Intelligent acid rain sensing and monitoring method and system based on total-current nano-generator

Abstract

The invention belongs to the technical field of intelligent sensing monitoring and micro-nano energy, and discloses an intelligent acid rain sensing monitoring method and system based on a total-current nano generator. The method comprises the following steps: preparing a total-current nano-generator as a sensing core element, and generating a characteristic electric signal corresponding to a pH value through the action with acid rain; constructing an intelligent algorithm processing module based on a ResNet 18-1D convolutional neural network (CNN), and performing multi-dimensional feature extraction and pH value intelligent classification on the collected electric signals; and displaying the electric signal waveform characteristics and the pH recognition result in real time through a visual interaction interface. According to the invention, the full-current nano generator is combined with the deep learning technology, the technical bottlenecks that a traditional acid rain detection method depends on a large professional instrument, needs external power supply support and is complex in operation process are broken through, and in-situ, rapid and intelligent monitoring of the acid rain pH value is realized. And an efficient and reliable solution is provided for real-time monitoring of acid rain in industrial areas and outdoor environments.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent sensing monitoring and micro-nano energy technology, and in particular relates to an intelligent acid rain sensing monitoring method and system based on a full-current nanogenerator. Background Art

[0002] Acid rain pollution has become a global environmental problem, posing a significant threat to the atmosphere, water resources, soil quality, and human health. Therefore, real-time and effective monitoring and detection of acid rain are crucial. Currently, traditional monitoring technologies face three key bottlenecks: First, limitations in sensing mechanisms. Current electrochemical sensors rely on ion mobility detection, require continuous external power, and are susceptible to electrolyte aging, resulting in sensitivity degradation exceeding 30% with long-term monitoring. Online monitoring systems based on spectral analysis, while capable of achieving an accuracy of ±0.05 pH, are bulky (typically >1 m³) and expensive (over 500,000 RMB per unit), making them difficult to implement for wide-area distributed monitoring in remote areas such as plateaus and transboundary river basins. Second, energy supply constraints exist. Existing automatic samplers are mostly powered by lithium batteries or solar energy, but their battery life drops by over 60% in extreme temperature and humidity environments. A 2024 Acid Rain Sampler Market Report indicates that 78% of equipment failures are due to power system failures, severely restricting continuous field monitoring. Third, intelligent processing is a key bottleneck. Although deep learning has been applied to acid rain prediction, traditional feature extraction methods have difficulty capturing the transient electrical signal characteristics generated by droplet impact, resulting in low pH recognition resolution.

[0003] In view of the problems of the above-mentioned acid rain monitoring technology, such as complex equipment (dependence on chemical reagents), low sensitivity and requirement of external power supply, the present invention proposes an intelligent acid rain sensing and monitoring method and system based on full-current nanogenerator. Summary of the Invention

[0004] To overcome the problems existing in the related art, the disclosed embodiments of the present invention provide an intelligent acid rain sensing and monitoring method and system based on a full-current nanogenerator, and more particularly, a self-driven sensing design based on a full-current droplet nanogenerator. Combined with the ResNet 18-1D deep learning model, an energy-self-supplied real-time acid rain monitoring system is constructed, which is suitable for distributed real-time monitoring and prediction of outdoor industrial acid rain.

[0005] The technical solution is as follows: an intelligent acid rain sensing and monitoring method based on full-current nanogenerators, comprising the following steps: S1, preparing a full-current nanogenerator to generate electrical signals of acid rain with different pH values, using a process of photocuring 3D printing to achieve biomimetic structure reproduction to prepare the full-current nanogenerator, so that it generates electrical signals with different characteristics when exposed to acid rain with different pH values; S2 collects the electrical signals of acid rain with different pH values ​​generated by the full-current nanogenerator and preprocesses them into standardized time series data. The preprocessed time series data is input into an algorithm processing module based on a ResNet 18-1D convolutional neural network. The network's residual block structure extracts multi-level time series features and classifies and identifies the corresponding pH values. The algorithm processing module uses the ResNet 18-1D CNN to extract and classify the electrical signals of acid rain with different pH values. S3, the waveform diagram, feature extraction results and pH value classification and identification results of the extracted and classified acid rain electrical signals are displayed in real time through the interactive interface, so as to enable users to dynamically query and analyze the monitoring data.

