A radio frequency power transmission control method and device based on deep learning

Abstract

This invention discloses a wireless power transfer control method and device based on deep learning. In this invention, a pre-trained two-stage residual neural network model is used to effectively predict the changed mutual inductance and load resistance values. At the same time, a pre-established dynamic optimal impedance matching surface is used to find the optimal equivalent load resistance and optimal equivalent AC voltage that balance high efficiency and changes. The duty cycle of the secondary-side Buck-Boost step-up / step-down circuit is controlled to achieve closed-loop control of the equivalent load resistance. Then, the duty cycle of the primary-side Buck-Boost step-up / step-down circuit is controlled to achieve closed-loop control of the equivalent input voltage, thereby achieving constant power output of the entire system with high efficiency within a preset error range.

Description

Technical Field

[0001] This invention relates to the field of wireless power transmission technology, and in particular to a wireless power transmission control method and apparatus based on deep learning. Background Technology

[0002] Wireless power transfer is a novel power transmission solution distinct from traditional wired power transfer. It utilizes intangible energy carriers such as magnetic fields, electric fields, and electromagnetic waves to achieve contactless power transmission from the transmitter to the receiver, effectively avoiding the component aging problems caused by frequent plugging and unplugging in traditional wired power transfer solutions. In practical applications, such as induction heating, the heating power of the wireless power transfer system directly affects the material's heating rate and process consistency. Therefore, even if the coil distance changes or the permeability of the metal load fluctuates, the heating power must be kept as constant as possible. However, during wireless transmission, positional offsets or angular changes between the transmitter and receiver can easily cause mutual inductance fluctuations; simultaneously, changes in ambient temperature can affect device parameters, leading to changes in load resistance. These factors can easily cause the wireless power transfer system to deviate from a constant power output state. Furthermore, traditional impedance matching networks often only achieve optimal matching at specific load points. When mutual inductance or load changes, system impedance mismatch leads to a significant decrease in transmission efficiency, making it difficult to balance constant power output and high-efficiency operation, thus affecting operational stability and control accuracy. Summary of the Invention

[0003] This invention provides a wireless power transmission control method and apparatus based on deep learning, which solves the technical problem that existing technologies cannot efficiently achieve constant power control in wireless power transmission systems.

[0004] This invention provides a deep learning-based wireless power transfer control method for an LCC-S topology wireless power transfer system. The LCC-S topology wireless power transfer system includes a primary-side Buck-Boost converter, a primary-side full-bridge inverter, a primary-side compensation network, a secondary-side compensation network, a secondary-side full-bridge rectifier, a secondary-side Buck-Boost converter, and a load resistor. The method includes:

[0005] The current and voltage values ​​of the primary-side series compensation inductor and the primary-side parallel compensation capacitor of the primary-side Buck-Boost step-up / step-down circuit are collected in real time.

[0006] The collected current and voltage values ​​are input into a pre-trained two-stage residual neural network model for parameter prediction, and the mutual inductance value and load resistance value are predicted.

[0007] The predicted mutual inductance value, preset output power target value and preset efficiency target value are input into the preset dynamic optimal impedance matching three-dimensional surface to determine the optimal equivalent load resistance value.

[0008] The duty cycle of the secondary-side Buck-Boost step-up / step-down circuit is adjusted according to the optimal equivalent load resistance value and the predicted load resistance value, so that the equivalent load resistance of the secondary-side full-bridge rectifier circuit approaches the optimal equivalent load resistance value.

[0009] The optimal equivalent AC voltage value is calculated based on the optimal equivalent load resistance value, the predicted mutual inductance value, the preset output power target value, and the system fixed parameters.

[0010] The duty cycle of the primary-side Buck-Boost step-up / step-down circuit is adjusted according to the actual input voltage and the optimal equivalent AC voltage value, so that the equivalent AC voltage of the primary-side full-bridge inverter circuit approaches the optimal equivalent AC voltage value.

[0011] Optionally, the process of obtaining the pre-trained two-stage residual neural network model includes:

[0012] Obtain the operating current and voltage values ​​of the primary-side series compensation inductor and the primary-side parallel compensation capacitor of the LCC-S topology wireless power transmission system, and collect the operating mutual inductance value and operating load resistance value associated with the operating current and voltage values.

[0013] The operating current and voltage values ​​are subjected to feature engineering processing to obtain a multi-dimensional input feature vector;

[0014] An initial two-stage residual neural network model is constructed, using the multidimensional input feature vector, the operating mutual inductance value, and the logarithmically transformed operating load resistance value as inputs. The initial two-stage residual neural network model is trained using a two-stage training strategy, wherein the two-stage training strategy includes a first-stage training strategy and a second-stage training strategy.

[0015] Under the first-stage training strategy, the multidimensional input feature vector, the running mutual inductance value, and the logarithmically transformed running load resistance value are used as inputs, and the initial two-stage residual neural network model is trained with only the associated running mutual inductance value as the learning target to obtain the two-stage residual neural network model.

[0016] Under the second-stage training strategy, the multidimensional input feature vector, the operating mutual inductance value, and the logarithmized operating load resistance value are used as inputs, and the associated operating mutual inductance value and the logarithmized operating load resistance value are used as joint learning targets to obtain a pre-trained two-stage residual neural network model.

[0017] Optionally, the process of constructing the preset dynamic optimal impedance matching three-dimensional surface includes:

[0018] Set a preset range for the mutual inductance value and the equivalent load resistance value, and establish a high-density parameter grid based on the preset range;

[0019] The output power and efficiency of all grid points in the high-density parameter grid are calculated based on the system output power calculation formula and the system efficiency calculation formula, respectively, so as to obtain the three-dimensional surface of output power and the three-dimensional surface of efficiency of mutual inductance value-equivalent load resistance value.

[0020] A preset dynamic optimal impedance matching three-dimensional surface is established using the output power three-dimensional surface and the efficiency three-dimensional surface;

[0021] The formula for calculating the system output power is as follows:

[0022]

[0023] In the formula: For output power, This is the equivalent AC voltage of the primary-side full-bridge inverter circuit. The system's operating frequency, This represents the mutual inductance between the transmitting coil of the primary-side compensation network and the receiving coil of the secondary-side compensation network. For the primary-side series compensation inductor L P impedance, This is the equivalent load resistance of the secondary-side full-bridge rectifier circuit. This is the equivalent internal resistance of the secondary edge of the secondary edge compensation network. This represents the equivalent internal resistance of the original edge of the original edge compensation network.

[0024] The system efficiency calculation formula is expressed as follows:

[0025]

[0026] In the formula: This is the efficiency value.

[0027] Optionally, the step of adjusting the duty cycle of the secondary-side Buck-Boost step-up / step-down circuit according to the optimal equivalent load resistance value and the predicted load resistance value, so that the equivalent load resistance of the secondary-side full-bridge rectifier circuit approaches the optimal equivalent load resistance value, includes:

[0028] According to the preset secondary-side mapping relationship, the duty cycle of the secondary-side Buck-Boost step-up / step-down circuit is adjusted by the PI controller based on the optimal equivalent load resistance value and the predicted load resistance value, so that the equivalent load resistance of the secondary-side full-bridge rectifier circuit approaches the optimal equivalent load resistance value.

