corresponding to the optimal fitness value is sent to the actual inverter control system. Then, in the actual inverter control system, the target control loop type is determined according to the control loop selection identifier, and the parameter value set is converted into parameter configuration instructions according to the mapping rules. The adjustable parameters in the control algorithm are updated by the DSP, and a PWM control signal is generated to drive the power switching device, thereby completing the parameter adjustment. Afterwards, a high-precision sensor is used to collect the inverter output waveform data after parameter adjustment. The data is filtered and normalized by the data preprocessing unit to generate a new original multi-dimensional operating data sequence. Finally, the new original data sequence is fed back to the historical database in real time to update the data, thereby forming a closed-loop optimization.

[0036] Based on the above technical solution, firstly, a symmetric deep autoencoder neural network is constructed to perform unsupervised learning on high-dimensional, nonlinear original historical operating data. This automatically extracts low-dimensional feature vectors that can highly summarize and characterize the core operating state of the system from massive amounts of data, discarding redundant information and providing accurate and efficient input for subsequent optimization. This avoids the reliance on expert experience to manually select features or use linear dimensionality reduction methods in existing technologies, making the extraction of highly nonlinear data from power electronic systems faster and more representative. Secondly, by initializing the probability amplitude of each qubit, the quantum superposition property is utilized to ensure that a population of size N initially covers all possible combinations of control loops and parameter values ​​with a certain probability. This also improves the diversity and global search capability of the initial population, laying a solid foundation for finding the global optimal solution. Thirdly, by constructing a quantum genetic optimization algorithm model, the extracted system state feature vectors, real-time control parameters, and the solution sequence decoded by the quantum population are input into a pre-trained fitness evaluation function. Through three sub-functions—dynamic, steady-state, and harmonic—the system performance is comprehensively evaluated, and a comprehensive fitness value is output. This deeply couples the optimization process with the actual system state, making the optimization objective no longer a single metric, but rather the optimal overall system performance. The search process also possesses both globality and fast convergence. This effectively addresses the shortcomings of existing optimization algorithms that operate under fixed system models or conditions, failing to integrate with real-time operating states and exhibiting poor adaptability. Finally, the introduction of a preset performance threshold creates a closed-loop system of "optimization-application-feedback-re-optimization." This allows the historical database and deep learning model to be continuously updated and evolved, enabling the entire optimization system to adapt to unknown situations such as inverter aging and changes in the external power grid environment, possessing a continuous self-improvement learning capability.

[0037] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 2 As shown, initializing a population containing N quantum chromosomes can be achieved through step 201, which is explained in detail below: Step 201: Each quantum chromosome consists of a gated unit containing m qubits and a parameter encoding unit containing n qubits connected in series. The gated unit is used to select the control loop type of the inverter through the qubit states, and the parameter encoding unit is used to encode the parameter values ​​of the selected control loop through the qubit states. Each qubit is a superposition of probability amplitudes. Where α and β are complex probability amplitudes, satisfying .

[0038] In this context, the qubit is the fundamental unit of quantum computing, existing as a superposition of probability amplitudes, represented as... Where α and β are complex probability amplitudes, satisfying The control loop type refers to the loop category of the inverter control system, such as voltage control loop, current control loop, or power control loop; the parameter value refers to the specific numerical parameters of the selected control loop.

[0039] In some implementations, each quantum chromosome first connects the gating unit and the parameter encoding unit in series, so that the qubit is in a state of... state and The superposition probability of the states is equal, so the control loop type of the inverter is selected after decoding the state of the qubit. Then, the parameter encoding unit contains n qubits, which are also initialized to a superposition state to encode the parameter values ​​of the selected control loop. This makes the entire series structure enable the quantum chromosome to cover all possible combinations of control loops and parameter values ​​at the same time during initialization, thereby enhancing the search efficiency by utilizing the quantum superposition property.

[0040] For example, assuming a gating unit contains 2 qubits, quantum state decoding may represent 4 types of control loops, such as 00 corresponding to a voltage control loop, 01 corresponding to a current control loop, etc. Meanwhile, a parameter encoding unit contains 3 qubits, which may represent 8 parameter values, such as 000 corresponding to a certain range of parameter values. Thus, a quantum chromosome can encode a specific control loop selection and its parameter settings, which are then collapsed into a specific solution for evaluation during the optimization process through quantum observation operations.

[0041] The probability amplitude of each qubit in each quantum chromosome Initialize to Used to put qubits into a state of state and The probabilities of superposition states are equal.

[0042] For example, suppose a quantum chromosome contains one qubit, and its probability amplitude is initialized to... During quantum observation, there is a 50% probability that the qubit will collapse into... (Representing binary 0), with a 50% probability of collapsing to (Representing binary 1), for example in control loop selection, this can fairly explore both states and avoid initial bias.

[0043] Based on the above technical solution, each qubit exists in a superposition state of probability amplitudes, and the probability amplitude (α,β) of each qubit is initialized to... This makes and Both are equal to 0.5, ensuring that the qubit collapses to 0.5 during observation. state or Each state has a 50% probability, thus fairly covering all possible combinations of control loops and parameter values ​​during population initialization. This enables efficient solution space exploration, avoids premature convergence, and leverages quantum superposition to enhance the algorithm's global search capability and convergence speed, ultimately allowing the optimization process to quickly find the solution with optimal overall performance (such as a balance between dynamic performance, steady-state performance, and harmonic distortion rate). This addresses the problem that existing technologies in grid-connected inverter parameter optimization rely on fixed models or static operating conditions, failing to integrate with real-time operating states, resulting in poor adaptability, and the initialization process is prone to deviations, causing the search to get trapped in local optima and failing to cover the entire solution space.

[0044] In one possible implementation of this application embodiment, the quantum genetic optimization algorithm model can be implemented through the following steps 301 to 303, which are described in detail below: Step 301: The quantum genetic optimization algorithm model is a three-layer evaluation network used to calculate the fitness value F of the solution represented by each quantum chromosome. It includes an input layer, a quantum solution mapping layer, and a fitness calculation layer, where the input layer is connected to the feature vector V. feat and the real-time control parameter set C of the inverter real .

[0045] In some implementations, the input layer receives the feature vector V. feat and the real-time control parameter set C of the inverter real As input.

[0046] Step 302: In each iteration, the quantum solution mapping layer modifies the population... Each quantum chromosome q in i The quantum observation operation collapses the sequence into a binary solution sequence B. i And the binary solution sequence B iThe middle gate control unit is decoded into the control loop selection identifier G. i The binary solution sequence B i The parameter encoding unit is partially decoded into a parameter value set P. i Summary control loop selection identifier G i、 Parameter value set P i、 Feature vector V feat and real-time control parameter set C real Both are output to the fitness calculation layer.

