A method for automatic satellite command generation based on self-attention mechanism

Abstract

The present invention discloses a method for automatically generating satellite commands based on a self-attention mechanism, belonging to the field of satellite data processing technology. The method comprises the following steps: collecting satellite command interference data and processing it to obtain a command interference assessment value; obtaining a satellite command sequence, converting the satellite command sequence into a command representation matrix, encoding the command sequence position, and adding the position code to the command representation matrix to generate an command sequence representation matrix; using a scaled dot product attention component to integrate global information into the command sequence representation matrix to obtain an updated command sequence representation matrix; outputting an interval time series based on the updated command sequence representation matrix through a fully connected neural network, and correcting the interval time series based on the command interference assessment value. This method solves the problems of low efficiency and insufficient accuracy of traditional methods when dealing with complex task command sequences.

Description

Technical Field

[0001] The present invention relates to the field of satellite data processing technology, and in particular to a method for automatically generating satellite instructions based on a self-attention mechanism. Background Art

[0002] In modern aerospace engineering, the instruction sequence interval specifies the time interval between two instructions. A reasonable and effective interval can improve instruction quality and reduce the total time cost of the mission. Therefore, automatic generation of instruction sequence intervals is one of the core technologies in spacecraft mission planning.

[0003] Existing satellite command automatic generation methods are mainly implemented through various means such as meta-task decomposition and structural variables, basic instruction codes and verification, metadata models and databases, and satellite autonomous mission planning technology to improve the efficiency and accuracy of command generation.

[0004] For example, the patent application with publication number CN117632430A discloses a method for autonomously generating agile satellite imaging instruction sequences based on mapping tables, which includes: matching the task system mapping table, the business instruction mapping table, and the time constraint mapping table according to the basic task information, business information, and time information of the current task in sequence; obtaining the constraints for executing the current task; and generating a satellite imaging task instruction sequence according to the basic task information, system, execution content, and constraints, wherein the satellite imaging task instruction sequence is used to instruct the system to execute the current task according to the execution content under the constraints.

[0005] For example, the patent application with publication number CN118276941A discloses a method, apparatus, device and storage medium for generating an instruction sequence, comprising: receiving an instruction generation operation command; wherein the instruction generation operation command includes an instruction category; calling an instruction generation architecture according to the instruction generation operation command; calling an instruction set corresponding to the instruction category from the instruction generation architecture; and generating an instruction sequence according to the order in which instructions in the instruction set are selected.

[0006] However, in the process of implementing the technical solutions of the invention in the embodiments of the present application, the present application found that the above technology has at least the following technical problems:

[0007] In the existing technology, under the resource-constrained satellite software environment, data processing capabilities are limited, resulting in low efficiency and insufficient accuracy when facing complex mission instruction sequences. Summary of the Invention

[0008] The present invention provides a method for automatically generating satellite commands based on a self-attention mechanism, which solves the problem in the prior art of limited data processing capabilities under resource-constrained satellite software environments, resulting in low efficiency and insufficient accuracy when faced with complex task command sequences, and achieves significant optimization of timeliness, availability and time overhead.

[0009] The present invention provides a method for automatically generating satellite commands based on a self-attention mechanism, comprising the following steps: collecting satellite command interference data and obtaining a command interference evaluation value through processing; obtaining a satellite command sequence, converting the satellite command sequence into a command representation matrix, encoding the command sequence position, and adding the position code to the command representation matrix to generate an command sequence representation matrix; using a scaled dot product attention component to fuse the command sequence representation matrix with global information to obtain an updated command sequence representation matrix; outputting an interval time series through a fully connected neural network based on the updated command sequence representation matrix, and correcting the interval time series according to the command interference evaluation value.

[0010] Furthermore, the steps of collecting satellite command interference data and obtaining a command interference evaluation value through processing include: the satellite command interference data includes satellite equipment data, satellite attitude evaluation value and command interference data; obtaining a satellite equipment evaluation value based on a comprehensive analysis of the satellite equipment data, and performing adjustment feedback based on the satellite equipment evaluation value; obtaining a satellite attitude evaluation value based on a comprehensive analysis of the satellite attitude data; and obtaining a command interference evaluation value based on a comprehensive analysis of the command interference data, the satellite attitude evaluation value and the satellite equipment evaluation value.

[0011] Furthermore, the step of obtaining a satellite equipment evaluation value based on a comprehensive analysis of satellite equipment data includes: the satellite equipment data includes transmission power, receiving sensitivity and bit error rate; obtaining critical transmission power, critical receiving sensitivity and critical bit error rate from an instruction generation database; and obtaining a satellite equipment evaluation value through comprehensive analysis.

[0012] Furthermore, the step of obtaining the instruction interference evaluation value through comprehensive analysis of instruction interference data, satellite attitude evaluation value and satellite equipment evaluation value includes: the instruction interference data includes electromagnetic interference intensity and the number of adjacent signals; obtaining the critical electromagnetic interference intensity, critical number of adjacent signals, critical satellite attitude evaluation value and critical satellite equipment evaluation value from the instruction generation database; and obtaining the instruction interference evaluation value through comprehensive analysis.

[0013] Furthermore, the instruction interference evaluation value is obtained as follows:

[0014]

[0015] Wherein, ξI represents the command interference evaluation value, α4 represents the command interference evaluation influence factor corresponding to the electromagnetic interference intensity, α5 represents the command interference evaluation influence factor corresponding to the number of adjacent signals, α6 represents the command interference evaluation influence factor corresponding to the satellite equipment evaluation value, α7 represents the command interference evaluation influence factor corresponding to the satellite attitude evaluation value, E1 represents the electromagnetic interference intensity, E0 represents the critical electromagnetic interference intensity, C1 represents the number of adjacent signals, C0 represents the critical number of adjacent signals, ξE1 represents the satellite equipment evaluation value, ξE0 represents the critical satellite equipment evaluation value, ξA1 represents the satellite attitude evaluation value, ξA0 represents the critical satellite attitude evaluation value, and e is a natural constant.