[0029] The calculation formula for the preset secondary edge mapping relationship is expressed as follows:

[0030]

[0031] In the formula: The duty cycle of the secondary-side Buck-Boost step-up / step-down circuit. This represents the load resistance value.

[0032] Optionally, the step of adjusting the duty cycle of the primary-side Buck-Boost step-up / step-down circuit according to the actual input voltage and the optimal equivalent AC voltage value, so that the equivalent AC voltage of the primary-side full-bridge inverter circuit approaches the optimal equivalent AC voltage value, includes:

[0033] According to the preset primary-side mapping relationship, the duty cycle of the primary-side Buck-Boost step-up / step-down circuit is adjusted by the PI controller based on the actual input voltage and the optimal equivalent AC voltage value, so that the equivalent AC voltage of the primary-side full-bridge inverter circuit approaches the optimal equivalent AC voltage value.

[0034] The calculation formula for the preset original edge mapping relationship is expressed as follows:

[0035]

[0036] In the formula: This is the actual input voltage. This represents the duty cycle of the primary-side Buck-Boost step-up / step-down circuit.

[0037] Optionally, the pre-trained two-stage residual neural network model includes an input layer, an input mapping layer, multiple residual blocks, and an output layer connected in sequence.

[0038] Each residual block includes a first linear transformation unit, a first batch normalization unit, a first ReLU activation function unit, an SE attention mechanism unit, a second linear transformation unit, a second batch normalization unit, a jump connection branch that is jump-connected to the first linear transformation unit, and a second ReLU activation function unit that is jump-connected to the jump connection branch.

[0039] Optionally, the primary-side Buck-Boost step-up / step-down circuit is connected to the primary-side full-bridge inverter circuit to perform step-up / step-down conversion on the actual input voltage. By adjusting the duty cycle of its power switch, the DC voltage amplitude output to the primary-side full-bridge inverter circuit is controlled, thereby adjusting the equivalent AC voltage of the primary-side full-bridge inverter circuit.

[0040] The primary-side full-bridge inverter circuit is connected to the primary-side compensation network and is used to invert the DC voltage output by the primary-side Buck-Boost step-up / step-down circuit into a high-frequency AC voltage, thereby providing high-frequency AC excitation for the transmitting coil of the primary-side compensation network.

[0041] The primary-side compensation network and the secondary-side compensation network are connected by magnetic field coupling, and are used to transfer energy from the transmitting coil to the receiving coil of the secondary-side compensation network through an alternating magnetic field;

[0042] The secondary-side compensation network is connected to the secondary-side full-bridge rectifier circuit and is used to transmit the high-frequency AC power induced by the receiving coil to the secondary-side full-bridge rectifier circuit.

[0043] The secondary-side full-bridge rectifier circuit is connected to the secondary-side Buck-Boost buck-boost circuit, and is used to rectify the high-frequency AC power output by the secondary-side compensation network into pulsating DC power and send it to the secondary-side Buck-Boost buck-boost circuit.

[0044] The secondary-side Buck-Boost step-up / step-down circuit is connected to the load resistor and is used to perform step-up / step-down conversion on the pulsating DC power. By adjusting the duty cycle of its power switch, the voltage and current output to the load resistor are controlled. At the same time, the equivalent load resistance of the secondary-side full-bridge rectifier circuit is dynamically adjusted to match the optimal equivalent load resistance value.

[0045] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of any of the wireless power transmission control methods described above.

[0046] This invention also provides a computer-readable storage medium storing a computer program or instructions thereon, which, when executed by a processor, implement the steps of any of the wireless power transmission control methods described above.

[0047] This invention also provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the steps of any of the wireless power transmission control methods described above.

[0048] As can be seen from the above technical solutions, the present invention has the following advantages:

[0049] This invention provides a deep learning-based wireless power transfer control method and apparatus. The LCC-S topology wireless power transfer system includes a primary-side Buck-Boost converter circuit, a primary-side full-bridge inverter circuit, a primary-side compensation network, a secondary-side compensation network, a secondary-side full-bridge rectifier circuit, a secondary-side Buck-Boost converter circuit, and a load resistor. The method includes:

[0050] The system acquires the current and voltage values ​​of the primary-side series compensation inductor and the primary-side parallel compensation capacitor of the primary-side Buck-Boost step-up / step-down circuit in real time. The acquired current and voltage values ​​are input into a pre-trained two-stage residual neural network model for parameter prediction, predicting the mutual inductance and load resistance values. The predicted mutual inductance, preset output power target, and preset efficiency target values ​​are input into a preset dynamic optimal impedance matching three-dimensional surface to determine the optimal equivalent load resistance value. Based on the optimal equivalent load resistance value and the predicted load resistance value, the duty cycle of the secondary-side Buck-Boost step-up / step-down circuit is adjusted so that the equivalent load resistance of the secondary-side full-bridge rectifier circuit approaches the optimal equivalent load resistance value. Based on the optimal equivalent load resistance value, the predicted mutual inductance value, the preset output power target value, and system fixed parameters, the optimal equivalent AC voltage value is calculated. Based on the actual input voltage and the optimal equivalent AC voltage value, the duty cycle of the primary-side Buck-Boost step-up / step-down circuit is adjusted so that the equivalent AC voltage of the primary-side full-bridge inverter circuit approaches the optimal equivalent AC voltage value.

[0051] In this invention, a pre-trained two-stage residual neural network model is used to effectively predict the changed mutual inductance and load resistance values. At the same time, a pre-established dynamic optimal impedance matching surface is used to find the optimal equivalent load resistance and optimal equivalent AC voltage that balance high efficiency and changes. The duty cycle of the secondary-side Buck-Boost step-up / step-down circuit is controlled to achieve closed-loop control of the equivalent load resistance. Then, the duty cycle of the primary-side Buck-Boost step-up / step-down circuit is controlled to achieve closed-loop control of the equivalent input voltage. Thus, the constant power output of the entire system is achieved with high efficiency within a preset error range. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 A flowchart illustrating the steps of a deep learning-based wireless power transfer control method provided in an embodiment of the present invention;

[0054] Figure 2 A flowchart illustrating the implementation of constant power control using a wireless power transfer control method based on a residual neural network, as provided in an embodiment of the present invention.

[0055] Figure 3 A topology circuit diagram of the LCC-S topology wireless power transmission system provided in an embodiment of the present invention;

[0056] Figure 4 The equivalent circuit diagram of the LCC-S topology wireless power transfer system provided in the embodiments of the present invention;

[0057] Figure 5 The equivalent circuit diagram of the transmitter of the LCC-S topology wireless power transmission system provided in the embodiments of the present invention;

[0058] Figure 6 The LCC-S topology wireless power transfer system provided in the embodiments of the present invention Equivalent U in Circuit diagram;

[0059] Figure 7 R of the LCC-S topology wireless power transmission system provided in the embodiments of the present invention E Equivalent R O Circuit diagram;

[0060] Figure 8 This is a model structure diagram of the pre-trained two-stage residual neural network model provided in an embodiment of the present invention;

[0061] Figure 9 A simplified simulation circuit diagram of the LCC-S topology wireless power transfer system provided in this embodiment of the invention;

[0062] Figure 10 A flowchart for generating a dataset provided in an embodiment of the present invention;

[0063] Figure 11 This is a flowchart illustrating the model training process of a two-stage residual neural network model provided in an embodiment of the present invention.

