An ipt system multi-parameter identification method based on a random forest prediction model

Abstract

The application discloses an IPT system multi-parameter identification method based on a random forest prediction model, and comprises the following steps: determining a relationship expression of a primary side current coil current and N items of to-be-identified parameters according to an equivalent circuit model of a system, wherein the N items of to-be-identified parameters comprise system mutual inductance and N-1 items of secondary side parameters, and the value determined according to the relationship expression represents a primary side coil current value under different frequency conditions that needs to be measured during identification; constructing N random forest prediction models, wherein each random forest prediction model is used to output an item of parameter identification value in the N items of to-be-identified parameters; training the N random forest prediction models by using simulation sample data sets, collecting primary side coil current measurement values under different frequency conditions and inputting the N random forest prediction models after training, and obtaining all identification values.

Description

Technical Field

[0001] This invention belongs to the field of wireless power transmission technology, and more specifically, relates to a multi-parameter identification method for IPT systems based on a random forest prediction model. Background Technology

[0002] In recent years, wireless power transfer technology has attracted much attention and research from scholars both at home and abroad due to its advantages such as flexible power supply, safe power extraction, and no wire connections. Among them, inductive power transfer (IPT) systems have the characteristics of high transmission efficiency and high transmission power under near-field conditions, and therefore have been applied in many fields such as electric vehicles, smart home appliances, and rail transportation.

[0003] In practical applications, such as the charging process of mobile devices like electric vehicles, the load and mutual inductance parameters of the IPT system constantly change due to the mobile nature of the charging equipment. However, these changes in system load and mutual inductance parameters lead to frequency drift and reduced transmission power and efficiency. To maintain good transmission performance, appropriate feedback control mechanisms are needed to improve system robustness. However, feedback control methods typically require real-time acquisition of system mutual inductance, load, and other parameter information to ensure efficient and stable operation. Therefore, the problem of system parameter identification has become a current research hotspot in the field of wireless power transfer.

[0004] Currently, most methods in the field of IPT system parameter identification require prior knowledge of the actual parameters of each component in the IPT system. Based on this, the load and mutual inductance parameters of the system are estimated by monitoring parameters such as current and voltage in the primary circuit. However, obtaining the actual parameters of each component in the IPT system in advance is very difficult. Due to the influence of factors such as differences in manufacturing processes and component aging, the actual parameter values ​​of the system may deviate from the design nominal values. This will cause errors in the mutual inductance and load parameter values ​​obtained by parameter estimation methods based on the system parameter design values, thus affecting the efficient and stable operation of the system. Summary of the Invention

[0005] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a multi-parameter identification method for IPT system based on random forest prediction model, thereby solving the technical problem of inaccurate parameter identification.

[0006] To achieve the above objectives, this invention provides a multi-parameter identification method for IPT systems based on a random forest prediction model, comprising the following steps: Based on the equivalent circuit model of the system, the relationship between the primary current, coil current, and N parameters to be identified is determined. These N parameters include the system mutual inductance and N-1 secondary parameters. The range of values ​​for , This indicates that measurements are required when identifying N parameters to be identified. Primary coil current values ​​under different frequency conditions; Construct N random forest prediction models, each random forest prediction model is used to predict based on The primary coil current value under different frequency conditions outputs the identification value of one of the N parameters to be identified, and the N random forest prediction models output all the identification values ​​of the N parameters to be identified. Construct a simulation sample dataset, and use the simulation sample dataset to train N random forest prediction models, collecting data... The primary coil current measurements under different frequency conditions are input into N trained random forest prediction models to obtain all the identification values ​​of N parameters to be identified.

[0007] Preferably, the step of determining based on the relational formula... The range of values ​​includes the following steps: Perform a first-order Taylor expansion on the relation for any value within the parameter variation range of the N parameters to be identified; Based on the relations derived from the first-order Taylor expansion, the following is listed: A system of non-homogeneous linear equations under different frequency conditions; When determining that a nonhomogeneous linear system of equations has a unique solution The range of values ​​for .

