CN117118855A - Data link SPMA access method based on machine learning priority prediction - Google Patents

Abstract

The invention discloses a data link SPMA access method based on machine learning priority prediction, and belongs to the technical field of battlefield network communication. The invention prioritizes the service according to the service attribute when the unknown service arrives, and ensures the transmission of the high-priority service. The SPMA protocol is adopted to ensure the transmission quality of the high-priority service, and the SPMA protocol function is verified aiming at the service with the self-adaptive allocation priority. The invention uses the method of combining machine learning and SPMA to predict service priority according to the service attributes such as bandwidth, time delay and the like on different data link nodes, and uses the priority result obtained by prediction to access the nodes by utilizing the SPMA protocol in the TTNT data link network, thereby improving the stability and reliability of communication in the data communication method based on the SPMA access protocol and guaranteeing the quality of data communication.

Description

Data link SPMA access method based on machine learning priority prediction

Technical Field

The invention belongs to the technical field of battlefield network communication, and particularly relates to a data link SPMA access method based on machine learning priority prediction.

Background

The TTNT data link is a novel battlefield network technology, is a high-speed dynamic Ad-Hoc network, has no fixed infrastructure between mobile nodes, can move freely, and can mutually send and receive messages through multi-hop wireless links. The TTNT data link adopts a higher-order modulation-demodulation mode, a higher coding and decoding rate and an omnidirectional antenna switched at a high speed. Under the technology, the data throughput of the system is improved by approximately 20 times compared with Link-16, the sensitive target which moves rapidly can be positioned more accurately, and the anti-interference performance of the TTNT system is extremely strong. This technique provides a wide range of benefits, particularly in defense and government applications. It not only provides a low latency, ad-Hoc, IP-based network function, but also has the ability to "self-form" and "self-heal" itself. The TTNT data chain also allows for immediate and secure sharing of voice, video and data transmissions of the types described above, for secure communications between high speed flying aircraft, for military-based applications. The real advantage of TTNT is the priority of the data during transmission, employing advanced statistical priority based multiple access (SPMA) protocols to ensure that critical data is always sent first, and that all lower priority data is prevented from being transmitted until needed.

The SPMA access protocol has a special operation mechanism, and compared with other MAC protocols, the SPMA access protocol provides better system performance and ensures service quality. The conventional MAC access protocol lacks adaptability to dynamic changes of network topology, and often needs to spend a lot of resources to solve the problems of network synchronization and time slot allocation. However, in the SPMA protocol, the node determines whether to send a data packet by comparing the channel load rate with each priority threshold, and resources are not required to allocate time slots and reserve channels for the node. In addition, it ensures high-priority data packets with high delivery rate and low delay, and such a mechanism meets the service requirements of different emergency degrees in the remote and has better performance. Thus, with the rapid development of tactical network communications, SPMA is receiving increasing attention. At present, the demand of the tactical data link in China is continuously increasing, so in order to improve the combat environment and ensure the task requirements and improve the military combat practice as soon as possible, the core technology in the tactical data link must be continuously improved and optimized.

Therefore, how to predict service priority according to the requirements of the service on bandwidth, time delay and the like on different data link nodes, and using the priority result obtained by prediction for accessing the nodes by using the SPMA protocol in the TTNT data link network, thereby improving the stability and reliability of communication in the data communication method based on the SPMA access protocol, guaranteeing the quality of data communication, and being the technical problem to be solved by the technical personnel in the field at present.

Disclosure of Invention

Aiming at the problem that the SPMA protocol service priority in a TTNT data chain needs to be manually divided, the invention provides the capability of reasonably dividing the priority of unknown services in a complex network environment, thereby improving the stability and reliability of communication in a data communication method based on an SPMA access protocol.

In order to achieve the above purpose, the present invention adopts the following technical scheme, and is a data link SPMA access method based on machine learning priority prediction, the method comprises the following steps:

step 1: establishing a business attribute model;

the higher the security level s of the service, the higher the priority of the service;

the maximum time delay of the service allowable transmission is delay_m, and the service priority is lower as the maximum time delay of the service allowable transmission is larger;

the service demand bandwidth B, the larger the service demand bandwidth is, the higher the service priority is;

service data size l, the larger the service data size is, the higher the service priority is;

the service type comprises voice, data, images and situation awareness, and the priority of the service type is increased from left to right at one time;

step 2: training a machine learning model;

step 2.1: the method for calculating the base-Ni index of the training set data comprises the following steps:

wherein C is _k Is a subset of samples belonging to class K, K being the number of classes;

step 2.2: comparing the base index obtained by calculation in the step 2.1 with a set threshold value, randomly adjusting samples in training set data if the base index is larger than the threshold value, and returning to the step 2.1; otherwise, enter step 2.3;

step 2.3: according to the training data set obtained in the step 2.2, a service attribute model is used as the characteristic input of the random forest model, the service priority is used as the output of the random forest model, and the self-adaptive distribution model of the service priority is obtained through training of the existing data; the random forest model is defined as:

y＝TreeGenerate(D,A) (1)

wherein a= (s, delay_m, B, l, type) is the feature input of the random forest model; d is training set data; y is the classification result, namely the service priority;

step 3: priority self-adaptive allocation;

aiming at the service priority self-adaptive allocation model obtained through random forest model training in the step 2, when an unknown service arrives, allocating priority to the unknown service according to service attributes (s, delay_m, B, l, type) of the unknown service;

step 4: protocol verification;

step 4.1: when a network has data packets arriving, the data packets enter different priority queues according to different priorities;

step 4.2: firstly checking whether the queue with the highest priority is empty or not, and if so, checking the next priority queue; otherwise, the head data packet of the queue with the highest priority is sent to a sending judging stage;

step 4.3: when a data packet enters a sending judging stage, judging whether the data packet is expired or not, and if not, comparing a priority threshold of the data packet with current channel occupation statistics; otherwise, the data packet is taken out of the queue, and step 4.2 is carried out;

step 4.4: if the current channel occupation statistics is smaller than the priority threshold of the data packet, the data packet is dequeued and sent out; otherwise, the data packet is retracted for a period of time;

step 5: and (5) ending.

The invention provides a data link SPMA access method based on machine learning priority prediction. The invention selects the specific service attribute including the service security level, the maximum time delay allowed to be transmitted, the required bandwidth, the data size and the service type as the characteristic input of the random forest model by utilizing the relation between the service attribute and the service priority, and the service priority self-adaptive distribution model finally obtained by training a large number of training sets can carry out priority distribution according to the unknown service attribute. The traditional MAC access protocol lacks adaptability to dynamic changes of a network topology structure, and a large amount of resources are often required to solve the problems of network synchronization, time slot allocation and the like, so that in order to ensure high-priority data packet transmission with high delivery rate and low delay and the requirements of services with different emergency degrees, the invention adopts an SPMA access method to solve the problems. The invention combines the machine learning priority prediction with the SPMA access method, predicts the service priority according to the requirements of the service on bandwidth, time delay and the like on different data link nodes, uses the priority result obtained by prediction for accessing the nodes by utilizing the SPMA protocol in the TTNT data link network, further improves the stability and reliability of communication in the data communication method based on the SPMA access protocol, and ensures the quality of data communication. The method provided by the invention has the advantages that the accuracy rate of the priority allocation of the unknown service can reach about 85%, the transmission success rate of the high-priority service can reach 99%, the transmission delay in 100 sea is less than 2ms, and the established TTNT data link network has good dynamic characteristics and can be rapidly accessed into/removed from the network. The method provided by the invention has certain advantages in theory and practical application, and provides a direction for the study of the data link.

Drawings

Fig. 1 is a TTNT tactical data chain network simulation platform architecture.

FIG. 2 is a schematic diagram of a connection of two platforms in a simulation platform.

Fig. 3 is a TTNT data chain protocol architecture.

Fig. 4 is a SPMA packet transmission decision.

Fig. 5 is a SPMA access flow chart.

Fig. 6 is a SPMA operating state transition diagram.

FIG. 7 is a graph of a random forest algorithm priority prediction result;

(a) The relation between the service priority prediction accuracy and different parameters of the random forest model is obtained;

(b) The relation between the service priority prediction accuracy and different sampling sizes of the random forest model is obtained;

(c) Comparing the real priority with the predicted priority.

Fig. 8 is a basic structure of a TTNT simulation module.

FIG. 9 is a diagram of network simulation parameters.

FIG. 10 is a simulation result of three service types of SPMA;

(a) Time delay is three service types;

(b) The throughput is three service types;

(c) System throughput for three traffic types.

FIG. 11 is a simulation result of five service types of SPMA;

(a) Time delay is five service types;

(b) The throughput is five service types;

(c) System throughput for five traffic types.

Detailed Description

And building a machine learning environment by using the tensorflow, and training a random forest model to predict the service priority. And using an NS-3 network simulation tool to perform TTNT data chain networking and SPMA access method simulation, sending the service priority parameters predicted by the random forest model to the NS-3, and establishing corresponding network topology by the NS-3 to acquire network simulation parameters. The joint simulation platform architecture and the connection relation are shown in fig. 1 and fig. 2. The method comprises the following specific steps:

step one: random forest model training

And building a machine learning environment by using tensorsurface, training a random forest model, and predicting service priority through service attributes. In the embodiment of the invention, five business attributes are selected as the characteristic inputs of the random forest model: the higher the security level of the service, the higher the priority of the service; maximum time delay of the allowed transmission of the service, and the service has lower priority when the maximum time delay of the allowed transmission of the service is larger; the service demand band, the larger the service demand bandwidth is, the higher priority is given to the service; the service data size is larger, and the service has higher priority; traffic types include voice, data, image, situational awareness, etc., with priorities rising once from left to right.