[0064] Figure 12 Mutual inductance M provided in the embodiments of the present invention 12 Error distribution histogram;

[0065] Figure 13 The load resistor R provided in the embodiments of the present invention O Error distribution histogram;

[0066] Figure 14 Mutual inductance M provided in the embodiments of the present invention 12 Determine the coefficient distribution diagram;

[0067] Figure 15 The load resistor R provided in the embodiments of the present invention O Determine the coefficient distribution diagram;

[0068] Figure 16 (a) is a model diagram of the three-dimensional surface of dynamic optimal impedance matching of mutual inductance value-equivalent load output power provided in an embodiment of the present invention; Figure 16 (b) is a model diagram of the three-dimensional surface of the system power dynamic optimal impedance matching of mutual inductance value and equivalent load provided in the embodiment of the present invention. Detailed Implementation

[0069] This invention provides a deep learning-based wireless power transfer control method to solve the technical problem that existing technologies cannot efficiently achieve constant power control in wireless power transfer systems.

[0070] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0071] It should be noted that, in the optional embodiments of the present invention, the data related to object information, etc., requires the permission or consent of the object when the embodiments of the present invention are applied to specific products or technologies. Furthermore, the collection, use, and processing of the relevant data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. In other words, if the embodiments of the present invention involve data related to an object, it needs to be obtained with the object's authorization and consent, the authorization and consent of relevant departments, and in accordance with the relevant laws, regulations, and standards of the country and region. If the embodiments involve personal information, the acquisition of all personal information requires the individual's consent. If sensitive information is involved, the separate consent of the information subject is required. The embodiments also need to be implemented with the object's authorization and consent.

[0072] Please see Figure 1 and Figure 2 This invention provides a deep learning-based wireless power transfer control method, the method comprising:

[0073] Step 101: Real-time acquisition of the current and voltage values ​​of the primary-side series compensation inductor and the primary-side parallel compensation capacitor of the primary-side Buck-Boost step-up / step-down circuit;

[0074] It should be noted that you should refer to [link / reference]. Figure 3This invention relates to an LCC-S topology wireless power transfer system, which includes a primary-side Buck-Boost step-up / step-down circuit, a primary-side full-bridge inverter circuit, a primary-side compensation network, a secondary-side compensation network, a secondary-side full-bridge rectifier circuit, a secondary-side Buck-Boost step-up / step-down circuit, and a load resistor.

[0075] The primary-side Buck-Boost step-up / step-down circuit is connected to the primary-side full-bridge inverter circuit to perform step-up / step-down conversion on the actual input voltage. By adjusting the duty cycle of its power switch, the DC voltage amplitude output to the primary-side full-bridge inverter circuit is controlled, thereby adjusting the equivalent AC voltage of the primary-side full-bridge inverter circuit.

[0076] The primary-side full-bridge inverter circuit is connected to the primary-side compensation network to invert the DC voltage output by the primary-side Buck-Boost step-up / step-down circuit into a high-frequency AC voltage, thereby providing high-frequency AC excitation for the transmitting coil of the primary-side compensation network.

[0077] The primary-side compensation network and the secondary-side compensation network are connected by magnetic field coupling, which is used to transfer energy from the transmitting coil to the receiving coil of the secondary-side compensation network through an alternating magnetic field;

[0078] The secondary-side compensation network is connected to the secondary-side full-bridge rectifier circuit and is used to transmit the high-frequency AC power induced by the receiving coil to the secondary-side full-bridge rectifier circuit.

[0079] The secondary-side full-bridge rectifier circuit is connected to the secondary-side Buck-Boost buck-boost circuit to rectify the high-frequency AC power output from the secondary-side compensation network into pulsating DC power and send it to the secondary-side Buck-Boost buck-boost circuit.

[0080] The secondary-side Buck-Boost converter circuit is connected to the load resistor and is used to perform buck-boost conversion on pulsating DC power. By adjusting the duty cycle of its power switch, the voltage and current output to the load resistor are controlled. At the same time, the equivalent load resistance of the secondary-side full-bridge rectifier circuit is dynamically adjusted to match the optimal equivalent load resistance value.

[0081] Specifically, the primary-side Buck-Boost circuit consists of a MOSFET switch Q. A Freewheeling diode D A Power inductor L A and filter capacitor C A Composition; The input voltage of the primary-side full-bridge inverter circuit is controlled by controlling the duty cycle d1;

[0082] The primary-side full-bridge inverter circuit consists of MOSFET switches Q1, Q2, Q3 and Q4. The first bridge arm is composed of MOSFET switches Q1 and Q2, and the second bridge arm is composed of MOSFET switches Q3 and Q4.

[0083] The primary-side compensation network consists of the primary-side transmitting coil L T Primary-side series compensation inductor L P Primary-side series compensation capacitor C T Primary-side parallel compensation capacitor C P The equivalent internal resistance R of the primary side T Composition; including the primary-side transmitting coil L T The upper end is connected in series with a primary-side compensation capacitor C. T Series compensation inductor L on the primary side P Connected to the positive end of the midpoint of the first bridge arm, while the primary side transmitting coil L... T The lower end passes through the primary side's equivalent internal resistance R. T Connected to the negative terminal of the midpoint of the second bridge arm; wherein the primary side is connected in series with the compensation inductor L. P Primary-side series compensation capacitor C T The compensation capacitor C is connected in parallel with the primary side. P Forming a T-shaped structure;

[0084] The secondary-side full-bridge rectifier circuit includes four fast recovery diodes D1, D2, D3, and D4, with a filter capacitor C connected in parallel at its output. L ;

[0085] The secondary-side compensation network consists of the secondary-side receiving coil L. S Secondary-side series compensation capacitor C S and the equivalent internal resistance R of the secondary side S Composition; wherein, the secondary receiving coil L S The upper end is connected in series with a secondary compensation capacitor C S Connected to the positive end of the midpoint of the bridge arm, while the secondary receiving coil L... S The lower end passes through the secondary side with an equivalent internal resistance R. S Connect to the negative end of the midpoint of the bridge arm;

[0086] The secondary-side Buck-Boost circuit consists of a MOSFET switch Q. B Freewheeling diode D B Power inductor L B and filter capacitor C B It consists of a parallel load resistor R at its output. O The equivalent load resistance of the secondary full-bridge rectifier circuit is controlled by controlling the duty cycle d2.

[0087] For ease of understanding, the fundamental wave equivalent method is used to... Figure 3 The complex topology of the LCC-S topology wireless power transfer system is equivalent to, for example, the following: Figure 4The equivalent circuit diagram is shown. Specifically, the output of the DC voltage after passing through the primary-side Buck-Boost step-up / step-down circuit and the primary-side full-bridge inverter circuit is equivalent to an equivalent AC voltage. The input resistance of the entire secondary-side full-bridge rectifier circuit and the secondary-side Buck-Boost buck-boost circuit can be equivalent to R. E Although the polarity of the equivalent AC voltage changes continuously, it remains constant within a sufficiently short time period. Consequently, the direction of the current in the circuit also remains unchanged, which facilitates subsequent circuit analysis.