[0008] Preferably, the step of constructing a simulation sample dataset and using the simulation sample dataset to train N random forest prediction models includes the following steps: Construct simulation sample datasets for each of the N random forest prediction models. Each sample in the simulation sample dataset includes... The primary coil current values ​​under different frequency conditions and the simulated values ​​of the corresponding parameters to be identified; The corresponding random forest prediction model is trained using the simulation sample dataset corresponding to each random forest prediction model.

[0009] Preferably, ≥N.

[0010] Preferably, the IPT system includes a DLCC compensation network, and the DLCC compensation network includes a primary-side series compensation inductor. Primary-side series compensation capacitor Primary-side parallel compensation capacitor Secondary series compensation inductor Secondary-side series compensation capacitor Parallel compensation capacitor with secondary side N=7 ≥7.

[0011] Preferably, the seven parameters to be identified include coil mutual inductance. and secondary side parameters , Indicates the secondary coil inductance. Indicates the internal resistance of the secondary coil. This is the equivalent resistance of the load.

[0012] Preferably, the relationship between the primary current, the coil current, and the N parameters to be identified is as follows: ; in, Indicates the inverter's operating frequency. Indicates the inverter's resonant frequency. Primary coil Current on This represents the fundamental component of the inverter's output voltage. To represent a complex unit, Indicates the inductance of the primary coil. This indicates the internal resistance of the primary coil.

[0013] Overall, compared with the prior art, the above-described technical solutions conceived in this invention can obtain estimated values ​​of multiple parameters to be determined in the system by simply inputting a set of measured effective values ​​of the primary coil current into a trained prediction model during online identification. Attached Figure Description

[0014] Figure 1 This is a circuit diagram of the DLCC-IPT system according to an embodiment of the present invention; Figure 2 This is an equivalent circuit diagram of the DLCC-IPT system according to an embodiment of the present invention; Figure 3 This is the training random forest prediction model of this invention embodiment; Figure 4 This is a multi-parameter identification based on a random forest prediction model, as described in this embodiment of the invention. Figure 5 This is the data preprocessing process in an embodiment of the present invention; Figure 6 This is the basic structure of the regression tree model in this embodiment of the invention; Figure 7 This is the single regression tree prediction process according to an embodiment of the present invention; Figure 8 This is a simulation model of the DLCC-IPT system according to an embodiment of the present invention; Figure 9 These are the input and output voltage and current waveforms under resonant conditions according to an embodiment of the present invention; Figure 10This is the relationship between the prediction error of each system parameter and the number of regression trees in the model in this embodiment of the invention; Figure 11 This is a distribution diagram of the prediction errors of various system parameters in an embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0016] In the description of the embodiments of this application, the term "multiple" means two or more.

[0017] In the embodiments of the present invention, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, apparatus, product or device that includes a series of steps or modules is not necessarily limited to those steps or modules that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such process, method, product or device.

[0018] The naming or numbering of steps in the embodiments of the present invention does not mean that the steps in the method flow must be executed in the time / logical order indicated by the naming or numbering. The execution order of the named or numbered process steps can be changed according to the technical purpose to be achieved, as long as the same or similar technical effect can be achieved.

[0019] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0020] This invention provides a multi-parameter identification method for IPT systems based on a random forest prediction model, comprising the following steps: Based on the equivalent circuit model of the system, the relationship between the primary current, coil current, and N parameters to be identified is determined. These N parameters include the system mutual inductance and N-1 secondary parameters. The range of values ​​for , This indicates that measurements are required when identifying N parameters to be identified. Primary coil current values ​​under different frequency conditions; Construct N random forest prediction models, each random forest prediction model is used to predict based on The primary coil current value under different frequency conditions outputs the identification value of one of the N parameters to be identified, and the N random forest prediction models output all the identification values ​​of the N parameters to be identified. Construct a simulation sample dataset, and use the simulation sample dataset to train N random forest prediction models, collecting data... The primary coil current measurements under different frequency conditions are input into N trained random forest prediction models to obtain all the identification values ​​of N1 parameters to be identified.

