Epitaxial antenna adjustment method and device, equipment, storage medium and program product

Abstract

The invention discloses an epitaxial antenna adjustment method and device, equipment, a storage medium and a program product. The method comprises the steps of obtaining perception data of a target object; inputting the sensing data and the historical adjustment data into a trained hybrid enhancement model to predict a target motion track of the target object, predicting adjustment parameters of the epitaxial antenna according to the target motion track of the target object, and outputting a target adjustment angle and a target power value of the epitaxial antenna; and adjusting the epitaxial antenna based on the target adjustment angle and the target power value. The defect that a traditional gateway static antenna design cannot adapt to dynamic scenes such as personnel movement and furniture shielding is overcome, and real-time matching of signal coverage and user positions is achieved. According to the embodiment of the invention, through a technical chain of dynamic perception-intelligent decision-accurate execution, the core defects of static design, passive optimization, resource waste, function splitting and the like in the prior art are solved, and the technical effect of improving the adaptive ability and user experience in a wireless network is achieved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of terminal control of home gateways, and particularly relates to an extension antenna adjustment method and device, equipment, a storage medium and a program product. BACKGROUND

[0002] The existing home gateway and wireless network adopt a static antenna design, which cannot adjust the radiation direction according to dynamic environmental changes such as personnel movement and furniture shielding, resulting in coverage blind spots and multipath interference. At the same time, due to the passive optimization mechanism, the parameters are adjusted depending on the preset channel or simple rules, and there is a lack of real-time perception and intelligent decision-making for complex scenes. In view of the fragmentation of gateway perception functions, the environmental perception, signal optimization and hardware control modules are independent of each other, and it is difficult to form a closed-loop system, which greatly affects the user experience of the related home gateway. SUMMARY

[0003] The application embodiment provides an extension antenna adjustment method, device, equipment, storage medium and program product, which can break through the defects of the traditional gateway static antenna design that cannot adapt to dynamic scenes such as personnel movement and furniture shielding, realize real-time matching of signal coverage and user position, and achieve the technical effects of improving the adaptive ability and user experience in the wireless network through the technical chain of dynamic perception-intelligent decision-making-precise execution.

[0004] In a first aspect, the application embodiment provides an extension antenna adjustment method based on a hybrid reinforcement model, applied to a home gateway, including: acquiring perception data of a detected target object, the perception data including position data and time-frequency power data; inputting the perception data and historical adjustment data into a trained hybrid reinforcement model to predict a target motion trajectory of the target object, and predicting adjustment parameters of the extension antenna according to the target motion trajectory of the target object, outputting a target adjustment angle and a target power value of the extension antenna, the historical adjustment data representing historical perception data adjusted by the hybrid reinforcement model; adjusting the extension antenna based on the target adjustment angle and the target power value.

[0005] In a possible implementation, the acquiring of the perception data of the target object includes: scanning a target environment by a target perception module at a set time interval to obtain time domain signal data; performing fast Fourier transform processing on the time domain signal data to obtain time-frequency power data; identifying direct signals and reflection paths corresponding to the target object by a multipath component separation algorithm to obtain multipath data; performing positioning processing on the multipath data to obtain position data; and determining the perception data based on the position data and the time-frequency power data.

[0006] In a possible implementation, the fast Fourier transform processing on the time domain signal data obtains time-frequency power data, including: performing fast Fourier transform processing on the time domain signal data to obtain frequency domain signal data; extracting a channel impulse response result of the frequency domain signal data to obtain a channel frequency point power value, and taking the channel frequency point power value as the time-frequency power data.

[0007] In a possible implementation, the positioning processing on the multipath data obtains position data, including: performing fusion positioning processing on the multipath data based on a parallax positioning algorithm and an angle of arrival estimation algorithm to obtain a target azimuth angle of the extended antenna and distance data corresponding to the target object; and generating an environmental spatial feature map by splicing fusion based on the target azimuth angle and the distance data, and taking the environmental spatial feature map as the position data.

[0008] In a possible implementation, the hybrid reinforcement model is obtained by: performing feature extraction on input perception training data through a convolutional neural network layer to obtain spatial training features; extracting motion trajectory data of a training object in position data of the perception training data through a long short-term memory network layer; predicting a target motion trajectory of the training object based on the spatial training features and the motion trajectory data through a hybrid reinforcement training model; predicting initial adjustment parameters of the extended antenna based on the target motion trajectory of the training object to obtain an initial adjustment angle and an initial power value of the extended antenna; constructing a reward function corresponding to the initial adjustment angle and the initial power value, the reward function being used to represent deviations of an adjustment angle and an adjustment power of the extended antenna; and performing optimization processing on network parameters contained in the reward function to obtain a trained hybrid reinforcement model.

[0009] In a possible implementation, the reward function includes a throughput parameter, a signal strength parameter, and a multipath interference parameter; and the optimization processing on the network parameters contained in the reward function to obtain the trained hybrid reinforcement model includes: adjusting parameter weight values corresponding to the throughput parameter, the signal strength parameter, and the multipath interference parameter by using a proximal policy optimization algorithm; updating the hybrid reinforcement training model based on the adjusted parameter weight values; comparing a dominance relationship of the throughput parameter, the signal strength parameter, and the multipath interference parameter based on the updated hybrid reinforcement training model by using a non-dominated sorting genetic algorithm to obtain corresponding non-dominated front data; and performing sorting based on the non-dominated front data to obtain a target solution set and the trained hybrid reinforcement model.

[0010] In a possible implementation, the reward function R_t is obtained by the following equation: Wherein, a represents a throughput parameter weight value, r throughput represents a throughput parameter, r signal represents a signal strength parameter, β represents a signal strength parameter weight value, r multipath represents a multipath interference parameter, and γ represents a multipath interference parameter weight value.

[0011] In a possible implementation, the adjusting the extended antenna based on the target adjustment angle and the target power value comprises: adjusting the extended antenna according to a biaxial adjustment by driving a motor and a target reducer, so that the extended antenna is adjusted to the target adjustment angle; and adjusting the power of the extended antenna to the power value by a power amplifier.

