Mechanism knowledge and data fusion driven dc converter modeling method and system

Abstract

A mechanism knowledge and data fusion driven direct current converter modeling method and system, the method comprises combining the mechanism knowledge of the direct current converter circuit, obtaining the direct current converter simulation data set; the pre-established data driven model is trained by using the obtained direct current converter simulation data set; the target circuit parameters corresponding to the highest efficiency of the data driven model under different working conditions are found by using the trained data driven model and adopting the particle swarm algorithm; the pre-established BP neural network is trained by using the target circuit parameters, and the trained BP neural network is used to output the duty cycle of the direct current converter control signal when the external working condition changes, and the modeling of the direct current converter is completed. The present application solves the problems of poor accuracy and heavy artificial burden in the previous modeling method based on mechanism knowledge, and the defects of weak generalization ability of the existing optimization algorithm, and can give the optimal duty cycle of the converter switch tube driving control signal in time under variable working condition.

Description

Technical Field

[0001] This invention belongs to the field of microgrid control technology, specifically relating to a DC-DC converter modeling method and system driven by mechanism knowledge and data fusion. Background Technology

[0002] Microgrids are an important way to utilize distributed energy resources, better able to cope with fluctuations and changes in renewable energy, thereby enhancing the reliability and security of the power supply system. For example... Figure 1 As shown, a microgrid combines distributed energy resources, transmission and distribution systems, energy storage devices, and loads to form a small power grid. As a power system, it can operate independently or be connected to a larger power grid. Among these, the DC / DC converter, as one of the energy conversion devices, includes boost, buck, and buck-boost converters, which can convert DC power to different voltage levels, enabling the interconnection of distributed power sources with the DC bus and promoting the development and application of distributed power sources. Through accurate modeling and optimization of DC / DC converters, the operational reliability and efficiency of microgrids can be improved, energy losses reduced, energy conservation and emission reduction promoted, and intelligent control of microgrids achieved. Therefore, research on DC / DC converters is particularly important.

[0003] Traditional DC / DC converter modeling methods are based on mechanistic knowledge, requiring circuit modeling and tedious formula derivations, resulting in a high degree of human intervention. Furthermore, neglecting parasitic parameters of circuit components, linearly simplifying device characteristics, and approximating details in the derivation process can all reduce model accuracy, necessitating the search for a high-precision modeling method. To simplify DC / DC converter modeling, many researchers have begun to adopt data-driven modeling approaches. Data-driven modeling is a black-box model acquisition method that does not focus on the internal circuit topology of the converter but utilizes its port characteristics to build the circuit model. Compared to mechanistic knowledge-based methods, this approach helps to avoid the limitations of low model accuracy and heavy human workload associated with traditional modeling methods. Currently, a common method in data-driven modeling of power electronic converters is to use data-intensive basic networks, trained with large datasets, which is time-consuming.

[0004] With the advancement of computer technology, some optimization algorithms have been applied to the field of power electronics, such as the classic Particle Swarm Optimization (PSO) algorithm, Genetic Algorithm (GA), and Ant Colony Optimization (ACO) algorithm. In traditional modeling methods, these algorithms are often used to tune controller parameters to improve system performance. In data-driven modeling, these optimization algorithms typically aim to maximize the efficiency or minimize the current stress of the DC / DC converter, seeking the optimal parameters within the rated operating range. After continuous iteration, they obtain the circuit parameters corresponding to the highest efficiency or lowest current stress. While this method can obtain the most efficient set of circuit parameter combinations, when external operating conditions change, the optimization algorithm needs to re-find the circuit parameters that maximize efficiency, and it cannot perform real-time efficiency-optimal control of the converter. Therefore, current optimization algorithms can only optimize for specific operating conditions, meaning that the versatility of these optimization methods needs to be improved.

[0005] Purpose of the invention

[0006] The purpose of this invention is to address the problems in the prior art by providing a DC-DC converter modeling method and system driven by mechanistic knowledge and data fusion. Under varying operating conditions, it can promptly provide the optimal duty cycle of the converter switching transistor drive control signal, thereby improving the accuracy and training speed of DC-DC converter modeling, as well as the generalization ability of the model optimization process.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] Firstly, a DC-DC converter modeling method driven by the fusion of mechanistic knowledge and data is provided, including:

[0009] By combining the mechanistic knowledge of DC-DC converter circuits, a DC-DC converter simulation dataset is obtained;

[0010] The pre-established data-driven model is trained using the acquired DC-DC converter simulation dataset;

[0011] Using a trained data-driven model, the particle swarm optimization algorithm is used to find the target circuit parameters that achieve the highest efficiency of the data-driven model under different operating conditions.