[0088] Because the primary transmitting coil receives an alternating current in the complete circuit under the equivalent AC voltage, an alternating magnetic field is generated that passes through the secondary receiving coil. Due to the principle of electromagnetic induction, the secondary receiving coil induces an electromotive force with a magnitude of... Furthermore, since the secondary side is a complete circuit with a load, a secondary current I will be generated. S The secondary current, after passing through the secondary receiving coil, generates its own magnetic field. This magnetic field couples back to the primary coil, generating a reverse electromotive force on the primary side, with a magnitude of... .

[0089] To reduce the reactive power of the entire system and improve its efficiency, the system must be brought into a resonant state. This requires:

[0090] (1)

[0091] Among them, the secondary receiving coil L S impedance Secondary side series compensation capacitor C S impedance Primary transmitting coil L T impedance Primary-side series compensation inductor L P impedance Primary-side parallel compensation capacitor C P impedance ;

[0092] Therefore, the operating frequency of the entire system satisfy:

[0093] (2)

[0094] For the secondary side, according to Kirchhoff's Voltage Law (KVL), we can obtain:

[0095] (3)

[0096] For the reverse induced electromotive force on the primary side, its equivalent resistance is:

[0097] (4)

[0098] Combining formulas (1), (3), and (4), the reflection impedance R can be obtained. eq :

[0099] (5)

[0100] Therefore, the principle can be understood. Figure 4 Further simplified to Figure 5 ,for Figure 5 The simplified schematic diagram, based on KCL and KVL, is as follows:

[0101] (6)

[0102] By simplifying formula (1), we can obtain:

[0103] (7)

[0104] (8)

[0105] (9)

[0106] for Figure 4 For the secondary loop, we have:

[0107] (10)

[0108] Substituting formulas (1) and (8) into formula (10), we can obtain:

[0109] (11)

[0110] For the system as a whole, we have:

[0111] (12)

[0112] Substituting formula (11) into formula (12) yields:

[0113] (13)

[0114] In the formula: This refers to the output power.

[0115] Then, substituting formula (5) into formula (13) will establish the formula for calculating the system output power, namely:

[0116] (14)

[0117] Furthermore, for the system input power of the entire system In other words:

[0118] (15)

[0119] Substituting formula (7) into formula (15) yields:

[0120] (16)

[0121] Dividing formula (14) by formula (16) yields the overall system efficiency. :

[0122] (17)

[0123] Understandably, the overall efficiency of the computing system... It can help find the optimal operating point that ensures constant output power while minimizing system energy consumption and heat generation.

[0124] Since formula (14) is derived from the simplified equivalent circuit diagram, the formula contains... From Figure 3 The primary-side Buck-Boost converter circuit and the primary-side full-bridge inverter circuit are equivalent simplifications, and From Figure 3 The secondary-side full-bridge rectifier circuit and the secondary-side Buck-Boost buck-boost circuit are equivalent simplifications, which will be discussed in detail below:

[0125] for Specifically, its equivalent circuit is as follows: Figure 6 As shown, from Figure 6 It can be observed that the MOSFET Q A Diode D A Power inductor L A and filter capacitor C A This forms a DC / DC converter circuit, namely a Buck-Boost converter circuit, whose output voltage U O With input voltage U I The ratio is:

[0126] (18)

[0127] Where d1 is the MOSFET Q A The period duty cycle, so according to the relationship in formula (18) from Figure 5 The input voltage of the inverter circuit can be obtained. :

[0128] (19)

[0129] The input voltage of the primary-side inverter circuit The relationship between the amplitude of the fundamental wave and the equivalent AC voltage is as follows:

[0130] (20)

[0131] Substituting formula (19) into formula (20) yields U. in and The preset primary edge mapping relationship is:

[0132] (twenty one)

[0133] And for Specifically, its equivalent circuit is as follows: Figure 7 As shown, the equivalent input resistance R of the secondary full-bridge rectifier circuit can be obtained from the circuit diagram. E The equivalent load resistance R of the secondary-side Buck-Boost buck-boost circuit B The relationship is:

[0134] (twenty two)

[0135] According to the ideal state of power conservation in the secondary circuit, that is, for the secondary Buck-Boost buck-boost circuit:

[0136] (twenty three)

[0137] In the formula: This is the input voltage of the Buck-Boost step-up / step-down circuit. This refers to the output voltage of the Buck-Boost buck-boost circuit.

[0138] Combining the input-output voltage relationship of the Buck-Boost buck-boost circuit with formula (18), we can obtain:

[0139] (twenty four)

[0140] Combining equations (22) and (24), the load resistance R can be obtained. O The equivalent input resistance R of the secondary full-bridge rectifier circuit E The preset secondary edge mapping relationship is as follows:

[0141] (25)

[0142] Observe formula (14), angular frequency Primary resistance R T and secondary side internal resistance R S These parameters are determined during circuit design and can be considered constants during normal system operation, serving as fixed parameters for subsequent system operation. However, the mutual inductance M... 12 The load resistance R may change due to foreign objects between the receiving device and the transmitting end, or due to external forces causing the receiving coil to shift from the fixed transmitting coil; O Because, for example, when a mobile phone is charging, the battery switches from requiring a large current at a low load resistance to requiring a small current at a high load resistance, causing the load resistance to rise from a few ohms to tens of ohms. This change in load resistance can disrupt the system's condition of achieving optimal impedance matching only under a "specific load resistance." How can the mutual inductance M be obtained in real time throughout the entire system's operation? 12 and load resistance R O The changed value is the key to the entire system achieving constant power output.

[0143] To address this problem, this invention proposes a method for real-time prediction of mutual inductance M during system operation based on a two-stage residual neural network model. 12 and load resistance R O The algorithm, the logarithmic two-stage residual neural network, is a residual neural network improved from the traditional residual neural network based on the circuit topology characteristics of this invention. It is a residual neural network that extracts the relationship between the input and output features of a large amount of data and continuously modifies the internal parameters of the network through multiple forward and backward propagation processes, thereby fitting a complex formula for the input and output.

[0144] In this embodiment, considering the topological characteristics of the LCC-S topology wireless power transfer system, and in order to reduce sampling costs and improve the prediction reliability of the two-stage residual neural network model, this embodiment selects the primary-side series compensation inductor L of the primary-side Buck-Boost buck-boost circuit. P The compensation capacitor C is connected in parallel with the primary side. P The current and voltage values ​​are used as input features of the two-stage residual neural network model; where the primary side series compensation inductor L P The compensation capacitor C is connected in parallel with the primary side. P The current and voltage values ​​are high-frequency AC signals, and their amplitude, phase, and harmonic content contain rich information. They are easier to extract features from than simple DC signals. At the same time, single-sided (i.e., input-side) sampling eliminates the need for position sensors, Hall sensors, and additional communication links, which greatly improves the reliability and response speed of the system and reduces sampling costs.