[0047] Quantum observation is a process that causes a qubit to collapse from a superposition state to a classical 0 or 1 state. Binary solution sequence B i It is a binary string obtained after quantum observation. Control loop selection identifier G i This is the identifier obtained after decoding the gating unit, indicating the type of control loop, such as voltage, current, or power. Parameter value set P i It is the set of parameter values ​​obtained after decoding the parameter encoding unit. Real-time control parameter set C real It is the set of control parameters used by the inverter during its current operation.

[0048] It should be noted that quantum observation operations are based on the probability amplitude of qubits to perform random collapse, ensuring the exploration of the solution space; the decoding process relies on predefined mapping rules to convert binary sequences into meaningful control identifiers and parameter values; the aggregated data includes system state characteristics and real-time parameters, providing comprehensive input for fitness calculation.

[0049] For example, suppose that in some iteration, a quantum chromosome q i After being collapsed into a binary sequence Bi="101001" by quantum observation, the gated unit part is decoded into G. i This indicates a voltage control loop; the parameter encoding unit part is decoded as P. i Parameters include a proportional gain of 0.5 and an integration time of 0.1 seconds; these are related to the feature vector Vfeat (such as extracted steady-state features) and the real-time control parameter set C. real (Such as the current voltage value) is summarized and output to the fitness calculation layer.

[0050] Step 303: Obtain the fitness evaluation function by supervising the training of the neural network model using historical data and corresponding system performance indicators. The fitness evaluation function Embedded within the fitness computation layer, the fitness computation layer receives the output of the quantum solution mapping layer and feeds it into the pre-trained fitness evaluation function. In the calculation, the corresponding quantum chromosome q is obtained. i fitness value F iIt is a comprehensive index evaluation value used to characterize the dynamic performance, steady-state performance and harmonic distortion rate of a system.

[0051] Fitness evaluation function This includes dynamic performance evaluation subfunctions, steady-state performance evaluation subfunctions, and harmonic distortion rate evaluation subfunctions: The dynamic performance evaluation subfunction selects the identifier G through the analog control loop. i and parameter value set P i The overshoot M of the system's step response is obtained from the system's dynamic response. p and the settling time t of the system step response s ,pass Calculate the dynamic performance index f d , of which M p,max and t s,max This is the preset maximum allowable overshoot and settling time. p and w s The preset dynamic performance weighting coefficients, and w p +w s =1.

[0052] The steady-state performance evaluation subfunction selects the identifier G through the analog control loop. i and parameter value set P i The steady-state response of the system under the given conditions yields the steady-state error e. ss和 System efficiency η, through Calculate the steady-state performance index f s , where e ss,max and η max For the preset maximum allowable steady-state error and ideal efficiency, w e and w η The preset steady-state performance weighting coefficients are w. e +w η =1.

[0053] The harmonic distortion rate evaluation subfunction selects the identifier G through an analog control loop. i and parameter value set P i The system output waveform was analyzed and Fourier analysis was performed to obtain the THD. Calculate the harmonic performance f of the integrated system h THD max This is the preset maximum permissible total harmonic distortion rate.

[0054] Dynamic performance index f d Steady-state performance index f s and the overall system harmonic performance f h Weighted fusion yields the fitness value F i .

[0055] Among them, overshoot M p This is the maximum deviation of the system response from its steady-state value. Settling time t s This is the time required for the system response to enter the steady-state error band. Steady-state error e ss This is the difference between the system's steady-state output and the target value. System efficiency η is the ratio of output power to input power. Total harmonic distortion (THD) is the ratio of the total effective harmonic distortion (THD) to the effective fundamental harmonic distortion (THD). Weighting coefficient w p w s w e w η This is a coefficient used to adjust the importance of various performance indicators. Maximum allowable value M p,max t s,max e ss,max η max THD max It is the preset performance threshold upper limit.

[0056] In some implementations, the neural network model is trained under supervised conditions using historical data and corresponding system performance metrics to obtain a fitness evaluation function; this function is then embedded into the fitness calculation layer; the fitness calculation layer receives output from the quantum solution mapping layer, including the control loop selection identifier G. i Parameter value set P i eigenvector V feat and real-time control parameter set C real Next, these data are input into the pre-trained fitness evaluation function; the fitness evaluation function contains three sub-functions: a dynamic performance evaluation sub-function, a steady-state performance evaluation sub-function, and a harmonic distortion rate evaluation sub-function, which respectively calculate the dynamic performance index f. d Steady-state performance index f s and the overall system harmonic performance f h The fitness value F is obtained by weighted fusion. i .

[0057] Based on the above technical solution, a three-layer network structure consisting of an input layer, a quantum solution mapping layer, and a fitness calculation layer is constructed, allowing the input layer to receive the feature vector V. feat and real-time control parameter set C real This ensures the optimization process is based on the current system state, enhancing real-time performance; the quantum demapping layer uses quantum observation operations to map the quantum chromosome q... i Collapse into binary solution sequence B i And decoded into the control loop selection identifier G i and parameter value set P iThe results are then aggregated and output to the fitness calculation layer, which maps quantum information to classical solutions, improving search efficiency and diversity. The fitness calculation layer embeds a pre-trained fitness evaluation function, which calculates the index f through dynamic performance evaluation sub-functions, steady-state performance evaluation sub-functions, and harmonic distortion rate evaluation sub-functions, respectively. d f s and f h The fitness value F is obtained by weighted fusion. i This achieves multi-objective comprehensive optimization, ensuring a balanced improvement in system performance. The overall structure supports iterative evolution, driving the population to converge toward the optimal solution.

[0058] In one possible implementation of this application embodiment, the quantum rotating gate operation can be implemented through steps 401 to 403, which are described in detail below: Step 401: Obtain the best individual q in the current population. best With the current individual q i Observe the binary value b of the j-th bit in the solution sequence after collapse. best,j and b i,j ,pass Calculate the fitness value F of the current best individual best With the current individual fitness value F i The difference F.

[0059] Among them, the optimal individual q best It is the quantum chromosome with the highest fitness value in the current population. Current individual q i This is a quantum chromosome being processed within the population. Observation collapse refers to the process of converting a qubit from a superposition state to a classical 0 or 1 state through quantum observation operations. The solution sequence is the binary string obtained after the observation collapse of the quantum chromosome. The binary value of the j-th bit is b. best,j and b i,j Let F represent the values ​​at the j-th position in the solution sequence for the optimal individual and the current individual, respectively. Fitness value F best Fi is a comprehensive performance evaluation value calculated from the pre-trained fitness evaluation function.

