A weighted fuzzy process neural network (WFPNN) model, and its corresponding classification algorithm, are proposed for quantitative and qualitative mixed-process signal classification problems. Based on predicate logic, fuzzy process neurons (FPN) are expressed as weighted fuzzy process inference rules on the semantic. Among them, the premise and conclusion are fuzzy predicates defined on a fuzzy set containing process information. Considering the applicability of rules to practical problems, confidence factor and applicable threshold are introduced to make the rules based on FPN have adaptive application ability. Based on FPN, the WFPNN was constructed with a process signal input layer, an FPN hidden layer, and a Takagi-Sugeno fuzzy classifier. It can integrate the learning properties and classification mechanisms of process neural networks (P-NNs) for time-vary signals with the logical inference abilities of fuzzy systems. So then to realize the quantitative and qualitative mixed process signal processing and direct classification. This study proposes a fusion analysis model that combining fuzzy decision theory with the process information processing method of P-NNs. It can automatically deduce the membership function and the fuzzy predicate logic rules from a set of given training examples, quickly build a fuzzy expert system prototype. And it can provide a new method for the construction and application of an expert system based on models or rules. Meanwhile, due to the adaptive learning property of WFPNN and the representation mechanism of inference rules based on the FPN, it is easy to realize the extension of the rule bases, model bases and application fields of the expert systems. This paper analyzes the properties of WFPNN, and establishes a corresponding learning algorithm. The discrimination of reservoir water-flooded states based on well-logging curves was used as an example for experimental analysis. The continuous measurement values of five physical quantities, such as resistivity, acoustic and radioactivity levels in the oil layer, were taken as input. The output was the discriminating result for four types of water-flooded states. The accuracy of the proposed method was 82.5%. Our accuracy was more than 7% higher than other existing automatic identification methods. These results demonstrate the viability and effectiveness of the proposed model and algorithms.