Neural Network-Based Adaptive AC / DC Power Control Method and System

Abstract

This invention discloses an adaptive AC / CDC power supply control method and system based on neural networks, belonging to the field of power supply control technology. By deeply integrating neural network technology with wide voltage impedance matching control and overall dynamic control of the AC / CDC power supply, it divides voltage characteristic intervals and constructs nonlinear correlation models of input impedance, grid voltage, and output power within each interval using neural networks. This achieves adaptive impedance matching by dynamically generating impedance matching control parameters based on real-time operating conditions within each interval. A smooth switching mapping relationship of control parameters between adjacent voltage characteristic intervals is established, enabling seamless adjustment of control parameters when the input voltage switches between intervals. Using a power factor correction circuit as the core regulating component, its parameters are dynamically adjusted via neural networks to achieve real-time and accurate matching of input impedance with the grid side. Combined with output closed-loop regulation, this meets the output requirements of electrical equipment, completing the overall adaptive control of the AC / CDC power supply.

Description

Technical Field

[0001] This invention relates to the field of power control technology, specifically to an adaptive AC-DC power control method and system based on neural networks. Background Technology

[0002] As the core equipment for power conversion, AC / DC power supplies are widely used in power electronics, industrial control, consumer electronics, new energy power generation and other fields. Their core function is to convert AC power from the power grid into DC power that meets the needs of the electrical equipment. In application scenarios where the power grid voltage fluctuates greatly, AC / DC power supplies with a wide input voltage range have become the mainstream choice. The input impedance matching control and adaptive control of the overall operation within the wide voltage range directly determine the power supply's operational stability and energy utilization efficiency.

[0003] In existing technologies, ACDC power supply control methods mostly employ control logic with fixed control parameters, making it difficult to adapt to control requirements arising from dynamic changes in operating conditions. For ACDC power supplies with a wide input voltage range, existing impedance matching control often adopts a fixed segmented strategy, that is, pre-dividing the wide input voltage range into several intervals and preset fixed impedance matching control parameters for each interval. During power supply operation, the corresponding fixed parameters are called according to the interval to which the input voltage belongs. Although this method achieves basic segmented impedance matching, it has significant drawbacks: First, the fixed segmented parameters cannot be dynamically adapted to the actual operating conditions of the ACDC power supply, which can easily lead to input impedance mismatch with the grid side, resulting in a sharp drop in energy conversion efficiency. Second, the control parameter switching between intervals is a step change without a smooth transition design. When the input voltage switches between intervals, the control parameters are prone to sudden changes, causing circuit oscillations and further aggravating impedance mismatch. Third, the preset segmented mathematical model cannot fully fit the nonlinear relationship between input impedance, grid voltage, and output power within each interval, resulting in low impedance matching accuracy across the entire input voltage range.

[0004] Currently, some technologies have attempted to fine-tune impedance matching parameters under wide voltage input conditions. However, such fine-tuning is still based on a fixed segmented framework and has not broken through the limitations of preset parameters. It cannot fundamentally solve the impedance mismatch problem under dynamic operating conditions, nor can it achieve adaptive control of ACDC power supplies under all operating conditions. It is still difficult to meet the requirements for power supply operating efficiency and stability under wide grid voltage fluctuation scenarios. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies by providing an adaptive ACDC power supply control method and system based on neural networks. This method deeply integrates neural network technology with wide-voltage impedance matching control and overall dynamic control of the ACDC power supply. By dividing voltage characteristic intervals and constructing nonlinear correlation models of input impedance, grid voltage, and output power within each interval using neural networks, it achieves adaptive impedance matching by dynamically generating impedance matching control parameters based on real-time operating conditions within each interval. It establishes a smooth switching mapping relationship for control parameters between adjacent voltage characteristic intervals, completing abrupt adjustments of control parameters when the input voltage switches between intervals, ensuring the continuity of impedance matching across intervals and the stability of power supply operation. Using a power factor correction circuit as the core regulating component, its parameters are dynamically adjusted through neural networks to achieve real-time and accurate matching between the input impedance and the grid side. Combined with output closed-loop regulation, it meets the output requirements of electrical equipment, completing the overall adaptive control of the ACDC power supply.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On one hand, an adaptive AC / DC power supply control method based on neural networks, which includes the following specific steps: S1: Real-time collection of multi-dimensional operating condition information during the operation of ACDC power supply to obtain raw operating condition information set; S2: Divide the wide input voltage range of the ACDC power supply to obtain multiple continuous and non-overlapping voltage characteristic intervals, and preprocess the original operating condition information set to obtain a standard operating condition information set that meets the input requirements of the neural network model. S3: Construct a neural network model. Through the neural network model, construct nonlinear correlation models of input impedance, grid voltage, and output power in each voltage characteristic interval. At the same time, establish a smooth switching mapping relationship of control parameters between adjacent voltage characteristic intervals. Complete the training and convergence of the neural network model to obtain a pre-trained piecewise adaptive impedance matching neural network model. S4: Input the standard operating condition information set into the pre-trained segmented adaptive impedance matching neural network model, and identify the voltage characteristic range to which the input voltage belongs in real time through model inference calculation, and obtain the adaptive impedance matching control parameter set that matches the current operating condition. S5: Based on the adaptive impedance matching control parameter set, a corresponding electrical signal form of control command is generated and transmitted to the power factor correction circuit control terminal; S6: The main circuit of the ACDC power supply adjusts the operating parameters of the power factor correction circuit according to the received control commands, realizes real-time matching between the input impedance and the grid side, and completes the adaptive control of the ACDC power supply.

