Research On Key Technology Of Data Quality Assurance In Mobile Crowd Sensing

In recent years,Mobile Crowd Sensing(MCS)has received extensive attention and has become a very attractive research paradigm in the field of urban perception.In terms of data collection,the Mobile Crowd Sensing relies on the contribution of mobile devices from a large number of participants or groups.Smartphones,tablets and wearables are widely used and already have many sensors,making them excellent sources of information.Compared with traditional sensor networks,human mobility and intelligence ensure higher coverage and better contextual awareness.However,the data reported by the perception node may be unintentionally inaccurate or deliberately falsified.For example,the location of the perception task will expose the sensitive information of the perception node,which will affect the privacy and security of the perception node,causing the perception node to reject the perception or upload false data.The nodes involved in perception have different purposes.Some malicious perception nodes upload false information,which results in wrong perception results and affects the quality of perception data.In task allocation,the perceptual nodes assigned the perceptual task have different preferences,and are limited by their ability of perceptual nodes.When their perceptual conditions or preferences are not met,they cannot complete the perceptual task or upload false perceptual data,resulting in the degradation of perceptual data quality.Therefore,this dissertation studies the protection of location privacy,trusted recommendation of perceptual nodes and task assignment in mobile group intelligence perception.The following work was done:Firstly,in view of the location privacy problem of participants in mobile crowd sensing,this dissertation proposes a method to protect the location of participants based on local differential privacy preference.First of all,the map is discretized and mapped from two-dimensional space to one-dimensional space bymeans of MHC,which can guarantee the spatial correlation,and the map is segmented based on the density of participants using genetic algorithm;then,according to the personal privacy needs of current location,two different local differential privacy perturbation methods,RAPPOR and k-RR,are chosen by participants;finally,the chose local differential privacy perturbation method is used to perturb the location of each participant in the region after segmentation,and the perturbed location data is sent to the data collection server to protect the participants' locations.Secondly,aiming at the problem of unreliable data quality caused by sensing node uncertainty in mobile crowd sensing,a cross-domain collaborative filtering trusted sensing node recommendation method based on SDN is proposed.Firstly,SDN is introduced to decouple the user surface and the control surface,and it is convenient to manage sensing nodes and reduce the burden of server for task allocation;Then,through cross-domain collaborative filtering method,find sensing nodes which show similar credibility in the historical task allocation and complete some similar tasks with target sensing nodes;Finally,according to the current sensing ability of sensing node,and distance from target task,and similar sensing nodes' credibility in the target task and time decay,the recommendation value of the sensing node in the target task is obtained,and then,the trusted sensing node is selected.Finally,for task allocation of mobile crowd sensing,the low quality of sensing data is caused by the unsatisfied sensor perception willingness and the change of perception state which causes the perception task cannot be completed normally,the task allocation method of active learning mobile crowd sensing based on normal cloud model is proposed.Firstly,the data quality,perception environment and network state of the sensor are evaluated,and the threshold is set according to the battery capacity of the sensor and the total amount of data sent,and then the sensor perception ability is monitored in real time;secondly,when the state is changed,the multi-granularity standard distribution cloud and sensor state cloud are established by using the normal cloud model,and the willingness score of the sensor to the perception task is obtained by cosine method;finally,according to the obtained sensor willingness list and task willingness list,tasks are allocated by the maximum sensor willingness and theminimum number of sensing sensor.