The physical and chemical properties of particle materials are closely related to their particle size that the measurement of particle size distribution plays an important role in a wide of applications in chemical industry, environmental protection and other fields. Among the various particle size measurement methods, the forward laser scattering technique has been widely used because of its advantages, including high efficiency, high precision, good repeatability and non-intrusive measurement. The problem of retrieving particle size distribution based on spatial distribution of light scattering intensity belongs to the first Fredholm integral problem, which is difficult to give an analytical solution, which was typically solved by numerical inversion methods. Due to the influence of ambient light noise and circuit noise on light energy distribution signal, the inversion of particle size distribution with high accuracy is one of the key issues in this technique. The inversion methods are generally divided into two kinds: independent mode method and non-independent mode method. The independent model method does not need to know the particle size distribution information in advance, which can theoretically obtain the particle size distribution of any particle system. However, such methods suffers from its sensitivity to noise and the width of the particle size distribution, which might cause distortion in the results. The non-independent mode method needs to assume that the particle size distribution meets a certain distribution function based on the prior information of particle size distribution, and then solve the problem on the basis of the set distribution function model. Compared to the independent mode method, this kind of method has advantages in computation efficiency and anti-noise ability. The traditional intelligent algorithm can obtain good inversion results when dealing with unimodal distribution particle system. However, for the cases where the particle size distribution of the particles presents a bimodal or multimodal pattern, the optimization parameters in the inversion process increases a lot, resulting in an exponential increase in inversion computation. Traditional inversion algorithms suffer from problems such as rapid decline in optimization efficiency, rapid deterioration of robustness and inversion accuracy. Regarding the inversion problem of particle size distribution in the forward laser scattering measurement technique, an improved whale optimization algorithm was proposed in this paper by using a logarithmic form of adaptive probability threshold and non-linear convergence factor, which balancing its ability in global search and local optimization in the inversion optimization process. By using reverse learning method for initialization and greedy principle for individual updates, an accurate and fast inversion of particle size distribution can be achieved. The simulation results show that the algorithm has a good robustness and inversion accuracy for unimodal and multimodal distributions that follow normal distribution, Rosin-Rammler distribution, or Johnson's SB distribution under different levels of random noise. The proposed algorithm was applied to the practical experimental measurement using standard spherical polystyrene particles, very good inversion results were obtained, verifying the effectiveness of the algorithm in terms of noise resistance and measurement accuracy.