[0012] The target circuit parameters are used to train a pre-built BP neural network. The trained BP neural network is then used to output the duty cycle of the DC-DC converter control signal when the external operating conditions change, thus completing the modeling of the DC-DC converter.

[0013] As a preferred approach, for a Buck-Boost DC-DC converter, the corresponding DC-DC converter simulation dataset includes:

[0014] Input DC voltage Output DC voltage Duty cycle of the switching transistor pulse width modulation signal Load resistance Input power Output power and circuit transmission efficiency ;

[0015] Based on the principles of DC-DC converter circuits, in a Buck-Boost circuit, given a set of input DC voltages... Duty cycle of the switching transistor pulse width modulation signal and load resistance Then there is a corresponding input power. Output power and output DC voltage After determining the rated operating range of the circuit, input DC voltage. Duty cycle of the switching transistor pulse width modulation signal and load resistance By combining different values, the output DC voltage is obtained according to the following formula. Scope:

[0016]

[0017] The output power is obtained by the following formula. Scope:

[0018]

[0019] For input DC voltage With output DC voltage Both are measured using a voltmeter; for input power With output power The instantaneous values ​​of the input and output voltages and currents are measured using a voltmeter and an ammeter, respectively, and then calculated using the following formula:

[0020]

[0021] From input power With output power The circuit transmission efficiency can be obtained by the following formula. :

[0022] .

[0023] As a preferred embodiment, the data-driven model employs the LightGBM model. When training the LightGBM model using the acquired DC-DC converter simulation dataset, the input DC voltage is... Output DC voltage Load resistance As input to the LightGBM model, the duty cycle of the pulse width modulation signal of the switching transistor is used. With circuit transmission efficiency As the output of the LightGBM model, the output of the LightGBM model should be as consistent as possible with the true value.

[0024] As a preferred approach, in the step of training the pre-established data-driven model using the acquired DC-DC converter simulation dataset, the hyperparameters of the LightGBM model are adjusted using a grid search method. In the process of providing hyperparameters, the range and value interval of the hyperparameters are gradually reduced until the model evaluation score meets the requirements after multiple runs. Then, regularization parameters are added, and the grid search method is used to continue optimization until the model evaluation score meets the requirements. The hyperparameters corresponding to this point are the target hyperparameters.

[0025] As a preferred embodiment, the step of using a trained data-driven model and employing a particle swarm optimization algorithm to find the target circuit parameters corresponding to the highest efficiency of the data-driven model under different operating conditions includes:

[0026] The velocities and positions of the particles are randomly initialized, the fitness value of each particle is calculated, and the individual best historical position of each particle is determined. With the historical best position of the entire particle swarm ;

[0027] Update the velocity and position of each particle using the following formula:

[0028]

[0029]

[0030] In the formula, Indicates the first During the nth iteration, the 1st The velocity of each particle; Indicates the first During the nth iteration, the 1st The position of each particle; Indicates the first The particle reached the [number]th [particle]. The optimal position in the next iteration; Indicates up to the number The optimal positions of all particles in the next iteration; , The inertia weight factor is calculated using a linear decreasing strategy based on the following formula: (The random number between 0 and 1 is used.) :

[0031]

[0032] In the formula, For the number of iterations, The maximum number of iterations, , Individual learning coefficient and global learning coefficient;

[0033] After updating the particle velocity and position, recalculate the fitness function for each particle to determine its individual best historical position. With the historical best position of the entire particle swarm Update until the maximum number of iterations is reached;

[0034] In the process of using the particle swarm optimization algorithm for optimization, the load will be... By varying the input DC voltage within the rated range, we can find the input DC voltage corresponding to the highest efficiency for different values. Duty cycle of the switching transistor pulse width modulation signal and output DC voltage The circuit parameters that achieve the highest efficiency under different operating conditions are combined to form the target circuit parameters.