[0012] In a possible implementation, after the adjusting the extended antenna based on the target adjustment angle and the target power value, the method further comprises: calculating a throughput gain of a throughput parameter corresponding to the target adjustment angle and the target power value; feeding back the throughput gain to the hybrid reinforcement model in a case where the throughput gain is lower than a set gain threshold; and updating the hybrid reinforcement model by updating a model parameter based on the throughput gain.

[0013] In a second aspect, an embodiment of the present application provides an extended antenna adjustment device based on a hybrid reinforcement model, comprising: a perception module configured to acquire perception data of a target object detected by a target perception module, the perception data comprising position data and time-frequency power data; an AI model processing module configured to input the perception data and historical adjustment data into a trained hybrid reinforcement model to predict a target motion trajectory of the target object, and to predict an adjustment parameter of the extended antenna according to the target motion trajectory of the target object, and to output a target adjustment angle and a target power value of the extended antenna, the historical adjustment data representing historical perception data adjusted by the hybrid reinforcement model; and a hardware adjustment module configured to adjust the extended antenna based on the target adjustment angle and the target power value.

[0014] In a third aspect, an embodiment of the present application provides an extended antenna adjustment device based on a hybrid reinforcement model, comprising: a processor and a memory storing computer program instructions; and the processor executes the computer program instructions to implement the extended antenna adjustment method based on a hybrid reinforcement model according to any one of the first aspect.

[0015] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, the computer readable storage medium storing computer program instructions, and the computer program instructions are executed by a processor to implement the extended antenna adjustment method based on a hybrid reinforcement model according to any one of the first aspect.

[0016] In a fifth aspect, an embodiment of the present application provides a computer program product, instructions in the computer program product being executed by a processor of an electronic device to cause the electronic device to perform the method for adjusting an extension antenna based on a hybrid reinforcement model according to any one of the first aspect.

[0017] The method, device, equipment, storage medium and program product for adjusting an extension antenna provided by the embodiments of the present application can collect perception data of a target object, eliminate a coverage blind area and multipath interference caused by a static antenna, predict a target action trajectory of the target object and an adjustment parameter of the extension antenna by using a trained hybrid reinforcement model, and output a target adjustment angle and a target power value of the extension antenna, thereby improving the accuracy of adjusting the angle and power of the extension antenna, achieving real-time observation and dynamic adjustment of the target object, and breaking through the defects of a traditional gateway static antenna design that cannot adapt to dynamic scenes such as personnel movement and furniture shielding, achieving real-time matching of signal coverage and user position, and achieving the technical effects of improving the adaptive capability and user experience in a wireless network. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required to be used in the embodiments of the present application will be briefly introduced as follows, and other drawings can be obtained by those of ordinary skill in the art without creative labor on the basis of these drawings.

[0019] Figure 1 is a flowchart of a method for adjusting an extension antenna based on a hybrid reinforcement model provided by an embodiment of the present application; Figure 2 is a flowchart of a method for determining perception data provided by an embodiment of the present application; Figure 3 is a flowchart of another method for determining perception data provided by an embodiment of the present application; Figure 4 is a flowchart of still another method for determining perception data provided by an embodiment of the present application; Figure 5 is a flowchart of a method for adjusting an extension antenna provided by an embodiment of the present application; Figure 6 is a flowchart of a training process of a hybrid reinforcement model provided by an embodiment of the present application; Figure 7 is a flowchart of another training process of a hybrid reinforcement model provided by an embodiment of the present application; Figure 8 is a flowchart of still another training process of a hybrid reinforcement model provided by an embodiment of the present application; Figure 9is a flowchart of another training process of a hybrid reinforcement model provided by an embodiment of the present application; Figure 10 is a flowchart of another epitaxial antenna adjustment method based on a hybrid reinforcement model provided by an embodiment of the present application; Figure 11 is a structural diagram of an epitaxial antenna adjustment device based on a hybrid reinforcement model provided by an embodiment of the present application; Figure 12 is a structural diagram of an epitaxial antenna adjustment device based on a hybrid reinforcement model provided by an embodiment of the present application. DETAILED DESCRIPTION

[0020] The features and exemplary embodiments of various aspects of the present application will be described in detail below with reference to the accompanying drawings and specific embodiments. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, but not to limit the present application. The present application can be implemented without some of these specific details by those skilled in the art. The following description of the embodiments is only to provide a better understanding of the present application by showing examples of the present application.

[0021] It should be noted that, in this paper, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the elements defined by the statement "include" do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0022] It should be noted that the acquisition, storage, use and processing of data in the embodiments of the present application comply with the relevant provisions of national laws and regulations.

[0023] It should be noted that in the embodiments of the present application, some industry existing schemes of software, components, models, etc. may be mentioned, which should be considered as exemplary, and the purpose is only to illustrate the feasibility of the implementation of the technical solutions of the present application, but does not mean that the applicant has or will necessarily use the scheme.

[0024] An AI model is a mathematical model system that simulates human intelligent behavior through computer algorithms and data training. Its core lies in using machine learning techniques to automatically learn rules from data and complete specific tasks. AI models are based on neural network architectures such as Transformers, which process input data by adjusting parameters and weights to achieve tasks such as classification, prediction, or generation. For example, large language models (LLMs) use self-attention mechanisms to capture long-range dependencies in text, enabling natural language understanding and generation capabilities.

[0025] Proximal Policy Optimization (PPO) is a reinforcement learning algorithm proposed by OpenAI in 2017, belonging to the policy gradient method family. It has become a mainstream algorithm in the field of reinforcement learning due to its efficiency, stability, and ease of implementation. Its core is to limit the amplitude of policy updates to avoid drastic fluctuations during training, while using importance sampling to improve sample efficiency. By clipping the objective function (PPO-Clip) or adding a penalty term (PPO-Penalty) to limit the difference between the new and old policies, it prevents unstable updates caused by excessive updates. Using importance sampling techniques, historical data is efficiently utilized.

