Multi-band RF energy collection efficiency optimization method based on adaptive neural network

Abstract

The invention relates to the technical field of multi-band RF energy collection efficiency optimization methods, in particular to a multi-band RF energy collection efficiency optimization method based on an adaptive neural network, comprising the following steps: monitoring radio frequency signal intensity distribution of multiple bands in a frequency domain in real time based on a reconfigurable antenna array to obtain signal feature data; processing the signal feature data based on a neural network model to predict an energy collection parameter of the target frequency band; parameters of the matching network are dynamically adjusted, and the collection efficiency of the radio frequency energy is optimized; transmission and energy collection of multi-band RF signals are completed based on optimized parameters, network parameters can be adaptively adjusted and matched according to real-time channel changes through a time sequence adaptive dynamic network and introduction of a dynamic adjustment mechanism, so that the dynamic adaptive capacity of energy collection is improved, and the energy collection efficiency is improved through a collaborative band aggregation network. Cooperative information among multi-band signals is integrated, the overall radio frequency signal intensity prediction result is optimized, and the energy collection efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-band RF energy collection efficiency optimization methods, and in particular to a multi-band RF energy collection efficiency optimization method based on an adaptive neural network. Background Art

[0002] Currently, multi-band RF energy harvesting technology has been widely used in wireless communications, the Internet of Things, smart sensing, and other fields. With the rapid development of emerging communication technologies such as 5G and Wi-Fi 6, the frequency range and dynamic complexity of RF signals have increased significantly. Traditional RF energy harvesting methods typically rely on fixed-parameter antennas and matching networks, which often make it difficult to achieve efficient energy capture and transmission in complex and changing signal environments.

[0003] In existing technologies, multi-band RF energy harvesting generally uses the following two methods:

[0004] 1. Single frequency band optimization method: This method optimizes the matching network parameters of a single frequency band signal to maximize the RF energy within the frequency band. For example, Chinese patent publication number CN111628576A discloses a RF energy harvesting system that obtains RF energy in a specific frequency band in the environment in real time and converts the RF energy into a RF signal corresponding to the frequency band, thereby reducing costs and power consumption. However, in actual applications, there are usually complex coupling and synergistic relationships between multi-band signals. Optimizing only a single frequency band makes it difficult to fully utilize the signal resources of other frequency bands, resulting in low overall energy harvesting efficiency.

[0005] 2. Static multi-band signal optimization method, such as the Chinese patent with publication number CN113302816A discloses a multi-band energy harvesting system. This technology introduces static analysis of multi-band signals and improves energy harvesting efficiency through multi-band feature extraction and static matching network adjustment. For another example, the Chinese patent with publication number CN108683340B discloses a multi-band RF rectifier circuit, which uses different branches to process RF signals of different frequency bands, effectively realizing the processing of RF signals of different frequency bands. This technology relies on the linear relationship in the frequency domain and lacks the ability to adapt to dynamic channel changes in real time. When the signal environment changes, the system cannot respond in time, resulting in a significant decrease in energy harvesting efficiency and signal transmission quality.

[0006] In summary, the above methods have the following technical problems when dealing with complex dynamic signal environments:

[0007] 1. Insufficient feature extraction capabilities: Traditional methods for extracting multi-band signal features are mostly limited to linear relationships within the frequency domain, making it difficult to capture high-order synergistic relationships between multi-band signals. This limitation in feature extraction directly leads to reduced RF energy capture efficiency.

[0008] 2. Lack of dynamic adaptive capabilities: Existing technologies are unable to adjust the parameters of the matching network in real time and cannot adapt to dynamically changing channel conditions, resulting in large fluctuations in energy collection efficiency under different signal environments, affecting the overall performance of the system.

[0009] 3. Insufficient collaborative optimization: The synergistic relationship between multi-band signals is not fully utilized. Existing methods generally process each frequency band signal independently, failing to improve the prediction accuracy of the total signal strength and energy collection efficiency through collaborative optimization. Summary of the Invention

[0010] To address these technical issues, the present invention proposes a method for optimizing the efficiency of multi-band RF energy harvesting. By introducing a multi-band feature enhancement network (MFEN), a time-adaptive dynamic network (TADN), and a collaborative band aggregation network (CBAN), the present invention addresses the existing issues of multi-band signal feature extraction, insufficient dynamic adaptability, and inadequate collaborative optimization.