[0007] Furthermore, in step S1, the multi-dimensional operating condition information includes at least one of the following: grid-side input voltage information, input impedance information, output power information, and circuit operating temperature information.

[0008] Furthermore, in step S2, the specific steps for dividing the wide input voltage range of the ACDC power supply are as follows: Define the upper and lower limits of the rated wide input voltage of the ACDC power supply; simulate various grid voltage input scenarios by building a test platform; collect data on the power supply's input impedance, output power, and energy conversion efficiency under different input voltage conditions in real time; after data fitting, feature extraction, and effectiveness analysis, locate the impedance matching efficiency inflection point; combine the continuous change gradient of the input voltage amplitude with the impedance adaptation characteristics of the power supply's internal circuit; divide the wide input voltage range into multiple continuous and non-overlapping voltage characteristic intervals based on the efficiency inflection point and the principle of continuous voltage amplitude transition; the boundary values ​​of each interval take into account both the normal fluctuation of grid voltage and the power supply's control response characteristics; ensure that the nonlinear correlation law between input impedance, grid voltage, and output power in each interval is stable and fitable; and ensure smooth connection between adjacent interval boundaries, providing a basic support for the subsequent seamless transition of control parameters across intervals.

[0009] Furthermore, in step S3, the specific technical solution for constructing the neural network model is as follows: Collect sample data from the full-width input voltage range of the ACDC power supply, including full-condition sample sets within each voltage characteristic interval and operating condition sample sets during cross-interval switching. Simultaneously collect corresponding tag data, namely, the optimal impedance matching control parameter sample set within each interval and the smooth transition control parameter sample set during cross-interval switching. The control parameter samples are the switching frequency, duty cycle, and modulation depth parameters of the power factor correction circuit. Using the operating condition sample set of each voltage characteristic interval as input and the corresponding impedance matching control parameter sample set as output, the nonlinear correlation model in each interval is trained, so that the model learns the nonlinear correlation relationship between input impedance, grid voltage and output power in each interval. Using cross-interval switching condition samples as input and corresponding smooth transition control parameter sample sets as output, the smooth switching mapping relationship between intervals is trained, enabling the model to learn the non-abrupt adjustment law of control parameters when the input voltage crosses intervals. The deviation between the model output and the actual sample label is calculated using a loss function. Based on this deviation, the connection weights inside the model are adjusted. The model is iterated repeatedly until the output meets the preset convergence condition, thus obtaining a pre-trained piecewise adaptive impedance matching neural network model.

[0010] Furthermore, in step S4, the preprocessed standard operating condition information set is constructed into a multi-dimensional feature vector, which is then input into the pre-trained piecewise adaptive impedance matching neural network model in real time, and the feature vector is determined by the interval identification discriminant formula. Complete the accurate identification of the voltage characteristic range to which the current input voltage belongs, among which, This represents the total number of voltage characteristic intervals. For the first Feature matching degree function for each voltage feature interval The current input voltage range is assigned a number, and then a nonlinear correlation inference model is called to perform inference calculations, outputting a set of control parameters that are adapted to the current real-time operating conditions. If a cross-range switching of the input voltage is detected by the voltage change discriminant, the smooth transition logic is triggered, and the control parameters are adjusted without sudden changes through the smooth transition parameter adjustment formula to make the adaptation decision.

[0011] Furthermore, in step S4, if the voltage change discriminant is used... If a cross-range switching of the input voltage is detected, a smooth transition logic is triggered, wherein... For the current moment, For the previous sampling time, The threshold voltage fluctuation at the interval boundary is determined by the smooth transition parameter adjustment formula. The control parameters were adjusted without sudden changes, among which for Smooth transition control parameters at different times The weighting coefficients and linearly change over time This is the set of control parameters for adjacent target intervals.