[0026] NSGA-II (Non-dominated Sorting Genetic Algorithm II) is an efficient multi-objective optimization algorithm designed to solve complex problems involving multiple conflicting objectives. Its core simulates the natural selection process to generate a set of non-dominated solutions. It implements non-dominated sorting to quickly stratify the population and prioritize individuals that are not dominated by other solutions. It also implements crowding distance to measure the distribution density of solutions in the objective space, ensuring the diversity of the Pareto front. It has an elitist strategy that preserves excellent individuals for the next generation, avoiding information loss caused by genetic operations.

[0027] Current home gateway antennas mostly use fixed or manually adjustable structures, with fixed polarization direction and radiation angle, unable to dynamically adjust according to user position changes or furniture obstructions. When users move to signal blind areas, fixed antennas cannot eliminate multipath interference through polarization switching or angle compensation, leading to signal attenuation and throughput reduction. In addition, although omnidirectional antennas can cover 360°, signal strength distribution is uneven, easily forming "apple-shaped" coverage blind areas. When switching to idle channels based on historical data, it cannot respond to dynamic scenarios such as personnel movement and furniture layout changes in real time. When obstacles suddenly block signal paths, traditional gateways need several seconds to several tens of seconds to re-scan and adjust parameters, causing user experience interruption.

[0028] Current MIMO antennas improve capacity through spatial diversity, but require multiple antennas to work independently, occupy a large physical space, and the correlation between antennas is difficult to control. For example, signal superposition at a specific angle of co-polarized antennas can cause interference and reduce spectral efficiency. Current smart home systems can only sense the presence of devices and cannot deeply combine environmental data (such as human position and obstacle distribution) with wireless signal optimization. When a user is sitting, a traditional gateway cannot identify the user's position and optimize signal coverage accordingly.

[0029] Existing AI-driven wireless optimization solutions rely on cloud computing, resulting in high decision-making delay (seconds) and inability to adapt to local privacy protection needs.

[0030] To solve the problems in the prior art, the embodiments of the present application provide an extension antenna adjustment method, device, equipment, storage medium and program product. By collecting the sensing data of the target object, the coverage blind area and multipath interference caused by the static antenna are eliminated, the target action trajectory of the target object is predicted by the trained hybrid reinforcement model, and the adjustment parameters of the extension antenna are predicted. By outputting the target adjustment angle and target power value of the extension antenna, the accuracy of adjusting the angle and power of the extension antenna is improved, real-time observation and dynamic adjustment of the target object are realized. Through the technical chain of dynamic perception-intelligent decision-making-precise execution, the defects of the traditional gateway static antenna design that cannot adapt to dynamic scenes such as personnel movement and furniture obstruction are broken through, real-time matching of signal coverage and user position is realized, and the technical effects of improving the adaptive ability and user experience in the wireless network are achieved. First, the extension antenna adjustment method based on the hybrid reinforcement model provided by the embodiments of the present application is introduced.

[0031] Figure 1 is a flowchart of an extension antenna adjustment method based on a hybrid reinforcement model provided by the embodiments of the present application. According to the provided diagram, the steps of the extension antenna adjustment method based on the hybrid reinforcement model specifically include S101-S103. Figure 1

[0032] S101, acquire the sensing data of the detected target object, and the sensing data includes position data and time-frequency power data.

[0033] The present application is applied in the environment of a home gateway. The sensing data mentioned here can be understood as the position and moving speed information of the moving target collected by Wi-Fi. The target object mentioned here can be understood as a moving object in the home environment, a moving person.

[0034] Further, the sensing data of the moving target in the home environment is collected by an image collection device such as a camera and Wi-Fi sensing control, which is used as reference data for judging the moving direction and moving speed of the moving target in the next step. ​

[0035] S102, input the perception data and the historical adjustment data into the trained hybrid reinforcement model to predict the target motion trajectory of the target object, and predict the adjustment parameter of the epitaxial antenna according to the target motion trajectory of the target object, output the target adjustment angle and the target power value of the epitaxial antenna, and the historical adjustment data represents the historical perception data adjusted by the hybrid reinforcement model.

[0036] The hybrid reinforcement model mentioned here can be understood as an AI model for predicting environmental change trends, completing antenna angle optimization in milliseconds, and realizing real-time signal compensation prediction in dynamic scenarios.

[0037] The adjustment parameter mentioned here can be understood as the movement adjustment angle and the movement adjustment power of the epitaxial antenna.

[0038] Further, the collected perception data and the previously processed perception data are input into the hybrid reinforcement model in the AI model, and the hybrid reinforcement model is used to predict the motion change trend and the motion power of the target object. The movement trajectory of the target object is calculated to correspond to the movement adjustment angle and the movement adjustment power of the camera as the output result.

[0039] S103, adjust the epitaxial antenna based on the target adjustment angle and the target power value.

[0040] On the premise that the movement angle and the movement power of the target object are calculated by the corresponding hybrid reinforcement model of the AI model, the tracking adjustment angle and the movement power of the camera are calculated, and the corresponding angle and power are executed through the hardware adjustment structure, so as to achieve real-time accurate tracking and improve the adjustment accuracy.

[0041] The embodiment of the application provides an epitaxial antenna adjustment method, which collects perception data of a target object, eliminates coverage blind area and multipath interference caused by a static antenna, predicts a target action trajectory of the target object through a trained hybrid reinforcement model, and predicts an adjustment parameter of the epitaxial antenna. The accuracy of adjusting the angle and power of the epitaxial antenna is improved by outputting the target adjustment angle and the target power value of the epitaxial antenna, and real-time observation and dynamic adjustment of the target object are realized. Through the technical chain of dynamic perception-intelligent decision-accurate execution, the defects of traditional gateway static antenna design that cannot adapt to dynamic scenarios such as personnel movement and furniture shielding are overcome, real-time matching of signal coverage and user position is realized, and the technical effects of improving the adaptive ability and user experience in the wireless network are achieved. First, the epitaxial antenna adjustment method based on the hybrid reinforcement model provided by the embodiment of the application is introduced.