[0011] The present invention provides a multi-band RF energy collection efficiency optimization method based on an adaptive neural network, comprising the following steps:

[0012] Based on a reconfigurable antenna array, the RF signal strength distribution in multiple frequency bands in the frequency domain is monitored in real time to obtain signal characteristic data;

[0013] Processing the signal characteristic data based on a neural network model to predict energy harvesting parameters of a target frequency band;

[0014] Dynamically adjust the parameters of the matching network to optimize the efficiency of RF energy collection;

[0015] Based on the optimized parameters, the transmission and energy collection of multi-band RF signals are completed.

[0016] Preferably, the processing based on the neural network model includes obtaining an enhanced feature matrix through a multi-band feature enhancement network, specifically including:

[0017] F (1) =Φ(T·X+B1)

[0018] Among them, F (1) is the enhanced feature matrix, X∈R M×N is the input frequency domain signal strength distribution matrix, M is the number of signal channels, N is the number of frequency bands, T∈R P×M×Nis a high-order feature tensor, B1∈R P×N is the bias matrix, and Φ is the Bessel function activation function.

[0019] Preferably, the enhanced feature matrix is further optimized by a time-series adaptive dynamic network, and the optimization process includes the following formula:

[0020] F (2) =Ψ(W2·F (1) +B2)+Ω(F (1) ,H t )

[0021] Among them, F (2) is the optimized feature matrix, W2∈R Q×P is the weight matrix, B2∈R Q×N is the bias matrix, Ψ is the Laguerre polynomial activation function, Ω(F (1) ,H t ) is a dynamic adjustment item, H t is the time-dependent hidden state matrix.

[0022] Preferably, the optimized feature matrix is subjected to multi-band collaborative optimization through a collaborative band aggregation network to obtain a total signal strength prediction value, specifically including the following formula:

[0023]

[0024] Among them, y total is the predicted value of total signal strength, λ i is the frequency band weight coefficient, is the nonlinear frequency band intensity function, μ j is the topological weight, κ(A j ,F (2) ) is the collaborative optimization function based on the topology graph, A j is the frequency band coordination topology matrix.

[0025] Preferably, the high-order characteristic tensor T is constructed by the following integral formula:

[0026] T ijk =∫ Ω f i (x)g j (y)h k (z)dx dy dz

[0027] Among them, f i (x), g j (y), h k (z) is the basis function of signal distribution, Ω is the signal intensity normalization interval, T ijkIt is an element of the high-order feature tensor, which is used to describe the synergistic relationship between signals in different frequency bands.

[0028] Preferably, the process of dynamically adjusting the matching network parameters is based on the following feedback optimization algorithm:

[0029]

[0030] Among them, θ t is the parameter of the matching network at the tth iteration, η is the learning rate, is the gradient of the loss function with respect to the parameters.

[0031] As an advantage, the loss function L(θ,F (2) ) is of the following form:

[0032]

[0033] Among them, y i is the target frequency band signal strength, is the prediction function.

[0034] As an advantage, the frequency band cooperative topology matrix A j The Laplacian matrix is calculated by the following graph:

[0035] L=DA

[0036] Where L is the graph Laplacian matrix, D is the degree matrix, and A is the adjacency matrix.

[0037] Preferably, the frequency band weight λ i Determined by the following optimization function:

[0038]

[0039] Among them, w i is the weight parameter of the i-th frequency band.

[0040] Preferably, the multi-band RF signal transmission is implemented through a fixed parameter matching network, specifically including signal modulation and demodulation, and the signal modulation is based on the following formula:

[0041]

[0042] Among them, s(t) is the modulation signal, A i is the signal amplitude, f i is the frequency, φ i For phase.