[0060] In some implementations, the optimal individual q with the highest fitness value is identified from the current population in each iteration. best and the individual q currently being evaluated i Then, regarding q... best and q i Perform quantum observation operations on each of them to collapse their quantum chromosomes into a binary solution sequence; then extract the binary value b of the j-th bit from the solution sequence. best,j and b i,j Finally, the fitness value F of the current best individual is calculated by arithmetic subtraction. bestWith the current individual fitness value F i The difference F.

[0061] Step 402, through Determine the quantum rotation gate pair q i Each quantum bit The rotation angle θj, where The rotation angle calculated for the j-th qubit in generation t. For the preset rotation angle lookup table, The preset adaptive adjustment coefficient η(t) is used, where the adaptive adjustment coefficient η(t) dynamically decreases as the iteration number t increases, and T max η is the maximum number of iterations. max and η min These are the preset maximum and minimum coefficient values.

[0062] The quantum rotation gate is a matrix operation used to update the probability amplitude of a qubit, adjusting the quantum state by rotating the angular angle θ. j η(t) is the rotation angle applied to the j-th qubit in the quantum rotation gate operation, used to drive population evolution. The adaptive adjustment coefficient η(t) is a coefficient that varies with the iteration number t, used to scale the size of the rotation angle. The iteration number t is the current evolution generation of the algorithm, representing the number of rounds in the optimization process.

[0063] In some implementations, ΔF is obtained through a formula. Calculate the rotation angle θ j This allows for the preset rotation angle values ​​under different ΔF and bit difference conditions based on the statistical results of the relationship between ΔF and binary bit changes in historical optimization data; according to The adaptive adjustment coefficient η(t) decreases dynamically as the number of iterations t increases, thus ensuring that η(t) is large in the early stage of evolution to promote global search, and small in the later stage to promote local convergence. Finally, the probability amplitude of the qubit is updated by the quantum rotation gate matrix to complete the evolution driving of the population.

[0064] Step 403: Through the quantum rotation gate matrix Probability amplitude of updating qubit , where α j and β j For the current individual q i The probability amplitude of the j-th qubit. and This is the new probability amplitude after being updated by the quantum rotation gate.

[0065] The construction of the preset rotation angle lookup table includes: based on the statistical results of the relationship between ΔF and binary bit changes in historical optimization data, preset rotation angle values ​​for different ΔF and bit differences, and summarizing them to form the preset rotation angle lookup table. The rotation angle value refers to the value that allows the population to converge to the global optimum.

[0066] In some implementations, the rotation angle θ of the j-th qubit is used as the basis. j Constructing the quantum rotation gate matrix R(θ_j) yields the current probability magnitude vector. Perform matrix multiplication with the rotating gate matrix to update the probability amplitude. and Applied to the current individual q i The j-th qubit completes the refresh of the probability amplitude.

[0067] It should be noted that the quantum rotation gate matrix is ​​a unitary matrix, ensuring that the updated probability amplitude still satisfies the normalization condition.

[0068] Based on the above technical solution, through parameter η max η min and T max A linearly decreasing adaptive adjustment coefficient η(t) is constructed. In the early stages of iteration, a larger η(t) allows the quantum rotation gate to generate a larger angle, enhancing the exploratory nature of the population and avoiding premature convergence. In the later stages of iteration, a smaller η(t) reduces the rotation angle, accelerating convergence to the optimal solution, thereby improving the algorithm's global search efficiency and convergence stability. Simultaneously, the rotation angle lookup table ξ is a pre-defined mapping table whose input is ΔF and the binary bit difference, yielding the optimized rotation angle value. This makes the determination of the rotation angle no longer fixed or heuristic but data-driven, thus improving the scientific rigor and adaptability of the rotation angle. This allows the algorithm to guide the population to converge to the global optimum more quickly, reducing blind searches. Finally, updating the probability amplitude of each qubit through the quantum rotation gate matrix combines the individualized rotation angle θ. j (t) precisely rotates the probability amplitude of the quantum state, thereby enhancing the algorithm's local fine search capability, enabling individuals to effectively utilize the information of the optimal individual, and improving the quality of the solution.

[0069] In one possible implementation of this application embodiment, the deep autoencoder network performs dimensionality reduction and feature extraction through steps 501 to 502, which are described in detail below: Step 501: Construct a deep autoencoder network comprising an input layer, multiple encoding layers, a bottleneck layer, multiple decoding layers, and an output layer. The input layer receives the original multi-dimensional operating data sequence obtained from the historical database of the grid-connected inverter. The multiple encoding layers consist of at least two fully connected layers. Each encoding layer uses a non-linear activation function to perform feature transformation and gradually reduce the data dimension. Finally, a low-dimensional feature vector is output at the bottleneck layer. The decoding layer is symmetrical to the encoding layer and is used to reconstruct the data from the feature vector of the bottleneck layer. The data dimension is gradually restored through multiple fully connected layers and activation functions. Finally, the reconstructed data is output at the output layer. The original multi-dimensional operating data sequence includes time series data of voltage, current, and power parameters. Each data sample contains measurements at multiple time steps.

[0070] The network consists of several layers: the input layer (first layer) receives input data; multiple encoding layers, each consisting of at least two fully connected layers, use a non-linear activation function to transform features and progressively reduce data dimensionality; a bottleneck layer, located between the encoder and decoder, outputs compressed low-dimensional feature vectors; multiple decoding layers, symmetrical to the encoding layers, reconstruct data from the feature vectors; the output layer, the last layer, outputs the reconstructed data; a data sample is a single data instance containing measurements at multiple time steps; multiple time steps indicate that each sample has records at multiple time points; measurements are the specific numerical values ​​of the parameters; fully connected layers are neural network layers where each neuron is connected to all neurons in the previous layer; non-linear activation functions, such as ReLU or sigmoid, introduce non-linear transformations; feature transformation is the process of changing the representation of data features through network layers; reducing data dimensionality is the operation of reducing the number of data features; and reconstructed data is an approximate reconstruction of the original data from low-dimensional features.

[0071] In some implementations, a deep autoencoder network is first constructed, consisting of an input layer, multiple encoding layers, a bottleneck layer, multiple decoding layers, and an output layer. The input layer receives raw, multi-dimensional operational data sequences containing time-series data of voltage, current, and power parameters from a historical database of grid-connected inverters, with each data sample composed of measurements from multiple time steps. Then, the multiple encoding layers consist of at least two fully connected layers, each applying a non-linear activation function to perform feature transformation, thereby gradually reducing the data dimensionality. Finally, a low-dimensional feature vector is output at the bottleneck layer. Next, multiple decoding layers are symmetrically arranged with the encoding layers, using fully connected layers and activation functions to reconstruct the data starting from the low-dimensional feature vector of the bottleneck layer, gradually restoring the data dimensionality. Finally, the reconstructed data is output at the output layer, completing the entire network process.