[0035] As a preferred embodiment, the step of training a pre-built BP neural network using target circuit parameters, and then using the trained BP neural network to output the duty cycle of the DC-DC converter control signal when external operating conditions change, thereby completing the modeling of the DC-DC converter, includes:

[0036] With load resistor As input to the BP neural network, the duty cycle of the switching transistor pulse width modulation signal is used. Output DC voltage Input DC voltage As output, the BP neural network is initialized, including the number of inputs in the input layer, the number of hidden layers, the number of nodes in each hidden layer, and the number of outputs in the output layer. Weights are then assigned to each neuron in the hidden layers. The number of nodes in the hidden layers is determined by the following formula:

[0037]

[0038] In the formula, This represents the number of hidden layer nodes. The number of nodes in the input layer. This represents the number of nodes in the output layer. This is the adjustment constant;

[0039] The hidden layer neurons are computed through forward propagation, and their outputs are represented as follows:

[0040]

[0041] In the formula, This represents the activation function. For the weight parameters in the network, use gradient descent to optimize the network weight parameters according to the cost function;

[0042] The trained BP neural network is exported and used as the controller of the Buck-Boost circuit, under load resistance... When changes occur, the controller provides the optimal duty cycle corresponding to the maximum efficiency, thus completing the online optimization of the data-driven model.

[0043] As a preferred approach, a DC-DC converter simulation dataset is obtained through MATLAB or Simulink. After online optimization of the data-driven model, the neural network controller in the MATLAB or Simulink simulation is converted into C language code and downloaded to a digital signal processor (DSP) to control the DC-DC converter.

[0044] Secondly, a DC-DC converter modeling system driven by the fusion of mechanism knowledge and data is provided, including:

[0045] The DC-DC converter simulation dataset acquisition module is used to acquire DC-DC converter simulation datasets by combining the mechanistic knowledge of DC-DC converter circuits.

[0046] The data-driven model training module is used to train a pre-established data-driven model using the acquired DC-DC converter simulation dataset.

[0047] The target circuit parameter acquisition module is used to use a trained data-driven model and a particle swarm optimization algorithm to find the target circuit parameters corresponding to the highest efficiency of the data-driven model under different operating conditions.

[0048] The DC-DC converter control signal duty cycle output module is used to train a pre-built BP neural network using the target circuit parameters. When the external operating conditions change, the trained BP neural network outputs the DC-DC converter control signal duty cycle to complete the modeling of the DC-DC converter.

[0049] Thirdly, an electronic device is provided, comprising:

[0050] A memory that stores at least one instruction; and a processor that executes the instructions stored in the memory to implement the mechanism knowledge and data fusion-driven DC-DC converter modeling method.

[0051] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the DC-DC converter modeling method driven by the fusion of the underlying mechanism knowledge and data.

[0052] Compared with the prior art, the present invention has at least the following beneficial effects:

[0053] This invention employs a data-driven modeling approach based on mechanistic knowledge and lightweight data fusion to model DC-DC converters. It utilizes a particle swarm optimization (PSO) algorithm to find the target circuit parameters that maximize efficiency under different operating conditions. The target circuit parameters are then used to train a backpropagation (BP) neural network (BPNN) for online optimization of the DC-DC converter model. This approach addresses the issues of poor accuracy and heavy manual workload associated with previous mechanistic-based modeling methods, as well as the limitations of existing optimization algorithms in generalization. It can provide the optimal duty cycle for the converter switching control signals in a timely manner under varying operating conditions. First, based on the mechanistic knowledge of the DC-DC converter circuit, this invention collects operating data for data-driven modeling, avoiding the low accuracy issues caused by approximations in traditional model construction and derivation, thus improving model accuracy and reducing manual workload. Second, it uses maximum transmission efficiency as the optimization objective and employs a PSO algorithm to search for the optimal circuit parameters under different operating conditions. These parameters are then used to train a backpropagation (BP) neural network, enabling the BP neural network to output the optimal duty cycle for the converter control signal, thereby optimizing the controller design. The resulting BP neural network can provide the optimal duty cycle for the corresponding control signal online and in a timely manner after changes in external operating conditions.

[0054] Furthermore, this invention employs the LightGBM model for data-driven modeling of the converter. Compared with neural networks, the LightGBM model can reduce the data requirements of the data-driven model, efficiently process data during training, and improve the training speed of DC-DC converter modeling. Attached Figure Description

[0055] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0056] Figure 1 Schematic diagram of a DC microgrid structure;

[0057] Figure 2 A schematic diagram of the LightGBM model training process according to an embodiment of the present invention;

[0058] Figure 3 Flowchart of the DC-DC converter modeling method driven by mechanism knowledge and data fusion in this invention embodiment;

[0059] Figure 4 Statistical chart of efficiency error percentage of LightGBM model in embodiments of the present invention;

[0060] Figure 5 A statistical chart of the duty cycle error percentage of the LightGBM model in an embodiment of the present invention. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, those skilled in the art can obtain other embodiments without creative effort.