[0012] Furthermore, in step S4, the adaptive impedance matching control parameter set includes the switching frequency, duty cycle, and modulation depth of the power factor correction circuit.

[0013] Furthermore, in step S5, if the input voltage does not switch across intervals, the control command is directly transmitted to the power factor correction circuit control terminal of the ACDC power supply main circuit; if the input voltage switches across intervals, the control parameters are adjusted without sudden changes based on the smooth switching mapping relationship between intervals to generate a smooth transition control command.

[0014] Furthermore, in step S6, based on the switching frequency, duty cycle, and modulation depth parameters contained in the instruction, and combined with the real-time operating conditions such as the current grid input voltage, input impedance, and output power, the operating state is dynamically adjusted. Real-time and accurate matching of the ACDC power supply input impedance with the grid side is achieved through parameter adaptation of the power factor correction circuit. If a cross-range smooth transition control instruction is received, the operating parameters of the power factor correction circuit are gradually optimized according to the gradient adjustment logic. Simultaneously, based on the completed impedance matching adjustment, the power supply output voltage and output current are monitored in real time, and closed-loop feedback adjustment is performed against the rated parameters of the electrical equipment. By fine-tuning the relevant parameters of the inverter circuit and filter circuit in the main circuit, the output parameters are stably matched to the needs of the electrical equipment, achieving adaptive control of the ACDC power supply.

[0015] On the other hand, an adaptive ACDC power control system based on neural networks, the system comprising: The operating condition acquisition module is used to collect multi-dimensional operating condition information during the operation of the ACDC power supply in real time to obtain the original operating condition information set. The voltage range division module is used to divide the wide input voltage range of the ACDC power supply into multiple continuous and non-overlapping voltage characteristic ranges. The preprocessing module is used to preprocess the original working condition information set to obtain a standard working condition information set; The neural network processing module has a pre-trained segmented adaptive impedance matching neural network model built in, which is used to receive standard operating condition information set, identify the voltage characteristic range to which the input voltage belongs in real time, and obtain the adaptive impedance matching control parameter set through inference calculation, and output smooth transition control parameters when the input voltage switches across ranges. The control signal generation module is used to generate regular control commands based on an adaptive impedance matching control parameter set, or to generate smooth transition control commands without abrupt changes based on smooth transition control parameters. The ACDC power supply main circuit module is used to receive control commands and adjust its own operating parameters to achieve real-time matching of input impedance with the power grid and smooth transition of interval switching. The model update module is used to collect new operating condition data and impedance matching control effect data during the operation of the ACDC power supply, and transmit them as incremental samples to the neural network processing module to complete the online fine-tuning training and real-time update of the model.

[0016] Compared with existing technologies, this neural network-based adaptive AC-DC power control method and system have the following advantages: This invention abandons the traditional fixed segmented impedance matching framework by dividing voltage characteristic intervals and constructing a nonlinear correlation model within each interval and a smooth switching mapping relationship between intervals. It can dynamically generate impedance matching control parameters according to real-time operating conditions, realizing adaptive impedance matching and initial power supply control in each interval. At the same time, it completes a smooth transition of control parameters across intervals without sudden changes, avoiding circuit oscillations and ensuring the continuity of impedance matching and the stability of power supply operation across intervals. With a power factor correction circuit as the core adjustment component, combined with neural network dynamic adjustment and output closed-loop adjustment, it achieves real-time and accurate matching of input impedance with the grid side, meeting the output requirements of electrical equipment and completing the overall adaptive control of the power supply. It also supports online incremental training and updating of the neural network model to adapt to changes in long-term operating conditions of the power supply, continuously ensuring impedance matching accuracy and adaptive control effect. It organically combines wide-voltage segmented adaptive impedance matching control with overall dynamic control of the power supply, breaking through the traditional design concept, realizing dual protection on the grid side and the electrical equipment side, significantly improving the overall operating performance of ACDC power supplies under wide grid voltage fluctuation scenarios, and improving energy conversion efficiency and power supply operation stability across the entire input voltage range.