[0043] The present invention has the following beneficial effects:

[0044] The present invention uses a multi-band feature enhancement network and a high-order feature tensor decomposition method to capture high-order synergistic relationships between multi-band signals in the frequency domain, thereby significantly improving feature extraction capabilities. The present invention introduces a dynamic adjustment mechanism through a time-series adaptive dynamic network, which can adaptively adjust the matching network parameters according to real-time channel changes to improve the dynamic adaptability of energy collection. Through a collaborative band aggregation network, the present invention integrates the synergistic information between multi-band signals and optimizes the overall RF signal strength prediction results, thereby further improving energy collection efficiency. The present invention significantly improves the overall efficiency of multi-band RF energy collection and achieves high-precision signal prediction and energy optimization in complex dynamic channel environments. Specifically, the multi-band feature enhancement network can fully extract the synergistic information between multi-band signals, and through a high-order feature tensor decomposition method, the features of signals in different frequency bands are effectively fused in the input layer. This synergistic effect improves the characterization capability of input features through feature superposition and enhancement. The time-series adaptive dynamic network can dynamically adjust network parameters based on real-time feedback, allowing the system to quickly adapt to changes in the signal environment. By introducing dynamic adjustment items, the present invention achieves rapid response to dynamic channel changes and significantly reduces fluctuations in energy collection efficiency caused by channel changes. The collaborative band aggregation network aggregates the prediction results of signals in each frequency band and further optimizes the total signal strength prediction results by modeling the collaborative relationship between multiple frequency bands through the topological Laplace matrix. This module is closely connected with the first two-stage network and achieves an overall improvement in energy collection efficiency and signal prediction accuracy through multi-level collaboration of feature input, dynamic feedback, and optimization results.

[0045] In summary, the present invention effectively solves the technical bottleneck of the existing technology in optimizing the efficiency of multi-band RF energy collection, and provides a new technical means for efficient energy collection in complex signal environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of the data collection and preprocessing method of the present invention.

[0047] Figure 2 This is a flow chart of the multi-band feature enhancement network of the present invention.

[0048] Figure 3 Flowchart of the timing adaptive dynamic network of the present invention.

[0049] Figure 4 This is a flow chart of the collaborative frequency band aggregation network of the present invention.

[0050] Figure 5 This is a flow chart of the dynamic matching network parameter adjustment of the present invention. DETAILED DESCRIPTION

[0051] To further illustrate the technical means and effects employed by the present invention to achieve its intended objectives, the following detailed description, along with the accompanying drawings and preferred embodiments, includes a detailed description of the specific implementations, structures, features, and effects thereof. In the following description, references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0052] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0053] See Figure 1-5 , the present invention proposes a multi-band RF energy collection efficiency optimization method based on an adaptive neural network, which aims to dynamically optimize the RF energy collection efficiency. The method first obtains the RF signal strength distribution data of multiple frequency bands in the frequency domain based on a reconfigurable antenna array. The reconfigurable antenna array can be dynamically adjusted in real time to adapt to the signal reception requirements in different scenarios. By real-time monitoring and recording of the frequency band signal strength distribution, comprehensive signal coverage and efficient energy collection can be ensured. Preferably, in specific application scenarios, the adjustment frequency of the antenna array can be set to 1000 times per second, which can quickly respond to changes in the channel environment, thereby improving the coverage range and energy collection efficiency of multi-band signals.

[0054] The method then uses a neural network model to predict and evaluate the signal strength distribution data. This model can efficiently analyze the dynamic changes in multi-band signals and determine the optimal energy harvesting parameters for each band. In practical deployments, a neural network with three hidden layers, each containing 128 neurons, is preferably used to balance computational efficiency and prediction accuracy.

[0055] On this basis, the matching network parameters are dynamically adjusted to further optimize RF energy capture efficiency. Dynamic adjustments to the matching network can be based on real-time feedback data streams to ensure optimal performance during the energy capture process. Specifically, gradient descent can be used to adjust the matching network parameters, with a learning rate of 0.01 being preferred to ensure rapid convergence and optimization accuracy.