[0145] Specifically, the primary-side series compensation inductor L of the primary-side Buck-Boost buck-boost circuit is acquired in real time through a current and voltage sampling circuit. P The compensation capacitor C is connected in parallel with the primary side. P The current and voltage values ​​are collected and used as input data for the subsequent two-stage residual neural network model.

[0146] Step 102: Input the collected current and voltage values ​​into the pre-trained two-stage residual neural network model for parameter prediction, and predict the mutual inductance value and load resistance value.

[0147] Understandably, after the system reaches steady-state operation, the current and voltage values ​​of the primary-side series compensation inductor and the primary-side parallel compensation capacitor can be periodically sampled and input into the pre-trained two-stage residual neural network model in step 102 above for parameter prediction, in order to monitor the mutual inductance value M. 12 and load resistance R O If the two do not change, continue periodic sampling; if they change, proceed to step 103.

[0148] It should be noted that you should refer to [link / reference]. Figure 8 The pre-trained two-stage residual neural network model includes an input layer, an input mapping layer, multiple residual blocks, and an output layer connected in sequence. Each residual block includes a first linear transformation unit, a first batch normalization unit, a first ReLU activation function unit, an SE attention mechanism unit, a second linear transformation unit, a second batch normalization unit, a jump connection branch connected to the first linear transformation unit, and a second ReLU activation function unit connected to the jump connection branch. The first linear transformation unit performs a linear transformation on the input features; the first batch normalization unit performs batch normalization on the transformed features output by the first linear transformation unit; the first ReLU activation function unit introduces non-linear features into the batch normalized features; the SE attention mechanism unit adaptively allocates weights for different feature channels; and the jump connection branch directly adds the input features of the residual block to the output of the second ReLU activation function unit, forming a residual connection. ,in For the second batch of normalization units, the input features The mapping result after batch normalization.

[0149] In the pre-trained two-stage residual neural network model, the collected current and voltage values ​​are input to the input mapping layer through the input layer, mapping the collected current and voltage values ​​to a 128-dimensional space to initially extract high-level representations. Batch normalization is used to accelerate convergence, and ReLU activation function is used for nonlinear processing to obtain the mapped data. Then, four residual blocks are used to retain the original information of the mapped data, alleviate gradient vanishing, and enable the network to stably train deep structures. Finally, the output layer maps the 128-dimensional features output from the last residual block to a 2-dimensional output, predicting the normalized mutual inductance value M. 12 Logarithm of load resistance To facilitate subsequent calculations, the logarithm of the load resistance value is... Convert to the corresponding load resistance value .

[0150] The logarithmic two-stage residual neural network structure established in this invention is simple yet highly expressive. After training, the forward propagation is a deterministic computation without an iterative process, making it suitable for embedded deployment. It can be scaled to multiple outputs and can be transferred to systems with different power levels or topologies, exhibiting strong scalability.

[0151] In one specific implementation, the process of obtaining the pre-trained two-stage residual neural network model includes the following steps:

[0152] S11. Obtain the operating current and voltage values ​​of the primary-side series compensation inductor and the primary-side parallel compensation capacitor of the LCC-S topology wireless power transmission system, and collect the operating mutual inductance value and operating load resistance value associated with the operating current and voltage values.

[0153] S12. Perform feature engineering processing on the operating current and voltage values ​​to obtain a multi-dimensional input feature vector;

[0154] S13. Construct an initial two-stage residual neural network model. Use the multidimensional input feature vector, the operating mutual inductance value, and the logarithmically transformed operating load resistance value as inputs. Train the initial two-stage residual neural network model using a two-stage training strategy. The two-stage training strategy includes a first-stage training strategy and a second-stage training strategy.

[0155] S14. Under the first-stage training strategy, the multi-dimensional input feature vector, the running mutual inductance value and the logarithmized running load resistance value are used as inputs. The initial two-stage residual neural network model is trained with only the associated running mutual inductance value as the learning target to obtain the two-stage residual neural network model.

[0156] S15. Under the second-stage training strategy, the multi-dimensional input feature vector, the running mutual inductance value, and the logarithmized running load resistance value are used as inputs, and the associated running mutual inductance value and the logarithmized running load resistance value are used as joint learning targets to obtain a pre-trained two-stage residual neural network model.

[0157] To improve the training cost and speed of the model, in the data acquisition step S11 above, this embodiment uses Simulink in Matlab to build a simulation model of the LCC-S topology wireless power transfer system. To simplify the model, the Buck-Boost buck-boost circuits on the primary and secondary sides are ignored in the simulation model. The simplified simulation model is as follows: Figure 9 As shown.

[0158] When building the model, the range of values ​​for three key physical parameters can be preset, namely the input voltage U. in (30-90V), mutual inductance value M 12 (30-90uH) and load resistance R O (25-50Ω), the mutual inductance value M under different operating conditions of the system in actual operation is simulated by three sets of varying quantities. 12 and operating load resistance value R O The changes are due to various factors; then, the code loops to randomly sample these three sets of data and substitute them into the simulation model to run the simulation, while simultaneously adjusting the primary-side series compensation inductor L. P The compensation capacitor C is connected in parallel with the primary side. P The operating current and voltage values ​​are sampled. Thus, each generated dataset is a 6-dimensional vector, which for the two-stage residual neural network model is data with 4 input features and 2 output features, i.e.:

[0159] (26)

[0160] Such an operation can be run efficiently multiple times within a code loop (e.g., ten thousand times) to obtain multiple sets of data. To improve model robustness and avoid overfitting, this embodiment also adds zero-mean Gaussian white noise to the ideal simulation values ​​to simulate errors caused by parasitic inductance, parasitic capacitance, etc., in actual circuits. Thus, multiple datasets with 4 input features and 2 output features are constructed under the condition of simulating actual circuits with noise, as shown below. Figure 10 As shown.

[0161] Next, raw physical quantities such as operating voltage and current values ​​are read from the dataset, and the operating load resistance R with a large range is then analyzed. O A logarithmic transformation is performed to smooth the data distribution; then, StandardScaler is used to refine the multidimensional input feature vector and mutual inductance M. 12and the logarithmic load logR O Standardization is performed separately to eliminate differences in units; finally, the processed data is divided into training and testing sets according to a preset ratio (e.g., 8:2), and packaged into PyTorch's DataLoader to achieve efficient batch reading of the model.

[0162] During the training phase, due to mutual inductance M 12 The main influence on the system's resonant frequency is strongly correlated with characteristics such as the phase and amplitude ratio of the input voltage and current. The model can more easily capture these patterns from the data, and the load resistance R... O The impact is more complex, and factors involving energy transfer efficiency and load matching can easily lead to higher learning difficulty for the model. Therefore, this embodiment divides the training of the initial two-stage residual neural network model with batch normalization fully connected residual neural network structure into two stages based on a two-stage training strategy, such as... Figure 11 As shown:

[0163] Under the first-stage training strategy, only mutual inductance values ​​are used. 12 The network model undergoes 100 rounds of pre-training with labels to initially learn feature representations. The loss function for the first stage is as follows:

[0164] (27)

[0165] In the formula: The loss function value for the first stage of training. Mean square error, and These are the mutual inductance values ​​predicted by the model. Mutual inductance value and actual measurement , The total number of training samples, and The first The predicted mutual inductance value and the actual measured mutual inductance value for each sample.