[0017] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0019] Figure 1 The flowchart shows the adaptive AC-DC power supply control method based on neural networks. Figure 2 This is a block diagram of an adaptive ACDC power control system based on a neural network. Figure 3 A flowchart illustrating the construction of a neural network model for an adaptive AC-CDC power supply control method based on neural networks. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0021] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an adaptive ACDC power supply control method and system based on neural networks. This method deeply integrates neural network technology with wide-voltage impedance matching control and overall dynamic control of the ACDC power supply. By dividing voltage characteristic intervals and constructing nonlinear correlation models of input impedance, grid voltage, and output power within each interval using neural networks, it achieves adaptive impedance matching by dynamically generating impedance matching control parameters based on real-time operating conditions within each interval. It establishes a smooth switching mapping relationship for control parameters between adjacent voltage characteristic intervals, completing abrupt adjustments of control parameters when the input voltage switches between intervals, ensuring the continuity of impedance matching across intervals and the stability of power supply operation. Using a power factor correction circuit as the core regulating component, its parameters are dynamically adjusted through neural networks to achieve real-time and accurate matching between the input impedance and the grid side, and combined with output closed-loop regulation to meet the output requirements of electrical equipment, thus completing the overall adaptive control of the ACDC power supply.

[0022] First, preliminary testing and calibration work is carried out to clarify the upper and lower limits of the rated wide input voltage of the AC / DC power supply to be controlled. Assuming that the rated wide input voltage range of the power supply is 85V-265V, a special test platform is built based on this. The test platform must be able to simulate different grid voltage fluctuation scenarios and cover all typical operating conditions within the rated input voltage range. At the same time, a variable load adapted to the power supply output is configured to simulate different output power demand scenarios. In the test platform, an impedance detection element and a high-precision voltage and current sensor are connected in series on the grid input side, and voltage and current sensors are connected in parallel on the power supply output side. A temperature acquisition element is placed in the core circuit of the power supply. The sampling frequency of all acquisition elements is set to a value that matches the power supply switching frequency to ensure accurate capture of dynamic changes in operating parameters. The acquired data is stored in real time to the data processing terminal through a data transmission link.

[0023] The input voltage was gradually adjusted using a test platform, starting from the lower limit of 85V and increasing in 1V increments to the upper limit of 265V. After each voltage adjustment, the input impedance, output power, and energy conversion efficiency data were collected under this condition after 30 seconds of stable operation. Three sets of data were collected for each voltage level, and the average value was taken. Outliers were removed. The collected data were then processed, using a filtering algorithm to denoise the raw data, eliminating the influence of power grid interference and environmental electromagnetic interference. Finally, a data fitting algorithm was used to fit the relationship curve between input voltage and energy conversion efficiency, extracting the efficiency inflection point. The judgment criterion was... When the input voltage change causes the energy conversion efficiency to decrease by more than 5%, the corresponding voltage value is the impedance matching efficiency inflection point. Combining the continuous change gradient of the input voltage amplitude and the impedance adaptation characteristics of the power supply's internal circuit, the wide input voltage range of 85V-265V is divided into three continuous and non-overlapping voltage characteristic intervals according to the efficiency inflection point. The boundary voltage values ​​of each interval take into account both the normal fluctuation of the grid voltage and the power supply control response characteristics to ensure smooth connection between adjacent interval boundaries. The final intervals are, for example, 85V-120V, 120V-200V, and 200V-265V (the interval boundary voltages are efficiency inflection points and do not overlap).

[0024] The raw operating condition information set collected during the experiment was standardized: a mean filtering algorithm was used to denoise the signal and remove high-frequency interference components from the data; normalization was used to convert parameters of different dimensions to the [0,1] interval to ensure that the magnitude of the feature parameters input to the neural network was consistent; parameters strongly related to impedance matching (input voltage, input impedance, output power, circuit temperature) were selected through feature extraction algorithms, redundant parameters were eliminated, and a standard operating condition information set was formed. Subsequently, a piecewise adaptive impedance matching neural network model was constructed. The model adopted a multi-output structure, with the number of nodes in the input layer corresponding to 4 feature parameters, 3 hidden layers (the number of nodes in each layer was optimized and adjusted according to the sample size), and the number of nodes in the output layer corresponding to the power factor correction circuit. Three core control parameters (switching frequency, duty cycle, and modulation depth) are used. During model training, standard operating condition data within each voltage characteristic range are used as the training sample set, and the corresponding optimal control parameters are used as the label set. At the same time, operating condition data and smooth transition control parameters during cross-range switching are collected as supplementary training samples. During training, the deviation between the model output and the sample labels is calculated using a loss function. The internal connection weights of the model are adjusted through gradient descent algorithm. Iterative training continues until the loss function value is lower than a preset threshold, at which point the model converges, resulting in a pre-trained piecewise adaptive impedance matching neural network model. After training, the model is validated to ensure that the parameter inference error within each range does not exceed 3%, and that the parameter transition is smooth and without abrupt changes during cross-range switching.

