Powerful radio galaxies are excellent candidates for investigating and ultimately understanding the formation and evolution of galaxies. These beacons are now observed out to z>5 and are commonly associated with the massive early-type galaxies observed in the local universe. While the radio emission reveals the presence of a supermassive black hole, a dusty parsec-scale torus acts like a natural coronograph, making it easier to study the properties of the host galaxy. The aim of this PhD thesis is to characterise the nature and evolution of the stellar population and the relationship between the stellar population and the active galactic nucleus (AGN). To reach our scientific goals, we use the galaxy evolution code, PEGASE, combined with a AGN model which both consider the radiative transfer of the UV, optical, and IR photons through dust. To begin, we present the HeRGE project consisting of 70 radio galaxies which have been observed with Herschel. These IR observations allow us to calculate the total infrared luminosities and reveal that our sample belongs to the ULIRG regime. We decompose the infrared SED into an AGN and starburst components using observational templates. Converted into accretion and star formation rate, their relative luminosities indicate that the black holes are growing proportionally faster than are the host galaxies.In addition, we constrain the configuration of the jet and torus by combining the results from mid-infrared spectral energy distribution (SED), and the radio emission from the lobes (isotropic at 500MHz) and the core (anisotropic at 20GHz). In agreement with the unified scheme, these observations allow us to estimate the absorption Av, the inclination of the torus, and provides a constraint on the Lorentz factor for the radio jet.A subsample of 12 radio galaxies observed from the UV to sub-mm is also analysed with PEGASE.3 and an AGN torus model. While one stellar component is clearly insufficient to fit the observations, two stellar components are necessary to successfully reproduce the SED (one evolved and massive, about 10^12 msun, formed over a reasonably short time, <1Gyr at high redshift; and a much younger component, <40Myr, that is also less massive, about 10^11 msun. Such a star formation history suggests rapid growth at high redshift of longer duration followed much by another period of rapid, stochastic growth.These results put strong constraints on galaxy formation models. Unfortunately, the crudeness of some of our data and theoretical understanding the IR emission from AGN, means that the relation of the galaxy to its AGN is still not well constrained. Additional observations at optical through millimeter wavelengths are needed to extend our findings.