[0056] Finally, based on the optimized parameters, the method achieves the transmission and collection of multi-band RF signals. This step dynamically adjusts the signal modulation and demodulation methods to ensure stability and efficiency during signal transmission. During transmission, the signal bit error rate is preferably controlled below 0.01 to ensure signal quality.

[0057] First, the method uses a reconfigurable antenna array to monitor the distribution of radio frequency signal strength across multiple frequency bands in real time. Preferably, the reconfigurable antenna array consists of multiple programmable antenna elements, each of which can dynamically adjust its reception parameters based on changes in ambient signals to ensure optimal signal reception performance.

[0058] Through the above steps, the system can obtain a complete frequency domain signal strength distribution, providing data support for subsequent signal prediction and optimization. The antenna array of the present invention also includes a matching network. The dynamic reconfiguration characteristics of this network enable it to adapt to the signal characteristics of different frequency bands, thereby further improving the efficiency of RF energy collection.

[0059] Furthermore, the step includes extracting features from the signal strength distribution data through a multi-band feature enhancement network to obtain enhanced frequency domain features. The multi-band feature enhancement network obtains an enhanced feature matrix based on the following formula:

[0060] F (1) =Φ(T·X+B1)

[0061] Among them, F (1) is the enhanced feature matrix; X is the input frequency domain RF signal strength distribution matrix, whose rows represent the number of signal channels and columns represent the number of frequency bands. In a specific embodiment, the number of signal channels M is set to 32 and the number of frequency bands N is set to 128. This parameter setting can cover most common RF signal scenarios. The high-order feature tensor T is used to capture the high-order synergistic relationship between multi-band signals. Its dimension is preferably 64×32×128, so as to establish complex associations between different dimensions of the signal. The activation function Φ uses the Bessel function J n (x), this function can enhance the characteristic expression of high-frequency signals, especially under complex channel conditions, and can significantly improve the accuracy of signal prediction. P×N is the bias matrix used to compensate for the deviation in signal characteristics.

[0062] In one embodiment of the present invention, the method further dynamically optimizes the enhanced features through a time-adaptive dynamic network. Specifically, the network introduces time-dependent correlation and can respond to dynamically changing signal environments. The optimization process of the feature matrix is as follows:

[0063] F (2) =Ψ(W2·F (1) +B2)+Ω(F (1) ,H t )

[0064] Among them, F (2)is the optimized feature matrix; W2 is the weight matrix of the temporal network, and its dimension can be set to 128×64 to ensure efficient mapping between feature enhancement and optimization. Dynamic adjustment term Ω(F (1) ,H t ) adjusts the matching network parameters in real time based on historical signal characteristics, allowing the system to quickly respond to dynamic channel changes.

[0065] Preferably, the Laguerre polynomial L n (x) is used for the adaptive transformation function ψ, which has a high performance in capturing the dynamic changes of nonlinear signals, thereby improving the prediction accuracy and adaptability of the model.

[0066] In one embodiment of the present invention, the optimized features are subjected to multi-band collaborative optimization through a collaborative frequency band aggregation network. The collaborative frequency band aggregation network calculates the total signal strength prediction value using the following formula:

[0067]

[0068] Among them, y total is the predicted value of total signal strength; i and μ j They are frequency band weight and topology weight, which are dynamically optimized through model training. Topology A j Used to model the collaborative relationship between multiple frequency bands. Its calculation is based on the graph Laplacian matrix, which can effectively capture the efficient collaborative information between frequency bands.

[0069] In one embodiment, the frequency band weight λ i By calculating the soft maximization function to ensure that its sum is 1, this setting can optimize the weight distribution among frequency bands, thereby improving the accuracy of the total signal strength prediction.

[0070] In one embodiment, the high-order feature tensor T is constructed by the following integral formula:

[0071] T ijk =∫ Ω f i (x)g j (y)h k (z)dx dy dz

[0072] In the above formula, f i (x), g j (y), h k (z) are basis functions used to describe signal distribution, preferably Chebyshev polynomials, which have advantages in signal representation and compression; Ω is the integration range, which is usually set to the standardized interval [0,1] of signal intensity to ensure uniform expression between different signal dimensions.