[0062] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0063] This invention proposes a DC-DC converter modeling method driven by the fusion of mechanism knowledge and data, which can provide the optimal duty cycle of the converter switching control signal in a timely manner under varying operating conditions.

[0064] In DC / DC converters, this embodiment of the invention uses a buck-boost converter as an example for modeling and optimization analysis. For other DC / DC converters, only the input and output variables of the LightGBM model need to be changed; the remaining steps are consistent with the buck-boost modeling and optimization process. Specifically, this embodiment of the invention includes the following steps:

[0065] Step 1: Obtaining the Buck-Boost Transformer Dataset

[0066] Step 1.1: Building the Buck-Boost Converter Simulation Model

[0067] In DC-DC converters, the Buck-Boost converter has advantages such as simple structure and small size. By adjusting the duty cycle of the power transistors, it can achieve both boost and buck functions, and is widely used for voltage level conversion in DC microgrids. For example... Figure 1 As shown in the circuit on the right, the topology of the Buck-Boost converter consists of a power supply and power switching devices. (In this embodiment of the invention, an IGBT is used), freewheeling diode Output filter inductor Output filter capacitor Composed of a load, the input DC voltage and the output DC voltage are: and The load resistance is To better reflect real-world operation, parameters such as line resistance and the internal resistance of the power supply and power switching devices were included in the simulation. Inductance and resistance Diode resistor capacitors and resistors .

[0068] Step 1.2: Determine the input / output range

[0069] As we know from the converter mechanism, in a Buck-Boost circuit, given a set of input voltages... Duty cycle of the switching transistor pulse width modulation (PWM) signal Load resistance There will be a corresponding input power. Output power Output voltage After determining the rated operating range of the circuit, , , By combining different values, the range of output voltage for each combination can be known from formula (1), and the range of output power can be known from formula (2).

[0070] (1)

[0071] (2)

[0072] Step 1.3: Obtain simulation data

[0073] The dataset to be acquired includes: , , , , , Each data point is output to the MATLAB workspace via the "To Workplace" module in Simulink. and Both were measured using a voltmeter. and The instantaneous values ​​of the input and output voltage and current are measured using a voltmeter and an ammeter, respectively. After the circuit stabilizes, the values ​​are obtained using formula (3):

[0074] (3)

[0075] Depend on and The circuit transmission efficiency can be obtained using formula (4). :

[0076] (4)

[0077] Step 2: LightGBM Model Training and Saving

[0078] exist , , , , The data set will consist of , , As input to the LightGBM model, and As output, the model is trained using the dataset to make the model output as consistent as possible with the true value.

[0079] Step 2.1: Model Training Process

[0080] The core idea of ​​Gradient Boosting Decision Tree (GBDT) is to build an optimal model by iteratively training weak classifiers (decision trees). This model exhibits good performance during training and is less prone to overfitting. LightGBM is a framework for implementing the GBDT algorithm, consisting of a series of weak decision trees, supporting efficient parallel training. Compared to other algorithms, LightGBM offers faster training speed, lower memory consumption, and higher accuracy, enabling it to quickly process large amounts of data.

[0081] After importing the data, it is randomly divided into training and test sets. For the input data, LightGBM uses a histogram algorithm to discretize the continuous floating-point features. _ discrete_values, and construct a width of _ The histogram algorithm is used for feature selection. It only requires iterating through the discrete values ​​of the histogram to find the optimal split point, resulting in lower memory usage and computational cost. When processing high-dimensional sparse data, a mutually exclusive feature bundling algorithm is used. Features are sorted according to the number of non-zero values, and then the conflict ratio between different features is calculated. Each feature is iterated through, and attempts are made to merge features to minimize the conflict ratio.

[0082] When growing the decision tree, a depth-constrained leaf-wise growth algorithm is used instead of a level-wise growth strategy. With the same number of splits, the leaf-wise algorithm is more effective at reducing error, thus improving model accuracy. LightGBM also employs a one-sided gradient sampling (GOSS) algorithm, which calculates information gain using only the remaining samples by excluding most samples with small gradients. This method achieves a good balance between reducing data volume and maintaining accuracy. The GOSS algorithm sorts the absolute values ​​of the gradients in the data and selects the top... The sample was then randomly sampled from the remaining data. The sample size is calculated by multiplying the small gradient data by the total number of small gradient samples and dividing by the number of randomly sampled small gradient samples.