[0016] Furthermore, the step of converting the satellite instruction sequence into an instruction representation matrix includes: inputting the satellite instruction sequence and marking each satellite instruction with a digital number; using one-hot encoding to generate a corresponding feature representation for each satellite instruction, marking it as a high-dimensional feature vector; and mapping each high-dimensional feature vector to the required low-dimensional space to obtain an embedding matrix, which is the instruction representation matrix.

[0017] Furthermore, the instruction sequence representation matrix is ​​fused with global information using the scaled dot product attention component to obtain the updated instruction sequence representation matrix. The steps include: performing three linear transformations on the instruction sequence representation matrix to obtain a query matrix, a key matrix, and a value matrix, which are marked as q, k, and v respectively; obtaining three weight matrices to be learned, which are marked as W respectively. q 、W k and W v ; Use the weight matrix to map the input to the query matrix, key matrix and value matrix, and obtain the corresponding query embedding matrix, key embedding matrix and value embedding matrix by performing matrix multiplication with the instruction sequence representation matrix, which are marked as Q, K and V respectively; According to the query embedding matrix, key embedding matrix and value embedding matrix, the updated instruction sequence representation matrix is ​​obtained.

[0018] Furthermore, the step of obtaining an updated instruction sequence representation matrix based on the query embedding matrix, the key embedding matrix and the value embedding matrix includes: calculating the dot product of q and k to obtain an attention score matrix; scaling the attention score matrix and then normalizing it to obtain a normalized attention weight matrix, labeled A; multiplying the attention weight matrix A with the value embedding matrix V, and the output matrix obtained is the updated instruction sequence representation matrix, labeled Z.

[0019] Furthermore, according to the matrix represented by the updated instruction sequence, the step of outputting the interval time series through the fully connected neural network includes: passing the input matrix Z through the first intermediate layer of the fully connected neural network, performing a linear transformation and applying an activation function to obtain a matrix Z1; passing Z1 through the second intermediate layer of the fully connected neural network, also performing a linear transformation and applying an activation function to obtain a matrix Z2; and linearly transforming Z2 to obtain an output vector with a dimension of n-1, which is the interval time series.

[0020] Furthermore, the step of correcting the interval time series according to the instruction interference evaluation value includes: matching the instruction interference evaluation value with the interval time series correction vector corresponding to each instruction interference evaluation value interval preset in the instruction generation database to obtain the interval time series correction vector; summing the interval time series and the interval time series correction vector to obtain the corrected interval time series.

[0021] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0022] 1. The present invention provides a satellite command automatic generation method based on a self-attention mechanism, which can efficiently collect and process satellite command interference data, obtain accurate command interference evaluation values, and correct the command interval time series based on the command interference evaluation values, thereby achieving improved accuracy and efficiency in satellite command generation.

[0023] 2. By utilizing one-hot encoding and multi-layer perception mechanisms, the present invention can effectively reduce dimensionality and overcome sparsity processing while retaining the important features of the original instruction sequence, thereby achieving efficient conversion of feature representation.

[0024] 3. The present invention adds the position code and adds the instruction representation matrix to the position code embedding matrix, so that the generated instruction sequence representation matrix contains the characteristics of each instruction and the positional relationship between the instructions in the instruction sequence, thereby improving the accuracy of the generated results. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A flowchart of a method for automatically generating satellite instructions based on a self-attention mechanism is provided in an embodiment of the present application.

[0026] Figure 2 A diagram showing changes in satellite equipment evaluation values ​​based on a method for automatically generating satellite instructions according to an embodiment of the present application.

[0027] Figure 3 This is a structural diagram of the method for automatically generating satellite instructions based on the self-attention mechanism provided in an embodiment of the present application. DETAILED DESCRIPTION

[0028] The embodiment of the present application provides a method for automatically generating satellite commands based on a self-attention mechanism, which solves the problem in the prior art that, in a resource-constrained satellite software environment, data processing capabilities are limited, resulting in low efficiency and insufficient accuracy when facing complex task command sequences. Satellite command interference data is collected and processed to obtain a command interference evaluation value; a satellite command sequence is obtained, converted into a command representation matrix, the command sequence position is encoded, and the position code is added to the command representation matrix to generate an command sequence representation matrix; the command sequence representation matrix is ​​fused with global information using a scaled dot product attention component to obtain an updated command sequence representation matrix; an interval time series is output through a fully connected neural network based on the updated command sequence representation matrix, and the interval time series is corrected according to the command interference evaluation value, thereby achieving significant optimization of timeliness, availability, and time overhead.

[0029] The technical solution in the embodiments of the present application is to solve the problem of limited data processing capability under resource-constrained satellite software environment, which leads to low efficiency and insufficient accuracy when facing complex task instruction sequences. The overall idea is as follows:

[0030] By collecting satellite command interference data and processing it to obtain the command interference evaluation value; obtaining the satellite command sequence, converting the satellite command sequence into a command representation matrix, encoding the command sequence position, and adding the position code to the command representation matrix to generate the command sequence representation matrix; using the scaled dot product attention component to fuse the command sequence representation matrix with global information to obtain the updated command sequence representation matrix; based on the updated command sequence representation matrix, a fully connected neural network is used to output the interval time series, and the interval time series is corrected according to the command interference evaluation value, thereby achieving accurate prediction of the interval time series.

