A suitable pH value of the slurry is the key for efficient froth flotation. In the industrial process, it is difficult to measure the pH value online so that control of the pH value is delayed. To solve this problem, pH-associated sensitive image features of the froth are obtained; a soft sensor model-multi-model LSSVM (least squares support vector machine) based on affinity propagation clustering (AP) is then introduced. Then, an predictive control strategy based on online support vector regression (OSVR) and differential evolution (DE) optimization for the pH is proposed. The prediction model is built offline and corrected online; a DE optimization method is used to solve the predictive control problem to find the optimal decision variables, so as to achieve the real-time control of the slurry pH value. The industrial test results in antimony flotation show that the proposed control strategy can stabilize the pH value, reduce the chemical consumption.