[0072] Step 502: Use the original multi-dimensional running data sequences from the historical database as the training dataset, making each data sample both the input and the expected output. Optimize the network parameters using the backpropagation algorithm, minimizing the reconstruction loss function to train the deep autoencoder network. After training, discard the decoder part and retain only the encoder part as the feature extractor. The reconstruction loss function is... Where N is the number of training samples, x i It is the i-th original data sample, x i ′ is the mean squared error loss defined as the i-th data sample in the network reconstruction. It is the Euclidean norm.

[0073] Among them, the backpropagation algorithm is an algorithm used to optimize the parameters of a neural network by updating the weights by calculating gradients; network parameters are adjustable parameters such as weights and biases in a neural network; and the reconstruction loss function is a function that measures the difference between the network output and the expected output.

[0074] In some implementations, the original multi-dimensional running data sequence from the historical database is used as the training dataset, where each data sample serves as both input and expected output. The parameters of the deep autoencoder network are then optimized through backpropagation to minimize the reconstruction loss function, which is defined as mean squared error loss. The difference between the original data sample and the network-reconstructed data sample is calculated using the Euclidean norm. After training, the decoder part can be discarded, and only the encoder part can be retained as a feature extractor for subsequent feature extraction tasks.

[0075] For example, suppose there are 1000 data samples in the historical database, each sample contains 100 time-step measurements of voltage, current and power. During training, each sample is used as input and desired output. The loss function is minimized through backpropagation, such as calculating the mean square error of each sample. After training, the encoder outputs a 10-dimensional feature vector for subsequent optimization tasks.

[0076] Based on the above technical solution, firstly, a deep autoencoder network is used for unsupervised feature extraction, avoiding the limitations of traditional methods that rely on expert experience to manually select features and improving the automation of feature learning. Secondly, a symmetrical encoder-decoder structure is used to achieve data dimensionality reduction and reconstruction, ensuring the accuracy and efficiency of feature extraction and capturing complex patterns in power electronic systems using nonlinear transformations. At the same time, the decoder part is discarded after training, and only the encoder is retained as a lightweight feature extractor, optimizing computational resources and simplifying the model structure. Finally, deep learning is applied to the field of grid-connected inverters to specifically handle high-dimensional nonlinear time series data, improving the robustness and adaptability of system operating state representation.

[0077] In one possible implementation of this application embodiment, the parameter adjustment of the actual inverter control system can be achieved through steps 601 to 604, which are described in detail below: Step 601: Select identifier G according to the control loop. i Determine the target control loop type, which may include voltage control loop, current control loop, or power control loop.

[0078] Among them, the control loop selection identifier G i This refers to the identifier extracted from the binary sequence obtained after decoding the quantum chromosome, used to specify the type of inverter control loop. The target control loop type refers to the category of loops that need to be optimized in the inverter control system, including voltage control loops, current control loops, or power control loops. A voltage control loop is used to regulate the inverter's output voltage. A current control loop is used to regulate the inverter's output current. A power control loop is used to regulate the inverter's output power.

[0079] In some implementations, the identifier G is selected based on the control loop. i The numerical value or code is used to determine the corresponding target control loop type by querying the preset mapping rules or logical judgment process; then the target control loop type is specified as one of voltage control loop, current control loop or power control loop, thus completing the loop type selection process.

[0080] Step 602: Set the parameter value set P i The parameters are converted into parameter configuration instructions that can be recognized by the target control loop according to the preset mapping rules. The mapping rules are based on the control loop parameter specification table defined in the historical database, which stores the parameter addresses and scaling factors corresponding to different control loop types.

[0081] The preset mapping rules refer to predefined conversion rules used to convert parameter value sets into an instruction format recognizable by the control loop. Parameter configuration instructions are parameter setting commands that the control loop can directly recognize and execute. The mapping rules are defined based on a control loop parameter specification table in a historical database. This specification table is a data structure that stores parameter addresses and scaling factors corresponding to different control loop types. A parameter address refers to the memory address or unique identifier of a parameter stored in a digital signal processor or control system. A scaling factor is a coefficient used to adjust the range of parameter values ​​to ensure that the parameter values ​​adapt to the actual needs of the control loop.

[0082] In some implementations, the selection identifier G based on the control loop is obtained. iOnce the target control loop type is determined, the control loop parameter specification table in the historical database is queried to obtain the parameter addresses and scaling factors corresponding to that loop type; then, using a preset mapping rule, the parameter value set P is mapped... i Each parameter value in the process is linearly transformed according to a scaling factor (e.g., multiplying the original parameter value by the scaling factor to obtain the actual parameter value); the transformed parameter value can then be mapped to the corresponding parameter address, generating parameter configuration instructions (binary or specific protocol format) that can be recognized by the target control loop, ensuring that they can be directly sent to the control system for execution.

[0083] Step 603: Send the parameter configuration command to the digital signal processor (DSP) of the actual inverter control system. The DSP updates the adjustable parameters in the control algorithm according to the parameter configuration command, and calculates the duty cycle or frequency through the PID control law to generate the PWM control signal.

[0084] In this context, the actual inverter control system refers to the hardware system that controls the operation of the grid-connected inverter. A digital signal processor (DSP) is a dedicated processor used to execute control algorithms. Adjustable parameters in the control algorithm refer to parameter values ​​that can be adjusted through optimization. PID control law refers to the proportional-integral-derivative control algorithm. Duty cycle refers to the proportion of the high-level time in the pulse-width modulation (PWM) signal to the entire cycle. Frequency refers to the switching frequency of the PWM signal. The PWM control signal is a pulse-width modulation signal used to drive the inverter's power switching devices.

[0085] In some implementations, parameter configuration instructions are sent to the digital signal processor (DSP) of the actual inverter control system via a communication interface. The DSP receives and parses the parameter configuration instructions and updates the adjustable parameters in the control algorithm. Based on the updated adjustable parameters, the DSP calculates the duty cycle or frequency in real time using a PID control law. Finally, the DSP generates a PWM control signal based on the calculation results to complete the output of the control signal.

[0086] Step 604: Drive the inverter power switching devices with PWM control signals to adjust the inverter output waveform.

[0087] Forming a closed-loop optimization includes: High-precision sensors are used to collect time-series data of voltage, current, and power at the inverter output.

[0088] The collected data is filtered and normalized by the data preprocessing unit to generate a new original multi-dimensional running data sequence.

[0089] New raw data sequences are fed back to the historical database in real time to update the training data of the deep autoencoder network, forming a closed-loop optimization cycle.

