Bolt connections, characterized by high reliability and ease of installation and maintenance, are extensively employed in transmission tower structures. Due to the long-term influence of static and dynamic loads, bolt connections are prone to loosening, resulting in a decline in the structural bearing capacity and endangering the safety of the power system. In response to the abovementioned problems, an intelligent detection method based on multi-feature information of audio signals was proposed. A typical bolt connection node in the transmission tower was selected, and audio signals in different pre-tightening force states were collected through fixed-point hammering. Based on the wavelet packet decomposition signal processing technology, the energy value loosening index of the signals was obtained, and the optimization of the penalty factor and kernel function parameters was achieved using the particle swarm algorithm. A support vector machine model integrating multi-feature information of audio signals was proposed, realizing the quantitative identification of bolt pre-tightening force. The research results indicate that the particle swarm optimization support vector machine model has witnessed a notable improvement in classification accuracy, with the maximum accuracy reaching 93.42%. Compared with that before optimization, the model's recognition accuracy has increased by 9.2%, providing an effective approach for the detection of bolt pre-tightening force in transmission towers.