[0166] Under the second-stage training strategy, mutual intuition values ​​are used. 12 and logarithmically converted operating load resistance value The model is trained using joint labels and fine-tuned for 200 rounds using a weighted mean squared error loss function. It employs the Adam optimizer and a learning rate decay strategy, and saves the optimal model based on validation loss, ultimately achieving [the desired result]. 12 and High-precision synchronous prediction. The loss function for the second stage is shown below:

[0167] (28)

[0168] In the formula: The loss function value for the second stage of training. Logarithmic load resistance value predicted by the model , The actual measured logarithmic load resistance value .

[0169] It should be noted that traditional calculation methods require precise knowledge of coil dimensions, number of turns, material, distance, angle, etc., to calculate mutual inductance. 12 The logarithmic two-stage residual neural network proposed in this invention can automatically learn the relationship between the input signal and the input voltage, provided that the training data includes the current and voltage of the series compensation inductor and the parallel compensation capacitor on the primary side of the input side. 12 and The mapping relationship is particularly suitable for complex nonlinear scenarios.

[0170] The second-stage training strategy can achieve synchronous high precision. 12 and After the pre-trained two-stage residual neural network model is completed, it is necessary to use the test set to test and evaluate the pre-trained two-stage residual neural network model.

[0171] During the testing phase, MAE and R² can be used to evaluate the pre-trained two-stage residual neural network model. MAE (Mean Absolute Error) measures the average absolute difference between the predicted and actual values. A smaller MAE indicates a smaller gap between the model's predictions and the actual values, signifying better model performance. It assigns equal weight to all errors, thus being unaffected by outliers. R² (Coefficient of Determination) represents the proportion of variability explained by the model relative to the total variability. R² values ​​range from negative infinity to 1; an R² value close to 1 indicates that the model fits the data well.

[0172] For example, please refer to Table 1 and Figures 12-13 As shown in the chart, the predicted mutual inductance M 12 The average absolute error is only 1.054uH, while the predicted load resistance R O The mean absolute error was only 2.606Ω, and the model's prediction performance met expectations.

[0173] The MAE evaluation metrics for the model are as follows:

[0174] Table 1 Mean Absolute Error Table

[0175]

[0176] For example, please refer to Table 2 and Figures 14-15 As shown in the chart, the predicted mutual inductance value M 12 The coefficient of determination is as high as 0.9936, indicating that the model explains M. 12 99.4% of the variable variance indicates that the model is effective for M. 12 The prediction is excellent and highly accurate; and the predicted load resistance R... O The coefficient of determination is 0.7716, indicating that the model can explain R. O The variance of the variable is approximately 77.39%.

[0177] The R² evaluation metric for the model is as follows:

[0178] Table 2. Deterministic Coefficients Table

[0179]

[0180] In practical applications, the pre-trained two-stage residual neural network model is written into the DSP chip using C language. Figure 3 When the circuit is working, the pre-trained two-stage residual neural network model compensates for the inductor L by sampling the primary side in series. P The compensation capacitor C is connected in parallel with the primary side. P The current and voltage values ​​predict the real-time mutual inductance value M. 12 and load resistance value R O Achieve adaptive impedance matching.

[0181] Step 103: Input the predicted mutual inductance value, preset output power target value and preset efficiency target value into the preset dynamic optimal impedance matching three-dimensional surface to determine the optimal equivalent load resistance value.

[0182] It should be noted that if only the constant power function is considered for the overall circuit while ignoring the efficiency, it may lead to severe overheating of the overall circuit and a sharp increase in input voltage pressure.

[0183] Substituting formula (5) into formula (17) yields:

[0184] (29)

[0185] From formula (29), it can be seen that when efficiency When set to a higher value, mutual inductance M 12 Changes will lead to the optimal impedance matching point, i.e., R. E The impedance changes. It can be seen that in actual circuit operation, in order to maintain high efficiency, the optimal impedance matching point R... EIt will keep changing; it's not a fixed value. How to determine the mutual inductance M... 12 Finding the optimal impedance matching point during changes is key to maintaining high system efficiency.

[0186] To address these issues, this invention proposes a dynamic optimal impedance matching three-dimensional surface method. The predicted mutual inductance value, the preset output power target value, and the preset efficiency target value are input into the preset dynamic optimal impedance matching three-dimensional surface to determine the optimal equivalent load resistance value.

[0187] In one specific implementation, the process of constructing the preset dynamic optimal impedance matching three-dimensional surface includes the following steps:

[0188] S21. Set the preset variation range of mutual inductance value and equivalent load resistance value, and establish a high-density parameter grid according to the preset variation range;

[0189] S22. Calculate the output power and efficiency of all grid points in the high-density parameter grid based on the system output power calculation formula and the system efficiency calculation formula, respectively, so as to obtain the three-dimensional surface of output power and the three-dimensional surface of efficiency of mutual inductance value-equivalent load resistance value.

[0190] S23. Establish a preset dynamic optimal impedance matching three-dimensional surface using the output power three-dimensional surface and the efficiency three-dimensional surface.

[0191] In this specific embodiment, a preset range of variation for the mutual inductance value and the equivalent load resistance value is set according to actual needs, such as the mutual inductance value M. 12 (30-90) and equivalent load resistance R E (0.1-30) The range of variation of ( ). Simultaneously, obtain the system's fixed parameters, namely the operating frequency. Primary resistance Secondary resistance Series compensation inductor and reactance of the primary side Inherent parameters, etc.

[0192] Subsequently, based on mutual inductance M 12 (30-90) and equivalent load resistance R E (0.1-30) The range of variation generates a 120x120 high-density grid. The output power of all grid points in the high-density grid is calculated according to the system output power calculation formula (14) and the system efficiency calculation formula (29). and efficiency This yields the three-dimensional surface of output power and the three-dimensional surface of efficiency based on mutual inductance and equivalent load resistance. The output power three-dimensional surface is defined by the mutual inductance value minus the equivalent load resistance value. and equivalent load resistance As the independent variable, with output power The dependent variable is a three-dimensional surface, while the efficiency three-dimensional surface is based on the mutual inductance value. and equivalent load resistance As the independent variable, with system efficiency The three-dimensional surface is the dependent variable.

[0193] Finally, a preset dynamic optimal impedance matching three-dimensional surface is established using the output power three-dimensional surface and the efficiency three-dimensional surface.

[0194] After establishing the preset dynamic optimal impedance matching three-dimensional surface, the predicted mutual inductance value, preset output power target value and preset efficiency target value are input into the preset dynamic optimal impedance matching three-dimensional surface. According to the preset engineering error range and preset variation range, the set of equivalent load resistance values ​​that meet the preset output power target value and preset efficiency target value are selected from the preset dynamic optimal impedance matching three-dimensional surface, and the optimal equivalent load resistance value is determined from the set.

[0195] For example, a preset engineering error range is set for mutual inductance value ±2μH and efficiency ±1%, with a target efficiency value. =0.95, target output power value =500W, predicted mutual inductance M 12 =60 Input to such Figure 16 In the preset dynamic optimal impedance matching three-dimensional curves shown in (a) and (b), the optimal equivalent load resistance R is obtained. E =19.20 Accurate within the error range =500W, accurate within the error range =0.952.