[0021] The above method is applicable to various IPT systems. When applied to different IPT systems, the differences lie in the different system components, the different secondary side parameters to be identified, and the different equivalent circuit models. Therefore, the relationship between the primary side current, the coil current, and the N1 term parameters to be identified is different, but the working principle is similar.

[0022] Using the DLCC-IPT system as an example, this paper derives the relationships between the primary coil current and the system's mutual inductance, as well as multiple parameters of the secondary side. Based on this, it analyzes the fundamental conditions required for multi-parameter identification by measuring the primary coil current. Secondly, based on the mapping relationship between the primary coil current and the multiple parameters to be identified, several random forest prediction models are established for identifying these parameters. Finally, simulation data is input into the prediction models for offline training. During online identification, only a set of measured effective values ​​of the primary coil current needs to be input into the trained prediction model to obtain estimated values ​​of the multiple parameters to be identified in the system.

[0023] (1) DLCC-IPT system modeling and parameter relationships Figure 1 This is a circuit diagram of a DLCC-IPT system, also known as an IPT system with a bilateral LCC compensation network. It features a symmetrical primary and secondary structure, easy tuning of the compensation topology, and output current independent of the load. It also exhibits good offset resistance and is widely used in wireless charging systems for electric vehicles. The DLCC-IPT system includes a DLCC compensation network, which comprises a primary-side series compensation inductor. Primary-side series compensation capacitor Primary-side parallel compensation capacitor Secondary series compensation inductor Secondary-side series compensation capacitor Parallel compensation capacitor on the secondary side Let the inductance of the primary coil in the IPT system be denoted as... The internal resistance of the primary coil is denoted as . The inductance of the secondary coil is denoted as The internal resistance of the secondary coil is denoted as , Figure 2 This is the equivalent circuit diagram of the DLCC-IPT system.

[0024] According to Kirchhoff's voltage law Figure 1 The equivalent circuit model shown has the primary and secondary voltage and current equations as shown in equation (1): ; in, Indicates the inverter's operating frequency. This represents the fundamental component of the inverter's output voltage. This represents the fundamental component of the inverter's output current. Primary coil Current on Secondary coil Current on Input current to the rectifier, The load equivalent resistance, To represent a complex unit, This indicates mutual inductance between coils.

[0025] N=7, the 7 parameters to be identified include coil mutual inductance. and secondary side parameters .

[0026] According to existing research, when the output voltage of the DLCC type IPT system inverter... The initial phase angle is At that time, the primary coil current for: ; In the formula The resonant frequency of the system. This is the reflection impedance of the secondary circuit. It can be represented as: ; Substituting equation (3) into equation (2), we can obtain the primary coil current. The relationship between the parameters of the system compensation network and the parameters is shown in equation (4): ; in, This indicates the inverter's resonant frequency.

[0027] Typically, the primary side parameters in an IPT system These parameters can be obtained through measurement and will not change due to aging or other reasons in a short period of time. Therefore, this article will focus on the various parameters of the primary edge. Inverter output voltage and its operating frequency It is considered a known constant. At this time, the primary coil current... Only affected by mutual induction and secondary side parameters The influence of this can be observed by measuring the primary coil current. The parameters mentioned above are identified. However, the primary coil current is measured only at one frequency. The value of the parameter cannot simultaneously identify all seven parameters mentioned above. Therefore, the primary coil current can be measured under multiple different frequency conditions. The value of .