[0090] In this context, inverter power switching devices refer to the power semiconductor devices used in the inverter to achieve power conversion, such as IGBTs or MOSFETs. Inverter output waveform refers to the shape of the voltage or current waveform at the inverter's output terminal.

[0091] In some implementations, the inverter power switching devices are driven by PWM control signals to adjust the on and off times of the switching devices, thereby controlling the shape and parameters of the inverter output waveform. Then, in the closed-loop optimization, high-precision sensors are used to collect time-series data of voltage, current, and power at the inverter output to ensure the accuracy and real-time performance of the data acquisition. Next, the collected data is filtered by a data preprocessing unit to remove noise and normalized to unify the data scale, generating a new original multi-dimensional operating data sequence. Finally, the new original data sequence is fed back to the historical database in real time to update the training data of the deep autoencoder network, enabling the feature extraction model to adapt to the latest operating state and forming a continuous closed-loop optimization cycle.

[0092] Based on the above technical solution, the target control loop type can be determined by the control loop selection identifier Gi based on the binary sequence decoded from the quantum chromosome. This allows for dynamic switching of control strategies according to real-time operating status, avoiding the limitations of fixed loops and improving the inverter's adaptability and performance optimization potential under different operating conditions. Simultaneously, by converting the parameter value set into parameter configuration instructions recognizable by the target control loop through preset mapping rules, the data-driven characteristics of the specification table can be utilized to ensure hardware compatibility after parameter scaling and address mapping. This achieves standardization and automation of parameter configuration, reducing errors from manual intervention and improving the accuracy and efficiency of parameter optimization. Furthermore, by sending parameter configuration instructions to the digital signal processor (DSP), which updates the adjustable parameters in the control algorithm and calculates the duty cycle or frequency using a PID control law to generate a PWM control signal, the fast processing power of the DSP and the stability of PID control can be combined to achieve direct conversion from parameters to control signals. This ensures precise adjustment of the inverter's output waveform, fast response speed, excellent dynamic performance, and reduced harmonic distortion. Finally, a closed loop of "optimization-application-feedback-re-optimization" is formed, enabling the deep learning model to continuously learn the latest data, thereby adapting to unknown factors such as inverter aging and environmental changes, and improving the robustness of long-term operation.

[0093] The above primarily describes the solutions of the embodiments of this application from the perspective of device implementation. It is understood that each device, such as a grid-connected inverter parameter optimization device based on deep learning and quantum genetics, includes at least one of the hardware structures and software modules corresponding to each function in order to achieve the above-mentioned functions. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is implemented in hardware or by computer software driving hardware 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 implementation should not be considered beyond the scope of this application.

[0094] This application embodiment can divide the grid-connected inverter parameter optimization device based on deep learning and quantum genetics into functional units according to the above method example. For example, each function can be divided into a separate functional unit, or two or more functions can be integrated into the same processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0095] When using integrated units, Figure 7 A possible structural schematic diagram of the grid-connected inverter parameter optimization device based on deep learning and quantum genetics (referred to as grid-connected inverter parameter optimization device 70 based on deep learning and quantum genetics) involved in the above embodiments is shown. The grid-connected inverter parameter optimization device 70 based on deep learning and quantum genetics includes a processing unit 701 and a communication unit 702, and may also include a storage unit 703. Figure 7 The schematic diagram shown can be used to illustrate the structure of the grid-connected inverter parameter optimization device based on deep learning and quantum genetics involved in the above embodiments.

[0096] when Figure 7 The schematic diagram shown illustrates the structure of the grid-connected inverter parameter optimization device based on deep learning and quantum genetics involved in the above embodiments. The processing unit 701 is used to control and manage the operation of the grid-connected inverter parameter optimization device based on deep learning and quantum genetics. The communication unit 702 is used for the grid-connected inverter parameter optimization device based on deep learning and quantum genetics to communicate with other devices. The storage unit 703 is used to store the program code and data of the grid-connected inverter parameter optimization device based on deep learning and quantum genetics.

[0097] For example, communication unit 702 is used to acquire the original multi-dimensional operation data sequence in the historical database of the grid-connected inverter; Processing unit 701 is used to perform dimensionality reduction and feature extraction through a deep autoencoder network to obtain feature vectors representing the system's operating state. A population containing N quantum chromosomes is initialized. Each quantum chromosome consists of a gate unit representing the control loop selection state and a parameter encoding unit representing the control loop parameter values, connected in series. The superposition property of quantum states is utilized to simultaneously cover possible solutions during population initialization. The feature vectors and the inverter's real-time control parameter set are input into the quantum genetic optimization algorithm model. Population evolution is driven by probability amplitude-based quantum rotating gate operations, iteratively calculating the fitness value of the solution represented by each chromosome in the population. The fitness value is calculated by a pre-trained fitness evaluation function and is used to comprehensively evaluate the system's dynamic performance, steady-state performance, and harmonic distortion rate. The optimal fitness value in the current population is compared with a preset performance threshold. If the fitness value is better than the performance threshold, the optimized control parameter set corresponding to the optimal fitness value is sent to the actual inverter control system for parameter adjustment. The inverter output waveform data after parameter adjustment is fed back to the historical database as new raw data, forming a closed-loop optimization.

[0098] In one possible implementation, the processing unit 701 is further configured to initialize a population containing N quantum chromosomes, wherein each quantum chromosome is composed of a gating unit containing m qubits and a parameter encoding unit containing n qubits connected in series, wherein the gating unit is used to select the control loop type of the inverter through the qubit state, the parameter encoding unit is used to encode the parameter value of the selected control loop through the qubit state, and each qubit is a superposition state of probability amplitudes. Where α and β are complex probability amplitudes, satisfying The probability amplitude of each qubit in each quantum chromosome. Initialize to Used to put qubits into a state of state and The probabilities of superposition states are equal.

[0099] In one possible implementation, the processing unit 701 is further configured to include a quantum genetic optimization algorithm model: the quantum genetic optimization algorithm model is a three-layer evaluation network used to calculate the fitness value F of the solution represented by each quantum chromosome, including an input layer, a quantum solution mapping layer, and a fitness calculation layer, wherein the input layer is connected to the feature vector V. feat and the real-time control parameter set C of the inverter real The quantum solution mapping layer, in each iteration, modifies the population... Each quantum chromosome q in i The quantum observation operation collapses the sequence into a binary solution sequence B.i And the binary solution sequence B i The middle gate control unit is decoded into the control loop selection identifier G. i The binary solution sequence B i The parameter encoding unit is partially decoded into a parameter value set P. i Summary control loop selection identifier G i、 Parameter value set P i、 Feature vector V feat and real-time control parameter set C real Both outputs are fed to the fitness calculation layer. The neural network model is then trained under supervised conditions using historical data and corresponding system performance metrics to obtain the fitness evaluation function. The fitness evaluation function Embedded within the fitness computation layer, the fitness computation layer receives the output of the quantum solution mapping layer and feeds it into the pre-trained fitness evaluation function. In the calculation, the corresponding quantum chromosome q is obtained. i fitness value F i It is a comprehensive index evaluation value used to characterize the dynamic performance, steady-state performance and harmonic distortion rate of a system.

