Gree (China)

As distributed PV capacity expands rapidly, its inherent variability and partial observability create operational challenges for secure and stable distribution-network operation. Therefore, accurate PV power identification becomes a cornerstone for optimizing flexible interconnection power regulation strategies. To address the limitations of existing model-driven and data-driven methods in capturing nonlinear features and generalization capabilities, this paper proposes an unsupervised identification model named IHRTHAVOA-BP, based on the Improved Hybrid Red-tailed Hawk and African Vultures Optimization Algorithm (IHRTHAVOA). By integrating the Red-tailed Hawk Algorithm (RTH) and African Vultures Optimization Algorithm (AVOA), the model employs Circle chaotic mapping for population initialization, a Composite Opposition-based Learning Strategy (COBL) to enhance global search capabilities, and Gaussian mutation to avoid local optima, significantly improving optimization efficiency. Across diverse test functions, simulations show that the proposed model lowers Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) by over 90% relative to traditional BP Neural Networks and single-algorithm optimizers, confirming its superior accuracy in precise power identification.

The column photobioreactor has become the predominant approach for carbon sequestration by microalgae in power plant settings, owing to its capacity for high-density cultivation and efficient light energy utilization. Due to the dense arrangement of the columnar photobioreactor and its height, insufficient light became one of the main factors limiting the carbon sequestration rate of microalgae growth. In this paper, a light resource optimization method of reflective baffle and top diffusing glass was proposed. When the angle of reflective baffle on the north and east walls was 35°, and the angle of reflective baffle on the west and south floors was 0°, the overall light radiation intensity of the reactor array became the largest, reaching up to 916.81 W/m2, which was 14.39% higher than that before the optimization. The replacement of the top glass with diffusing material converted the direct radiation of solar radiation into scattered radiation. When the transmittance was 95% and the haze was 95%, the overall average light radiation intensity of the algal solution reached 830.93 W/m2, which was an increase of 3.7%. Four new exhaust air distribution methods were proposed, in which the three-entrance staggered-arrangement type glasshouse had the lowest algal liquid temperature.

With the rapid development of e-commerce, the modern logistics industry has put forward higher requirements for the detection efficiency and accuracy of express packages. To address this challenge, this paper proposes a deep learningbased package automatic detection system for express delivery. The system adopts the improved YOLOv8 algorithm and integrates the Global Attention Module (GAM), Bidirectional Feature Pyramid Network (BiFPN) and Concat2 modules. It is designed to improve the accuracy and real-time of express package detection. Through in-depth training with 300 Epochs, our model has demonstrated superior performance in key performance indicators such as accuracy, recall, mAP50, and F1 scores, significantly exceeding existing technologies. The experimental results show that the model can effectively identify the irregular shape and variable size of the logistics package, which provides an efficient and reliable solution for logistics automation.