[0028] To investigate the minimum number of frequency changes required to simultaneously identify the seven parameters mentioned above, this paper... Expanding equation (4) by a first-order Taylor series, we can write: ; In the formula, For equation (4) in Place The value, Variables Any value within the parameter variation range. For the remaining items, They are respectively about The first-order partial derivative is shown in equation (6): ; By adjusting the frequency of the inverter control signal, it is possible to measure Primary coil current under different frequency conditions The value of . At this time, according to equation (5), the non-homogeneous linear equation system can be listed as shown in equation (7): ; In equation (7), Indicates the first The primary coil current measured at a certain frequency The value, Indicates the first The primary coil current measured at a certain frequency The value, Then, the meaning of the first At each frequency about First-order partial derivative, similar They represent the first At each frequency about The first-order partial derivative, Then it means the first The remainder after Taylor expansion of equation (4) for each frequency. Then it means the first The remainder after Taylor expansion of equation (4) at each frequency. Representing (7) in matrix form, it can be written as: ,in: ; As can be seen from (6), at the inverter output voltage frequency Under the condition of matrix The rank is: ; when hour, At this point, equation (7) has a unique solution. Therefore, the primary coil current must be measured under at least seven different frequency conditions. value By solving equation (7), the various parameters of the DLCC type IPT system can be obtained. The precise value.

[0029] (2) Multi-parameter identification method based on random forest prediction model As can be deduced from the previous section, solving equation (7) can yield the mutual inductance in the IPT system. Secondary coil self-inductance Secondary coil internal resistance Secondary resonant capacitor Parallel compensation capacitor on the secondary side Secondary compensation inductor and equivalent resistance The exact value of . However, as a system of multivariate nonlinear equations, equation (7) is difficult to solve directly using algebraic formulas.

[0030] To address the aforementioned issues, this paper proposes a multi-parameter identification method based on a random forest regression model. This method first preprocesses the many-to-many mapping relationship shown in equation (7) into seven many-to-one mapping relationships. Then, it trains seven random forest prediction models using simulation data. Each of the seven random prediction models can be trained independently, ensuring that each model can represent one many-to-one mapping relationship. When a new set of samples is available... Each model provides a predicted value for the corresponding mapping relationship as input, and ultimately, the seven models each provide the identification results for the seven parameters to be determined. The training and identification process of multi-parameter identification of the DLCC-IPT system based on the random forest prediction model is as follows: Figure 3 , Figure 4 As shown.

[0031] Data preprocessing According to equation (7), the parameters to be determined can be... and The mapping relationship can be represented as: ; Since equation (8) satisfies Therefore, the mapping relationship represented by equation (10) can be rewritten as: ; In order for the random forest prediction model to fit the mapping relationship represented by equation (11), the many-to-many mapping relationship in equation (11) can be decomposed into 7 many-to-one mapping relationships. The decomposed mapping relationship can be expressed as: ; At this point, prior samples representing the mapping relationship shown in equation (12) can be obtained through data preprocessing. , This represents the primary coil current values ​​under seven different frequency conditions. express Corresponding parameters The identification value is split into 7 prior samples representing many-to-one mapping relationships, namely: ; Data preprocessing process such as Figure 5 As shown: After bootstrap sampling data preprocessing, each training set representing a many-to-one mapping relationship... Both contain the same number of prior samples, and each sample Each contains a set of characteristic vectors composed of primary coil currents measured under different frequency conditions. and the corresponding parameters to be determined , can be represented as: ; The Bootstrap resampling method is used to sample from the total dataset. Sampling with replacement, each time randomly selected Each sample is used as a subset of the training set for a single regression tree in the random forest. The number of samples in the sub-training set is required to meet the following conditions. And samples in the subset The features can be no more than 7, that is: ; Building a random forest prediction model to achieve After using a subset of training sets, a regression tree prediction model is constructed using the subsets. Total number of buildings Each regression tree is independent of the others. All the completed regression trees are then... By forming a set, we obtain a random forest regression model, which can be represented as: ; A regression tree is a tree-structured weak predictive model and the basic unit of a random forest regression model. A regression tree consists of a root node, several internal nodes, and several leaf nodes. Each internal node contains a split point that divides the dataset into two subsets. Each leaf node contains one subset and corresponds to a prediction result. Its structure is as follows: Figure 6 As shown.