[0042] For the process of determining the perception data in step S101, Figure 2 It is a flowchart for determining perception data provided by the embodiment of the application.Figure 2 is introduced on the basis of the above embodiment. According to Figure 2 The provided diagram, the step of determining the perception data includes: S201-S205.

[0043] S201, the target environment is scanned by the target perception module according to the set time interval, and the time domain signal data is obtained.

[0044] S202, the time domain signal data is processed by fast Fourier transform to obtain time-frequency power data.

[0045] The time interval mentioned here can be understood as a scanning period, for example, 100ms-1s is a period.

[0046] The integrated Wi-Fi perception module containing orthogonal frequency division multiple access (OFDMA) modulation technology transmits a probe sequence according to the set period time to obtain time domain data. The channel impulse response is extracted by fast Fourier transform (i.e. FFT transform) on the time domain data. The obtained channel impulse response is converted into time-frequency power data.

[0047] S203, the direct signal and the reflection path corresponding to the target object are identified by the multipath component separation algorithm to obtain multipath data.

[0048] S204, the position data is obtained by positioning the multipath data.

[0049] The time-frequency power data is obtained by transmitting a probe signal through multiple-input multiple-output (MIMO) technology, and the multipath reflection signal corresponding to the direct signal and the reflection path is received by the multipath component separation algorithm, and the path delay and attenuation coefficient are calculated as the multipath data. Based on the fusion positioning of time difference of arrival (TDOA) and angle of arrival (AOA), the phase difference of three groups of extension antenna arrays is used to calculate the azimuth angle of the device, and the RSSI value is combined to estimate the distance, to obtain the position data after positioning processing.

[0050] S205, the perception data is determined based on the position data and the time-frequency power data.

[0051] By taking the collected position data and time-frequency power data as real-time tracking perception data, generating a dynamic spatial feature map containing position data and time-frequency power data from the perception data, and taking the dynamic spatial feature map as input data of a hybrid reinforcement model in an AI model, the position of the target object and the distribution of obstacles are perceived in real time by the Wi-Fi perception module, accurate data is provided for the AI model to dynamically adjust the pointing direction of the extension antenna, and the coverage blind area and multipath interference caused by the static antenna are eliminated.

[0052] The method for adjusting the extension antenna based on the hybrid reinforcement model provided in the embodiments of the application breaks through the defects of the traditional gateway static antenna design that cannot adapt to dynamic scenes such as personnel movement and furniture shielding, and achieves the technical effect of real-time matching of signal coverage and user position.

[0053] For the process of obtaining the time-frequency power data in step S202, Figure 3 is another flowchart for determining perception data provided in the embodiments of the application. Figure 3 is introduced on the basis of the above embodiment. According to Figure 3 The determination step of the time-frequency power data in the perception data includes S301-S302, as shown in the provided diagram.

[0054] S301, the time-domain signal data is subjected to fast Fourier transform processing to obtain frequency-domain signal data.

[0055] S302, the channel impulse response result of the frequency-domain signal data is extracted to obtain a channel frequency point power value, and the channel frequency point power value is taken as the time-frequency power data.

[0056] The fast Fourier transform (FFT transform) is used to convert the time-domain signal into frequency-domain information data. The channel impulse response result corresponding to the extracted frequency-domain signal data is taken as the time-frequency power data, and the perception data is extracted and processed.

[0057] In a possible example scenario, the Wi-Fi perception module is used to scan at a period of 100 ms to generate an environment feature vector X_t containing position data and time-frequency power data; the environment feature vector X_t is normalized and denoised (wavelet transform threshold method), and effective features F_t are extracted as input data of the AI model.

[0058] The embodiment of the application breaks through the defect that the traditional gateway static antenna design cannot adapt to dynamic scenes such as personnel movement and furniture shielding, and realizes the technical effect of real-time matching of signal coverage and user position, by sensing technology, processing and calculating the moving position and moving power of the target object through a Wi-Fi sensing module to obtain corresponding time-frequency power data.

[0059] For the process of obtaining the position data in step S204, Figure 4 is another flowchart for determining sensing data provided by the embodiment of the application. Figure 3 is introduced on the basis of the above embodiment. According to Figure 3 The provided diagram shows that the determination step of the position data in the sensing data includes S401-S402.

[0060] S401, based on the parallax positioning algorithm and the angle of arrival estimation algorithm, the multipath data is fused and positioned to obtain the target azimuth angle of the epitaxial antenna and the distance data corresponding to the target object.

[0061] S402, based on the target azimuth angle and the distance data, the environment space feature map is generated by splicing fusion, and the environment space feature map is taken as the position data.

[0062] The direct signal and the multipath reflection signal corresponding to the reflection path are received by the multipath component separation algorithm, and the time delay and the attenuation coefficient of each path are calculated as multipath data. In the fusion positioning based on the time difference of arrival (TDOA) and the angle of arrival (AOA), the phase difference of the three groups of epitaxial antenna arrays is used to calculate the device azimuth angle, the RSSI value is combined to estimate the distance data, and the environment space feature map is generated by splicing fusion based on the target azimuth angle and the distance data, to obtain the position data after positioning processing.

[0063] The embodiment of the application breaks through the defect that the traditional gateway static antenna design cannot adapt to dynamic scenes such as personnel movement and furniture shielding, and realizes the technical effect of real-time matching of signal coverage and user position, by sensing technology, processing and calculating the moving position and moving power of the target object through a Wi-Fi sensing module to obtain corresponding position data.

[0064] For the adjustment process of the epitaxial antenna in step S103, Figure 5 is a flowchart of an epitaxial antenna adjustment method provided by the embodiment of the application. Figure 5 is introduced on the basis of the first embodiment. According to Figure 5 The provided diagram shows that the epitaxial antenna adjustment method includes S501-S502.

[0065] S501, adjust the epitaxial antenna according to the double-axis by driving the motor and the target reducer, so that the epitaxial antenna is adjusted to the target adjustment angle.