[0083] For ease of explanation, Figure 3 Described with The Boosting training process of a LightGBM model with decision trees. During the training of LightGBM, the... The decision trees are trained sequentially, with the target value being... residual It is the target value The difference between the sum of all previous tree outputs and the sum of the previous tree outputs. For example, the predicted value of decision tree 1. Will try to follow residual This will become the training target for decision tree 2 learning. Similarly, given Decision tree 2 will output To get close The corresponding residuals This will be used as the training target for decision tree 3. This sequential training process of residuals is called Boosting learning and is applied to all subsequent trees. The output of the LightGBM model is obtained by summing the outputs of all decision trees.

[0084] Step 2.2: Determine the optimal parameters using the grid search method.

[0085] In the LightGBM model, there are several manually adjustable hyperparameters, commonly including the learning rate, maximum depth, minimum number of samples per leaf node, number of iterations, number of leaves, and regularization parameter. These parameters have a wide range of adjustment options, and they may interact with each other. Manually adjusting these parameters requires a certain level of knowledge and experience, and the results are often unsatisfactory, time-consuming, and labor-intensive. Therefore, this invention employs a grid search method for automatic hyperparameter tuning. This method provides a list of candidate hyperparameter values ​​and performs a comprehensive search of all possible parameter combinations. For each set of parameters, model training and evaluation are performed. During the parameter tuning process, the parameter range and value intervals are gradually narrowed until the model evaluation score remains essentially constant after multiple runs. Then, a regularization parameter is added, and the grid search method is used to continue optimization until the model evaluation score remains essentially constant; the parameters at this point are considered the optimal hyperparameters. This method significantly reduces the time and effort required for manual hyperparameter tuning, improving model performance and accuracy.

[0086] Step 2.3: Model Training and Saving

[0087] Will , , As input to the LightGBM model, and As output, explore parameters , , and , The relationship between the parameters is as follows: The optimal parameters found by the grid search method are substituted into the LightGBM model, and the model is trained using the training set. The percentage error results for efficiency and duty cycle are obtained as follows: Figure 4 and Figure 5 As shown. Save the trained model for easy access by the particle swarm optimization algorithm when optimizing the model.

[0088] Step 3: Particle Swarm Optimization Search for Optimal Parameters

[0089] The purpose of using the particle swarm optimization algorithm is to obtain the highest efficiency of the Buck-Boost circuit under different operating conditions. Therefore, the fitness function of the algorithm is the efficiency output by the trained LightGBM model.

[0090] Step 3.1: Particle Swarm Optimization Algorithm

[0091] First, the velocities and positions of the particles are randomly initialized. Then, the fitness value of each particle is calculated, and the individual best historical position of each particle is determined. With the historical best position of the entire particle swarm Then update the velocity and position of each particle, and the velocity and position update formulas are (5) and (6).

[0092] (5)

[0093] (6)

[0094] in, Indicates the first During the nth iteration, the 1st The velocity of each particle; Indicates the first During the nth iteration, the 1st The position of each particle; Indicates the first The particle reached the [number]th [particle]. The optimal position in the next iteration; Indicates up to the number The next iteration determines the optimal positions of all particles; , A random number between 0 and 1.

[0095] The inertia weight adopts a linear decreasing strategy, and its expression is as follows:

[0096] (7)

[0097] in, As the inertia weighting factor, For the number of iterations, The maximum number of iterations, , These represent the individual learning coefficient and the global learning coefficient.

[0098] After updating the particle velocity and position, recalculate the fitness function for each particle. and Update until the maximum number of iterations is reached.

[0099] Step 3.2: Obtain the optimal dataset

[0100] In practical applications, situations such as sudden load changes may occur, requiring the identification of optimal circuit parameters that maximize efficiency under different operating conditions. Therefore, when using the particle swarm optimization algorithm for optimization, it is necessary to perform optimization separately for each different operating condition. In this invention, the load... Within the rated range, find the value that yields the highest efficiency for different values. , , Then, the circuit parameters corresponding to the highest efficiency under different operating conditions are combined to form the optimal dataset.

[0101] Step 4: Online optimization implementation

[0102] Step 4.1: Construction of BP Neural Network

[0103] by As input to the BP neural network, , , As output, the BP neural network is initialized by initializing the number of inputs in the input layer, the number of hidden layers, the number of nodes in each hidden layer, and the number of outputs in the output layer. The weights of each neuron in the hidden layer are assigned, and the number of nodes in the hidden layer is determined according to formula (8).

