The paper presents a relational model of random process prognosis related to the description of ground points dislocations within the area of underground exploitation influence. The prognosis of the analysed nonstationary process refers to the moment tn+1 with the restriction, that both the process state until moment tn (survey results) as well as underground exploitation development generating given process are known. Moreover, an assumption has been introduced, that the investigated process is a composition of deterministic component and of random quantity. The process of post-mining dislocations has been analysed as random walk of rockmass particle - Markov chain - leading to diffusion process. In result partial differential equation has been obtained. What has been carried out for utilitarian purposes is the optimization of presented model with regard to several temporal horizons of surveying within the period of nonstationary dislocations formation. The quality of the description has been characterised by mean square error in relation to dislocations survey results the actual mining - geological situation.