[0066] S502, adjust the power of the epitaxial antenna to the power value by the power amplifier.

[0067] For the hardware structure of the adjustment execution, a stepper motor (precision ±0.5°), a precision reducer and a dual-polarized epitaxial antenna array (horizontal / tilt double-axis adjustable) are contained. After receiving the adjustment instruction, the motor driving circuit controls the rotation speed through the algorithm, and the angle encoder feedbacks the position error in real time.

[0068] The stepper motor adopts CAN bus communication, the instruction frame format contains target angle, rotation speed and check code. The motor driving current closed loop control ensures smooth rotation. When it is detected that the angle fluctuation exceeds the adjustable set value ±1° for continuous multiple times (for example, 3 times), the secondary calibration is triggered (based on the Hall sensor detection motor rotor position). The power amplifier adopts dynamic voltage frequency adjustment, and enters sleep mode (power consumption <1mW) when idle.

[0069] The double-axis adjustment is realized by driving the motor and the target reducer to adjust the epitaxial antenna. The dual-polarized epitaxial antenna is driven by the stepper motor and the precision reducer (reduction ratio 1:100), which supports horizontal / tilt double-axis ±90° continuous rotation. And through the PID algorithm and the Hall sensor calibration, the angle error is less than 0.1°, and the feedback regulation is realized. Through the hardware support adjustment, the problem of traditional manual adjustment antenna unable to realize accurate dynamic pointing, reducing multipath interference and signal attenuation is solved.

[0070] Optionally, the application relates to an edge-cloud collaborative computing and privacy protection method. Through a distributed computing architecture, the edge end executes real-time decision (delay <50ms), the cloud end performs model training and parameter synchronization. And the sensitive data (such as device position) is processed locally, and only the encrypted feature vector is uploaded to the cloud end, so that the privacy processing is realized. The problem of high decision delay and privacy leakage risk in the traditional cloud solution is solved.

[0071] The embodiment of the application solves the problem of traditional manual adjustment antenna unable to realize accurate dynamic pointing, reducing multipath interference and signal attenuation by adopting the stepper motor and the power amplifier to execute the real-time adjustment of the epitaxial antenna.

[0072] For the training process of the mixed reinforcement model in step S102, Figure 6 is a flowchart of a mixed reinforcement model training process provided by the embodiment of the application. Figure 6 is described on the basis of the first embodiment. According to Figure 6The provided diagram, the mixed reinforcement model training process includes: S601~S606.

[0073] S601, feature extraction is performed on the input perception training data through a convolutional neural network layer to obtain spatial training features.

[0074] S602, the motion trajectory data of the training object in the position data of the perception training data is extracted through a long short-term memory network layer.

[0075] S603, the target motion trajectory of the training object is predicted by the mixed reinforcement training model on the spatial training features and the motion trajectory data.

[0076] The convolutional neural network layer mentioned here can be understood as a CNN layer. The long short-term memory network layer mentioned here can be understood as an LSTM layer.

[0077] First, the Wi-Fi data (100x100 grid environment map) is input into the CNN layer to extract spatial training features. The LSTM layer is used as a long short-term memory network layer to extract the motion trajectory data of the training object as a time-varying feature, and the feature extraction process is realized. The mixed reinforcement learning framework is adopted, and the extracted spatial training features and motion trajectory data are used as processing data to train the model and predict the target motion trajectory of the training object. The predicted target motion trajectory is used as intermediate processing data in the model, which provides a reference for calculating the adjustment parameters of the epitaxial antenna in the next step.

[0078] S604, based on the target motion trajectory of the training object, the initial adjustment parameters of the epitaxial antenna are predicted, and the initial adjustment angle and initial power value of the epitaxial antenna are obtained.

[0079] S605, the reward function corresponding to the initial adjustment angle and initial power value is constructed, and the reward function is used to represent the deviation of the adjustment angle and adjustment power of the epitaxial antenna.

[0080] S606, the network parameters contained in the reward function are optimized to obtain the trained mixed reinforcement model.

[0081] Further, after obtaining the target motion trajectory of the training object, the initial adjustment parameters of the epitaxial antenna are predicted through the mixed reinforcement training model to obtain the initial output result of the mixed reinforcement training model. The deviation of the adjustment angle and adjustment power of the epitaxial antenna is adjusted through the set reward function, so that the output result after reward function processing is more accurate. The network parameters contained in the reward function are optimized through the NSGA-II algorithm to obtain the trained mixed reinforcement model.

[0082] The mixed reinforcement training model provided by the embodiment of the application adopts an edge computing chip framework to deploy an AIDeepseek 1.5B large model, adopts an edge computing unit and a cloud end collaborative architecture, realizes local real-time decision through a lightweight AI model, and gives consideration to low delay and data privacy. The model size is compressed after quantization precision, and the real-time requirement of an edge device (single inference delay < 50 ms) is met. A lightweight neural network model is deployed, the input layer receives time domain signal strength of Wi-Fi sensing, device location data and historical network load information, and the output layer generates an optimal angle of an antenna, a polarization direction and a power allocation strategy. The model training data includes a laboratory simulation environment and a real home scene dataset. The adaptive adjustment strategy scene self-learning is supported, the model is adapted to different house types through transfer learning, the mixed gun training model explores the optimal strategy through reinforcement learning in the initial stage, and is fine-tuned based on transfer data in the later period. When the same frequency interference is detected (for example, the adjacent frequency channel occupancy rate is greater than 70%), the polarization diversity mode is automatically switched to, and the switching time of the extension antenna horizontal / vertical polarization is less than 200 ms, so that the purpose of interference is achieved.

[0083] The mixed reinforcement model training process provided by the embodiment of the application is a lightweight training model, Wi-Fi sensing data is used as input data, the motion trajectory of a target object is predicted, the adjustment angle and the adjustment power of the extension antenna are obtained through calculation optimization, the limitation that a traditional passive optimization mechanism (such as channel switching and fixed polarization) cannot cope with complex dynamic scenes is overcome, and the technical effects of millisecond-level response and adaptive resource allocation are realized.