[0006] In step S1, a full-current nanogenerator is prepared, including: Using light-curing 3D printing technology, a resin external model was designed and printed to collect the acid rain droplets to be tested; A base substrate is provided at the lower part of the resin outer model, and a friction layer surface is opened on the base substrate; a top electrode and a bottom electrode are respectively installed on the top and bottom of the friction layer surface, and a charge collecting needle is installed on the top electrode.

[0007] In step S1 , electrical signals of acid rain with different pH values ​​are generated, including: acid rain droplets to be measured fall naturally through the acid rain circulation pipe, and electrical signals are generated by the acid rain droplets to be measured hitting the surface of the friction layer.

[0008] In step S2, collecting electrical signals of acid rain with different pH values ​​generated by the full-current nanogenerator includes: The electrical signal generated by the all-current nanogenerator is collected through a NI acquisition card. The signal duration is 0.1-5 seconds and includes current pulses caused by the impact of acid rain droplets. The generated electrical signal is denoised, normalized, and subjected to time-frequency analysis. The time-frequency analysis includes applying short-time Fourier transform (STFT) to the signal to extract frequency components and time-domain variation characteristics to form a two-dimensional time-frequency spectrum.

[0009] In step S2, the ResNet 18-1D convolutional neural network includes: The input layer receives a one-dimensional electrical signal vector; the input one-dimensional electrical signal is normalized to a fixed length and processed by sliding window segmentation; Initial convolution block, the convolution kernel is combined with batch normalization BatchNorm and ReLU activation function for preliminary feature extraction; the initial convolution block uses a one-dimensional convolution kernel to perform local feature scanning on the input signal; The residual module consists of 8 residual blocks, each of which consists of two convolutional layers and a residual connection; Global average pooling is used to reduce the dimensionality of feature maps; The fully connected layer FC outputs the probability distribution of multiple pH value categories through two layers of fully connected networks.

[0010] In step S2, the feature extraction process is implemented through the residual module and convolutional layer of ResNet 18-1D, which efficiently captures features and classifies pH values ​​of the one-dimensional time-series electrical signals generated by the full-current nanogenerator. Specifically, it includes: Time domain features, using residual blocks to capture local and global patterns of signals; residual blocks capture local timing patterns and global patterns of TENG signals through multi-layer convolution; Multi-scale feature fusion: The eight residual blocks of ResNet 18-1D achieve multi-scale feature fusion by gradually increasing the number of channels. Low-level convolutions retain signal details, while high-level convolutions capture abstract patterns. Residual connections ensure that the deep network retains original signal information during the fusion process. Nonlinear feature enhancement: Each convolutional layer is followed by a ReLU activation function and batch normalization. ReLU improves the model's adaptability to the non-stationary characteristics of TENG signals by introducing nonlinear transformations. BatchNorm ensures the model's robustness to different droplet volumes or impact velocities by standardizing feature distributions.

[0011] In step S2, feature classification includes mapping the FC layer output to a probability distribution through the softmax function, which is expressed as: ; Where, is the predicted probability, For the The logit value of each pH category (i.e., the original output of the fully connected layer FC), usually not normalized; For the Logit values ​​for pH categories, summation in the denominator Normalize the index output of all categories to ensure the properties of probability distribution; The cross entropy loss function is used for optimization, and the expression is: ; Where, Cross-Entropy Loss is a function that measures the difference between the model's predicted probability distribution and the true label distribution. The smaller its value, the closer the prediction is to the true distribution. For the True labels of categories; The model is The predicted probability of each category is converted into a probability distribution through the Softmax function (all The sum is 1).

[0012] In step S2, after feature extraction and classification, model training and verification are also performed.