[0025] The configured acquisition element is connected to the actual operating circuit of the ACDC power supply. The acquisition frequency of input voltage, impedance, output power, voltage, current, and core circuit temperature is kept consistent with the previous test. The real-time acquired data is transmitted to the preprocessing module to perform noise reduction, normalization, and feature extraction to form a real-time standard operating condition feature vector. ,in The input voltage of the power grid is collected in real time. For real-time input impedance, For real-time output power, This represents the real-time circuit temperature.

[0026] The real-time standard operating condition feature vector is input into the pre-trained neural network model, and the model determines the public interest through interval recognition. ( =3, meaning 3 voltage characteristic intervals. The feature matching degree function for each interval (calculated using the interval feature weights learned during model training) determines the voltage feature interval to which the current input voltage belongs in real time. For example, when the real-time input voltage is 110V, the model calculates that this voltage has the highest feature matching degree with the 85V-120V interval, and determines that it belongs to the interval k=1. Then, the nonlinear correlation inference model for this interval is called. (in For the first Interval neural network inference mapping function For the first The adaptive impedance matching control parameter set of the interval), and the fast inference output of the switching frequency, duty cycle, and modulation depth control parameter set adapted to the current operating conditions, if determined by voltage change discriminant... ( The setting is 2V, meaning that when the fluctuation of the input voltage at adjacent sampling times exceeds 2V, it is determined that a cross-range switching may occur. If a cross-range change in the input voltage is detected, such as a fluctuation from 118V to 122V, crossing the 120V range boundary, the smooth transition logic is immediately triggered, and the smooth transition parameter adjustment formula is used. Calculate real-time control parameters, where The weighting coefficient starts from 1 and gradually changes linearly to 0 over time, with the duration of the change depending on the rate of change of the input voltage. ( (The sampling time interval is set to 1ms in this embodiment) is adjusted accordingly; the faster the voltage change rate, the better. The shorter the transition time, but no less than 5ms, the better to ensure that the parameter transition is smooth and avoid circuit oscillation.

[0027] The control signal generation module converts the digital control parameters output by the neural network into electrical signal control commands that are compatible with the control port of the power factor correction circuit. The transmission method adopts wired or wireless communication, and the transmission delay is controlled within 2ms to ensure the real-time performance and stability of the control commands.

[0028] After receiving control commands, the ACDC power supply main circuit dynamically adjusts its operating status based on the control parameters in the commands and the current real-time operating conditions. For example, when the input voltage is low (85V-120V range), the duty cycle is appropriately increased and the switching frequency is optimized to ensure that the input impedance matches the grid side. When the input voltage is high (200V-265V range), the modulation depth is adjusted to suppress input current harmonics and ensure energy conversion efficiency. If a cross-range smooth transition control command is received, the power factor correction circuit gradually optimizes the parameters according to the gradient adjustment logic, adjusting the parameter value every 1ms, with the adjustment range not exceeding the rated parameters. To prevent circuit oscillations caused by parameter mutations, the power supply output voltage and current are monitored in real time at a frequency of 1ms / time. The monitored data are compared with the rated parameters of the electrical equipment (e.g., rated output voltage 24V, rated output current 5A). If the output voltage deviation exceeds ±0.5V or the output current deviation exceeds ±0.1A, the relevant parameters of the inverter circuit and filter circuit are immediately fine-tuned through closed-loop feedback adjustment logic. For example, the output voltage is stabilized by adjusting the switching on time of the inverter circuit, and the output ripple is suppressed by optimizing the impedance characteristics of the filter circuit, ensuring that the output parameters are stable and meet the needs of the electrical equipment.

[0029] During the long-term operation of the power supply, new operating condition data is collected every 24 hours, including operating data and control effect data under different input voltages, output power, circuit temperatures. The new data is used as incremental samples to fine-tune the neural network model in batches. The sample size of each batch is 10% of the previous training sample size. The nonlinear correlation model of each interval and the smooth switching mapping relationship between intervals are updated. During the fine-tuning process, the original core weights of the model are retained, and only the edge weights are optimized to avoid model overfitting and ensure that the model can continuously adapt to the changes in operating conditions of the power supply during long-term operation (such as changes in impedance characteristics caused by circuit aging), and maintain control accuracy and operational stability.

