Satellite internet task planning method based on intention understanding and large model

Abstract

The invention relates to a satellite internet task planning method based on intention understanding and a large model, which belongs to the field of task planning and comprises the following steps: constructing a satellite internet intention library and a satellite internet knowledge tensor library based on RAG enhanced retrieval; receiving a communication intention demand report, and performing intention acceptance according to a user communication intention; performing RAG enhanced retrieval through the accepted communication intention and the knowledge tensor database to obtain a tensor retrieval result, and performing reasoning to obtain an intention analysis result; communication intention verification is carried out according to the intention analysis result and a satellite internet intention library, if legality is met, execution is continued, and otherwise, the process is ended and abnormal warning is carried out; task planning and scheduling are carried out through a satellite internet intention library and an equipment and resource library, and a network planning parameter result is obtained; and then network parameter configuration is carried out on the application layer to realize satellite planning, station planning and terminal planning. According to the method, the total resource occupancy rate and the service capacity can be effectively improved, and meanwhile load balancing is achieved.

Description

Technical Field

[0001] The present invention relates to a satellite Internet mission planning method based on intent understanding and a large model in the field of satellite Internet, and is particularly suitable for mission requirements of complex networking in a large-scale satellite Internet system. Background Art

[0002] In recent years, AI algorithms represented by large model technology have made breakthrough progress and have been effectively applied in various vertical fields. In the field of communications, some domestic and foreign companies have carried out relevant research on mobile communication large models, trying to build a communication large model integrated with 6G networks, but in the field of satellite Internet mission planning, especially in the field of high-orbit and low-orbit integrated satellite Internet mission planning, research on large models has not yet been made public.

[0003] When the network scale is small, network management focuses on detailed configurations such as network parameters and terminal configurations. As the network scale expands, satellite mission planning and scheduling are generally used to achieve overall control of different tasks and network systems. However, in satellite Internet, as the network scale continues to expand and the business dimension continues to expand, traditional satellite communication mission planning based on greedy, heuristic, reinforcement learning and other methods requires a higher level of abstraction, that is, a mission planning method for intent understanding. Large model technology is naturally capable of understanding and parsing intent inputs, so large model technology can be used to parse satellite Internet mission planning for intent understanding. In addition, due to the complexity of satellite Internet scenarios, the application of existing mission planning algorithms has great limitations. Therefore, it is necessary to innovate the alignment of satellite Internet expertise with large models, and also to study satellite Internet mission planning methods for intent understanding.

[0004] Therefore, in order to meet the demand for larger network scale management and control in future satellite Internet systems, a satellite Internet mission planning method based on intent understanding and large models is needed, compared with the traditional satellite communication mission planning method that cannot adapt to complex network configurations. Summary of the invention

[0005] To solve the above problems, the present invention proposes a satellite Internet mission planning method based on intent understanding and large models. In the field of satellite Internet management and control, the present invention realizes the system's management and control needs for a larger network scale, and can understand the communication intent based on the large model or further perform mission planning, which solves the problem of low efficiency of traditional mission planning for large-scale constellation management, improves the overall utilization efficiency of satellite resources, and improves user service quality.

[0006] The present invention adopts the following technical solution:

[0007] A satellite Internet mission planning method based on intent understanding and large models includes the following steps:

[0008] Step 1: construct a satellite Internet intent library and a satellite Internet knowledge tensor library based on RAG enhanced retrieval; the satellite Internet intent library contains a mission intent template, which consists of a mission type and a planning parameter type, and is used to match and verify the intent analysis result with the mission intent template; the satellite Internet knowledge tensor library is used to perform RAG enhanced retrieval on communication intent;

[0009] Step 2: receiving the user's communication intention demand submission, expressed in natural language, and the task and planning scheduling system accepts the user's communication intention; the intention acceptance is to confirm whether the demand is responded to;

[0010] Step 3: Perform RAG enhanced retrieval on the accepted communication intent and the satellite internet knowledge tensor database to obtain tensor retrieval results, and perform reasoning through the satellite internet expert big model to obtain intent analysis results; the satellite internet expert big model is obtained by incremental fine-tuning training of the pre-trained open source big model with satellite internet professional related data;

[0011] Step 4: According to the intent analysis result and the satellite Internet intent library, the legitimacy of the user's communication intent is verified to determine whether the intent analysis result meets the legitimacy. If so, step 5 is executed. If not, the communication intent task planning is terminated and an abnormal alarm is issued. The intent verification is the legitimacy verification of the user's communication intent, which is used to pre-verify the task planning conflict.