[0013] Another object of the present invention is to provide an intelligent acid rain sensing and monitoring system based on a full-current nanogenerator, which implements the intelligent acid rain sensing and monitoring method based on a full-current nanogenerator. The system comprises: Full-current nanogenerator, used to monitor acid rain at different pH values; An electrical signal generating unit, used for generating electrical signals of acid rain with different pH values; The electrical signal acquisition, feature extraction, and classification unit is used to collect the electrical signals of acid rain with different pH values ​​generated by the full-current nanogenerator and preprocess them into standardized time series data. The preprocessed time series data is input into an algorithm processing module based on the ResNet 18-1D convolutional neural network. The network's residual block structure is used to extract multi-level time series features and classify the corresponding pH values. The algorithm processing module uses the ResNet 18-1D CNN to perform feature extraction and classification on the electrical signals of acid rain with different pH values. The interactive interface is used to display the waveform diagram, feature extraction results and pH value classification and identification results of the extracted and classified acid rain electrical signals in real time through the interactive interface, so as to enable users to dynamically query and analyze the monitoring data.

[0014] Furthermore, the intelligent acid rain sensing and monitoring system based on the full-current nanogenerator is carried on a computer-readable storage medium, and the computer-readable storage medium stores a computer program. When the computer program is executed by the processor, it can realize the functions of the intelligent acid rain sensing and monitoring system based on the full-current nanogenerator.

[0015] In combination with all the above technical solutions, the beneficial effects of the present invention are as follows: First, the present invention introduces full-current nanogenerator technology, uses 3D printing to produce an external model for collecting acid rain, constructs a ResNet 18-1D convolutional neural network and a human-computer interaction interface, and realizes intelligent identification, sensing and monitoring of acid rain with different pH values.

[0016] The new self-driven system of the present invention, which integrates droplet nanogenerators (DNG) and intelligent sensors, demonstrates a breakthrough advantage. By regulating the charge transfer characteristics of the liquid-solid interface of DNG, its triboelectric signal is highly sensitive to pH value. Combined with deep learning convolutional network technology, it can simultaneously realize environmental mechanical energy collection and acid rain pH value monitoring without the need for external power supply, providing an innovative solution for establishing all-weather, self-powered intelligent acid rain monitoring.

[0017] Second, the present invention provides efficient, self-powered detection: When acid rain droplets of varying pH values ​​strike the friction layer of the all-current nanogenerator, charge transfer occurs through the solid-liquid contact electrodynamics. The synergistic effect of the conduction current and displacement current directly converts the droplet's kinetic energy into an electrical signal. This system requires no external power supply, boasts a simple structure (consisting solely of a substrate, two electrodes, and a charge collection probe), and offers low manufacturing costs. It can distinguish acid rain droplets of varying pH values, making it suitable for real-time, rapid detection.

[0018] High monitoring sensitivity: The current signals generated by acid rain droplets of different pH levels when in contact with the friction layer differ significantly. Experiments have shown that the peak current signals of pure water and acid rain differ by several times. A deep learning model can establish a mapping between the electrical signal characteristics and acid rain of different pH levels, enabling high-precision classification.

[0019] The system is lightweight and easy to use: The interactive interface is developed based on a lightweight framework and supports running on embedded devices such as the Raspberry Pi, with low CPU usage. The system triggers identification by simply dropping a drop of acid rain to be tested, making it easy to operate and integrate into portable devices, making it suitable for on-site rapid testing in outdoor industries.

[0020] This invention solves the technical bottleneck of traditional methods that rely on professional instruments and have complex detection processes by integrating full-current nanogenerators with deep learning, providing an innovative solution for the rapid on-site identification of the pH value of outdoor industrial acid rain.

[0021] Third, traditional monitoring equipment is difficult to deploy in large numbers due to its high cost and poor portability. This invention is based on a solution that combines full-current nanogenerators with deep learning. The cost of a single device is about 60% lower than that of traditional instruments, and it has self-powered and real-time detection features. It is estimated that if 100,000 monitoring points are deployed in industrial-intensive areas across the country, the market size is expected to exceed 5 billion yuan, and the annual operation and maintenance costs can be reduced by more than 40%, bringing significant economic benefits to enterprises and environmental protection departments. The global demand for acid rain pollution control is urgent. The existing automatic acid rain observation system, while the present invention integrates nanogenerators with pH sensors to achieve self-powered real-time monitoring, can replace traditional high-maintenance equipment.

