Image semantic communication method based on Beidou satellite in smart power grid

Abstract

The invention relates to the field of semantic communication and Beidou satellite communication, and particularly discloses an image semantic communication method based on a Beidou satellite in an intelligent power grid, and the method comprises the steps: enabling the image data of the intelligent power grid in a remote region to serve as the input of a system, inputting the data into a semantic encoder network, and extracting the semantic features of the data; after semantic features of data are obtained, the semantic features are transmitted at a transmitting end, shaping operation is performed on the semantic features before actual transmission, after a receiving end receives the data, the transmitted data need to be recovered, an end-to-end training mode of a combined satellite channel is adopted, a trained system is utilized, and transmission of the image data of the intelligent power grid is completed. The robustness of a communication system is improved, the performance of image transmission is ensured, and lower bandwidth resource occupation is achieved.

Description

Technical Field

[0001] The present invention relates to semantic communication and Beidou satellite communication industries, and in particular to an image semantic communication method based on Beidou satellite in a smart grid. Background Art

[0002] The power communication network is an important support for the safe and stable operation of the smart grid and the efficient operation of power companies. In the context of a large number of smart devices being connected to the smart grid, there are some problems with the power communication network. Specifically, in remote areas and areas with harsh terrain such as mountainous areas, there are problems with the difficulty and high cost of laying optical cables and communication networks. At the same time, due to the complex terrain, the demand for flexible access to power equipment cannot be met, and there are challenges in full coverage of the power communication network. Compared with traditional ground communication networks, Beidou satellites can effectively solve the problem of communication network coverage.

[0003] At present, Beidou satellite communication has application solutions in smart grids, but the existing technology still has technical problems such as not considering the limited bandwidth when a large amount of power grid image data is transmitted to the ground control station through Beidou short message technology, ignoring the processing of the transmitted data itself, only updating the transmission protocol, and not applying it to specific smart grid and Beidou satellite communication scenarios.

[0004] In general, when using Beidou satellite communication to transmit power grid image data in smart grids, it is easier to solve the problem of limited satellite communication bandwidth resources and meet the smart grid's needs for communication network coverage and flexible access to power equipment by extracting and transmitting the semantic features of the data with the help of semantic communication. Therefore, there may be a Beidou satellite-based image semantic communication method that can effectively solve the problem that traditional power communication networks are difficult to deploy in complex terrain areas such as mountainous areas and remote areas, and efficiently transmit power grid image data. Summary of the invention

[0005] The purpose of the present invention is to provide an image semantic communication method based on Beidou satellite in a smart grid to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] An image semantic communication method based on Beidou satellite in a smart grid comprises the following steps:

[0008] Step 1: Take the image data of smart grids in remote areas as the input S of the system, and input the data into the semantic encoder network to extract the semantic features T∈R of the data, which can be expressed as:

[0009] T = E(α; S);

[0010] Where E(α; ·) represents the semantic encoder network with parameter α;

[0011] Step 2: After obtaining the semantic features T of the data, the semantic features are transmitted at the transmitter. Before actual transmission, T is shaped and adjusted to a complex signal X∈C:

[0012] X = Reshape(T);

[0013] The transmitted signal is transmitted to the receiving end through the satellite channel. The received signal at the receiving end can be expressed as:

[0014] Y = HX + n;

[0015] Where H is the channel matrix, n is Gaussian white noise;

[0016] Step 3: After receiving the data, the receiver needs to restore the transmitted data Adjust the dimension of the recovered semantic features to obtain And input to the semantic decoder to get the original transmission data:

[0017]

[0018] Where D(β; ) represents a semantic decoder network with parameter β;

[0019] Step 4: Train the designed communication system using an end-to-end training method of a joint satellite channel;

[0020] During the training process, the neural network parameters α are first initialized (0) and β (0) ,The semantic encoder extracts semantic features, then encodes, transmits and decodes them. After that, the loss is calculated and the parameters are iteratively updated using the stochastic gradient descent method. The training continues until the termination condition is met.

[0021] During the training process, the mean square error (MSE) is selected as the loss function. The loss function can be expressed as follows:

[0022]

[0023] Where I is the total dimension of the image data;

[0024] Step 5: Use the trained system to complete the transmission of smart grid image data.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] 1) By extracting the semantic features of the data to reduce the amount of transmitted data and using Beidou satellites for data transmission, it is possible to achieve data transmission in areas outside the coverage of traditional power communication networks in smart grids.

[0027] 2) In order to mitigate the impact of transmission uncertainty on semantic information, the satellite channel and semantic codec are jointly trained end-to-end to ensure the robustness of the communication system.

[0028] 3) It can better ensure the performance of image transmission and lower bandwidth resource occupancy under low signal-to-noise ratio conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 The present invention is a flowchart of a Beidou satellite-based image semantic communication method in a smart grid according to an embodiment of the present invention.

[0030] Figure 2 This is a structural diagram of a semantic encoder of an image semantic communication method based on Beidou satellite in a smart grid in an embodiment of the present invention.