[0196] Step 104: Adjust the duty cycle of the secondary Buck-Boost step-up / step-down circuit according to the optimal equivalent load resistance value and the predicted load resistance value, so that the equivalent load resistance of the secondary full-bridge rectifier circuit approaches the optimal equivalent load resistance value.

[0197] After selecting the optimal equivalent load resistance value and the predicted load resistance value, a high-efficiency constant power strategy can be implemented based on them. For ease of understanding, refer to formula (14), where the angular frequency... The primary-side parallel compensation capacitor impedance Z CP The primary equivalent internal resistance R T and the equivalent internal resistance R of the secondary sideS The output power P remains constant. OUT When it is the set value, mutual inductance M 12 Equivalent input resistance R E and equivalent input voltage These are the three variables in the formula. To ensure the left side of the formula remains constant, we need to know the values ​​of two of the three variables on the right side and adjust the third variable. Specifically, step 102 can predict the mutual inductance value M of the first variable. 12 The optimal equivalent load resistance R can be obtained through step 103. E Thus, we can obtain the values ​​of the two variables in formula (14).

[0198] However, the equivalent load resistance R E and load resistance R O The changes are generally inconsistent. According to the preset secondary side mapping relationship of formula (25), the duty cycle of the secondary side Buck-Boost step-up / step-down circuit is adjusted in a closed loop by the PI controller based on the optimal equivalent load resistance value and the predicted load resistance value. This makes the actual equivalent load resistance approximate the optimal equivalent load resistance R. E Substituting formula (25) into formula (14) yields the output power. Duty cycle The following relationship exists between them:

[0199] (30)

[0200] Based on formula (30), the secondary impedance matching power adjustment can be obtained. Thus, according to the preset secondary mapping relationship in formula (25), the corresponding duty cycle adjustment command is generated to make the actual equivalent load resistance approach the optimal equivalent load resistance R. E .

[0201] It is understandable that in practical applications, when the mutual inductance M in the circuit... 12 Or load resistance R O When the impedance changes, the original impedance matching is broken, leading to a decrease in efficiency. This step controls the duty cycle of the secondary-side Buck-Boost step-up / step-down circuit. To change the equivalent load resistance This allows the system to rematch to the optimal equivalent load resistance value, restoring the initial impedance matching state of the system.

[0202] Step 105: Calculate the optimal equivalent AC voltage value based on the optimal equivalent load resistance value, the predicted mutual inductance value, the preset output power target value, and the system fixed parameters.

[0203] In this step, the optimal equivalent load resistance value, the predicted mutual inductance value, the preset output power target value, and the system fixed parameters are substituted into formula (14) to obtain the optimal equivalent AC voltage value. The value of .

[0204] Step 106: Adjust the duty cycle of the primary-side Buck-Boost step-up / step-down circuit according to the actual input voltage and the optimal equivalent AC voltage value, so that the equivalent AC voltage of the primary-side full-bridge inverter circuit approaches the optimal equivalent AC voltage value.

[0205] It is worth noting that frequently adjusting the equivalent input voltage artificially... That is impractical; therefore, this embodiment is based on the actual input voltage. The duty cycle of the primary-side Buck-Boost step-up / step-down circuit is adjusted according to the optimal equivalent AC voltage value, so that the equivalent AC voltage of the primary-side full-bridge inverter circuit approaches the optimal equivalent AC voltage value.

[0206] In one specific implementation, step 106 includes: adjusting the duty cycle of the primary-side Buck-Boost step-up / step-down circuit according to the preset primary-side mapping relationship, based on the actual input voltage and the optimal equivalent AC voltage value, so that the equivalent AC voltage of the primary-side full-bridge inverter circuit approaches the optimal equivalent AC voltage value.

[0207] Specifically, according to the set actual input voltage Based on the preset primary-side mapping relationship of formula (21), the closed-loop PI adjusts the duty cycle of the original Buck-Boost buck-boost circuit. This makes the actual equivalent input voltage approximate the calculated equivalent input voltage. Substituting formula (21) into formula (30) yields the output power. With duty cycle ( , The following relationship exists between them:

[0208] (31)

[0209] Based on formula (31), the primary-side voltage regulation and overall system control objectives can be obtained. Thus, according to the preset primary-side mapping relationship in formula (21), the corresponding duty cycle regulation command is generated to make the actual equivalent input voltage approximate the calculated equivalent input voltage. In this way, the entire system achieves highly efficient constant power control.

[0210] The wireless power transfer control method based on deep learning provided in this invention has the following advantages:

[0211] 1. The control method proposed in this invention uses a logarithmic two-stage residual neural network to predict the mutual inductance of the circuit and the load resistance, and uses a dynamic optimal impedance matching three-dimensional surface to find the optimal impedance matching point that balances high efficiency and output power.

[0212] 2. The control method proposed in this invention utilizes dual-sided Buck-Boost circuits to collaboratively control the optimal impedance matching point and equivalent input voltage under varying mutual inductance and load resistance, enabling the circuit to still achieve high-efficiency constant power output.

[0213] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of any of the wireless power transmission control methods described above.

[0214] This invention also provides a computer-readable storage medium storing a computer program or instructions thereon, which, when executed by a processor, implement the steps of any of the wireless power transmission control methods described above.

[0215] This invention also provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the steps of any of the wireless power transmission control methods described above.

[0216] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0217] The terms "first," "second," etc., used in this application's specification and the foregoing drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0218] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0219] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0220] Furthermore, the functional units in the various embodiments of the present invention 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.

[0221] 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, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0222] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to 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 the present invention.

Claims

1. A deep learning-based wireless power transfer control method, characterized by, The method relates to an LCC-S topology wireless power transfer system, which includes a primary-side Buck-Boost converter, a primary-side full-bridge inverter, a primary-side compensation network, a secondary-side compensation network, a secondary-side full-bridge rectifier, a secondary-side Buck-Boost converter, and a load resistor; the method includes: The current and voltage values ​​of the primary-side series compensation inductor and the primary-side parallel compensation capacitor of the primary-side Buck-Boost step-up / step-down circuit are collected in real time. The collected current and voltage values ​​are input into a pre-trained two-stage residual neural network model for parameter prediction, and the mutual inductance value and load resistance value are predicted. The predicted mutual inductance value, preset output power target value and preset efficiency target value are input into the preset dynamic optimal impedance matching three-dimensional surface to determine the optimal equivalent load resistance value. The duty cycle of the secondary-side Buck-Boost step-up / step-down circuit is adjusted according to the optimal equivalent load resistance value and the predicted load resistance value, so that the equivalent load resistance of the secondary-side full-bridge rectifier circuit approaches the optimal equivalent load resistance value. The optimal equivalent AC voltage value is calculated based on the optimal equivalent load resistance value, the predicted mutual inductance value, the preset output power target value, and the system fixed parameters. The duty cycle of the primary-side Buck-Boost step-up / step-down circuit is adjusted according to the actual input voltage and the optimal equivalent AC voltage value, so that the equivalent AC voltage of the primary-side full-bridge inverter circuit approaches the optimal equivalent AC voltage value.