[0030] In this embodiment, each module of the system is built using common technologies, such as... Figure 2 As shown, it includes a working condition acquisition module, a voltage range division module, a preprocessing module, a neural network processing module, a control signal generation module, and an ACDC power supply main circuit module, specifically: The operating condition acquisition module consists of voltage sensors, current sensors, impedance sensing elements, temperature acquisition elements, and a data acquisition link. The arrangement of each acquisition element must meet the requirements of accurate acquisition and anti-interference: the voltage sensor and impedance sensing element on the power grid input side are connected in series in the input circuit, maintaining a close connection to the power input terminal to reduce line interference; the voltage sensor and current sensor on the power output side are connected in parallel in the output circuit, placed close to the load end to ensure that the acquired output parameters reflect the actual load requirements; the core circuit temperature acquisition element is attached to the surface of the power device and fixed with insulating and thermally conductive material to avoid short-circuit risks. The accuracy of each acquisition element must meet the control requirements: the voltage sensor accuracy is not lower than 0.5 class, the current sensor accuracy is not lower than 1 class, the impedance sensing element measurement error does not exceed 2%, and the temperature acquisition element measurement range covers -20℃ to 150℃ with an accuracy of not less than ±1℃. The data acquisition link uses shielded cables to reduce electromagnetic interference and transmits the acquired raw data to the preprocessing module in real time.

[0031] The voltage range division module is connected to the operating condition acquisition module and the preprocessing module via signal lines. It receives test data and real-time data transmitted from the operating condition acquisition module, and completes the efficiency inflection point location and voltage characteristic range division through the built-in data analysis algorithm. The division results are stored in the local cache and transmitted to the neural network processing module. The division logic can be flexibly adjusted according to the rated wide input voltage range of different power supplies. For example, for power supplies with a rated input voltage range of 100V-300V, the number and boundary values ​​of voltage characteristic ranges can be adjusted by increasing the test voltage range, optimizing the efficiency inflection point judgment criteria, and adapting to the impedance characteristics of different power supplies.

[0032] The preprocessing module has bidirectional signal connections with the operating condition acquisition module, voltage range division module, and neural network processing module. It has built-in signal denoising, normalization, and feature extraction processing logic, which can standardize the raw data transmitted by the operating condition acquisition module, remove abnormal data caused by power grid interference and environmental electromagnetic interference, unify and normalize parameters of different dimensions, and filter out effective feature parameters that are strongly correlated with impedance matching. After forming a standard operating condition information set, it is transmitted to the neural network processing module. At the same time, it can receive feedback signals from the neural network processing module and adjust the preprocessing logic according to the model inference requirements to ensure the data quality of the input model.

[0033] The neural network processing module has a built-in pre-trained piecewise adaptive impedance matching neural network model. It can receive standard operating condition information sets transmitted by the preprocessing module and interval division results transmitted by the voltage interval division module through the signal line. It can complete functions such as interval identification, parameter inference, and calculation of smooth transition parameters across intervals. It outputs control parameter sets to the control signal generation module and has model storage, recall, and fine-tuning functions. It can store multiple sets of model parameters adapted to different scenarios and switch to recall them according to the actual application scenario. At the same time, it can receive incremental samples transmitted by the model update module to complete online fine-tuning and updating of the model. The updated model parameters automatically overwrite the original parameters without stopping and restarting, and do not affect the normal operation of the power supply.

[0034] The control signal generation module is connected to the neural network processing module and the ACDC power supply main circuit module. It converts digital control parameters into electrical signal control commands. The output electrical signal must be compatible with the control port of the power factor correction circuit (such as analog 0-10V or digital signal). It also has signal amplification and isolation functions to prevent electromagnetic interference from the main circuit from affecting the transmission of control signals. It can automatically adjust the command generation logic according to the parameter type output by the neural network (conventional control parameters or smooth transition control parameters) to ensure that the command matches the control requirements of the power factor correction circuit.

[0035] The main circuit module of the ACDC power supply consists of a power factor correction circuit, a rectifier circuit, an inverter circuit, and a filter circuit connected in sequence. The parameters of each circuit must be designed to adapt to the rated power and wide input voltage range of the power supply: The power factor correction circuit adopts a Boost topology, which has wide voltage input adaptability and can adjust the switching frequency, duty cycle, and modulation depth according to control commands to achieve adaptive adjustment of input impedance; the rectifier circuit adopts an uncontrolled rectifier topology to complete the initial conversion of AC to DC; the inverter circuit adopts a full-bridge topology, which can stabilize the output voltage by adjusting the switching on time; the filter circuit adopts an LC filter structure to suppress output ripple and ensure stable output voltage and current. The connections between the circuits use copper busbars or thick wires to reduce line losses, and overvoltage, overcurrent, and overheat protection functions are configured to ensure safe circuit operation.