[0084] For the process of obtaining the mixed reinforcement model in step S606, Figure 7 is a flowchart of another mixed reinforcement model training process provided by the embodiment of the application. Figure 7 is described on the basis of the above embodiment. According to Figure 7 The provided diagram shows that the reward function includes a throughput parameter, a signal strength parameter and a multipath interference parameter. The mixed reinforcement model training process includes S701-S704.

[0085] S701, adjusting the parameter weight values corresponding to the throughput parameter, the signal strength parameter and the multipath interference parameter by using a proximal policy optimization algorithm.

[0086] The throughput mentioned here can be understood as the amount of data successfully transmitted per unit time. The signal strength parameter mentioned here can be understood as a parameter measured by RSSI (received signal strength indication) or SNR (signal-to-noise ratio). The multipath interference parameter can be understood as inter-symbol interference caused by signal reflection, which causes the bit error rate to rise.

[0087] The reward function R_t is obtained through the following equation 1: Equation 1 Wherein, a represents the throughput parameter weight value, r_throughput represents the throughput parameter, r_signal represents the signal strength parameter, β represents the signal strength parameter weight value, r_multipath represents the multipath interference parameter, γ represents the multipath interference parameter weight value.

[0088] By adjusting the α, β, γ weights, the rate and stability are balanced. The path that transmits data with the maximum reward function R_t is selected, and the encoding scheme is dynamically adjusted according to the real-time reward function R_t. The contents of table 1 are referred to for adjusting α, β, and γ.

[0089] ‌Table 1 Further, usually α=0.6, β=0.3, γ=0.1 are set, the PPO algorithm is used to cut the objective function or the penalty term to limit the difference between the new and old strategies, and prevent the instability caused by too large update. The throughput parameter, signal strength parameter and multipath interference parameter in the reward function are adjusted by the PPO algorithm. The purpose of optimizing the mixed reinforcement training model is achieved.

[0090] S702, update the mixed reinforcement training model by the adjusted parameter weight value.

[0091] S703, based on the updated mixed reinforcement training model, use the non-dominated sorting genetic algorithm to compare the dominance relationship of the throughput parameter, signal strength parameter and multipath interference parameter, and obtain the corresponding non-dominated front data.

[0092] S704, sort the non-dominated front data to obtain a target solution set, and obtain the trained mixed reinforcement model.

[0093] The non-dominated sorting genetic algorithm mentioned here can be understood as the NSGA-II algorithm.

[0094] The non-dominated sorting genetic algorithm is used to compare the throughput parameter, signal strength parameter and multipath interference parameter optimization process. First, an initial population containing multiple individuals is randomly generated, each individual representing a set of potential parameter configurations. The diversity of the initial population directly affects the subsequent ability to trade off throughput and other objectives. By comparing the dominance relationship between individuals, the population is stratified into multiple non-dominated frontiers (Pareto front). By calculating the density (crowding distance) of individuals in the objective space, over-concentration of solutions is avoided, ensuring uniform distribution of high-throughput solutions on the Pareto front, and avoiding local optima. If individual population A is superior to individual population B in the throughput parameter, and other objectives (such as the signal strength parameter) are not inferior to individual population B, then individual population A dominates individual population B. Solutions with high throughput but high interference may be dominated by solutions with low throughput but stability, forming a hierarchical structure of multi-objective trade-offs. Individual populations with higher non-dominated levels or greater crowding distances are preferred. Even if a solution has low throughput, if its signal strength or interference resistance is significantly better, it may still be selected. New solutions are generated by combining parent parameters, and after merging parents and children, the non-dominated solutions are reordered to avoid the loss of high-throughput solutions. By iteratively approaching the Pareto front, a balanced set of solutions for the throughput parameter and other objective parameters is ultimately provided. Further, a trained hybrid reinforcement model is obtained.

[0095] In one possible example scenario, Figure 8 is a flowchart of another hybrid reinforcement model training process provided by the embodiments of the present application. According to Figure 2 According to the provided diagram, first, the perception training data is input into the hybrid reinforcement training model as the original signal data, and the spatial features of the original signal are extracted to obtain position data and time-frequency signal data. The signal analysis is divided into environment perception data, network performance data and device state data. The hybrid reinforcement training model is pre-trained using environment features, network features and historical features, and the environment perception data, network performance data and device state data are processed by the decision model optimization to predict the adjustment angle, polarization switching and power allocation results of the epitaxial antenna. Through the NSGA-II algorithm, the throughput, delay and energy consumption are balanced to achieve multi-objective optimization. Through closed-loop feedback and real-time calibration, the model training result is obtained, and the training of the hybrid reinforcement model is realized.

[0096] The embodiments of the present application balance the throughput, delay and energy consumption by introducing the NSGA-II algorithm, generate a Pareto optimal solution set, and finally select the antenna configuration scheme with the highest comprehensive score. The limitations of the traditional passive optimization mechanism cannot cope with complex dynamic scenarios, and the technical effects of millisecond-level response and adaptive resource allocation are realized.

[0097] Figure 9is a flowchart of another mixed reinforcement model training process provided by an embodiment of the present application. Figure 9 is described on the basis of the above embodiment. According to Figure 9 The mixed reinforcement model training process includes S901-S903 according to the diagram provided.

[0098] S901, the throughput gain of the throughput parameter corresponding to the target adjustment angle and the target power value is calculated.

[0099] S902, in the case where the throughput gain is lower than the set gain threshold, the throughput gain is fed back to the mixed reinforcement model.

[0100] S903, the model parameters are updated through the throughput gain, and the mixed reinforcement model is updated.

[0101] Further, on the basis of the target adjustment angle and the target power value of the epitaxial antenna output by the mixed reinforcement model, the throughput gain of the throughput parameter is calculated by the network performance monitoring module, and compared with the set threshold, and the mixed reinforcement model is updated through the comparison result. For example, the throughput gain G_t, if G_t<5%, re-optimization is triggered; the throughput gain G_t is fed back to the AI model engine, and the model parameters are updated through online incremental learning.