[0031] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0032] like Figure 1 、 Figure 3 As shown, a satellite instruction automatic generation method based on the self-attention mechanism is provided in an embodiment of the present application. The method includes the following steps: collecting satellite instruction interference data and obtaining an instruction interference evaluation value through processing; obtaining a satellite instruction sequence, converting the satellite instruction sequence into an instruction representation matrix, encoding the instruction sequence position, and adding the position code to the instruction representation matrix to generate an instruction sequence representation matrix; using a scaled dot product attention component to fuse the instruction sequence representation matrix with global information to obtain an updated instruction sequence representation matrix, outputting an interval time series through a fully connected neural network according to the updated instruction sequence representation matrix, and correcting the interval time series according to the instruction interference evaluation value.

[0033] In this embodiment, the present invention generates an optimized interval time sequence for a given instruction sequence through a series of processing steps based on instruction interval constraints. First, the instruction sequence is encoded and embedded to ensure that the position and context information of each instruction in the sequence are retained and transmitted. Then, through the scaled dot product attention mechanism, the system can effectively evaluate the relationship between instructions and accurately calculate the most appropriate time interval between each pair of instructions. Finally, the system performs interval time prediction. This step ultimately generates the optimal interval time sequence that maximizes time utilization, under the premise of ensuring that all constraints are met. This sequence not only ensures that all instructions can be executed normally as expected, but also has the shortest time interval between every two instructions, thereby maximizing time utilization. The present invention not only improves the efficiency of instruction execution, but also effectively reduces the waste of resources caused by excessively long instruction intervals.

[0034] In addition, the instruction generation database is used to store data related to the automatic generation method of satellite instructions based on the self-attention mechanism, including: critical transmission power, critical receiving sensitivity, critical bit error rate, satellite equipment evaluation impact factors corresponding to receiving sensitivity, and interval time series correction vectors corresponding to the interference evaluation value intervals of each instruction. The data in the instruction generation database can be obtained through satellite databases or websites such as NASA's satellite database, the European Space Agency's database, etc., and can also be obtained through professional space segment monitoring equipment or systems, such as satellite orbit monitoring stations, space debris monitoring networks, etc.

[0035] Furthermore, the steps of collecting satellite command interference data and obtaining a command interference evaluation value through processing include: the satellite command interference data includes satellite equipment data, satellite attitude evaluation value and command interference data; obtaining a satellite equipment evaluation value based on a comprehensive analysis of the satellite equipment data, and performing adjustment feedback based on the satellite equipment evaluation value; obtaining a satellite attitude evaluation value based on a comprehensive analysis of the satellite attitude data; and obtaining a command interference evaluation value based on a comprehensive analysis of the command interference data, the satellite attitude evaluation value and the satellite equipment evaluation value.

[0036] In this embodiment, satellite attitude data describes the attitude of the satellite relative to the Earth, including attitude angular velocity (the rate at which the satellite attitude angle changes over time), pitch angle (the angle at which the satellite rotates around the Earth's transverse axis (perpendicular to the orbital plane)), and yaw angle (the angle at which the satellite rotates around the Earth's longitudinal axis (orbital direction)). The pitch angle, yaw angle, and attitude angular velocity can all be monitored by attitude measurement instruments (such as infrared attitude measurement instruments, star cameras, gyroscopes, etc.). Command interference data may cause changes in satellite attitude, thereby affecting the accuracy of attitude assessment values. Command interference data may also reflect the performance and status of satellite equipment. Degraded equipment performance or failure may also cause instability or deviation in satellite attitude, thereby affecting the accuracy of attitude assessment values. Therefore, there is a mutual influence between the three, and a comprehensive analysis is required to accurately assess the health status of satellite equipment. By comprehensively analyzing satellite equipment data, satellite attitude data, and command interference data, the performance of satellite equipment and the impact of command interference can be comprehensively considered from multiple dimensions. At the same time, the data can be mutually verified to ensure the reliability and accuracy of the assessment results. In addition, when adjusting feedback based on the satellite equipment evaluation value, the satellite equipment evaluation threshold can be obtained from the instruction generation database, and the satellite equipment evaluation value can be compared with the satellite equipment evaluation threshold. When the satellite equipment evaluation value is less than the satellite equipment evaluation threshold, it means that the current satellite equipment performance status is poor, and it is necessary to immediately check the hardware and software configuration of the satellite equipment. If there is a problem with the hardware, consider replacing the damaged parts or upgrading the hardware to improve performance. When there is a problem with the software configuration, adjust the software configuration of the device according to the recommendations in the evaluation report, such as updating the firmware, optimizing parameter settings, etc. During the adjustment process, continuously monitor the performance of the device to ensure the effectiveness of the adjustment measures; when the satellite equipment evaluation value is greater than or equal to the satellite equipment evaluation threshold, no additional processing is performed.

[0037] When the satellite attitude evaluation value is obtained based on the comprehensive analysis of the satellite attitude data, it is necessary to first obtain the reference attitude angular velocity, the allowable deviation attitude angular velocity, the reference pitch angle, the allowable deviation pitch angle, the reference yaw angle, and the allowable deviation yaw angle from the instruction generation database. The satellite attitude evaluation value is obtained as follows:

[0038]

[0039] Where ξA1 represents the satellite attitude evaluation value, β1 represents the satellite attitude evaluation influence factor corresponding to the attitude angular velocity, β2 represents the satellite attitude evaluation influence factor corresponding to the pitch angle, β3 represents the satellite attitude evaluation influence factor corresponding to the yaw angle, A1 represents the attitude angular velocity, A0 represents the reference attitude angular velocity, A2 represents the allowable deviation attitude angular velocity, PA1 represents the pitch angle, PA0 represents the reference pitch angle, PA2 represents the allowable deviation pitch angle, Y1 represents the yaw angle, Y0 represents the reference yaw angle, and Y2 represents the allowable deviation yaw angle.