[0032] The process of constructing a single regression tree based on the CART algorithm is as follows: Step 1: Sub-training set Mark it as the root node of the regression tree, and denote it as... ; Step 2: Traverse the node dataset All of them , with each For the split point Divide into 2 subsets and And calculate 2 subsets of the dataset. and The sum of the mean squared deviations: ; In the formula, and Sub-datasets and Parameters to be determined The average value.

[0033] Step 3: Find the two subsets and The optimal split point of the sum of the mean squared errors : ; Will As the optimal split point, and accordingly divide the dataset Divided into and Two subsets of data; Step 4: Determine the two subsets and Is the mean square error less than the threshold? If the mean square error is less than the threshold If the subset is marked as a leaf node, the splitting stops; if the mean squared error is greater than the threshold... If so, mark the dataset as an internal node and repeat Step 2; Step 5: Determine if all subsets are marked as leaf nodes. If they are all marked as leaf nodes, the algorithm ends. If there are internal nodes, repeat Step 4.

[0034] Multi-parameter identification: After the random forest regression model is built, the optimal split point of each regression tree in the model is derived. Average value of the parameters to be determined at the leaf nodes When a new set of samples is measured Then, each regression tree in the model is based on the feature points derived above. and the average value of the parameter to be determined In microcontrollers, such as Figure 7 The discriminant operation is shown.

[0035] The predicted value of a single regression tree is obtained, and the average of the predicted values ​​of all regression trees is finally calculated as the prediction result of the random forest regression model, as shown in Equation (19): ; In the formula, For the training set The random forest regression model constructed through training can fit the training set. The mapping relationship represented For the model For new samples Regarding the parameters to be identified The final prediction results. Seven data points were used to fit the results. Random forest regression model For new samples By making predictions, we can obtain estimated values ​​for all the parameters to be calculated, such as... Figure 4 As shown.

[0036] (3) Simulation verification and result analysis To verify the feasibility and effectiveness of the proposed identification method, this paper refers to Figure 2 A simulation model of the DLCC-IPT system is built on the Matlab / Simulink platform, such as... Figure 8 As shown in Table 1, the parameters of each component in the model are shown in Table 1. The input and output voltage and current waveforms of the simulation model when it is stable under resonance conditions are shown in Table 1. Figure 9 As shown.

[0037] Table 1 System Parameters ; use Figure 8 The Simulink simulation verification model shown has a single simulation run time set to 0.02s, at which point the system is running stably. Automatic simulation is performed by writing an M-file, and the simulation is checked before each run. , , , , , , The parameters were changed once, and each set of parameters was obtained by measuring the inverter control signal by altering the inverter control signal. , , , , , , Under 7 different frequency conditions and The values ​​were obtained, resulting in a total of 20,000 sets of simulation model data.

[0038] The model was trained using 19,000, 15,000, and 10,000 sets of simulation data respectively. When the model was tested using the remaining data, the identification results based on the random forest prediction model were as follows: Figure 10 As shown in (a)-(c), the simulation results show that the more training data and the more regression trees that constitute the random forest prediction model, the lower the prediction error of the model for each parameter of the system.

[0039] When 19,000 sets of simulation data are used to train the model and the remaining 1,000 sets of data are used to test the model, the random forest prediction model composed of 30 regression trees predicts the various parameters of the system. The predicted error distributions are as follows: Figure 11 As shown in (a)-(g).

[0040] Depend on Figure 11 It can be seen that when the number of samples in the dataset is more than 20,000 and the number of regression trees in the prediction model is greater than 30, the prediction accuracy of the proposed method for each parameter of the DLCC-IPT system is usually less than 2%, which basically meets the accuracy requirements of real-time mutual inductance and load identification methods for the actual parameters of the system.

