Fig. 1. The framework diagram of the optoelectronic reservoir computing system. (a) The input layer includes a CW-laser, a 1×4 beam splitter, and a modulator array; (b) The 32-node plum-shaped reservoir chip serves as the reservoir layer (Black square represents vacant ports, red square represents input ports and green square represents output ports); (c) The output layer consists of a photo-detector array comprising photodetectors (PDs), transimpedance amplifiers (TIAs), and analogue-to-digital converters (ADCs)

Fig. 2. Predictions with varying sliding window lengths (The magnitude of the RMSE, NMSE, and MAE metrics is on the order of 10⁻²)

Fig. 3. Prediction results of 12-nodes, 32-nodes and 60-nodes reservoir chips for the DJI index

Fig. 4. Effect of random phase factor on prediction (The black line segments represent the error bars, indicating the optimal and worst results obtained from multiple sets of phases)

Fig. 5. Mach-Zehnder modulator transmission response curve （① Positive Linear (PL); ② Positive Nonlinear (PN); ③ Negative Linear (NL); ④ Negative Nonlinear (NN)）

Fig. 6. The forecasting results and absolute error for three stock indexes in test-dataset period. (a) SHSECI; (b) FTSE; (c) DJI (The red line represents the actual normalized data, the green line represents the predicted normalized data, and the blue line represents the absolute difference between the actual and predicted data)

Fig. 7. Evaluation results with different input strategies

Table 1. Evaluations of DJI for different IO settings

Table 2. DJI evaluation results for different interval of modulation

Table 3. Evaluations of stock indexes in different research