[0040] β1, β2, and β3 represent the numerical values ​​of the degree of influence of the attitude angular velocity, pitch angle, and yaw angle on the satellite attitude evaluation value, respectively, and can be directly obtained from the instruction generation database when used. For example, the attitude angular velocity and the satellite attitude evaluation influence factors corresponding to the attitude angular velocity preset in the instruction generation database form a mapping set, and the satellite attitude evaluation influence factors corresponding to the attitude angular velocity are obtained based on the attitude angular velocity input mapping set; the pitch angle and the satellite attitude evaluation influence factors corresponding to the pitch angle preset in the instruction generation database form a mapping set, and the satellite attitude evaluation influence factors corresponding to the pitch angle are obtained based on the pitch angle input mapping set; the yaw angle and the satellite attitude evaluation influence factors corresponding to the yaw angle preset in the instruction generation database form a mapping set, and the satellite attitude evaluation influence factors corresponding to the yaw angle are obtained based on the yaw angle input mapping set. The mapping relationship is a many-to-one or one-to-one relationship, and in this embodiment, the value range of the influence factor is between 0 and 1.

[0041] Furthermore, the step of obtaining a satellite equipment evaluation value based on a comprehensive analysis of satellite equipment data includes: the satellite equipment data includes transmission power, receiving sensitivity and bit error rate; obtaining critical transmission power, critical receiving sensitivity and critical bit error rate from an instruction generation database; and obtaining a satellite equipment evaluation value through comprehensive analysis.

[0042] The satellite equipment evaluation value is obtained as follows:

[0043]

[0044] Where ξE1 represents the satellite equipment evaluation value, α1 represents the satellite equipment evaluation impact factor corresponding to the transmit power, α2 represents the satellite equipment evaluation impact factor corresponding to the receive sensitivity, α3 represents the satellite equipment evaluation impact factor corresponding to the bit error rate, P1 represents the transmit power, P0 represents the critical transmit power, S1 represents the receive sensitivity, S0 represents the critical receive sensitivity, and B1 represents the bit error rate. In practical applications, due to the complexity and uncertainty of various factors, the bit error rate cannot be 0. B0 represents the critical bit error rate.

[0045] α1, α2, and α3 are satellite device evaluation impact factors corresponding to transmit power, receive sensitivity, and bit error rate preset in the instruction generation database, respectively. They represent the numerical values ​​of the degree of influence of transmit power, receive sensitivity, and bit error rate on the satellite device evaluation value, and can be directly obtained from the instruction generation database when used. For example, transmit power and the satellite device evaluation impact factors corresponding to transmit power preset in the instruction generation database form a mapping set, and the satellite device evaluation impact factor corresponding to the transmit power is obtained by inputting the transmit power mapping set; receive sensitivity and the satellite device evaluation impact factors corresponding to receive sensitivity preset in the instruction generation database form a mapping set, and the satellite device evaluation impact factor corresponding to the receive sensitivity is obtained by inputting the receive sensitivity mapping set; bit error rate and the satellite device evaluation impact factors corresponding to the bit error rate preset in the instruction generation database form a mapping set, and the satellite device evaluation impact factor corresponding to the bit error rate is obtained by inputting the bit error rate mapping set. The mapping relationship is a many-to-one or one-to-one relationship, and in this embodiment, the impact factor ranges from 0 to 1.

[0046] In this embodiment, the transmission power refers to the power of electromagnetic waves emitted by the satellite antenna into the surrounding space, and its value can be directly measured using a radio frequency power meter. The receiving sensitivity can be obtained by using a signal generator to generate a signal with known power and frequency, which is then received by the satellite's receiving antenna. A spectrum analyzer or network analyzer is used at the receiving end to measure the received signal power. The power of the signal generator is adjusted until the signal can be just detected at the receiving end. The power at this time is the receiving sensitivity of the satellite. Generally speaking, a smaller receiver sensitivity indicates a lower minimum signal power that the receiver can correctly receive, resulting in a higher satellite equipment evaluation value. However, in the satellite domain, since receiver sensitivity is a negative number, the larger the absolute value of the receiver sensitivity, the higher the satellite equipment evaluation value. The bit error rate (BER) can be measured by establishing a signal transmission link between the satellite's measurement and control subsystem and the BER test system. The satellite's downlink telemetry mode is set to BER test mode, outputting a downlink telemetry signal with a fixed telemetry frame data field. The telemetry demodulation module demodulates the downlink telemetry signal and outputs the telemetry frame to the BER test module. The BER test module automatically acquires reference data for comparison and compares the received telemetry frame data against the reference frame data for consistency. The BER is calculated by counting the number of bit errors and the ratio of the total number of transmitted bits. All other conditions remaining unchanged, an increase in transmit power generally results in an increase in signal strength at the receiver, thereby reducing the BER. Similarly, an increase in receiver sensitivity can also reduce the BER because it allows the receiver to detect weaker signals, reducing errors caused by insufficient signal strength. Transmit power, receive sensitivity, and bit error rate are three parameters that fully reflect the communication performance of satellite equipment. By comprehensively analyzing these three parameters, we can evaluate the satellite equipment's transmission capacity, reception capacity, and communication quality.