[0073] The high-order feature tensor constructed by the above method can fully capture the synergistic characteristics between multi-band signals, thereby providing a more accurate and efficient foundation for subsequent feature enhancement and optimization.

[0074] In summary, this invention, through a series of innovative steps, effectively addresses the existing problem of insufficient multi-band RF energy collection efficiency. Its key technologies, through high-dimensional feature extraction, dynamic time-series optimization, and multi-band collaborative aggregation, not only improve energy collection efficiency but also significantly enhance the accuracy and dynamic adaptability of signal prediction.

[0075] In one embodiment of the present invention, dynamic adjustment of matching network parameters is achieved through the following feedback optimization algorithm:

[0076]

[0077] Among them, θ t is the parameters of the matching network at the tth iteration, including weights and biases; η is the learning rate, preferably ranging from 0.001 to 0.01. According to empirical values, when η = 0.005, the optimization effect is the best, which can achieve a good balance between convergence speed and stability; is the gradient of the loss function with respect to the parameter γ, indicating the optimization direction.

[0078] Loss function L(θ t ,F (2) ) is used to quantify the deviation between the current model's prediction and the actual signal. In practice, this loss function can effectively guide the dynamic adjustment of matching network parameters, gradually approaching the optimal solution. In particular, through the above-mentioned optimization algorithm, the present invention can achieve real-time optimization under variable channel conditions, thereby improving the efficiency of RF energy capture.

[0079] In one embodiment of the present invention, the loss function L(θ,F (2) ) takes the following form:

[0080]

[0081] Among them, y i is the target frequency band signal strength, which is usually measured by experiment. In a preferred embodiment, when the target value of the frequency band signal strength is y i =0.8 (normalized value), the optimization effect of the model is the best; is the model's predicted value for the signal strength of the i-th frequency band, based on the feature matrix output in the second stage and matching network parameters θ are calculated.

[0082] Through the above-mentioned loss function design, the present invention can accurately evaluate the pros and cons of the current matching network parameters and dynamically adjust them in each iteration, thereby achieving adaptive determination of the optimal RF energy capture parameters.

[0083] Furthermore, the topology A j The Laplacian matrix is constructed by:

[0084] L=DA

[0085] Where L is the graph Laplace matrix, D is the degree matrix, and its diagonal elements D ii is equal to the degree of node i, that is, the number of edges directly connected to node i; A is the adjacency matrix, whose element A ij Indicates whether there is a connection between node i and node j. If there is a connection, A ij =1, otherwise 0.

[0086] Preferably, in the application scenario of the present invention, the topology diagram is used to describe the correlation between signals of different frequency bands. For example, when there is a synergistic effect between specific frequency bands, the connection weight of the corresponding node can be increased, usually set to A ij =0.9. By constructing the graph Laplacian matrix, the present invention can effectively integrate the collaborative information of multi-band signals, thereby further improving the accuracy of overall signal prediction.

[0087] Preferably, the frequency band weight λ i Determined by the following optimization function:

[0088]

[0089] Among them, λ i is the weight coefficient of the i-th frequency band, which is used to measure the contribution of the frequency band in the total signal strength prediction; w i is the weight parameter of frequency band i, which is dynamically optimized through model training. i The soft maximization function is calculated to ensure that the sum of the weights of all frequency bands is 1, thus forming a probability distribution.

[0090] In practical applications, the allocation of frequency band weights can dynamically reflect changes in the signal environment. For example, when the signal strength of a certain frequency band is significantly higher than that of other frequency bands, its corresponding w i Will be increased accordingly, thereby increasing the weight λ of this frequency band i , preferably, λ i It can reach above 0.6 when the signal intensity changes significantly.

[0091] Through the above-mentioned weight optimization, the present invention can more accurately aggregate the prediction results of multi-band signals, thereby improving the overall efficiency of RF energy collection.