[0012] Step 5: By matching the intent analysis results with the mission intent template in the satellite Internet intent library, the intent analysis results are converted into corresponding mission types and planning parameters, and the mission is planned and scheduled through the equipment and resource library to obtain the network planning parameter results;

[0013] Step 6: According to the network planning parameter results, configure the network parameters of the application layer to implement satellite planning, station planning and terminal planning.

[0014] Furthermore, in step 1:

[0015] The task types in the mission intent template are tasks related to the operation and control of the satellite Internet network, including communication accompaniment tasks, beam scheduling tasks, resource adjustment tasks, networking tasks and feed tracking tasks; the planning parameter types are parameters related to the operation and control of the satellite Internet network, including network type parameters, satellite parameters, transponder bandwidth parameters and beam working parameters.

[0016] Furthermore, the construction process of the satellite internet knowledge tensor library described in step 1 is as follows:

[0017] Collect satellite internet data and perform knowledge tensorization on the data. That is, use an embedded tensorization model to convert text data into tensor representation. The embedded tensorization model is trained based on the Transformer Encoder architecture.

[0018] Then build a database that supports HNSW and IVF index structures, and add a KNN plug-in to support tensor search;

[0019] Finally, different collections are created in the database to store different types of tensor data in the satellite Internet field, including technical documents, protocol interfaces, network logs, and network traffic.

[0020] Furthermore, the specific process of step 3 is as follows:

[0021] The accepted communication intention is encoded into a communication intention tensor through the Embedding model; then the tensor search is performed through RAG enhanced retrieval to search for the data block closest to the communication intention tensor, and the tensor search result is obtained. The tensor search result contains multiple blocks of data with the most similar feature distance to the communication intention tensor; the tensor search is specifically as follows: the feature relationship between the communication intention tensor and the satellite Internet knowledge tensor database is calculated using the Cross Encoder reordering model;

[0022] Then, the communication intention and tensor retrieval results are used as inputs of the satellite Internet expert big model, and the big model is inferred to obtain the intention analysis results; the intention analysis results include communication accompaniment intention, beam scheduling intention, resource adjustment intention, networking intention and feed tracking intention; the intention analysis results include multiple sub-task sets that require task planning and planning parameter analysis results.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] The satellite Internet task planning method based on intent understanding and large models proposed in the present invention aims at the characteristics of larger network control scale and high and low orbit integration in the satellite Internet system. By implementing the analysis and understanding of communication intent based on the large model, the communication intent is decomposed into subtasks and further task planning is performed, which solves the problem of low efficiency of traditional task planning for large-scale constellation management. It can effectively improve the total resource occupancy rate and service capacity while achieving load balancing, avoiding waste and conflict of on-board resources of the satellite Internet system, and thus improving the user service quality Qos. In addition, the satellite Internet task planning based on intent is studied to realize intelligent analysis of tasks, and the traditional task planning method of manually converting communication requirements or manually specifying the network type required for the task is upgraded to intelligent automatic task planning based on communication intent. Therefore, the original "artificial intent understanding" is upgraded to "people in bypass, full automatic analysis and planning". BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 The present invention is a flowchart of a satellite Internet mission planning method based on intent understanding and large models in one embodiment.

[0026] Figure 2 A schematic diagram of system relationships in one embodiment of a satellite Internet mission planning method based on intent understanding and a large model of the present invention. DETAILED DESCRIPTION

[0027] The technical solution of the present invention will be described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0028] like Figure 1 and Figure 2 As shown, the present invention provides a satellite Internet mission planning method based on intention understanding and a large model, and the method comprises the following steps:

[0029] Step 1: Construct a satellite Internet intent library and a satellite Internet knowledge tensor library based on RAG enhanced retrieval.

