The increasing digitalization of modern energy systems, and similarly the ongoing digital transformation of the textile industry under the paradigm of Textile 4.0, has expanded their cyber attack surfaces and heightened the risk of sophisticated intrusions. The complex nature of energy networks traffic, which is characterized by heterogeneous data, multiprotocol communications, and strong temporal dependencies, has resulted in substantial growth in both the volume and dimensionality of network traffic, posing significant challenges for traditional intrusion detection systems (IDS). This paper proposes a Hybrid Intrusion Detection Framework with Heterogeneous Feature Fusion (H2IDF). The framework leverages multi-scale Convolutional Neural Networks (CNNs) to extract fine grained temporal patterns and utilizes Transformer encoders to model structured tabular features. A cross-type attention mechanism is introduced to semantically align and deeply fuse heterogeneous types, thereby enhancing the model's ability to capture complex inter feature dependencies. Furthermore, a prediction aggregation mechanism is employed to consolidate frame level decisions across overlapping sliding windows, significantly improving detection stability and robustness to noise. Experiments on the AWID dataset, selected for its representation of heterogeneous wireless traffic patterns analogous to energy system communications, demonstrate that H2IDF achieves superior performance over competitive baselines. These results highlight the framework's potential for enhancing cybersecurity in energy networks and analogous industrial environments, such as those found in Textile 4.0

cybersecurity, intrusion detection, convolutional neural network, textile 4.0, smart textiles