[0100] In one possible implementation, the processing unit 701 is also used for the fitness evaluation function. This includes dynamic performance evaluation subfunctions, steady-state performance evaluation subfunctions, and harmonic distortion rate evaluation subfunctions. The dynamic performance evaluation subfunction selects the identifier G through an analog control loop. i and parameter value set P i The overshoot M of the system's step response is obtained from the system's dynamic response. p and the settling time t of the system step response s ,pass Calculate the dynamic performance index f d , of which M p,max and t s,max This is the preset maximum allowable overshoot and settling time. p and w s The preset dynamic performance weighting coefficients, and w p +w s =1. The steady-state performance evaluation subfunction selects the identifier G through the analog control loop. i and parameter value set P i The steady-state response of the system under the given conditions yields the steady-state error e. ss和 System efficiency η, through Calculate the steady-state performance index f s , where e ss,max and η max For the preset maximum allowable steady-state error and ideal efficiency, w e and w ηThe preset steady-state performance weighting coefficients are w. e +w η =1. The harmonic distortion rate evaluation subfunction selects the identifier G through an analog control loop. i and parameter value set P i The system output waveform was analyzed and Fourier analysis was performed to obtain the THD. Calculate the harmonic performance f of the integrated system h THD max The preset maximum permissible total harmonic distortion (THD). The dynamic performance index f... d Steady-state performance index f s and the overall system harmonic performance f h Weighted fusion yields the fitness value F i .

[0101] In one possible implementation, the processing unit 701 is also used for quantum rotation gate operations, including: obtaining the best individual q in the current population. best With the current individual q i Observe the binary value b of the j-th bit in the solution sequence after collapse. best,j and b i,j ,pass Calculate the fitness value F of the current best individual best With the current individual fitness value F i The difference F. Through Determine the quantum rotation gate pair q i Each quantum bit The rotation angle θj, where The rotation angle calculated for the j-th qubit in generation t. For the preset rotation angle lookup table, The preset adaptive adjustment coefficient η(t) is used, where the adaptive adjustment coefficient η(t) dynamically decreases as the iteration number t increases, and T max η is the maximum number of iterations. max and η min These are the preset maximum and minimum coefficient values. (Using a quantum rotation gate matrix) Probability amplitude of updating qubit , where α j and β j For the current individual q i The probability amplitude of the j-th qubit. and This is the new probability amplitude after being updated by the quantum rotation gate.

[0102] In one possible implementation, the processing unit 701 is further configured to construct a preset rotation angle lookup table by: based on the statistical results of the relationship between ΔF and binary bit changes in historical optimization data, preset rotation angle values ​​for different ΔF and bit differences, and summarizing them to form a preset rotation angle lookup table. The rotation angle value refers to the value that allows the population to converge to the global optimum.

[0103] In one possible implementation, the processing unit 701 is also used for dimensionality reduction and feature extraction of the deep autoencoder network, including: constructing a deep autoencoder network comprising an input layer, multiple encoding layers, a bottleneck layer, multiple decoding layers, and an output layer. The input layer receives the original multi-dimensional operating data sequence obtained from the historical database of the grid-connected inverter. The multiple encoding layers consist of at least two fully connected layers. Each encoding layer uses a non-linear activation function for feature transformation and gradually reduces the data dimensionality, finally outputting a low-dimensional feature vector at the bottleneck layer. The decoding layer is symmetrical to the encoding layer and is used to reconstruct the data from the feature vector of the bottleneck layer. The data dimensionality is gradually restored through multiple fully connected layers and activation functions, and finally the reconstructed data is output at the output layer. The original multi-dimensional operating data sequence includes time series data of voltage, current, and power parameters, and each data sample contains measurements at multiple time steps. The original multi-dimensional operating data sequence in the historical database is used as the training dataset, with each data sample serving as both input and expected output. The network parameters are optimized using a backpropagation algorithm, and the deep autoencoder network is trained by minimizing the reconstruction loss function. After training, the decoder part is discarded, and only the encoder part is retained as a feature extractor. The reconstruction loss function is... Where N is the number of training samples, x i It is the i-th original data sample, x i ′ is the mean squared error loss defined as the i-th data sample in the network reconstruction. It is the Euclidean norm.

[0104] In one possible implementation, the processing unit 701 is also used to perform parameter adjustment for the actual inverter control system, including: selecting the identifier G according to the control loop. i Determine the target control loop type, which may be a voltage control loop, current control loop, or power control loop. Set the parameter value set P... iThe parameter configuration instructions are converted into recognizable parameter configuration commands for the target control loop according to preset mapping rules. These rules are based on a control loop parameter specification table defined in a historical database, which stores parameter addresses and scaling factors corresponding to different control loop types. The parameter configuration commands are then sent to the digital signal processor (DSP) of the actual inverter control system. The DSP updates the adjustable parameters in the control algorithm based on the configuration commands and calculates the duty cycle or frequency using a PID control law to generate a PWM control signal. This PWM control signal drives the inverter's power switching devices, adjusting the inverter's output waveform.

[0105] In one possible implementation, the processing unit 701 is further configured to form a closed-loop optimization by: acquiring time-series data of voltage, current, and power at the inverter output using high-precision sensors; filtering and normalizing the acquired data through a data preprocessing unit to generate a new original multi-dimensional operating data sequence; and feeding the new original data sequence back to the historical database in real time to update the training data of the deep autoencoder network, thus forming a closed-loop optimization loop.

[0106] The processing unit 701 can be a processor or a controller, and the communication unit 702 can be a communication interface, transceiver, transceiver circuit, transceiver device, etc. The term "communication interface" is a general term and may include one or more interfaces. The storage unit 703 can be a memory. When the grid-connected inverter parameter optimization device 70 based on deep learning and quantum genetics is a chip, the processing unit 701 can be a processor or a controller, and the communication unit 702 can be an input interface and / or an output interface, pins, or circuits, etc. The storage unit 703 can be a storage unit within the chip (e.g., a register, cache, etc.) or a storage unit located outside the chip (e.g., read-only memory (ROM), random access memory (RAM, etc.)).