[0104] (8)

[0105] in, This represents the number of hidden layer nodes. The number of nodes in the input layer. This represents the number of nodes in the output layer. This is the adjustment constant.

[0106] The hidden layer neurons are computed through forward propagation, and their outputs can be represented as follows:

[0107] (9)

[0108] In the formula The tansig function is selected as the activation function in this invention. Given the weight parameters in the network, use gradient descent to optimize the weight parameters in the network based on the cost function.

[0109] Step 4.2: Online Optimization

[0110] The trained optimal BP neural network is exported as a Simulink module and used as the controller for the Buck-Boost circuit. Under external load... When changes occur, the controller can promptly provide the optimal duty cycle corresponding to the maximum efficiency, thus completing online optimization.

[0111] After the simulation verification is successful, the neural network controller in the Simulink simulation is converted into C language code, downloaded to the DSP, and online optimization is completed for practical application.

[0112] Another embodiment of the present invention proposes a DC-DC converter modeling system driven by mechanistic knowledge and data fusion, comprising:

[0113] The DC-DC converter simulation dataset acquisition module is used to acquire DC-DC converter simulation datasets by combining the mechanistic knowledge of DC-DC converter circuits.

[0114] The data-driven model training module is used to train a pre-established data-driven model using the acquired DC-DC converter simulation dataset.

[0115] The target circuit parameter acquisition module is used to use a trained data-driven model and a particle swarm optimization algorithm to find the target circuit parameters corresponding to the highest efficiency of the data-driven model under different operating conditions.

[0116] The DC-DC converter control signal duty cycle output module is used to train a pre-built BP neural network using the target circuit parameters. When the external operating conditions change, the trained BP neural network outputs the DC-DC converter control signal duty cycle to complete the modeling of the DC-DC converter.

[0117] Another embodiment of the present invention also provides an electronic device, comprising:

[0118] A memory that stores at least one instruction; and a processor that executes the instructions stored in the memory to implement the mechanism knowledge and data fusion-driven DC-DC converter modeling method.

[0119] Another embodiment of the present invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the DC-DC converter modeling method driven by the fusion of the aforementioned mechanistic knowledge and data.

[0120] For example, the instructions stored in the memory can be divided into one or more modules / units. These modules / units are stored in a computer-readable storage medium and executed by the processor to complete the mechanism-knowledge and data-fusion-driven DC-DC converter modeling method described in this embodiment of the invention. The one or more modules / units can be a series of computer-readable instruction segments capable of performing specific functions, which describe the execution process of the computer program in the server.

[0121] The electronic device may be a smartphone, laptop, PDA, or cloud server, among other computing devices. It may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the electronic device may also include more or fewer components, or combinations of certain components, or different components; for example, it may also include input / output devices, network access devices, buses, etc.

[0122] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0123] The memory can be an internal storage unit of the server, such as a hard drive or RAM. Alternatively, it can be an external storage device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or FlashCard. Furthermore, the memory can include both internal and external storage units. The memory is used to store computer-readable instructions and other programs and data required by the server. It can also be used to temporarily store data that has been output or will be output.

[0124] It should be noted that the information interaction and execution process between the above-mentioned module units are based on the same concept as the method embodiment. For details on their specific functions and technical effects, please refer to the method embodiment section. They will not be repeated here.