2. The wireless power transfer control method of claim 1, wherein, The process of obtaining the pre-trained two-stage residual neural network model includes: Obtain the operating current and voltage values ​​of the primary-side series compensation inductor and the primary-side parallel compensation capacitor of the LCC-S topology wireless power transmission system, and collect the operating mutual inductance value and operating load resistance value associated with the operating current and voltage values. The operating current and voltage values ​​are subjected to feature engineering processing to obtain a multi-dimensional input feature vector; An initial two-stage residual neural network model is constructed, using the multidimensional input feature vector, the operating mutual inductance value, and the logarithmically transformed operating load resistance value as inputs. The initial two-stage residual neural network model is trained using a two-stage training strategy, wherein the two-stage training strategy includes a first-stage training strategy and a second-stage training strategy. Under the first-stage training strategy, the multidimensional input feature vector, the running mutual inductance value, and the logarithmically transformed running load resistance value are used as inputs, and the initial two-stage residual neural network model is trained with only the associated running mutual inductance value as the learning target to obtain the two-stage residual neural network model. Under the second-stage training strategy, the multidimensional input feature vector, the operating mutual inductance value, and the logarithmized operating load resistance value are used as inputs, and the associated operating mutual inductance value and the logarithmized operating load resistance value are used as joint learning targets to obtain a pre-trained two-stage residual neural network model.

3. The wireless power transfer control method of claim 1, wherein, The process of constructing the preset dynamic optimal impedance matching three-dimensional surface includes: Set a preset range for the mutual inductance value and the equivalent load resistance value, and establish a high-density parameter grid based on the preset range; The output power and efficiency of all grid points in the high-density parameter grid are calculated based on the system output power calculation formula and the system efficiency calculation formula, respectively, so as to obtain the three-dimensional surface of output power and the three-dimensional surface of efficiency of mutual inductance value-equivalent load resistance value. A preset dynamic optimal impedance matching three-dimensional surface is established using the output power three-dimensional surface and the efficiency three-dimensional surface; The formula for calculating the system output power is as follows: In the formula: For output power, This is the equivalent AC voltage of the primary-side full-bridge inverter circuit. The system's operating frequency, This represents the mutual inductance between the transmitting coil of the primary-side compensation network and the receiving coil of the secondary-side compensation network. For the primary-side series compensation inductor L P impedance, This is the equivalent load resistance of the secondary-side full-bridge rectifier circuit. This is the equivalent internal resistance of the secondary edge of the secondary edge compensation network. This represents the equivalent internal resistance of the original edge of the original edge compensation network. The system efficiency calculation formula is expressed as follows: In the formulae: is the efficiency value.

4. The wireless power transfer control method of claim 1, wherein, The step of adjusting the duty cycle of the secondary-side Buck-Boost step-up / step-down circuit according to the optimal equivalent load resistance value and the predicted load resistance value, so that the equivalent load resistance of the secondary-side full-bridge rectifier circuit approaches the optimal equivalent load resistance value, includes: According to the preset secondary-side mapping relationship, the duty cycle of the secondary-side Buck-Boost step-up / step-down circuit is adjusted by the PI controller based on the optimal equivalent load resistance value and the predicted load resistance value, so that the equivalent load resistance of the secondary-side full-bridge rectifier circuit approaches the optimal equivalent load resistance value. The calculation formula for the preset secondary edge mapping relationship is expressed as follows: In the formula: is the duty cycle of the secondary side Buck-Boost step-up / down circuit, is the load resistance value.

5. The wireless power transfer control method of claim 1, wherein, The step of adjusting the duty cycle of the primary-side Buck-Boost step-up / step-down circuit according to the actual input voltage and the optimal equivalent AC voltage value, so that the equivalent AC voltage of the primary-side full-bridge inverter circuit approaches the optimal equivalent AC voltage value, includes: According to the preset primary-side mapping relationship, the duty cycle of the primary-side Buck-Boost step-up / step-down circuit is adjusted by the PI controller based on the actual input voltage and the optimal equivalent AC voltage value, so that the equivalent AC voltage of the primary-side full-bridge inverter circuit approaches the optimal equivalent AC voltage value. The calculation formula for the preset original edge mapping relationship is expressed as follows: In the formula: This is the actual input voltage. This represents the duty cycle of the primary-side Buck-Boost step-up / step-down circuit.

6. The wireless power transfer control method of claim 1, wherein, The pre-trained two-stage residual neural network model includes an input layer, an input mapping layer, multiple residual blocks, and an output layer connected in sequence. Each residual block includes a first linear transformation unit, a first batch normalization unit, a first ReLU activation function unit, an SE attention mechanism unit, a second linear transformation unit, a second batch normalization unit, a jump connection branch that is jump-connected to the first linear transformation unit, and a second ReLU activation function unit that is jump-connected to the jump connection branch.

7. The wireless power transmission control method according to claim 1, characterized in that, The primary-side Buck-Boost step-up / step-down circuit is connected to the primary-side full-bridge inverter circuit and is used to perform step-up / step-down conversion on the actual input voltage. By adjusting the duty cycle of its power switch, the DC voltage amplitude output to the primary-side full-bridge inverter circuit is controlled, thereby adjusting the equivalent AC voltage of the primary-side full-bridge inverter circuit. The primary-side full-bridge inverter circuit is connected to the primary-side compensation network and is used to invert the DC voltage output by the primary-side Buck-Boost step-up / step-down circuit into a high-frequency AC voltage, thereby providing high-frequency AC excitation for the transmitting coil of the primary-side compensation network. The primary-side compensation network and the secondary-side compensation network are connected by magnetic field coupling, and are used to transfer energy from the transmitting coil to the receiving coil of the secondary-side compensation network through an alternating magnetic field; The secondary-side compensation network is connected to the secondary-side full-bridge rectifier circuit and is used to transmit the high-frequency AC power induced by the receiving coil to the secondary-side full-bridge rectifier circuit. The secondary-side full-bridge rectifier circuit is connected to the secondary-side Buck-Boost buck-boost circuit, and is used to rectify the high-frequency AC power output by the secondary-side compensation network into pulsating DC power and send it to the secondary-side Buck-Boost buck-boost circuit. The secondary-side Buck-Boost step-up / step-down circuit is connected to the load resistor and is used to perform step-up / step-down conversion on the pulsating DC power. By adjusting the duty cycle of its power switch, the voltage and current output to the load resistor are controlled. At the same time, the equivalent load resistance of the secondary-side full-bridge rectifier circuit is dynamically adjusted to match the optimal equivalent load resistance value.

8. A computer apparatus comprising a memory, a processor, and a computer program stored on the memory, wherein the computer program, when executed by the processor, causes the processor to perform the method of any one of claims 1 to 7. The processor executes the computer program to implement the steps of the wireless power transfer control method as described in any one of claims 1-7.

9. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, When the computer program or instructions are executed by the processor, they implement the steps of the wireless power transfer control method as described in any one of claims 1-7.

10. A computer program product comprising computer programs or instructions, characterized in that, When the computer program or instructions are executed by the processor, they implement the steps of the wireless power transfer control method as described in any one of claims 1-7.

Images