Step two: NS-3 network simulation setup

In order to verify the performance of TTNT data link networking and SPMA access methods, an NS-3 network simulation tool is required to be used for simulation. The TTNT data link protocol architecture is shown in fig. 3, according to which a corresponding network model is built in NS-3. The SPMA access method has the biggest characteristic of ensuring the transmission of high-priority service, wherein the data packet sending judgment is shown in fig. 4, the SPMA access flow chart is shown in fig. 5, and the SPMA working state transition chart is shown in fig. 6. Since NS-3 is a Linux-based network simulation tool, the training of the random forest model is based on the Windows operating system. Therefore, the two platforms are connected through the MySQL database to carry out joint simulation, and the connection concrete mode is shown in figure 2. Under the machine learning environment built by tensorflow, a random forest model generated by training stores a service priority prediction result into a database, and network data is read in real time by calling NS-3 through a shell script program to obtain a simulation result.

Step three: scene simulation

In the joint simulation platform of tensorflow+NS-3, after connecting NS-3 through MySQL database, the accuracy and priority result of service priority predicted by the random forest model trained on tensorflow according to service attribute are shown in figures 7 (a), 7 (b) and 7 (c). It can be seen that the machine learning algorithm can predict the priority of traffic with an accuracy of around 85%. In the network simulation model built in NS-3, a TTNT simulation module is shown in fig. 8, in the TTNT tactical data link communication system, data throughput of each node in a network layer under a single hop scene under the same wireless channel is tested, network simulation parameters are shown in fig. 9, three service type simulation results are shown in fig. 10 (a), 10 (b) and 10 (c), and five service type simulation results are shown in fig. 11 (a), 11 (b) and 11 (c). As can be seen from the simulation graph, when the number of nodes configured in the network is less than 42, the throughput and the number of nodes are about linear, and the traffic type B, C curves overlap; when the number of nodes is 48, the throughput of traffic type C, B with higher priority will continue to increase linearly, while traffic type a with lower priority starts to decrease. As the number of nodes continues to increase, the system throughput becomes increasingly saturated, the lowest priority traffic type a will continue to drop, but the overall system throughput will also remain unchanged. When the number of nodes is less than 42, the time delay of the high priority service is lower than the time delay of the low priority service. As the number of nodes increases, the throughput of the system gradually approaches saturation, the service type a with the lowest priority will wait for a longer time, and the delay of the service type a will be greatly improved at this time, so that the delay of the service type B, C with the high priority is obviously lower than that of the service type a with the low priority. It can be seen that the method provided by the invention can adaptively allocate the priority to the service, and can ensure the success rate of transmitting the data packet with high priority.

Claims (1)

1. A data link SPMA access method based on machine learning priority prediction comprises the following steps:

step 1: establishing a business attribute model;

the higher the security level s of the service, the higher the priority of the service;

the maximum time delay of the service allowable transmission is delay_m, and the service priority is lower as the maximum time delay of the service allowable transmission is larger;

the service demand bandwidth B, the larger the service demand bandwidth is, the higher the service priority is;

service data size l, the larger the service data size is, the higher the service priority is;

the service type comprises voice, data, images and situation awareness, and the priority of the service type is increased from left to right at one time;

step 2: training a machine learning model;

step 2.1: the method for calculating the base-Ni index of the training set data comprises the following steps:

wherein C is _k Is a subset of samples belonging to class K, K being the number of classes;

step 2.2: comparing the base index obtained by calculation in the step 2.1 with a set threshold value, randomly adjusting samples in training set data if the base index is larger than the threshold value, and returning to the step 2.1; otherwise, enter step 2.3;

step 2.3: according to the training data set obtained in the step 2.2, a service attribute model is used as the characteristic input of the random forest model, the service priority is used as the output of the random forest model, and the self-adaptive distribution model of the service priority is obtained through training of the existing data; the random forest model is defined as:

y＝TreeGenerate(D,A) (1)

wherein a= (s, delay_m, B, l, type) is the feature input of the random forest model; d is training set data; y is the classification result, namely the service priority;

step 3: priority self-adaptive allocation;

aiming at the service priority self-adaptive allocation model obtained through random forest model training in the step 2, when an unknown service arrives, allocating priority to the unknown service according to service attributes (s, delay_m, B, l, type) of the unknown service;

step 4: protocol verification;

step 4.1: when a network has data packets arriving, the data packets enter different priority queues according to different priorities;

step 4.2: firstly checking whether the queue with the highest priority is empty or not, and if so, checking the next priority queue;

otherwise, the head data packet of the queue with the highest priority is sent to a sending judging stage;

step 4.3: when a data packet enters a sending judging stage, judging whether the data packet is expired or not, and if not, comparing a priority threshold of the data packet with current channel occupation statistics; otherwise, the data packet is taken out of the queue, and step 4.2 is carried out;

step 4.4: if the current channel occupation statistics is smaller than the priority threshold of the data packet, the data packet is dequeued and sent out; otherwise, the data packet is retracted for a period of time;

step 5: and (5) ending.