[0125] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0126] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0127] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0128] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A DC-DC converter modeling method driven by the fusion of mechanistic knowledge and data, characterized in that, include: By combining the mechanistic knowledge of DC-DC converter circuits, a DC-DC converter simulation dataset is obtained; The pre-established data-driven model is trained using the acquired DC-DC converter simulation dataset; Using a trained data-driven model, the particle swarm optimization algorithm is used to find the target circuit parameters that achieve the highest efficiency of the data-driven model under different operating conditions. The target circuit parameters are used to train a pre-built BP neural network. The trained BP neural network is then used to output the duty cycle of the DC-DC converter control signal when the external operating conditions change, thus completing the modeling of the DC-DC converter. The data-driven model uses the LightGBM model. When training the LightGBM model using the obtained DC-DC converter simulation dataset, the input DC voltage is... Output DC voltage Load resistance As input to the LightGBM model, the duty cycle of the pulse width modulation signal of the switching transistor is used. With circuit transmission efficiency As the output of the LightGBM model, the goal is to make the output of the LightGBM model as consistent as possible with the true value; The process of using a trained data-driven model and employing a particle swarm optimization algorithm to find the target circuit parameters corresponding to the highest efficiency of the data-driven model under different operating conditions includes: The velocities and positions of the particles are randomly initialized, the fitness value of each particle is calculated, and the individual best historical position of each particle is determined. With the historical best position of the entire particle swarm ; Update the velocity and position of each particle using the following formula: In the formula, Indicates the first During the nth iteration, the 1st The velocity of each particle; Indicates the first During the nth iteration, the 1st The position of each particle; Indicates the first The particle reached the [number]th [particle]. The optimal position in the next iteration; Indicates up to the number The optimal positions of all particles in the next iteration; , The inertia weight factor is calculated using a linear decreasing strategy based on the following formula: (The random number between 0 and 1 is used.) : In the formula, For the number of iterations, The maximum number of iterations, , Individual learning coefficient and global learning coefficient; After updating the particle velocity and position, recalculate the fitness function for each particle to determine its individual best historical position. With the historical best position of the entire particle swarm Update until the maximum number of iterations is reached; In the process of using the particle swarm optimization algorithm for optimization, the load will be... By varying the input DC voltage within the rated range, we can find the input DC voltage corresponding to the highest efficiency for different values. Duty cycle of the switching transistor pulse width modulation signal and output DC voltage The circuit parameters corresponding to the highest efficiency under different operating conditions are combined to form the target circuit parameters. The process of training a pre-built BP neural network using target circuit parameters, and then using the trained BP neural network to output the duty cycle of the DC-DC converter control signal when external operating conditions change, to complete the modeling of the DC-DC converter includes: With load resistor As input to the BP neural network, the duty cycle of the switching transistor pulse width modulation signal is used. Output DC voltage Input DC voltage As output, the BP neural network is initialized, including the number of inputs in the input layer, the number of hidden layers, the number of nodes in each hidden layer, and the number of outputs in the output layer. Weights are then assigned to each neuron in the hidden layers. The number of nodes in the hidden layers is determined by the following formula: In the formula, This represents the number of hidden layer nodes. The number of nodes in the input layer. This represents the number of nodes in the output layer. This is the adjustment constant; The hidden layer neurons are computed through forward propagation, and their outputs are represented as follows: In the formula, This represents the activation function. For the weight parameters in the network, use gradient descent to optimize the network weight parameters according to the cost function; The trained BP neural network is exported and used as the controller of the Buck-Boost circuit, under load resistance... When changes occur, the controller provides the optimal duty cycle corresponding to the maximum efficiency, thus completing the online optimization of the data-driven model.

2. The DC-DC converter modeling method driven by the fusion of mechanism knowledge and data according to claim 1, characterized in that, For a Buck-Boost DC-DC converter, the corresponding DC-DC converter simulation dataset includes: Input DC voltage Output DC voltage Duty cycle of the switching transistor pulse width modulation signal Load resistance Input power Output power and circuit transmission efficiency ; Based on the principles of DC-DC converter circuits, in a Buck-Boost circuit, given a set of input DC voltages... Duty cycle of the switching transistor pulse width modulation signal and load resistance Then there is a corresponding input power. Output power and output DC voltage After determining the rated operating range of the circuit, input DC voltage. Duty cycle of the switching transistor pulse width modulation signal and load resistance By combining different values, the output DC voltage is obtained according to the following formula. Scope: The output power is obtained by the following formula. Scope: For input DC voltage With output DC voltage Both are measured using a voltmeter; for input power With output power The instantaneous values ​​of the input and output voltages and currents are measured using a voltmeter and an ammeter, respectively, and then calculated using the following formula: From input power With output power The circuit transmission efficiency can be obtained by the following formula. : 。 3. The DC-DC converter modeling method driven by the fusion of mechanism knowledge and data according to claim 1, characterized in that, In the step of training the pre-established data-driven model using the acquired DC-DC converter simulation dataset, the hyperparameters of the LightGBM model are adjusted using a grid search method. In the process of giving the hyperparameters, the range and value interval of the hyperparameters are gradually reduced until the model evaluation score meets the requirements after multiple runs. Then, regularization parameters are added, and the grid search method is used to continue to optimize until the model evaluation score meets the requirements. The hyperparameters corresponding to this point are the target hyperparameters.