[0047] The satellite equipment evaluation impact factor corresponding to the transmit power is set to 0.2, the satellite equipment evaluation impact factor corresponding to the receive sensitivity is set to 0.4, the satellite equipment evaluation impact factor corresponding to the bit error rate is set to 0.4, the transmit power is set to 2500W, the critical transmit power is set to 2000W, the receive sensitivity is set to -100dBm, the critical receive sensitivity is set to -120dBm, and the critical bit error rate is set to 1.0*e -6 , under the condition that the bit error rate continues to increase, the satellite equipment evaluation value is calculated. As shown in Table 1, the satellite equipment evaluation value data table based on the satellite command automatic generation method.

[0048] Table 1 Satellite equipment evaluation value data based on the satellite command automatic generation method

[0049] serial number <![CDATA[B1]]> <![CDATA[ξE1]]> 1 <![CDATA[1.0*e -8 ]]> 14.381 2 <![CDATA[1.0*e -7 ]]> 2.110 3 <![CDATA[1.0*e -6 ]]> 0.894 4 <![CDATA[1.0*e -5 ]]> 0.580 5 <![CDATA[1.0*e -4 ]]> 0.475

[0050] like Figure 2As shown in Table 1 and Figure 2 It can be seen that when the satellite equipment evaluation influence factor corresponding to the transmit power, the satellite equipment evaluation influence factor corresponding to the receiving sensitivity, the satellite equipment evaluation influence factor corresponding to the bit error rate, the transmit power, the critical transmit power, the receiving sensitivity, the critical receiving sensitivity and the critical bit error rate remain unchanged, and the bit error rate continues to increase, the satellite equipment evaluation value continues to decrease.

[0051] Furthermore, the step of obtaining the instruction interference evaluation value through comprehensive analysis of instruction interference data, satellite attitude evaluation value and satellite equipment evaluation value includes: the instruction interference data includes electromagnetic interference intensity and the number of adjacent signals; obtaining the critical electromagnetic interference intensity, critical number of adjacent signals, critical satellite attitude evaluation value and critical satellite equipment evaluation value from the instruction generation database; and obtaining the instruction interference evaluation value through comprehensive analysis.

[0052] In this embodiment, the electromagnetic interference intensity can be obtained by directly measuring the satellite command receiving point using an interference monitoring receiver or oscilloscope. The greater the electromagnetic interference intensity, the more destructive the interference signal is to the satellite communication signal, which may lead to problems such as decreased communication quality and increased bit error rate. The number of adjacent signals refers to the number of signals adjacent to a specific signal in a certain frequency band or communication system. These adjacent signals may come from different communication sources or devices and may have spectrum overlap or interference with the target signal. The number of adjacent signals can be obtained by monitoring the current satellite command frequency band using radio spectrum monitoring equipment. The greater the number of adjacent signals, the more severe the potential adjacent-channel interference or co-channel interference, which will further affect the stability and reliability of the satellite communication signal. The higher the satellite equipment evaluation value, the better the performance of the satellite equipment, the stronger its anti-interference capability, and the better it can cope with electromagnetic interference and adjacent signal interference. By comprehensively considering the electromagnetic interference intensity, the number of adjacent signals, and the satellite equipment evaluation value, the interference situation of the satellite communication system in a specific environment can be comprehensively assessed. At the same time, it can more accurately reflect the performance of the satellite communication system in actual operation, providing strong support for optimizing system configuration and improving communication quality.

[0053] Furthermore, the instruction interference evaluation value is obtained as follows:

[0054]

[0055] Wherein, ξI represents the command interference evaluation value, α4 represents the command interference evaluation influence factor corresponding to the electromagnetic interference intensity, α5 represents the command interference evaluation influence factor corresponding to the number of adjacent signals, α6 represents the command interference evaluation influence factor corresponding to the satellite equipment evaluation value, α7 represents the command interference evaluation influence factor corresponding to the satellite attitude evaluation value, E1 represents the electromagnetic interference intensity, E0 represents the critical electromagnetic interference intensity, C1 represents the number of adjacent signals, C0 represents the critical number of adjacent signals, ξE1 represents the satellite equipment evaluation value, ξE0 represents the critical satellite equipment evaluation value, ξA1 represents the satellite attitude evaluation value, ξA0 represents the critical satellite attitude evaluation value, and e is a natural constant.

[0056] In this embodiment, α4, α5, α6 and α7 are the instruction interference assessment impact factors corresponding to the electromagnetic interference intensity, number of adjacent signals, satellite equipment assessment value and satellite attitude assessment value preset in the instruction generation database, respectively representing the numerical values ​​of the degree of influence of the electromagnetic interference intensity, number of adjacent signals, satellite equipment assessment value and satellite attitude assessment value on the instruction interference assessment value, and can be directly obtained from the instruction generation database when used. For example, the electromagnetic interference intensity and the instruction interference assessment impact factor corresponding to the electromagnetic interference intensity preset in the instruction generation database form a mapping set, and the instruction interference assessment impact factor corresponding to the electromagnetic interference intensity is obtained by inputting the mapping set according to the electromagnetic interference intensity; the number of adjacent signals and the instruction interference assessment impact factor corresponding to the number of adjacent signals preset in the instruction generation database form a mapping set, and the instruction interference assessment impact factor corresponding to the number of adjacent signals is obtained by inputting the mapping set according to the number of adjacent signals; the satellite equipment evaluation value and the instruction interference assessment impact factor corresponding to the satellite equipment evaluation value preset in the instruction generation database form a mapping set, and the instruction interference assessment impact factor corresponding to the satellite equipment evaluation value is obtained by inputting the mapping set according to the satellite equipment evaluation value; the satellite attitude evaluation value and the instruction interference assessment impact factor corresponding to the satellite attitude evaluation value preset in the instruction generation database form a mapping set, and the instruction interference assessment impact factor corresponding to the satellite attitude evaluation value is obtained by inputting the mapping set according to the satellite attitude evaluation value, wherein the mapping relationship is a many-to-one or one-to-one relationship, and the value range of the impact factor in this embodiment is between 0 and 1.