[0036] The model update module is signal-connected to the neural network processing module and the ACDC power supply main circuit module. It can collect new operating condition data and control effect data of the power supply in real time. The data acquisition frequency is consistent with that of the operating condition acquisition module. The collected data is stored in batches, and incremental samples are transmitted to the neural network processing module every 24 hours to trigger the model fine-tuning and update process. This module has a data filtering function to remove abnormal operating condition data, ensuring the validity of incremental samples. It can also record model update logs for easy traceability and optimization in the future.

[0037] In this embodiment, through the specific implementation of the above method and system, adaptive control of an ACDC power supply with a wide input voltage range is achieved. Actual testing has verified that within the wide input voltage range of 85V-265V, the energy conversion efficiency is improved by 3%-8% compared to the traditional fixed segmented control method. There is no significant oscillation in the circuit operation when switching between ranges, the output voltage deviation is controlled within ±0.5V, and the output current ripple does not exceed 5%. After the model is updated online, the control accuracy of the power supply does not decrease significantly during long-term operation (1000 hours of continuous operation). It can effectively adapt to the scenarios of grid voltage fluctuations and load changes, meet the high-efficiency and stable operation requirements of a wide-voltage input ACDC power supply, and verify the feasibility and superiority of the technical solution of this invention.

[0038] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An adaptive AC-DC power supply control method based on neural networks, characterized in that, The method includes: S1: Real-time collection of multi-dimensional operating condition information during the operation of ACDC power supply to obtain raw operating condition information set; S2: Divide the wide input voltage range of the ACDC power supply to obtain multiple continuous and non-overlapping voltage characteristic intervals, and preprocess the original operating condition information set to obtain a standard operating condition information set that meets the input requirements of the neural network model. S3: Construct a neural network model. Through the neural network model, construct nonlinear correlation models of input impedance, grid voltage, and output power in each voltage characteristic interval. At the same time, establish a smooth switching mapping relationship of control parameters between adjacent voltage characteristic intervals. Complete the training and convergence of the neural network model to obtain a pre-trained piecewise adaptive impedance matching neural network model. S4: Input the standard operating condition information set into the pre-trained segmented adaptive impedance matching neural network model, and identify the voltage characteristic range to which the input voltage belongs in real time through model inference calculation, and obtain the adaptive impedance matching control parameter set that matches the current operating condition. S5: Based on the adaptive impedance matching control parameter set, a corresponding electrical signal form of control command is generated and transmitted to the power factor correction circuit control terminal; S6: The main circuit of the ACDC power supply adjusts the operating parameters of the power factor correction circuit according to the received control commands, realizes real-time matching between the input impedance and the grid side, and completes the adaptive control of the ACDC power supply.

2. The adaptive AC-DC power supply control method based on neural networks according to claim 1, characterized in that, In step S1, the multi-dimensional operating condition information includes at least one of the following: grid-side input voltage information, input impedance information, output power information, and circuit operating temperature information.

3. The adaptive AC-DC power supply control method based on neural networks according to claim 1, characterized in that, In step S2, the specific steps for dividing the wide input voltage range of the ACDC power supply are as follows: Define the upper and lower limits of the rated wide input voltage of the ACDC power supply; simulate various grid voltage input scenarios by building a test platform; collect data on the input impedance, output power, and energy conversion efficiency of the power supply under different input voltage conditions in real time; after data fitting, feature extraction, and effectiveness analysis, locate the impedance matching efficiency inflection point; and combine the continuous change gradient of the input voltage amplitude with the impedance adaptation characteristics of the power supply's internal circuitry. Based on the efficiency inflection point and the principle of continuous voltage amplitude transition, the wide input voltage range is decomposed into multiple continuous and non-overlapping voltage characteristic intervals.

4. The adaptive AC-DC power supply control method based on neural networks according to claim 1, characterized in that, In step S3, the specific technical solution for constructing the neural network model is as follows: Collect sample data from the full-width input voltage range of the ACDC power supply, including full-condition sample sets within each voltage characteristic interval and operating condition sample sets during cross-interval switching. Simultaneously collect corresponding tag data, namely, the optimal impedance matching control parameter sample set within each interval and the smooth transition control parameter sample set during cross-interval switching. The control parameter samples are the switching frequency, duty cycle, and modulation depth parameters of the power factor correction circuit. Using the operating condition sample set of each voltage characteristic interval as input and the corresponding impedance matching control parameter sample set as output, the nonlinear correlation model in each interval is trained, so that the model learns the nonlinear correlation relationship between input impedance, grid voltage and output power in each interval. Using cross-interval switching condition samples as input and corresponding smooth transition control parameter sample sets as output, the smooth switching mapping relationship between intervals is trained, enabling the model to learn the non-abrupt adjustment law of control parameters when the input voltage crosses intervals. The deviation between the model output and the actual sample label is calculated using a loss function. Based on this deviation, the connection weights inside the model are adjusted. The model is iterated repeatedly until the output meets the preset convergence condition, thus obtaining a pre-trained piecewise adaptive impedance matching neural network model.