[0031] Figure 3 It is the peak signal-to-noise ratio of image transmission under different signal-to-noise ratios in an image semantic communication method based on Beidou satellite in a smart grid in an embodiment of the present invention.

[0032] Figure 4 It is a multi-scale structural similarity of image transmission under different signal-to-noise ratios in an image semantic communication method based on Beidou satellite in a smart grid in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The present invention is further described in detail below in conjunction with the embodiments and drawings.

[0034] The present invention proposes an image semantic communication method based on Beidou satellite in a smart grid; the present invention constructs an end-to-end communication scenario in which a smart grid in a remote area transmits data through a Beidou satellite; a semantic encoder is used to extract semantic features of grid image data to reduce the amount of transmitted data; then, the satellite channel is modeled to simulate the actual communication scenario; after receiving the signal, the receiving end semantically decodes the semantic features to restore the original image; finally, the semantic codec is combined with the satellite channel for end-to-end training to complete the image transmission.

[0035] like Figure 1 As shown, the specific steps are as follows:

[0036] Step 1: Take the image data of smart grids in remote areas as the input S of the system, and input the data into the semantic encoder network to extract the semantic features T∈R of the data, which can be expressed as:

[0037] T = E(α; S);

[0038] Where E(α; ·) represents a semantic encoder network with parameter α.

[0039] Specifically, the semantic encoder consists of a convolutional layer (ConvLayer) and a normalization layer (GDN). The convolutional layer is responsible for extracting the semantic features of the data and reducing the dimension; the normalization layer is responsible for stabilizing the learning process of the network. Taking four convolutional layers as an example, we assume that the convolution kernel size of each convolutional layer is 5, the stride is 2, and the number of output channels of the last convolutional layer is 128. Input image S∈R 256×256×3 (image height 256, image width 256, number of channels 3), the semantic feature T∈R output after passing through the semantic encoder 128×16×16 .

[0040] Step 2: Adjust the semantic features to complex signals; simulate the satellite channel.

[0041] The specific steps are as follows:

[0042] Step 201, adjusting the semantic feature to a complex signal;

[0043] Considering the use of complex channels to transmit semantic features, it is necessary to convert semantic features into complex signals. First, the semantic features are expanded in one dimension, where every two are used as the imaginary part and real part of the complex signal respectively:

[0044]

[0045] X={X1,X2…,X K} is the complex signal transmitted.

[0046] Step 202, simulating the satellite channel;

[0047] We consider the communication between BeiDou-3 (BD3) satellite in geostationary orbit and ground-based smart grid communication terminals. Due to the strong line-of-sight (LoS) component between the terminal and the satellite, the channel gain H between the terminal and the satellite is modeled as a Rayleigh channel and can be calculated by the following formula:

[0048]

[0049] Where κ is the Rayleigh factor, H′=e jφ and H″~CN(0,1) represent the LoS and NLoS components, respectively. φ~U[-π,π] is a uniform random variable representing the phase offset due to the mobility and propagation delay of the satellite and the receiver.

[0050] Large-scale fading and loss consists of three parts: L dist , L shad, L angl Indicates that L dist is the power loss caused by the distance between the receiver and the satellite, expressed as:

[0051]

[0052] Where d is the distance between the receiving end and the satellite. f and c are the carrier frequency and the speed of light, respectively. shad Represents shadow attenuation, which can be expressed by the following formula:

[0053]

[0054] L angl is the loss caused by the pitch angle and can be calculated by the following formula:

[0055]

[0056] Among them, where θ is the elevation angle and η is the antenna factor that determines the coverage radius. The final large-scale fading can be written as: Furthermore, we assume that propagation delay and Doppler shift have been compensated during the time and frequency synchronization process. This is a reasonable assumption since they depend mainly on the position and velocity of the satellite and the user, which can be determined using satellite orbit information. In addition, other advanced techniques can be used to compensate for Doppler shift in real time and achieve symbol synchronization without a priori satellite information.

[0057] Step 3: After receiving the data, the receiver needs to restore the transmitted data Adjust the dimension of the recovered semantic features to obtain And input to the semantic decoder to get the original transmission data:

[0058]

[0059] Where D(β; ·) represents a semantic encoder network with parameter β.

[0060] Step 4: Train the designed communication system. In order to improve the robustness of the system and reduce the interference of channel uncertainty on the data, an end-to-end training method of the joint satellite channel is adopted. During the training process, the neural network parameter α is first initialized. (0) and β (0) The semantic encoder extracts semantic features, which are then encoded, transmitted, and decoded. Afterwards, the loss is calculated and the parameters are iteratively updated using stochastic gradient descent. Training continues until a termination condition is met, such as reaching the maximum number of iterations. Finally, the outputs E(α; ·) and D(β; ·) are obtained.

[0061] For the choice of loss function, the mean square error (MSE) is considered as the loss function, which aims to minimize the average square difference between the estimated value and the true value, thereby ensuring pixel-level accuracy. The loss function can be expressed as follows:

[0062]

[0063] Where I is the total dimension of the image data, is the mean squared error distortion, and n is the dimension of the input image.

