Small-scale PV generation is particularly affected by the irradiance variability produced by cloud shadows. Accurate predictions of the clouds passing over the PV field are thus necessary for the optimized management and integration of this renewable power source. Two main elements are required to perform these predictions: the local irradiance field and the cloud motion vector. Irradiance sensor networks are positioning as a promising data source at the spatial and temporal scales of the problem, avoiding the irradiance inference problems and costs of image-based instrumentation. This paper proposes a method to infer the cloud shadow motion vector from small-scale irradiance sensor networks data. The method does not require specific network configuration or layout, and the algorithm is computationally simple: graphical solutions are obtained by aggregating mean absolute errors in a diagram/matrix with each element representing a possible displacement of the cloud shadows. The validation is conducted with a fractal cloud model that allows the generation of irradiance time series according to an arbitrary cloud motion vector. The most correlated pair and the linear cloud edge methods are used for benchmarking purposes. Gridded and non-gridded sensors layouts are tested with number of sensors ranging from 9 to 100, monitored areas from 100 to 10000 m(2), and sampling periods from 0.3 to 3.3 s. The results show the superiority of the proposed method with a reduction of 50-90% of the root mean square errors respect to the benchmark methods in 75% of the tests. Additionally, the proposed method maintains similar performance as sampling rate decreases, while the benchmark methods exhibit worsening results.