[0030] The satellite Internet intent library contains a mission intent template, which consists of a mission type and a planning parameter type, and is used to match and verify the intent analysis result with the mission intent template; the mission type includes communication companion tasks, beam scheduling tasks, resource adjustment tasks, networking tasks, feed tracking tasks and other tasks involving satellite Internet network operation and control; the planning parameter type includes network type parameters, satellite parameters, transponder bandwidth parameters, beam working parameters and other parameters involving satellite Internet network operation and control.

[0031] The satellite internet knowledge tensor library is used to perform RAG enhanced retrieval of communication intentions; the satellite internet knowledge tensor library construction process is:

[0032] Collect satellite internet data and perform knowledge tensorization on the data. That is, use an embedded tensorization model to convert text data into tensor representation. The embedded tensorization model is trained based on the Transformer Encoder architecture.

[0033] Then build a database that supports HNSW and IVF index structures, and add a KNN plug-in to support tensor search;

[0034] Finally, different collections are created in the database to store different types of tensor data in the satellite Internet field, including technical documents, protocol interfaces, network logs, and network traffic.

[0035] Step 2: Receive the user's communication intention demand submission, expressed in natural language, and the task and planning scheduling system accepts the user's communication intention; the intention acceptance is manually confirmed in a one-click manner whether the demand is responded to.

[0036] Step 3: Perform RAG enhanced retrieval on the accepted communication intent and the satellite internet knowledge tensor database to obtain tensor retrieval results, and perform reasoning through the satellite internet expert big model to obtain intent analysis results; the satellite internet expert big model is obtained by incremental fine-tuning training of the pre-trained open source big model with satellite internet professional related data;

[0037] The specific process of step 3 is as follows:

[0038] The accepted communication intention is encoded into a communication intention tensor through the Embedding model; then the tensor search is performed through RAG enhanced retrieval to search for the data block closest to the communication intention tensor, and the tensor search result is obtained. The tensor search result contains multiple blocks of data with the most similar feature distance to the communication intention tensor; the tensor search is specifically as follows: the feature relationship between the communication intention tensor and the satellite Internet knowledge tensor database is calculated using the Cross Encoder reordering model;

[0039] Then, the communication intention and tensor retrieval results are used as inputs of the satellite Internet expert big model, and the big model is inferred to obtain the intention analysis results; the intention analysis results include communication accompaniment intention, beam scheduling intention, resource adjustment intention, networking intention and feed tracking intention; the intention analysis results include multiple sub-task sets that require task planning and planning parameter analysis results.

[0040] Step 4: According to the intent analysis result and the satellite Internet intent library, the communication intent is verified to determine whether the intent analysis result meets the legality; the specific process is:

[0041] According to the intention analysis result and the satellite Internet intention library, the legitimacy of the user's communication intention is verified to determine whether the intention analysis result meets the legitimacy. If so, step 5 is executed; if not, the communication intention task planning is terminated and an abnormal alarm is issued; the intention verification is the legitimacy verification of the user's communication intention, which is used to pre-verify the task planning conflicts to avoid conflicts between the subtask set in the intention analysis result obtained in step 3 and the existing tasks of the system.

[0042] Step 5: Implement task matching based on intent analysis results, perform task planning and scheduling through the satellite Internet intent library and the equipment and resource library to obtain network planning parameter results;

[0043] The task matching according to the intent analysis result refers to the process of converting the intent analysis result into the corresponding task type and network planning parameters described in step 1 by matching the intent analysis result with the task intent template of the satellite Internet intent library.

[0044] Step 6: According to the network planning parameter results, configure the network parameters of the application layer to implement satellite planning, station planning and terminal planning.

[0045] (s6): According to the network planning parameter results, the network parameters of the application layer are configured to realize satellite planning, station planning and terminal planning.