[0057] Furthermore, the step of converting the satellite instruction sequence into an instruction representation matrix includes: inputting the satellite instruction sequence and marking each satellite instruction with a digital number; using one-hot encoding to generate a corresponding feature representation for each satellite instruction, marking it as a high-dimensional feature vector; and mapping each high-dimensional feature vector to the required low-dimensional space to obtain an embedding matrix, which is the instruction representation matrix.

[0058] In this embodiment, each satellite instruction is numerically labeled with a number from 1 to n for easier representation. One-hot encoding is a technique for converting categorical variables into a binary matrix. Each instruction is represented as a binary vector of length n, with only one position set to 1 and the remaining positions set to 0. The embedding matrix generated by mapping each high-dimensional feature vector into the desired low-dimensional space can be used for feature learning using a multilayer perceptron. A multilayer perceptron is a feedforward artificial neural network that, through a series of weighted input, activation functions, and output layers, can capture and extract complex feature relationships. Here, the multilayer perceptron maps the high-dimensional feature vectors generated by one-hot encoding into the desired low-dimensional space. Through the multilayer perceptron processing, an embedding matrix is ​​generated that contains the essential features of each instruction. The dimensionality of the embedding matrix is ​​typically lower than that of the original one-hot encoding, but it effectively represents the unique characteristics of each instruction. This not only effectively reduces the dimensionality of the data, but also preserves the important features of the original instruction sequence. The generation of a dense representation matrix significantly improves storage and computational efficiency, providing high-quality input data for subsequent deep learning processing.

[0059] For example, if task M contains N instructions, the one-hot encoding process generates a feature vector of length N for each instruction. One-hot encoding embedding can effectively avoid numerical relationships between instructions, providing effective data for subsequent deep learning algorithms. After one-hot encoding, the entire instruction sequence is converted into an instruction representation matrix of size n×N. Although the instruction representation matrix can effectively represent the instructions, the vectors generated by the one-hot encoding embedding are sparse, which may lead to low storage and computational efficiency. To overcome the sparsity, a multilayer perceptron is introduced to further learn and process the embedding results. The multilayer perceptron maps the high-dimensional, sparse one-hot encoding vector into a low-dimensional, dense space through a series of linear transformations and nonlinear activation functions. Specifically, the multilayer perceptron transforms the n×N sparse one-hot encoding embedding matrix into a dense representation matrix of size n×d, where d is the length of the feature vector after dimensionality reduction.

[0060] To ensure that the subsequent self-attention mechanism can accurately capture the position information in the sequence, position encoding is introduced after generating the instruction representation matrix. In order to accurately represent the position of the instruction in the sequence, sine and cosine functions are used to encode the information of each position. The method of introducing position encoding is as follows:

[0061]

[0062] Among them, PE i Represents the position encoding of the vector of instruction i, T represents the linear transformation matrix, ω d Represents the encoding frequency of dimension d, where the value range of d is model represents the latitude of the model embedding vector.

[0063] In this way, a positional encoding embedding matrix of size n×d is generated, in which each position has a unique encoding, ensuring that the position information is evenly distributed across different dimensions. The generated positional encoding embedding matrix has the same dimension n×d as the previously obtained instruction representation matrix. The introduction of the positional encoding matrix ensures that each instruction contains not only its feature information but also its position information in the sequence. The generated positional encoding embedding matrix is ​​added to the instruction representation matrix to obtain the final instruction sequence representation matrix. Since each input vector calculates the inner product with other vectors in the subsequent self-attention mechanism, the order information of the original input sequence is lost. By adding positional information to each instruction in this step, the instruction sequence representation matrix can distinguish the task features in the instruction embedding, and the model's perception of sequence order is enhanced, enabling it to more accurately capture the relationship and dependency between instructions when processing instruction sequences.

[0064] Furthermore, the instruction sequence representation matrix is ​​fused with global information using the scaled dot product attention component to obtain the updated instruction sequence representation matrix. The steps include: performing three linear transformations on the instruction sequence representation matrix to obtain a query matrix, a key matrix, and a value matrix, which are marked as q, k, and v respectively; obtaining three weight matrices to be learned, which are marked as W respectively. q 、W k and W v ; Use the weight matrix to map the input to the query matrix, key matrix and value matrix, and obtain the corresponding query embedding matrix, key embedding matrix and value embedding matrix by performing matrix multiplication with the instruction sequence representation matrix, which are marked as Q, K and V respectively; According to the query embedding matrix, key embedding matrix and value embedding matrix, the updated instruction sequence representation matrix is ​​obtained.

[0065] The query matrix q represents the vector of relevant information that needs to be obtained from other positions in the sequence, which is used to calculate the attention weight of the current instruction on other instructions. The key matrix k represents the feature vector of each instruction in the sequence, which is used to perform a dot product with the query vector, that is, to determine the degree of attention each instruction pays to other instructions, and then calculate the attention weight. The value matrix v also represents the feature vector of each instruction in the sequence, which is used for weighted summation to generate the final feature representation. Under the action of the attention weight, the features of the corresponding instruction are emphasized or suppressed. W q 、W k and W v is the weight matrix to be learned, with sizes of d q ×d,d k ×d,d v ×d. These weight matrices are used to map the input to a matrix of dimension d q The query q and dimension dk The key k and dimension d v The value v, where d q =d k By performing matrix multiplication with the input instruction sequence representation, we obtain the corresponding query, key, and value embedding matrices Q, K, and V.