[0064] Step 5: Use the trained system to complete the transmission of smart grid image data.

[0065] The data set used in the present invention is a substation data set obtained from actual power grid inspections, which contains approximately 8,300 images, which are divided into training and test sets in a ratio of 7:3. In order to ensure that the input image size is consistent, the image is preprocessed to a resolution of pixels. In the experiment, we set the Beidou satellite to a height of d = 35,786m from the ground, the carrier center frequency to f = 1561.098MHz, and the antenna factor η = 20. Throughout the training process, we used the Adam optimizer with a learning rate of 0.0001, a beta of 0.05, a batch size of 16, and a training cycle of 30. All training was performed on an NVIDIA GeForceRTX 4090GPU.

[0066] The image semantic communication method based on Beidou satellite in the smart grid designed by the present invention is compared with the traditional image transmission scheme based on JPEG: when transmitting the image, the image is compressed by JPEG, then channel encoded by LDPC, and finally transmitted to the receiving end through the constructed satellite channel using the 16QAM channel coding scheme. In order to verify the image transmission performance of the present invention, the peak signal-to-noise ratio (PSNR) and multi-scale structural similarity (MS-SSIM) are used as performance indicators.

[0067] The present invention proposes an image semantic communication method based on Beidou satellite in smart grid. The system first uses a semantic encoder to extract features to extract the features of the power grid image, then converts the features into complex-valued data and transmits them to the receiving end through a simulated satellite channel, and then uses a semantic decoder to restore the output image.

[0068] The PSNR of image transmission under different signal-to-noise ratios between the comparison scheme and the present invention is: Figure 3are given. Under low SNR conditions, the proposed scheme significantly outperforms the baseline scheme in terms of PSNR performance, and under high SNR conditions, it performs about the same as the baseline scheme. In addition, we observe that the proposed method does not exhibit the cliff effect seen in the baseline scheme under low SNR conditions. This phenomenon occurs because once the channel coding and modulation scheme are selected for a specific SNR, the number of bits used for compression is fixed. Therefore, the reconstructed image quality does not improve as the SNR increases. In addition, when the channel quality is lower than the target SNR, the increasing bit error rate cannot be handled, resulting in a significant degradation in image quality.

[0069] In order to analyze the bandwidth resource usage of the proposed method, we evaluate the image transmission performance under different k / n (bandwidth compression ratio). Figure 3 It can be clearly seen that the proposed method achieves comparable PSNR performance to the baseline scheme when k / n=1 / 24, while the baseline scheme achieves similar performance when k / n=1 / 14. This shows that the proposed scheme can achieve similar performance to the baseline scheme with 71.43% bandwidth resource reduction. The lower bandwidth resource occupancy also proves that the proposed scheme can increase the data transmission volume by 71.43% with the same bandwidth resources.

[0070] Figure 4 The MS-SSIM of image transmission under different signal-to-noise ratios is shown. We can see that the MS-SSIM performance curve is Figure 3 The PSNR curve trends in are similar. Our method is still better than the MS-SSIM of the baseline scheme. Although the baseline scheme still experiences the cliff effect under low SNR conditions, our method is able to maintain a certain level of image transmission performance.

[0071] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A Beidou satellite-based image semantic communication method in a smart grid, characterized in that: The amount of transmitted data is reduced by extracting the semantic features of the data, and the Beidou satellite is used for data transmission. The satellite channel and the semantic codec are combined for end-to-end training to reduce the impact of transmission uncertainty on semantic information, including the following steps: Step 1: Take the image data of smart grids in remote areas as the input S of the system, and input the data into the semantic encoder network to extract the semantic features T∈R of the data, which can be expressed as: T = E(α; S); Where E(α; ·) represents the semantic encoder network with parameter α; Step 2: After obtaining the semantic features T of the data, the semantic features are transmitted at the transmitter. Before actual transmission, T is shaped and adjusted to a complex signal X∈C: X = Reshape(T); The transmitted signal is transmitted to the receiving end through the satellite channel. The received signal at the receiving end can be expressed as: Y = HX + n; Where H is the channel matrix, n is Gaussian white noise; Step 3: After receiving the data, the receiver needs to restore the transmitted data Adjust the dimension of the recovered semantic features to obtain And input to the semantic decoder to get the original transmission data: Where D(β; ) represents a semantic decoder network with parameter β; Step 4: Train the designed communication system using an end-to-end training method of a joint satellite channel; During the training process, the neural network parameters α are first initialized (0) and β (0) ,The semantic encoder extracts semantic features, then encodes, transmits and decodes them. After that, the loss is calculated and the parameters are iteratively updated using the stochastic gradient descent method. The training continues until the termination condition is met. During the training process, the mean square error (MSE) is selected as the loss function. The loss function can be expressed as follows: Where I is the total dimension of the image data; Step 5: Use the trained system to complete the transmission of smart grid image data.

Images