Joint Scheduling For Space-based Maritime Moving Targets Surveillance

Space-based Maritime Moving Target Surveillance (SMMTS) can be of wide area coverage, high security, and unrestrained with airspace, national boundaries and geographical conditions. Thereby it has outstanding strategic meaning and civil value in military and civil applications. Nowadays it is emergent and important to research how to utilize satellites and their payloads to satisfy the informatic requirement of maritime moving target surveillance and enhance the surveillance efficiency of maritime moving target search, location and track. However, because of target motion and diversity of satellites and their payloads, the difficulties increase for satellite mission planning. Only if efficient operation control pattern and multiple satellite joint scheduling models and algorithms could be established, observation mission can be planned well and satellite resource can be allocated properly. Based on the research background above mentioned, the main work and innovative points in this paper are as follows.(1) System analysis, formal descriptions and general models of Joint Scheduling for Space-based Maritime Moving Target Surveillance (JSSMMTS).At first systematically analyze characteristics of maritime moving targets, surveillance environment and satellite scheduling resource. Then analyze JSSMMTS's uncertain factors, operational procedures and basic assumptions. Furthermore, qualitatively and quantitatively analyze and give out the formal descriptions about grid division in surveillance area, transit observation window, joint scheduling task, actions, observation action set, action strategy and optimization objective. According to optimization objective for short term or long term gain, respectively build Single Stage Joint Scheduling Model (SSJSM), Feedback Delayed Single Stage Joint Scheduling Model (FDSSJSM) and Multiple Stage Joint Scheduling Model (MSJSM).(2) Calculation method for temporal-spatial relationship between satellite and target.The proposed calculation method can be used for schedule preprocessing and influence relation between target and observation. First, Slew Looking Algorithm (SLA) computes the point location observed by center of FOV (field of vision), based on sub-satellite point and slew-looking angle known. Secondly, Slew-looking and Observing Time Algorithm (SOTA) determines observing time and its corresponding slew-looking angle when a target is visible. Third, Swath's Grid Coverage Detection Algorithm (SGCDA) is used to judge whether a grid in surveillance region is under observation coverage of a satellite's swath.(3) Maritime moving target motion predictions'improvement and extension.If targets are not easy to distinguish, it is proposed to use Uniform Velocity Motion Prediction (UVMP) with random disturbance. Otherwise if targets can be distinguished realizingly, present Multiple Model Prediction (MMP) by integrating UVMP, Track Change Prediction (TCP), Track-based Prediction (TP) and Potential Area Prediction (PAP). Based on comparison between target location and prediction during former stages, MMP evaluates recent performance of each candidate prediction method and decide the appropriate one.(4) Maritime Moving Target Tracking Particle Filter Algorithm (MMTTPFA).According to maritime moving target's characteristics and surveillance application requirement, present MMTTPFA without data association. MMTTPFA can adaptively select target particle sampling method based on distance between targets. By this way MMTTPFA can balance its performance between computation consumption for target distinction without data association and precision of target estimation. And MMTTPFA uses Kmeans cluster method to adjust the order of particle's partitions corresponding to different targets. Moreover, MMTTPFA introduces particle weight modified factor influenced by geographic constraints to deal with particle adjustment when particles are located on land, island or other area where they can not arrive at.(5) Informatic measurement based single stage joint scheduling algorithm (IMBSSJSA).In statistical sense, analyze different action policies of maximizing Kullback- Leibler (KL) discriminant, entropy and target state probability, and then argue that the policy based on KL discriminant is statistically preferable to other candidate policies. Define Single Stage Expected Information Gain (SSEIG) formulation represented by particles sampled by probability distribution, take the policy of maximizing SSEIG for each grid under observation swath coverage, and put forward IMBSSJSA represented by particles sampled from probability distribution. Moreover, consider feedback delay problem induced by adjacent or overlapped transit observation windows, devise Feedback Delayed Single Stage Joint Scheduling Algorithm (FDSSJSA).(6) Reinforcement Learning Based Multiple Stage Joint Scheduling Algorithm (RLBMSJSA).Study on MSJSA with action policy of long term observation gain maximization. Analyze reinforcement learning's applied advantage under research background. Take the expected information gain as state representation of reinforcement learning. Considering different satellite sensors with different control and observation parameters may induce discrete action with different scale, define action set on continuous action space and action selection method. And then give out state-action pair's value function and learning rules to approximate value function by neural networks, and present Reinforcement Learning Based Multiple Stage Joint Scheduling Algorithm (RLBMSJSA) with continuous action space.(7) Applied study based on simulations.By designing simulation constitutive structure, main control procedure, simulation scenario, and schedule preprocessing and schedule results, simulation instance close to application in practice is presented to verify the proposed methods'effectiveness.