[0102] In a possible example scenario, Figure 10 is a flowchart of another epitaxial antenna adjustment method based on a mixed reinforcement model provided by an embodiment of the present application. According to Figure 1 The epitaxial antenna adjustment method based on the mixed reinforcement model includes: 1, the environment perception stage: Wi-Fi perception scans at a period of 100ms to generate an environment feature vector X_t; 2, data preprocessing: normalizing and denoising (wavelet transform threshold method) X_t, and extracting effective features F_t; 3, AI decision: input F_t into the decision model, and output the optimal angle \\theta^* and power P^*; 3, execute adjustment: the motor drive module rotates the antenna to \\theta^*, and the power amplifier is adjusted to P^*; 4, effect evaluation: the network performance monitoring module calculates the throughput gain G_t, and if G_t<5\\%, re-optimization is triggered; 5, model feedback update: G_t is fed back to the AI engine, and the model parameters are updated through online incremental learning.

[0103] The method for adjusting an extension antenna based on a hybrid reinforcement model provided in the embodiment of the present application can collect perception data of a target object, eliminate a coverage blind area and multipath interference caused by a static antenna, predict a target movement trajectory of the target object and an adjustment parameter of the extension antenna through a trained hybrid reinforcement model, and output a target adjustment angle and a target power value of the extension antenna, thereby improving the accuracy of adjusting the angle and power of the extension antenna and realizing real-time observation and dynamic adjustment of the target object. Through the technical chain of dynamic perception-intelligent decision-precise execution, the method breaks through the defects of a traditional gateway static antenna design that cannot adapt to dynamic scenes such as personnel movement and furniture obstruction, realizes real-time matching of signal coverage and user position, and achieves the technical effects of improving the adaptive capability and user experience in a wireless network.

[0104] Figure 11 is a structural schematic diagram of a device for adjusting an extension antenna based on a hybrid reinforcement model provided in the embodiment of the present application. According to the device for adjusting an extension antenna based on a hybrid reinforcement model provided in the embodiment of the present application, Figure 11 the structure of the device for adjusting an extension antenna based on a hybrid reinforcement model includes: a perception module 111 configured to acquire perception data of a target object detected by a target perception module, the perception data including position data and time-frequency power data; an AI model processing module 112 configured to input the perception data and historical adjustment data into a trained hybrid reinforcement model to predict a target movement trajectory of the target object, and predict an adjustment parameter of the extension antenna according to the target movement trajectory of the target object, and output a target adjustment angle and a target power value of the extension antenna, the historical adjustment data representing historical perception data adjusted by the hybrid reinforcement model; a hardware adjustment module 113 configured to adjust the extension antenna based on the target adjustment angle and the target power value.

[0105] The device for adjusting an extension antenna based on a hybrid reinforcement model 100 provided in the embodiment can be the device for adjusting an extension antenna based on a hybrid reinforcement model 100 as shown in Figure 11 the device for adjusting an extension antenna based on a hybrid reinforcement model 100 can realize the technical effects of the method for adjusting an extension antenna based on a hybrid reinforcement model as shown in Figures 1-10 the corresponding description is not repeated here for brevity. Figures 1-10

[0106] Figure 12 is a structural schematic diagram of a device for adjusting an extension antenna based on a hybrid reinforcement model provided in the embodiment of the present application. The device for adjusting an extension antenna based on a hybrid reinforcement model can include a processor 1201 and a memory 1202 storing computer program instructions.

[0107] ​In particular, the processor 1201 can include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or can be configured to implement one or more integrated circuits that embody the embodiments of the present application.

[0108] The memory 1202 can include mass storage for data or instructions. As an example and not by way of limitation, the memory 1202 can include a hard disk drive (HDD), a floppy disk drive, flash memory, an optical disc (e.g., a compact disc (CD) or a digital versatile disc (DVD)), a tape drive, a USB drive, or a combination of two or more of these. The memory 1202 can include removable or non-removable (or fixed) media, where appropriate. The memory 1202 can be internal or external to the integrated gateway disaster recovery device, as appropriate. In particular embodiments, the memory 1202 is non-volatile, solid-state memory.

[0109] The memory can include read-only memory (ROM), random-access memory (RAM), magnetic disk storage mediums, optical storage mediums, flash memory devices, electrical, optical, or other physical / tangible memory storage devices. Thus, in general, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software that, when executed (by one or more processors), is operable to perform operations described with reference to the methods according to the first aspect of the present application.

[0110] The processor 1201 implements the above-mentioned any one of the epitaxial antenna adjustment method based on the hybrid reinforcement model by reading and executing the computer program instructions stored in the memory 1202.

[0111] In one example, the epitaxial antenna adjustment device based on the hybrid reinforcement model can further include a communication interface 1203 and a bus 1210. Wherein, as shown in the figure, the processor 1201, the memory 1202, the communication interface 1203 are connected through the bus 1210 and complete the communication between each other. Figure 12

[0112] The communication interface 1203 is mainly used to realize the communication between each module, device, unit and / or equipment in the embodiments of the present application.

[0113] ​Bus 1210 includes hardware, software, or both, to couple components of the online data traffic metering device to each other and to couple components to other components within the online data traffic metering device. While bus 1210 is shown as a single bus, alternative embodiments include two or more buses. Although the present embodiments describe and show a particular bus, the present embodiments contemplate any suitable bus or interconnect.

[0114] In addition, in combination with the epitaxial antenna adjustment method based on the hybrid reinforcement model in the above embodiments, the embodiments of the present application can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; the computer program instructions are executed by the processor to implement any one of the epitaxial antenna adjustment methods based on the hybrid reinforcement model in the above embodiments.

[0115] The embodiments of the present application also provide a computer program product, including a computer program, which is executed by the processor to implement any one of the epitaxial antenna adjustment methods based on the hybrid reinforcement model in the above embodiments.

[0116] It needs to be clear that the present application is not limited to the specific configurations and processes described above and shown in the drawings. For the sake of brevity, detailed descriptions of well-known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between steps, after understanding the spirit of the present application.