[0022] A comprehensive search of domestic and international patent databases and academic literature has revealed no technical solutions combining full-current nanogenerators with a ResNet 18-1D convolutional neural network for real-time acid rain pH monitoring. Existing technologies often employ single sensors, rely on traditional electrochemical methods, or require an external power source. This invention, by directly analyzing the nanogenerator output signal through a ResNet 18-1D CNN, integrates pH classification and energy harvesting. This represents a significant technological breakthrough in acid rain monitoring, and is unparalleled in similar technology reports. Its self-powered nature frees the device from reliance on external power sources, enabling long-term stable operation in remote areas, filling a critical gap in high-precision acid rain monitoring in outdoor environments without power.

[0023] Fourth, rapid on-site testing of acid rain pH levels has long faced three major industry challenges: First, reliance on laboratory equipment. Existing acid rain pH testing methods primarily use pH glass electrodes and ion chromatography, requiring sample collection, transportation, and laboratory analysis. This process is cumbersome and time-consuming. Second, the performance of portable devices is insufficient. While portable pH meters can be used on-site, they rely on batteries, have short battery life, and require frequent calibration, making continuous and stable monitoring difficult. Existing technologies require offline sample processing, taking several hours from sampling to feedback, making real-time acid rain warning impossible. This invention achieves a comprehensive breakthrough through three technological innovations: First, self-powered sensing. A fully current nanogenerator (TCNG) directly converts the chemical energy of acid rain into a high-voltage DC signal, eliminating the need for an external power supply. Driven by the synergy of displacement current and conduction current, this generator boasts an output efficiency three times higher than traditional AC nanogenerators, solving the energy supply challenge for outdoor monitoring. Second, millisecond-level real-time analysis. A ResNet18-1D CNN performs multi-level feature extraction on the time-series electrical signals output by the nanogenerator, using residual connections to prevent gradient vanishing. Third, adaptability to complex environments. In scenarios such as industrial areas (high sulfur emissions) and mountainous areas (low signal-to-noise ratio), the system uses anti-noise convolution kernels and channel attention mechanisms (SE modules) to suppress interference signals and maintain detection stability.

[0024] Fifth, the present invention overcomes the prejudice that "high-precision detection must rely on large laboratory instruments." In the field of acid rain monitoring, the traditional concept is that high-precision detection must rely on large laboratory instruments, and portable devices are difficult to balance accuracy and portability. In addition, micro-nano energy devices are often considered unsuitable as core sensing components due to their weak output signals and poor stability. This technical prejudice limits the direction of innovation in the industry. The present invention greatly improves the signal output intensity and stability by optimizing the structural design of the nanogenerator (displacement current + conduction current); at the same time, it utilizes the powerful feature extraction capability of the ResNet 18-1D network to effectively amplify the key features in the weak signal and realize accurate identification of the pH value. The present invention overcomes the prejudice that "acid rain monitoring requires complex instruments." The traditional view is that pH detection requires precision electrodes or spectral analysis. The present invention utilizes the strong correlation between the acid rain electrical signal and pH (H + Ion concentration affects charge transfer), and sensing can be completed with only a single nanogenerator, breaking the mindset that "sensing must be independent of energy supply." BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure; Figure 1 This is a flow chart of an intelligent acid rain sensing and monitoring method based on full-current nanogenerators provided by an embodiment of the present invention; Figure 2 This is a model diagram of a full-current nanogenerator system based on 3D printing provided by an embodiment of the present invention; Figure 3 Schematic diagram of current signals generated by acid rain solutions with different pH values ​​on a full-current nanogenerator provided by an embodiment of the present invention; Figure 4 is a switching response and recovery diagram between normal rainwater and simulated acid rain provided by an embodiment of the present invention; Figure 5 is a cluster diagram of confusion matrix for identifying different pH solutions provided by an embodiment of the present invention; In the figure: 11, friction layer surface; 12, top electrode; 13, bottom electrode; 14, charge collection needle; 15, substrate; 16, resin external model; 17, acid rain flow tube; 18, acid rain droplets to be measured. DETAILED DESCRIPTION