[0107] The communication unit can also be called a transceiver unit. The antenna and control circuit with transceiver functions in the grid-connected inverter parameter optimization device 70 based on deep learning and quantum genetics can be considered as the communication unit 702 of the grid-connected inverter parameter optimization device 70 based on deep learning and quantum genetics, and the processor with processing functions can be considered as the processing unit 701 of the grid-connected inverter parameter optimization device 70 based on deep learning and quantum genetics. Optionally, the device in the communication unit 702 used to implement the receiving function can be considered as the communication unit, which is used to execute the receiving steps in the embodiments of this application. The communication unit can be a receiver, a receiver circuit, etc. The device in the communication unit 702 used to implement the transmitting function can be considered as the transmitting unit, which is used to execute the transmitting steps in the embodiments of this application. The transmitting unit can be a transmitter, a transmitter, a transmitting circuit, etc.

[0108] Figure 7 If the integrated units in the process 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 solutions of the embodiments of this application, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, 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.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. Storage media for storing computer software products include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0109] In implementation, each step of the method provided in this embodiment can be completed by integrated logic circuits in the processor or by instructions in software form. The steps of the method disclosed in the embodiments of this application can be directly manifested as being executed by a hardware processor, or being executed by a combination of hardware and software modules in the processor.

[0110] The processor in this application may include, but is not limited to, at least one of the following: a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), a microcontroller unit (MCU), or an artificial intelligence processor, etc., and other computing devices that run software. Each computing device may include one or more cores for executing software instructions to perform calculations or processing. The processor may be a standalone semiconductor chip or integrated with other circuits into a single semiconductor chip. For example, it may form a System-on-a-Chip (SoC) with other circuits (such as encoding / decoding circuits, hardware acceleration circuits, or various bus and interface circuits), or it may be integrated as a built-in processor within an ASIC. The ASIC of this integrated processor may be packaged separately or together with other circuits. In addition to the cores for executing software instructions to perform calculations or processing, the processor may further include necessary hardware accelerators, such as field-programmable gate arrays (FPGAs), programmable logic devices (PLDs), or logic circuits that implement dedicated logic operations.

[0111] The memory in the embodiments of this application may include at least one of the following types: read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions; random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions; or electrically erasable programmable-only memory (EEPROM). In some scenarios, the memory may also be compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0112] This application also provides a computer-readable storage medium including instructions that, when run on a computer, cause the computer to perform any of the methods described above.

[0113] This application also provides a computer program product containing instructions that, when run on a computer, cause the computer to perform any of the methods described above.

[0114] This application also provides a chip including a processor and an interface circuit. The interface circuit is coupled to the processor. The processor is used to run computer programs or instructions to implement the above-described method. The interface circuit is used to communicate with other modules outside the chip.

[0115] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software programs, implementation can be, in whole or in part, in the form of a computer program product. This computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device containing one or more servers, data centers, etc., that can be integrated with the medium. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0116] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, the disclosure, and the appended claims, can understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0117] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely illustrative descriptions of the application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and variations.

Claims

1. A method for optimizing grid-connected inverter parameters based on deep learning and quantum genetics, characterized in that, include: The original multi-dimensional operation data sequence of the grid-connected inverter historical database is obtained, and dimensionality reduction and feature extraction are performed through a deep autoencoder network to obtain the feature vector representing the system operation status. Initialize a population containing N quantum chromosomes. Each quantum chromosome consists of a gating unit that characterizes the selection state of the control loop and a parameter encoding unit that characterizes the parameter values ​​of the control loop, connected in series. Utilize the superposition property of quantum states to simultaneously cover all possible solutions during population initialization. The feature vector and the real-time control parameter set of the inverter are input into the quantum genetic optimization algorithm model. The population evolution is driven by quantum rotating gate operation based on probability amplitude, and the fitness value of the solution represented by each chromosome in the population is calculated iteratively. The fitness value is calculated by a pre-trained fitness evaluation function and is used to comprehensively evaluate the system's dynamic performance, steady-state performance, and harmonic distortion rate. The optimal fitness value in the current population is compared with a preset performance threshold. If the fitness value is better than the performance threshold, the optimized control parameter set corresponding to the optimal fitness value is sent to the actual inverter control system for parameter adjustment. The inverter output waveform data after parameter adjustment is fed back to the historical database as new raw data to form a closed-loop optimization.

2. The method for optimizing grid-connected inverter parameters based on deep learning and quantum genetics according to claim 1, characterized in that, The initialization of the population comprising N quantum chromosomes includes: Each quantum chromosome consists of a gated unit containing m qubits and a parameter encoding unit containing n qubits connected in series. The gated unit is used to select the control loop type of the inverter through the qubit states, and the parameter encoding unit is used to encode the parameter values ​​of the selected control loop through the qubit states. Each qubit is a superposition of probability amplitudes. Where α and β are complex probability amplitudes, satisfying ; The probability amplitude of each qubit in each quantum chromosome Initialize to Used to put qubits into a state of state and The probabilities of superposition states are equal.

3. The method for optimizing grid-connected inverter parameters based on deep learning and quantum genetics according to claim 2, characterized in that, The quantum genetic optimization algorithm model includes: The quantum genetic optimization algorithm model is a three-layer evaluation network used to calculate the fitness value F of the solution represented by each quantum chromosome, including an input layer, a quantum solution mapping layer, and a fitness calculation layer, wherein the input layer is connected to the feature vector V. feat and the real-time control parameter set C of the inverter real ; The quantum solution mapping layer, in each iteration, will convert the population... Each quantum chromosome q in i The quantum observation operation collapses the sequence into a binary solution sequence B. i and the binary solution sequence B i The middle gate control unit is decoded into the control loop selection identifier G. i The binary solution sequence B i The parameter encoding unit is partially decoded into a parameter value set P. i Summary control loop selection identifier G i、 Parameter value set P i、 Feature vector V feat and real-time control parameter set C real Both outputs are sent to the fitness calculation layer; The fitness evaluation function is obtained by supervising the training of the neural network model using historical data and corresponding system performance indicators. The fitness evaluation function Embedded within the fitness computation layer, the fitness computation layer receives the output of the quantum demapping layer and feeds it into a pre-trained fitness evaluation function. In the calculation, the corresponding quantum chromosome q is obtained. i fitness value F i It is a comprehensive index evaluation value used to characterize the dynamic performance, steady-state performance and harmonic distortion rate of a system.