[0046] The satellite Internet task planning method based on intent understanding and large model proposed in this embodiment uses a large model based on satellite Internet experts to realize the analysis and understanding of communication intent. In view of the larger scale of network control and the integration of high and low orbits in the satellite Internet system, the communication intent is decomposed into subtasks and further task planning is carried out, which solves the problem of low efficiency of traditional task planning for large-scale constellation management, can effectively improve the total resource occupancy rate and business capacity, and realize load balancing, avoid waste and conflict of on-board resources of the satellite Internet system, and thus improve the user service quality QoS.

[0047] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A satellite Internet mission planning method based on intent understanding and large models, characterized in that: The following steps are involved: Step 1: construct a satellite Internet intent library and a satellite Internet knowledge tensor library based on RAG enhanced retrieval; the satellite Internet intent library contains a mission intent template, which consists of a mission type and a planning parameter type, and is used to match and verify the intent analysis result with the mission intent template; the satellite Internet knowledge tensor library is used to perform RAG enhanced retrieval on communication intent; Step 2: receiving the user's communication intention demand submission, expressed in natural language, and the task and planning scheduling system accepts the user's communication intention; the intention acceptance is to confirm whether the demand is responded to; Step 3: Perform RAG enhanced retrieval on the accepted communication intent and the satellite internet knowledge tensor database to obtain tensor retrieval results, and perform reasoning through the satellite internet expert big model to obtain intent analysis results; the satellite internet expert big model is obtained by incremental fine-tuning training of the pre-trained open source big model with satellite internet professional related data; Step 4: According to the intent analysis result and the satellite Internet intent library, the legitimacy of the user's communication intent is verified to determine whether the intent analysis result meets the legitimacy. If so, step 5 is executed. If not, the communication intent task planning is terminated and an abnormal alarm is issued. The intent verification is the legitimacy verification of the user's communication intent, which is used to pre-verify the task planning conflict. Step 5: By matching the intent analysis results with the mission intent template in the satellite Internet intent library, the intent analysis results are converted into corresponding mission types and planning parameters, and the mission is planned and scheduled through the equipment and resource library to obtain the network planning parameter results; Step 6: According to the network planning parameter results, configure the network parameters of the application layer to implement satellite planning, station planning and terminal planning.

2. According to claim 1, a satellite Internet mission planning method based on intention understanding and large model is characterized in that: In step 1: The task types in the mission intent template are tasks related to the operation and control of the satellite Internet network, including communication accompaniment tasks, beam scheduling tasks, resource adjustment tasks, networking tasks and feed tracking tasks; the planning parameter types are parameters related to the operation and control of the satellite Internet network, including network type parameters, satellite parameters, transponder bandwidth parameters and beam working parameters.

3. The satellite Internet mission planning method based on intention understanding and large model according to claim 1 is characterized in that: The construction process of the satellite internet knowledge tensor library described in step 1 is: Collect satellite internet data and perform knowledge tensorization on the data. That is, use an embedded tensorization model to convert text data into tensor representation. The embedded tensorization model is trained based on the Transformer Encoder architecture. Then build a database that supports HNSW and IVF index structures, and add a KNN plug-in to support tensor search; Finally, different collections are created in the database to store different types of tensor data in the satellite Internet field, including technical documents, protocol interfaces, network logs, and network traffic.

4. According to claim 3, a satellite Internet mission planning method based on intention understanding and large model is characterized in that: The specific process of step 3 is as follows: The accepted communication intention is encoded into a communication intention tensor through the Embedding model; then the tensor search is performed through RAG enhanced retrieval to search for the data block closest to the communication intention tensor, and the tensor search result is obtained. The tensor search result contains multiple blocks of data with the most similar feature distance to the communication intention tensor; the tensor search is specifically as follows: the feature relationship between the communication intention tensor and the satellite Internet knowledge tensor database is calculated using the Cross Encoder reordering model; Then, the communication intention and tensor retrieval results are used as inputs of the satellite Internet expert big model, and the big model is inferred to obtain the intention analysis results; the intention analysis results include communication accompaniment intention, beam scheduling intention, resource adjustment intention, networking intention and feed tracking intention; the intention analysis results include multiple sub-task sets that require task planning and planning parameter analysis results.

Images