[0026] To make the above-mentioned objects, features, and advantages of the present invention more readily apparent, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. The following description sets forth numerous specific details to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art may make similar modifications without departing from the scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0027] The innovation of this invention lies in its combination of a full-current nanogenerator with a ResNet 18-1D convolutional neural network, creating a new technical path for the collaborative application of micro-nano energy devices and deep learning for acid rain monitoring. The nanogenerator enables self-powered sensing, while the neural network performs intelligent analysis of electrical signals, achieving an integrated "sensing-computing-display" model and overcoming the limitations of traditional single technologies. This technology transcends the traditional reliance on large laboratory instruments and external power sources, eliminating the need for sample pretreatment and enabling real-time outdoor detection of acid rain pH values. It overcomes the poor stability and inability of existing portable devices to continuously monitor, and can maintain reliable operation in complex environments such as industrial waste gas outlets and mountainous areas.

[0028] Example 1, as Figure 1 As shown, the intelligent acid rain sensing and monitoring method based on full-current nanogenerator provided by the embodiment of the present invention includes: S1, preparing a full-current nanogenerator to generate electrical signals of acid rain with different pH values, using a process of photocuring 3D printing to achieve biomimetic structure reproduction to prepare the full-current nanogenerator, so that it generates electrical signals with different characteristics when exposed to acid rain with different pH values; The preparation of the full-current nanogenerator specifically includes: Using the light-curing 3D printing technology, a resin external model 16 is designed and printed for collecting the acid rain droplets 18 to be tested.

[0029] A base substrate 15 is provided at the lower part of the resin external model 16, and a friction layer surface 11 is opened on the base substrate 15; a top electrode 12 and a bottom electrode 13 are respectively installed on the top and bottom of the friction layer surface 11, and a charge collection needle 14 is installed on the top electrode 12; the top electrode 12 and the bottom electrode 13 are respectively connected to an external electrical signal detector through wires.

[0030] Among them, the top electrode 12 and the bottom electrode 13 are made of conductive materials selected from carbon (C), copper (Cu), aluminum (Al), titanium (Ti) or silver (Ag), the distance between the two electrodes is 0.04m, and the electrode leads are connected by metal conductive wires; the charge collection needle 14 is 4mm long and is made of platinum (Pt), copper (Cu), aluminum (Al) or silver (Ag) conductive materials; the base substrate 15 is a 3D printed white resin insulating material with a thickness of 4mm; the friction layer surface 11 is a hydrophobic solid material selected from polyvinyl chloride (PVC), Kapton, polytetrafluoroethylene (PTFE) or perfluoroethylene propylene copolymer (FEP), with a surface contact angle ≥90°, and has a certain degree of hydrophobicity; The electrical signals of acid rain with different pH values ​​are generated by: the volume of the acid rain droplets 18 to be tested naturally falling through the acid rain circulation pipe 17 is 80ul, and the external energy is realized by gravity drive; the electrical signals are generated by the acid rain droplets 18 to be tested hitting the friction layer surface 11; S2 collects the electrical signals of acid rain with different pH values ​​generated by the full-current nanogenerator and preprocesses them into standardized time series data. The preprocessed time series data is input into an algorithm processing module based on a ResNet 18-1D convolutional neural network. The network's residual block structure extracts multi-level time series features and classifies and identifies the corresponding pH values. The algorithm processing module uses the ResNet 18-1D CNN to extract and classify the electrical signals of acid rain with different pH values. The traditional ResNet18 model replaces the 2D convolution kernel with a 1D convolution, preserving the residual block structure. Skip connections are used to preserve shallow details and deep semantics, preventing gradient vanishing and improving sensitivity to subtle pH differences. Global average pooling (GAP) is used to compress temporal features into a fixed-length vector. This is then followed by a fully connected layer to output logits of seven pH categories, which are then converted to probability distributions using a softmax algorithm. A weighted cross-entropy loss is used to address pH category imbalance (e.g., a high number of neutral samples). The model's robustness to signal fluctuations is enhanced by adding Gaussian noise, time offsetting, or amplitude scaling.