5. The adaptive AC-DC power supply control method based on neural networks according to claim 1, characterized in that, In step S4, the preprocessed standard operating condition information set is constructed into a multi-dimensional feature vector, which is then input into the pre-trained piecewise adaptive impedance matching neural network model in real time, and the feature vector is determined by the interval identification discriminant formula. Complete the accurate identification of the voltage characteristic range to which the current input voltage belongs, among which, This represents the total number of voltage characteristic intervals. For the first Feature matching degree function for each voltage feature interval The current input voltage range is assigned a number, and then a nonlinear correlation inference model is called to perform inference calculations, outputting a set of control parameters that are adapted to the current real-time operating conditions. If a cross-range switching of the input voltage is detected by the voltage change discriminant, the smooth transition logic is triggered, and the control parameters are adjusted without sudden changes through the smooth transition parameter adjustment formula to make the adaptation decision.

6. The adaptive AC-DC power supply control method based on neural networks according to claim 1, characterized in that, In step S4, if the voltage change discriminant is used... If a cross-range switching of the input voltage is detected, a smooth transition logic is triggered, wherein... For the current moment, For the previous sampling time, The threshold voltage fluctuation at the interval boundary is determined by the smooth transition parameter adjustment formula. The control parameters were adjusted without sudden changes, among which for Smooth transition control parameters at different times The weighting coefficients and linearly change over time This is the set of control parameters for adjacent target intervals.

7. The adaptive AC-DC power supply control method based on neural networks according to claim 1, characterized in that, In step S4, the adaptive impedance matching control parameter set includes the switching frequency, duty cycle, and modulation depth of the power factor correction circuit.

8. The adaptive AC-DC power supply control method based on neural networks according to claim 1, characterized in that, In step S5, if the input voltage does not switch across intervals, the control command is directly transmitted to the power factor correction circuit control terminal of the ACDC power supply main circuit; if the input voltage switches across intervals, the control parameters are adjusted without sudden changes based on the smooth switching mapping relationship between intervals to generate a smooth transition control command.

9. The adaptive AC-DC power supply control method based on neural networks according to claim 1, characterized in that, In step S6, based on the switching frequency, duty cycle, and modulation depth parameters contained in the instruction, and combined with the real-time operating conditions such as the current grid input voltage, input impedance, and output power, the operating state is dynamically adjusted. Real-time and accurate matching between the AC / CDC power supply input impedance and the grid side is achieved through parameter adaptation of the power factor correction circuit. If a cross-range smooth transition control instruction is received, the operating parameters of the power factor correction circuit are gradually optimized according to the gradient adjustment logic. Simultaneously, based on the completed impedance matching adjustment, the power supply output voltage and output current are monitored in real time, and closed-loop feedback adjustment is performed against the rated parameters of the electrical equipment. By fine-tuning the relevant parameters of the inverter circuit and filter circuit in the main circuit, the output parameters are stably matched to the needs of the electrical equipment, achieving adaptive control of the AC / CDC power supply.

10. A neural network-based adaptive AC-DC power control system, the system being applicable to the neural network-based adaptive AC-DC power control method according to any one of claims 1-9, characterized in that, The system includes: The operating condition acquisition module is used to collect multi-dimensional operating condition information during the operation of the ACDC power supply in real time to obtain the original operating condition information set. The voltage range division module is used to divide the wide input voltage range of the ACDC power supply into multiple continuous and non-overlapping voltage characteristic ranges. The preprocessing module is used to preprocess the original working condition information set to obtain a standard working condition information set; The neural network processing module has a pre-trained segmented adaptive impedance matching neural network model built in, which is used to receive standard operating condition information set, identify the voltage characteristic range to which the input voltage belongs in real time, and obtain the adaptive impedance matching control parameter set through inference calculation, and output smooth transition control parameters when the input voltage switches across ranges. The control signal generation module is used to generate regular control commands based on an adaptive impedance matching control parameter set, or to generate smooth transition control commands without abrupt changes based on smooth transition control parameters. The ACDC power supply main circuit module is used to receive control commands and adjust its own operating parameters to achieve real-time matching of input impedance with the power grid and smooth transition of interval switching. The model update module is used to collect new operating condition data and impedance matching control effect data during the operation of the ACDC power supply, and transmit them as incremental samples to the neural network processing module to complete the online fine-tuning training and real-time update of the model.

Images