[0041] In summary, this invention proposes a method for identifying multiple parameters of the secondary side of an IPT system based on a random forest prediction model. By analyzing the basic working principle of the IPT system, the relationship between the primary coil current, system mutual inductance, and multiple parameters of the secondary side is derived. Based on this, the basic conditions required for multi-parameter identification by measuring the primary coil current are analyzed. Secondly, based on the mapping relationship between the primary coil current and multiple parameters to be identified, multiple random forest prediction models are established for identifying these parameters. Finally, simulation data is input into the prediction model for offline training. During online identification, only a set of measured effective values ​​of the primary coil current needs to be input into the trained prediction model to obtain estimated values ​​of multiple parameters in the system. Simulation experiments show that when the number of prior samples exceeds 20,000 and the prediction model has more than 30 regression trees, the accuracy of the proposed method in predicting various system parameters basically meets the usage requirements.

[0042] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-parameter identification method for IPT systems based on a random forest prediction model, characterized in that, Including the following steps: Based on the equivalent circuit model of the system, the relationship between the primary current, coil current, and N parameters to be identified is determined. These N parameters include the system mutual inductance and N-1 secondary parameters. The range of values ​​for , This indicates that measurements are required when identifying N parameters to be identified. Primary coil current values ​​under different frequency conditions; Construct N random forest prediction models, each random forest prediction model is used to predict based on The primary coil current value under different frequency conditions outputs the identification value of one of the N parameters to be identified, and the N random forest prediction models output all the identification values ​​of the N parameters to be identified. Construct a simulation sample dataset, and use the simulation sample dataset to train N random forest prediction models, collecting data... The primary coil current measurements under different frequency conditions are input into N trained random forest prediction models to obtain all the identification values ​​of N parameters to be identified.

2. The multi-parameter identification method for IPT systems based on a random forest prediction model as described in claim 1, characterized in that, The determination based on the relational formula The range of values ​​includes the following steps: Perform a first-order Taylor expansion on the relation for any value within the parameter variation range of the N parameters to be identified; Based on the relations derived from the first-order Taylor expansion, the following is listed: A system of non-homogeneous linear equations under different frequency conditions; When determining that a nonhomogeneous linear system of equations has a unique solution The range of values ​​for .

3. The multi-parameter identification method for IPT systems based on a random forest prediction model as described in claim 1, characterized in that, The steps for constructing the simulation sample dataset and training N random forest prediction models using the simulation sample dataset include: Construct simulation sample datasets for each of the N random forest prediction models. Each sample in the simulation sample dataset includes... The primary coil current values ​​under different frequency conditions and the simulated values ​​of the corresponding parameters to be identified; The corresponding random forest prediction model is trained using the simulation sample dataset corresponding to each random forest prediction model.

4. The multi-parameter identification method for IPT systems based on a random forest prediction model as described in claim 1, characterized in that, ≥N。 5. The multi-parameter identification method for IPT systems based on a random forest prediction model as described in claim 1, characterized in that, The IPT system includes a DLCC compensation network, which includes a primary-side series compensation inductor. Primary-side series compensation capacitor Primary-side parallel compensation capacitor Secondary series compensation inductor Secondary-side series compensation capacitor Parallel compensation capacitor with secondary side N=7 ≥7.

6. The multi-parameter identification method for an IPT system based on a random forest prediction model as described in claim 5, characterized in that, The seven parameters to be identified include coil mutual inductance. and secondary side parameters , Indicates the secondary coil inductance. Indicates the internal resistance of the secondary coil. This is the equivalent resistance of the load.

7. The multi-parameter identification method for IPT systems based on a random forest prediction model as described in claim 6, characterized in that, The relationship between the primary current, the coil current, and the N parameters to be identified is as follows: ； in, Indicates the inverter's operating frequency. Indicates the inverter's resonant frequency. Primary coil Current on This represents the fundamental component of the inverter's output voltage. To represent a complex unit, Indicates the inductance of the primary coil. This indicates the internal resistance of the primary coil.

Images