[0092] Furthermore, the multi-band RF signal transmission is achieved through a fixed parameter matching network, specifically including signal modulation and demodulation processes. Signal modulation is based on the following formula:

[0093]

[0094] Among them, s(t) is the modulation signal, A i is the signal amplitude of the i-th frequency band, which can be preferably set to 0.5 to 1.0 according to the actual scenario; f i is the center frequency of the ith frequency band, which usually ranges from 900 MHz to 2.4 GHz in RF communication systems; i The phase can preferably be dynamically adjusted according to the requirements of signal transmission.

[0095] Through the modulation process described above, the signal can be effectively converted into an RF signal for transmission. Signal demodulation is accomplished through a matching network, which includes signal reception, de-emphasis, filtering, amplification, and sampling, ultimately restoring the RF signal to a digital baseband signal. By using a fixed-parameter matching network, the present invention ensures signal stability during transmission and reception. This can effectively reduce the bit error rate to below 0.01, particularly when transmitting multiple frequency bands in parallel, significantly improving the overall system performance.

[0096] To verify the superiority of the present invention in optimizing multi-band RF energy harvesting efficiency, a rigorous experimental comparison was conducted. The test dataset used in this experiment is the ITU-R recommended channel model dataset, which covers typical RF signal distributions in multiple frequency bands, including a frequency range from 900 MHz to 2.4 GHz and a signal dynamic range from -90 dBm to -30 dBm.

[0097] The present invention was tested using an optimized multi-band feature enhancement network, a time-adaptive dynamic network, and a collaborative band aggregation network. The neural network model consisted of three hidden layers, each containing 128 neurons. The matching network parameters were optimized using gradient descent, with a learning rate set to 0.005. The data input consisted of the signal strength distribution in the frequency domain, and the output was the optimized total RF energy collection efficiency.

[0098] The comparative example used the existing technology method, which did not perform high-order feature tensor decomposition and time series adaptive dynamic adjustment. Instead, it used a common multi-layer perceptron (MLP) model with two hidden layers and 64 neurons per layer. The matching network used fixed parameters and no dynamic adjustment function.

[0099] In order to objectively compare the performance of the two methods, this experiment selected the following key indicators:

[0100] 1. RF Energy Efficiency (REE): This measures the ratio of the RF energy collected per unit time to the theoretical maximum value.

[0101] 2. Signal prediction error (PE): The mean square error between the predicted signal strength and the actual signal strength.

[0102] 3. Dynamic Adaptability (DA): The system’s response speed to changes in signal strength under dynamic channel conditions.

[0103] The detection method is as follows:

[0104] REE: The collected energy is measured by a radio frequency energy meter and the ratio to the theoretical value is calculated.

[0105] PE: Calculated using the standard mean square error formula.

[0106] DA: Calculates the time required for the system to stabilize to the new signal strength after a change in channel conditions.

[0107] The test results are shown in the following table:

[0108] Test indicators Example Comparative Example RF energy collection efficiency (%) 92.5 75.8 Signal prediction error (dB) 1.2 3.8 Dynamic adaptation performance (seconds) 0.35 1.10

[0109] It can be seen from the above experimental results that the present invention is significantly superior to the prior art in multiple key indicators.

[0110] 1. RF Energy Collection Efficiency: The embodiment of the present invention achieved an energy collection efficiency of 92.5%, significantly higher than the 75.8% of the comparative example. This is primarily due to the invention's innovative design in multi-band feature extraction and dynamic matching network adjustment, which better adapts to the characteristics of signals in different frequency bands and improves energy collection efficiency.

[0111] 2. Signal Prediction Error: The signal prediction error of the present invention is 1.2dB, while the error of the comparative example is as high as 3.8dB. This shows that the present invention has higher accuracy in signal strength prediction, mainly due to the introduction of the collaborative band aggregation network, which effectively integrates the collaborative information between multiple bands.

[0112] 3. Dynamic Adaptability: The response time of the present invention under dynamic channel conditions is only 0.35 seconds, while the comparative example requires 1.10 seconds. This significant improvement reflects the excellent performance of the present invention's timing-adaptive dynamic network in rapidly adapting to channel changes.