4. The method for optimizing grid-connected inverter parameters based on deep learning and quantum genetics according to claim 3, characterized in that, The fitness evaluation function Includes dynamic performance evaluation subfunctions, steady-state performance evaluation subfunctions, and harmonic distortion rate evaluation subfunctions: The dynamic performance evaluation subfunction selects the identifier G by simulating the control loop. i and parameter value set P i The overshoot M of the system's step response is obtained from the system's dynamic response. p and the settling time t of the system step response s ,pass Calculate the dynamic performance index f d , of which M p,max and t s,max The preset maximum allowable overshoot and settling time; w p and w s The preset dynamic performance weighting coefficients, and w p +w s =1; The steady-state performance evaluation subfunction selects the identifier G by simulating the control loop. i and parameter value set P i The steady-state response of the system under the given conditions yields the steady-state error e. ss和 System efficiency η, through Calculate the steady-state performance index f s , where e ss,max and η max For the preset maximum allowable steady-state error and ideal efficiency, w e and w η The preset steady-state performance weighting coefficients are w. e +w η =1; The harmonic distortion rate evaluation subfunction selects the identifier G by simulating the control loop. i and parameter value set P i The system output waveform was analyzed and Fourier analysis was performed to obtain the THD. Calculate the harmonic performance f of the integrated system h THD max This is the preset maximum permissible total harmonic distortion rate; The dynamic performance index f d Steady-state performance index f s and the overall system harmonic performance f h Weighted fusion yields the fitness value F i .

5. The method for optimizing grid-connected inverter parameters based on deep learning and quantum genetics according to claim 4, characterized in that, The quantum rotating door operation includes: Obtain the best individual q in the current population best With the current individual q i Observe the binary value b of the j-th bit in the solution sequence after collapse. best,j and b i,j ,pass Calculate the fitness value F of the current best individual best With the current individual fitness value F i The difference F; pass Determine the quantum rotation gate pair q i Each quantum bit The rotation angle θj, where The rotation angle calculated for the j-th qubit in generation t. For the preset rotation angle lookup table, The preset adaptive adjustment coefficient η(t) is used, wherein the adaptive adjustment coefficient η(t) dynamically decreases as the iteration number t increases, and T max η is the maximum number of iterations. max and η min These are the preset maximum and minimum coefficient values; Through quantum rotation gate matrix Probability amplitude of updating qubit , where α j and β j For the current individual q i The probability amplitude of the j-th qubit. and This is the new probability amplitude after being updated by the quantum rotation gate.

6. The method for optimizing grid-connected inverter parameters based on deep learning and quantum genetics according to claim 5, characterized in that, The construction of the preset rotation angle lookup table includes: based on the statistical results of the relationship between ΔF and binary bit changes in historical optimization data, preset rotation angle values ​​for different ΔF and bit differences, and summarizing them to form the preset rotation angle lookup table. The rotation angle value refers to the value that allows the population to converge to the global optimum.

7. The method for optimizing grid-connected inverter parameters based on deep learning and quantum genetics according to claim 6, characterized in that, The deep autoencoder network performs dimensionality reduction and feature extraction, including: A deep autoencoder network is constructed, comprising an input layer, multiple encoding layers, a bottleneck layer, multiple decoding layers, and an output layer. The input layer receives raw multi-dimensional operating data sequences obtained from a historical database of grid-connected inverters. The multiple encoding layers consist of at least two fully connected layers. Each encoding layer uses a nonlinear activation function to perform feature transformation and gradually reduce the data dimension, finally outputting a low-dimensional feature vector at the bottleneck layer. The decoding layer is symmetrical to the encoding layer and is used to reconstruct data from the feature vector of the bottleneck layer. The data dimension is gradually restored through multiple fully connected layers and activation functions, and finally the reconstructed data is output at the output layer. The raw multi-dimensional operating data sequences include time series data of voltage, current, and power parameters, and each data sample contains measurement values ​​at multiple time steps. The original multi-dimensional running data sequences from the historical database are used as the training dataset. Each data sample serves as both input and expected output. The network parameters are optimized using the backpropagation algorithm, and the deep autoencoder network is trained by minimizing the reconstruction loss function. After training, the decoder part is discarded, and only the encoder part is retained as the feature extractor. The reconstruction loss function is... Where N is the number of training samples, x i It is the i-th original data sample, where x i ′ is the mean squared error loss defined as the i-th data sample in the network reconstruction. It is the Euclidean norm.

8. The method for optimizing grid-connected inverter parameters based on deep learning and quantum genetics according to claim 7, characterized in that, The actual inverter control system performs parameter adjustments, including: Select identifier G based on control loop. i Determine the target control loop type, which includes a voltage control loop, a current control loop, or a power control loop; The parameter value set P i The parameter configuration instructions are converted into recognizable parameters of the target control loop according to the preset mapping rules. The mapping rules are based on the control loop parameter specification table defined in the historical database, where the control loop parameter specification table stores the parameter addresses and scaling factors corresponding to different control loop types. The parameter configuration command is sent to the digital signal processor (DSP) of the actual inverter control system. The DSP updates the adjustable parameters in the control algorithm according to the parameter configuration command and calculates the duty cycle or frequency through the PID control law to generate the PWM control signal. The inverter's output waveform is adjusted by driving the inverter's power switching devices with a PWM control signal.

9. The method for optimizing grid-connected inverter parameters based on deep learning and quantum genetics according to claim 8, characterized in that, The closed-loop optimization includes: High-precision sensors are used to collect time-series data of voltage, current and power at the inverter output. The collected data is filtered and normalized by the data preprocessing unit to generate a new original multi-dimensional running data sequence. New raw data sequences are fed back to the historical database in real time to update the training data of the deep autoencoder network, forming a closed-loop optimization cycle.

10. A grid-connected inverter parameter optimization device based on deep learning and quantum genetics, characterized in that, The device includes: a communication unit and a processing unit; The communication unit is used to acquire the original multi-dimensional operation data sequence from the historical database of the grid-connected inverter; The processing unit is used to perform dimensionality reduction and feature extraction through a deep autoencoder network to obtain a feature vector representing the system's operating state. Initialize a population containing N quantum chromosomes. Each quantum chromosome consists of a gating unit that characterizes the selection state of the control loop and a parameter encoding unit that characterizes the parameter values ​​of the control loop, connected in series. Utilize the superposition property of quantum states to simultaneously cover all possible solutions during population initialization. The eigenvectors and the real-time control parameter set of the inverter are input into the quantum genetic optimization algorithm model. Population evolution is driven by quantum rotating gate operation based on probability amplitude, and the fitness value of the solution represented by each chromosome in the population is calculated iteratively. The fitness value is calculated by a pre-trained fitness evaluation function and is used to comprehensively evaluate the dynamic performance, steady-state performance and harmonic distortion rate of the system. The optimal fitness value in the current population is compared with a preset performance threshold. If the fitness value is better than the performance threshold, the optimized control parameter set corresponding to the optimal fitness value is sent to the actual inverter control system for parameter adjustment. The inverter output waveform data after parameter adjustment is fed back to the historical database as new raw data to form a closed-loop optimization.

Images