[0066] In this embodiment, the self-attention mechanism allows attention to instructions at all other positions in the input sequence when computing the representation of a particular instruction. This feature can capture dependencies between distant elements in the input sequence, rather than relying solely on local information. At the same time, by calculating attention weights, the weight of each instruction can be dynamically assigned, allowing for more attention to be paid to instructions with higher relevance during computation. Its key idea is to calculate the attention weight of each position in the sequence with respect to other positions using a query, key, and value matrix, thereby performing a weighted sum. Therefore, in the second step of generating the interval time for the instruction sequence, a scaled dot product attention component is introduced to associate different positions in the sequence to update the representation of the sequence.

[0067] Furthermore, the step of obtaining an updated instruction sequence representation matrix based on the query embedding matrix, the key embedding matrix and the value embedding matrix includes: calculating the dot product of q and k to obtain an attention score matrix; scaling the attention score matrix and then normalizing it to obtain a normalized attention weight matrix, labeled A; multiplying the attention weight matrix A with the value embedding matrix V, and the output matrix obtained is the updated instruction sequence representation matrix, labeled Z.

[0068] In this embodiment, the updated instruction sequence representation matrix Z not only contains the characteristics and location information of each instruction, but also reflects the global dependencies. By introducing the scaled dot product attention mechanism, the system can more effectively capture the long-range dependencies in the instruction sequence, thus laying a solid foundation for generating the optimized interval time series. The full process formula of the above steps is as follows:

[0069]

[0070] Where Z is the updated instruction sequence representation matrix, q is the query matrix, k is the key matrix, T is the scaling factor of the attention score matrix, and d is the k is the normalized latitude, V is the value embedding matrix, and softmax() is the softmax function. The dimension of the output matrix Z is n×d v , where each query vector performs a weighted sum operation on the value vector according to its corresponding attention weight, and the resulting output vector contains global information about the instruction sequence.

[0071] Furthermore, according to the matrix represented by the updated instruction sequence, the step of outputting the interval time series through the fully connected neural network includes: passing the input matrix Z through the first intermediate layer of the fully connected neural network, performing a linear transformation and applying an activation function to obtain a matrix Z1; passing Z1 through the second intermediate layer of the fully connected neural network, also performing a linear transformation and applying an activation function to obtain a matrix Z2; and linearly transforming Z2 to obtain an output vector with a dimension of n-1, which is the interval time series.

[0072] In this embodiment, the ReLU activation function is a commonly used activation function in artificial neural networks. It defines a simple nonlinear transformation that maps input values ​​to output values. The instruction sequence interval prediction module receives the instruction sequence representation matrix that integrates global information, passes it through a three-layer fully connected neural network, and finally completes the prediction of the time interval sequence. The fully connected neural network consists of two intermediate layers and an output layer. v The matrix of size is gradually converted into the final prediction output. During this process, the weights and biases of each layer are optimized through training to minimize the prediction error. The ReLU activation function of each layer introduces nonlinearity, allowing the prediction part to capture more complex feature relationships. The specific conversion formula of the input matrix Z through the first and second intermediate layers of the fully connected neural network is as follows:

[0073] Z1 = ReLU(Z*W1+b1);

[0074] Z2 = ReLU(Z1*W2+b2);

[0075] Where W1 and W2 are weight matrices, and b1 and b2 are bias vectors, all obtained through training. Matrix Z1 is n × d1, and matrix Z2 is n × d2. The intermediate feature representation is obtained by transforming the two intermediate layers and applying activation functions.

[0076] The mathematical expression of the interval time series is as follows:

[0077] Output = Z2*W3+b3;

[0078] Here, Output is the interval time series, Z2 is the output matrix of the second intermediate layer, W3 is a weight matrix of size d2×(n-1), and b3 is the bias vector. The output layer converts the intermediate features into the final prediction results, and the final prediction output is obtained through linear transformation. By combining one-hot encoding embedding, multi-layer perceptron, and scaled dot product attention mechanisms, the instruction sequence time interval generation method ensures the rationality of the final generated time intervals. The entire process fully considers the inherent characteristics of each instruction in the instruction sequence and the order information of the sequence. Through instruction representation learning and self-attention mechanisms, the system can reasonably arrange the time intervals of the instruction sequence, ensuring smooth and efficient task completion.

[0079] Furthermore, the step of correcting the interval time series according to the instruction interference evaluation value includes: matching the instruction interference evaluation value with the interval time series correction vector corresponding to each instruction interference evaluation value interval preset in the instruction generation database to obtain the interval time series correction vector; summing the interval time series and the interval time series correction vector to obtain the corrected interval time series.

[0080] In this embodiment, the correction vector is a set of values ​​determined based on a preset instruction interference evaluation value interval and a corresponding correction strategy. The instruction interference evaluation value reflects the degree of interference that may occur during instruction execution. By matching the evaluation value with the preset interval and applying the corresponding correction vector, the impact of interference on the interval time series can be more accurately reflected, thereby improving the accuracy of the corrected interval time series; at the same time, the impact of these uncertain factors on the interval time series can be reduced, making the system more stable and reliable.

[0081] In summary, the embodiments of the present application collect satellite command interference data and obtain a command interference evaluation value through processing; obtain a satellite command sequence, convert the satellite command sequence into a command representation matrix, encode the command sequence position, and add the position code to the command representation matrix to generate an instruction sequence representation matrix; use the scaled dot product attention component to fuse the instruction sequence representation matrix with global information to obtain an updated instruction sequence representation matrix; according to the updated instruction sequence representation matrix, output an interval time series through a fully connected neural network, and correct the interval time series according to the command interference evaluation value, thereby achieving significant optimization of timeliness, availability, and time overhead.

