In this paper, we present a nutrient solution control system, designing a nutrient solution electrical conductivity (EC) sensing system composed of multiple long-range radio (LoRa) slave nodes, narrow-band Internet of Things (NB-IoT) master nodes, and a host computer, building a nutrient solution EC control model and using the particle swarm optimization (PSO) algorithm to optimize the initial weights of a back-propagation neural network (BPNN). In addition, the optimized best weights are put into the BPNN to adjust the proportional–integral–derivative (PID) control parameters Kp, Ki, and Kd so that the system performance index can be optimized. Under the same initial conditions, we input EC = 2 mS/cm and use the particle swarm optimization BP neural network PID (PSO-BPNN-PID) to control the EC target value of the nutrient solution. The optimized scale factors were Kp = 81, Ki = 0.095, and Kd = 0.044; the steady state time was about 43 s, the overshoot was about 0.14%, and the EC value was stable at 1.9997 mS/cm–2.0027 mS/cm. Compared with the BP neural network PID (BPNN-PID) and the traditional PID control approach, the results show that PSO-BPNN-PID had a faster response speed and higher accuracy. Furthermore, we input 1 mS/cm, 1.5 mS/cm, 2 mS/cm, and 2.5 mS/cm, respectively, and simulated and verified the PSO-BPNN-PID system model. The results showed that the fluctuation range of EC was 0.003 mS/cm~0.119 mS/cm, the steady-state time was 40 s~60 s, and the overshoot was 0.3%~0.14%, which can meet the requirements of the rapid and accurate integration of water and fertilizer in agricultural production.

To seek an appropriate stabilization and remediation scheme for cadmium (Cd) and arsenic (As) pollution in farmland, a typical polluted soil sample was selected from a mining area in Southwest China for a soil box simulation experiment. Biochar (BC), a modified type of biochar made from rice husk with different mass ratios of ferric chloride and rice husk, was set up (the mass ratio of ferric chloride to rice husk was 1:9 (defined as LFB), 3:7 (defined as MFB), and 5:5 (defined as HFB) and the control group (BL)) to explore the effects of soil water and fertilizer loss, the bioavailability of Cd and As, and the bioenrichment effects of plant organs during the growth period of rice. The results showed that the porous structure and large specific surface area of biochar effectively regulated soil aggregate composition and improved soil water holding capacity. Compared to the BL treatment, soil water storage under the four carbon-based material control modes increased from 8.98% to 14.52%. Biochar has a strong ion exchangeability and can absorb soil ammonium, nitrogen, and phosphoric acid groups, effectively inhibiting the loss of soil fertilizer. Biochar improves soil pH and reduces the specific gravity of exchangeable Cd. In addition, the oxygen-containing functional groups in biochar can react with metals in a complex manner. The diethylenetriaminepentaacetic acid (DTPA) concentrations of Cd in soils treated with BC, LFB, MFB, and HFB were 79.69%, 72.92%, 64.58%, and 69.27% lower, respectively, than those treated with BL. In contrast, the Fe³⁺ in ferric chloride combines with As after hydrolysis and oxidation to form amorphous ferric arsenate precipitates or insoluble secondary minerals. Therefore, the curing effect of the modified biochar on As was more potent than that of applied biochar alone. In conclusion, ferric chloride-modified biochar can effectively inhibit the effects of water and fertilizer loss in farmland soil and realize cross-medium long-term inhibition and control of combined Cd and As pollution.

Accurate monthly temperature and precipitation forecasting in Guizhou Province is challenging because of its complex karst terrain and substantial spatiotemporal variability in meteorological conditions. We propose a forecasting framework that integrates Seasonal-Trend Decomposition using Loess (STL), Variational Mode Decomposition (VMD), bidirectional Long Short-Term Memory (BiLSTM) encoding, dynamic graph learning, graph convolution, and spatiotemporal cross-attention. STL and VMD provide multiscale temporal features, which are combined with meteorological, temporal, and station-related variables for spatiotemporal representation learning. An auxiliary high-precipitation loss is introduced to improve the prediction of high-precipitation events. Experiments were conducted using monthly observations from 38 meteorological stations in Guizhou Province during 1961-2011 under a chronological, leakage-free evaluation protocol. The proposed model achieved an RMSE of 1.759 °C for temperature and 57.550 mm for precipitation. Although it did not obtain the lowest aggregate precipitation error among the evaluated baselines, it achieved an upper-tail precipitation RMSE of 93.096 mm and MAE of 73.053 mm lower than those of monthly climatology, STGCN, and Graph WaveNet. Moving-block bootstrap analysis further supported the reductions in upper-tail RMSE and MAE relative to Graph WaveNet. These results indicate that the primary advantage of the proposed framework lies in forecasting high-precipitation magnitudes rather than minimizing aggregate precipitation error alone.