4. The DC-DC converter modeling method driven by the fusion of mechanism knowledge and data according to claim 1, characterized in that, After obtaining the DC-DC converter simulation dataset through MATLAB or Simulink and completing the online optimization of the data-driven model, the neural network controller in the MATLAB or Simulink simulation is converted into C language code and downloaded to the digital signal processor (DSP) to control the DC-DC converter.

5. A DC-DC converter modeling system driven by the fusion of mechanistic knowledge and data, characterized in that, include: The DC-DC converter simulation dataset acquisition module is used to acquire DC-DC converter simulation datasets by combining the mechanistic knowledge of DC-DC converter circuits. The data-driven model training module is used to train a pre-established data-driven model using the acquired DC-DC converter simulation dataset. The target circuit parameter acquisition module is used to use a trained data-driven model and a particle swarm optimization algorithm to find the target circuit parameters corresponding to the highest efficiency of the data-driven model under different operating conditions. The DC-DC converter control signal duty cycle output module is used to train a pre-built BP neural network using the target circuit parameters. When the external operating conditions change, the trained BP neural network outputs the DC-DC converter control signal duty cycle to complete the modeling of the DC-DC converter. The data-driven model uses the LightGBM model. When training the LightGBM model using the obtained DC-DC converter simulation dataset, the input DC voltage is... Output DC voltage Load resistance As input to the LightGBM model, the duty cycle of the pulse width modulation signal of the switching transistor is used. With circuit transmission efficiency As the output of the LightGBM model, the goal is to make the output of the LightGBM model as consistent as possible with the true value; The process of using a trained data-driven model and employing a particle swarm optimization algorithm to find the target circuit parameters corresponding to the highest efficiency of the data-driven model under different operating conditions includes: The velocities and positions of the particles are randomly initialized, the fitness value of each particle is calculated, and the individual best historical position of each particle is determined. With the historical best position of the entire particle swarm ; Update the velocity and position of each particle using the following formula: In the formula, Indicates the first During the nth iteration, the 1st The velocity of each particle; Indicates the first During the nth iteration, the 1st The position of each particle; Indicates the first The particle reached the [number]th [particle]. The optimal position in the next iteration; Indicates up to the number The optimal positions of all particles in the next iteration; , The inertia weight factor is calculated using a linear decreasing strategy based on the following formula: (The random number between 0 and 1 is used.) : In the formula, For the number of iterations, The maximum number of iterations, , Individual learning coefficient and global learning coefficient; After updating the particle velocity and position, recalculate the fitness function for each particle to determine its individual best historical position. With the historical best position of the entire particle swarm Update until the maximum number of iterations is reached; In the process of using the particle swarm optimization algorithm for optimization, the load will be... By varying the input DC voltage within the rated range, we can find the input DC voltage corresponding to the highest efficiency for different values. Duty cycle of the switching transistor pulse width modulation signal and output DC voltage The circuit parameters corresponding to the highest efficiency under different operating conditions are combined to form the target circuit parameters. The process of training a pre-built BP neural network using target circuit parameters, and then using the trained BP neural network to output the duty cycle of the DC-DC converter control signal when external operating conditions change, to complete the modeling of the DC-DC converter includes: With load resistor As input to the BP neural network, the duty cycle of the switching transistor pulse width modulation signal is used. Output DC voltage Input DC voltage As output, the BP neural network is initialized, including the number of inputs in the input layer, the number of hidden layers, the number of nodes in each hidden layer, and the number of outputs in the output layer. Weights are then assigned to each neuron in the hidden layers. The number of nodes in the hidden layers is determined by the following formula: In the formula, This represents the number of hidden layer nodes. The number of nodes in the input layer. This represents the number of nodes in the output layer. This is the adjustment constant; The hidden layer neurons are computed through forward propagation, and their outputs are represented as follows: In the formula, This represents the activation function. For the weight parameters in the network, use gradient descent to optimize the network weight parameters according to the cost function; The trained BP neural network is exported and used as the controller of the Buck-Boost circuit, under load resistance... When changes occur, the controller provides the optimal duty cycle corresponding to the maximum efficiency, thus completing the online optimization of the data-driven model.

6. An electronic device, characterized in that, include: Memory, storing at least one instruction; and The processor executes instructions stored in the memory to implement the DC-DC converter modeling method driven by the mechanism knowledge and data fusion as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the DC-DC converter modeling method driven by the mechanistic knowledge and data fusion as described in any one of claims 1 to 4.

Images