[0082] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0083] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts 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, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0084] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0086] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0087] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for automatically generating satellite instructions based on a self-attention mechanism, characterized in that: The following steps are involved: Collect satellite command interference data and obtain command interference evaluation value after processing; Obtaining a satellite command sequence, converting the satellite command sequence into a command representation matrix, encoding the command sequence position, and adding the position code to the command representation matrix to generate the command sequence representation matrix; The scaled dot product attention component is used to fuse the instruction sequence representation matrix with global information to obtain an updated instruction sequence representation matrix. Based on the updated instruction sequence representation matrix, a fully connected neural network is used to output the interval time series, and the interval time series is corrected according to the instruction interference evaluation value. The step of using the scaled dot product attention component to fuse the instruction sequence representation matrix with global information to obtain an updated instruction sequence representation matrix includes: The instruction sequence representation matrix is ​​transformed into query matrix, key matrix and value matrix through three linear transformations, which are marked as q, k and v respectively; Get three weight matrices that need to be learned, marked as 、 and ; The weight matrix is ​​used to map the input to the query matrix, key matrix and value matrix. The corresponding query embedding matrix, key embedding matrix and value embedding matrix are obtained by matrix multiplication with the instruction sequence representation matrix, which are marked as Q, K and V respectively. The updated instruction sequence representation matrix is ​​obtained based on the query embedding matrix, key embedding matrix, and value embedding matrix, including: Calculate the dot product of q and k to get the attention score matrix; After scaling the attention score matrix, normalize it to obtain the normalized attention weight matrix, marked as A; Multiply the attention weight matrix A by the value embedding matrix V, and the resulting output matrix is ​​the updated instruction sequence representation matrix, marked as Z.

2. The method for automatically generating satellite commands based on the self-attention mechanism according to claim 1, characterized in that: The step of collecting satellite command interference data and obtaining a command interference evaluation value through processing includes: The satellite command interference data includes satellite equipment data, satellite attitude data and command interference data; Obtain satellite equipment evaluation values ​​based on comprehensive analysis of satellite equipment data, and provide adjustment feedback based on the satellite equipment evaluation values; The satellite attitude evaluation value is obtained based on comprehensive analysis of satellite attitude data; The command interference evaluation value is obtained by comprehensive analysis of the command interference data, satellite attitude evaluation value and satellite equipment evaluation value.

3. The method for automatically generating satellite commands based on the self-attention mechanism according to claim 2, characterized in that: The step of obtaining the satellite equipment evaluation value based on the comprehensive analysis of the satellite equipment data comprises: The satellite equipment data, including transmit power, receive sensitivity and bit error rate; Obtaining critical transmission power, critical receiving sensitivity, and critical bit error rate from a command generation database; Comprehensive analysis is performed to obtain the satellite equipment evaluation value.

4. The method for automatically generating satellite commands based on the self-attention mechanism according to claim 2, wherein: The step of obtaining the command interference evaluation value by comprehensive analysis of the command interference data, the satellite attitude evaluation value and the satellite equipment evaluation value comprises: The instruction interference data includes electromagnetic interference intensity and the number of adjacent signals; obtaining a critical electromagnetic interference intensity, a critical number of adjacent signals, a critical satellite attitude evaluation value, and a critical satellite equipment evaluation value from a command generation database; Comprehensive analysis is performed to obtain the instruction interference evaluation value.

5. The method for automatically generating satellite commands based on the self-attention mechanism according to claim 4, characterized in that: The instruction interference evaluation value is obtained as follows: ； Where, represents the instruction interference evaluation value, Indicates the instruction interference assessment impact factor corresponding to the electromagnetic interference intensity, Indicates the instruction interference evaluation impact factor corresponding to the number of adjacent signals, Indicates the command interference assessment impact factor corresponding to the satellite equipment assessment value, Indicates the command interference assessment impact factor corresponding to the satellite attitude assessment value, Indicates the electromagnetic interference intensity, Indicates the critical electromagnetic interference intensity, Indicates the number of adjacent signals, represents the number of critical adjacent signals, Indicates the satellite equipment evaluation value, represents the critical satellite equipment assessment value, represents the satellite attitude evaluation value, represents the critical satellite attitude evaluation value, and e is a natural constant.

6. The method for automatically generating satellite commands based on the self-attention mechanism according to claim 1, characterized in that: The step of converting the satellite command sequence into a command representation matrix comprises: Enter the satellite command sequence and mark each satellite command with a number; Use one-hot encoding to generate corresponding feature representations for each satellite command, marked as high-dimensional feature vectors; The embedding matrix obtained by mapping each high-dimensional feature vector to the required low-dimensional space is the instruction representation matrix.

7. The method for automatically generating satellite commands based on the self-attention mechanism according to claim 1, characterized in that: The step of outputting the interval time series through a fully connected neural network according to the updated instruction sequence representation matrix includes: The input matrix Z is passed through the first intermediate layer of the fully connected neural network, linearly transformed and activated, and the matrix is ​​obtained. ; Will Through the second middle layer of the fully connected neural network, the same linear transformation is performed and the activation function is applied to obtain the matrix ; Will Perform a linear transformation to obtain an output vector with a dimension of n-1, which is an interval time series.

8. The method for automatically generating satellite commands based on the self-attention mechanism according to claim 1, wherein: The step of correcting the interval time series according to the instruction interference evaluation value includes: Matching the instruction interference evaluation value with the interval time series correction vector corresponding to each instruction interference evaluation value interval preset in the instruction generation database to obtain the interval time series correction vector; The interval time series is summed with the interval time series correction vector to obtain the corrected interval time series.

Images