[0113] Experimental results fully demonstrate the technical advantages of this invention in optimizing multi-band RF energy collection efficiency. By introducing a multi-band feature enhancement network, a time-adaptive dynamic network, and a collaborative band aggregation network, this invention significantly improves the system's performance in energy collection, signal prediction accuracy, and dynamic adaptability. These improvements not only increase RF energy utilization but also provide technical support for efficient signal collection under complex dynamic channel conditions.

[0114] It should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-band RF energy harvesting efficiency optimization method based on an adaptive neural network, characterized in that: The following steps are involved: Based on a reconfigurable antenna array, the RF signal strength distribution in multiple frequency bands in the frequency domain is monitored in real time to obtain signal characteristic data; Processing the signal characteristic data based on a neural network model to predict energy harvesting parameters of a target frequency band; Dynamically adjust the parameters of the matching network to optimize the efficiency of RF energy collection; Based on the optimized parameters, the transmission and energy collection of multi-band RF signals are completed.

2. The method according to claim 1, characterized in that The processing based on the neural network model includes obtaining an enhanced feature matrix through a multi-band feature enhancement network, specifically including: F (1) =Φ(T·X+B1) Among them, F (1) is the enhanced feature matrix, X∈R M×N is the input frequency domain signal strength distribution matrix, M is the number of signal channels, N is the number of frequency bands, T∈R P×M×N is a high-order feature tensor, B1∈R P×N is the bias matrix, and Φ is the Bessel function activation function.

3. The method according to claim 2, characterized in that The enhanced feature matrix is further optimized by a time-series adaptive dynamic network, and the optimization process includes the following formula: F (2) =Ψ(W2·F (1) +B2)+Ω(F (1) ,H) Among them, F (2) is the optimized feature matrix, W2∈R Q×P is the weight matrix, B2∈R Q×N is the bias matrix, Ψ is the Laguerre polynomial activation function, Ω(F (1) ,H t ) is a dynamic adjustment item, H t is the time-dependent hidden state matrix.

4. The method according to claim 3, characterized in that The optimized feature matrix is subjected to multi-band collaborative optimization through a collaborative band aggregation network to obtain a total signal strength prediction value, specifically including the following formula: Among them, y total is the predicted value of total signal strength, λ i is the frequency band weight coefficient, is the nonlinear frequency band intensity function, μ j is the topological weight, κ(A j ,F (2) ) is the collaborative optimization function based on the topology graph, A j is the frequency band coordination topology matrix.

5. The method according to claim 4, characterized in that The high-order feature tensor T is constructed by the following integral formula: T ijk =∫ Ω f i (x)g j (y)h k (z)dxdydz Among them, f i (x), g j (y), h k (z) is the basis function of signal distribution, Ω is the signal intensity normalization interval, T ijk It is an element of the high-order feature tensor, which is used to describe the synergistic relationship between signals in different frequency bands.

6. The method according to claim 5, characterized in that The process of dynamically adjusting matching network parameters is based on the following feedback optimization algorithm: Among them, θ t is the parameter of the matching network at the tth iteration, η is the learning rate, is the gradient of the loss function with respect to the parameters.

7. The method according to claim 6, characterized in that The loss function L(θ,F (2) ) is of the following form: Among them, y i is the target frequency band signal strength, is the prediction function.

8. The method according to claim 7, characterized in that The frequency band cooperative topology matrix A j The Laplacian matrix is calculated by the following graph: L=DA Where L is the graph Laplacian matrix, D is the degree matrix, and A is the adjacency matrix.

9. The method according to claim 8, characterized in that The frequency band weight λ i Determined by the following optimization function: Among them, w i is the weight parameter of the i-th frequency band.

10. The method according to claim 9, characterized in that The multi-band RF signal transmission is achieved through a fixed parameter matching network, specifically including signal modulation and demodulation. The signal modulation is based on the following formula: Among them, s(t) is the modulation signal, A i is the signal amplitude, f i is the frequency, φ i For phase.

Images