[0117] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0118] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0119] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0120] The above merely describes a specific implementation of the present application. Those skilled in the art can clearly understand the specific working processes of the system, modules and units described above for the convenience and brevity of description, and can refer to the corresponding processes in the foregoing method embodiments, which will not be described herein again. It should be understood that the protection scope of the present application is not limited to this, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application.

Claims

1. An epitaxial antenna adjustment method based on a hybrid reinforcement model, applied to a home gateway, characterized in that, include: Acquire sensing data of the detected target object, the sensing data including location data and time-frequency power data; The sensing data and historical adjustment data are input into the trained hybrid reinforcement model to predict the target motion trajectory of the target object, and the adjustment parameters of the extension antenna are predicted based on the target motion trajectory of the target object. The target adjustment angle and target power value of the extension antenna are output. The historical adjustment data represents the historical sensing data that has been adjusted by the hybrid reinforcement model. The extension antenna is adjusted based on the target adjustment angle and the target power value.

2. The method according to claim 1, characterized in that, The acquisition of the perception data of the target object includes: The target perception module scans the target environment at set time intervals to obtain time-domain signal data; The time-domain signal data is processed by Fast Fourier Transform to obtain time-frequency power data; The direct signal and reflection path corresponding to the target object are identified by a multipath component separation algorithm to obtain multipath data; The multipath data is processed to obtain location data; The sensing data is determined based on the location data and the time-frequency power data.

3. The method according to claim 2, characterized in that, The step of performing a Fast Fourier Transform on the time-domain signal data to obtain time-frequency power data includes: The time-domain signal data is processed by Fast Fourier Transform to obtain the frequency-domain signal data; The channel impulse response result of the frequency domain signal data is extracted to obtain the channel frequency power value, and the channel frequency power value is used as time-frequency power data.

4. The method according to claim 2, characterized in that, The step of performing location processing on the multipath data to obtain location data includes: The multipath data is fused and located based on the disparity positioning algorithm and the angle of arrival estimation algorithm to obtain the target azimuth angle of the extended antenna and the distance data corresponding to the target object. An environmental spatial feature map is generated by stitching and fusing the target azimuth and distance data, and the environmental spatial feature map is used as location data.

5. The method according to claim 1, characterized in that, The hybrid reinforcement model is obtained through the following method: Spatial training features are obtained by extracting features from the input perceptual training data through convolutional neural network layers. The motion trajectory data of the training object in the location data of the perception training data is extracted by the long short-term memory network layer; The target motion trajectory of the training object is predicted by using the spatial training features and the motion trajectory data through a hybrid reinforcement training model. Based on the target motion trajectory of the training object, the initial adjustment parameters of the epitaxial antenna are predicted, and the initial adjustment angle and initial power value of the epitaxial antenna are obtained. Construct a reward function corresponding to the initial adjustment angle and the initial power value, the reward function being used to characterize the deviation between the adjustment angle and the adjustment power of the epitaxial antenna; The network parameters contained in the reward function are optimized to obtain a trained hybrid reinforcement model.

6. The method according to claim 5, characterized in that, The reward function includes throughput parameters, signal strength parameters, and multipath interference parameters; The optimization of the network parameters contained in the reward function to obtain the trained hybrid reinforcement model includes: The parameter weights corresponding to the throughput parameter, signal strength parameter, and multipath interference parameter are adjusted using a near-end strategy optimization algorithm. The hybrid reinforcement training model is updated using the adjusted parameter weight values; Based on the updated hybrid reinforcement training model, the non-dominated sorting genetic algorithm is used to compare the dominance relationships of the throughput parameter, signal strength parameter, and multipath interference parameter to obtain the corresponding non-dominated frontier data. The target solution set is obtained by sorting the non-dominated frontier data, and a trained hybrid reinforcement model is obtained.

7. The method according to claim 5, characterized in that, The reward function R_t is obtained through the following equation: ， Where α represents the throughput parameter weight value, r_throughput represents the throughput parameter, r_signal represents the signal strength parameter, β represents the signal strength parameter weight value, r_multipath represents the multipath interference parameter, and γ represents the multipath interference parameter weight value.

8. The method according to claim 1, characterized in that, The adjustment of the epitaxial antenna based on the target adjustment angle and the target power value includes: The extended antenna is adjusted along two axes using a drive motor and a target reducer to bring it to the target adjustment angle. The power of the extended antenna is adjusted to the specified power value using a power amplifier.

9. The method according to claim 6, characterized in that, After adjusting the epitaxial antenna based on the target adjustment angle and the target power value, the method further includes: Calculate the throughput gain of the throughput parameter corresponding to the target adjustment angle and the target power value; If the throughput gain is lower than the set gain threshold, the throughput gain is fed back to the hybrid enhancement model; The hybrid enhancement model is updated by updating the parameters of the throughput gain update model.

10. An epitaxial antenna adjustment device based on a hybrid reinforcement model, characterized in that, include: The perception module is used to acquire perception data of the target object detected by the target perception module. The perception data includes location data and time-frequency power data. The AI ​​model processing module is used to input the perception data and historical adjustment data into the trained hybrid reinforcement model to predict the target motion trajectory of the target object, and to predict the adjustment parameters of the extension antenna based on the target motion trajectory of the target object, and output the target adjustment angle and target power value of the extension antenna. The historical adjustment data represents the historical perception data that has been adjusted by the hybrid reinforcement model. A hardware adjustment module is used to adjust the epitaxial antenna based on the target adjustment angle and the target power value.

11. An epitaxial antenna adjustment device based on a hybrid reinforcement model, characterized in that, The device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the epitaxial antenna adjustment method based on the hybrid enhancement model as described in any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the epitaxial antenna adjustment method based on a hybrid enhancement model as described in any one of claims 1-9.

13. A computer program product, characterized in that, When the instructions in the computer program product are executed by the processor of the electronic device, the electronic device performs the epitaxial antenna adjustment method based on the hybrid enhancement model as described in any one of claims 1-9.

Images