[0031] S3, the waveform diagram, feature extraction results and pH value classification and identification results of the extracted and classified acid rain electrical signals are displayed in real time through the interactive interface, so as to enable users to dynamically query and analyze the monitoring data.

[0032] Through low-latency data collection and rendering technology, real-time visualization with end-to-end latency less than 0.5s is achieved. Combined with expression feedback images, the user interaction experience is significantly improved.

[0033] Illustratively, in step S2, the conduction current and displacement current act synergistically to enhance sensitivity to acid rain droplets; through the coupling of solid-liquid contact electrification and electrostatic induction, a transient current pulse containing pH characteristics is generated, capturing multi-dimensional time series features such as peak value (such as 100 nA for pH 3 and 600 nA for pH 7), rising edge (such as about 5 ms for pH 3 and about 10 ms for pH 7), falling edge and signal interval, significantly improving the discrimination of acid rain signals with different pH values, and providing rich information for the accurate classification of the ResNet 18-1D model.

[0034] For example, in step S2, to accurately map the electrical signal characteristics of acid rain to different pH values, the present invention uses a ResNet 18-1D convolutional neural network (CNN), specifically designed to process the one-dimensional time-series electrical signals generated by a full-current nanogenerator (TENG). The following details the model architecture, data processing flow, and classification mechanism.

[0035] The electrical signals of acid rain with different pH values ​​generated by the full-current nanogenerator are collected specifically including: The electrical signals generated by the full-current nanogenerator were collected using a NI acquisition card. The signal duration ranged from 0.1 to 5 seconds and included current pulses caused by the impact of acid rain droplets. To improve signal quality, the following preprocessing steps were used: Denoising: Wavelet transform (Daubechies wavelet, db4, decomposition level 4) is applied to remove high-frequency noise and retain the time-frequency characteristics of 0-5 kHz.

[0036] Normalization: Normalize the signal amplitude to [-1, 1] to ensure the robustness of the model to different droplet volumes or impact velocities.

[0037] Time-frequency analysis: Short-time Fourier transform (STFT, window size 256, hop size 128) is applied to the signal to extract the frequency components (0-5kHz) and time-domain variation characteristics, forming a two-dimensional time-frequency spectrum as auxiliary input.

[0038] An exemplary ResNet 18-1D convolutional neural network (CNN) architecture includes: The input layer receives a one-dimensional electrical signal vector of length 5000 (0.5 seconds, 10kHz sampling rate) or a 256×256 time-frequency spectrum (depending on the preprocessing mode).

[0039] The initial convolution block, combined with batch normalization (BatchNorm) and ReLU activation function, is used for preliminary feature extraction. As the first layer processing unit for signals entering the neural network, the initial convolution block undertakes the key task of capturing basic features from the original electrical signal. Its typical structure includes: 1) One-dimensional convolution layer (Conv1D): Use a larger convolution kernel size of 7 and a stride of 2 (kernel_size=7, stride=2), output 64 channels, and perform a preliminary scan of the time series signal.

[0040] 2) Batch Normalization Layer (BatchNorm1D): Normalizes the convolution output to accelerate training and enhance model stability.

[0041] 3) ReLU activation function: introduces nonlinear characteristics and filters negative signal components.

[0042] The residual module consists of 8 residual blocks (convolution kernel size 5, stride 1) and residual connections, and the number of channels gradually increases to 256 to enhance feature extraction capabilities.

[0043] Global average pooling is used to reduce the dimensionality of feature maps.

[0044] The fully connected layer FC outputs the probability distribution of multiple pH value categories through two layers of fully connected networks.

[0045] In step S2, the feature extraction process is implemented through the residual module and convolutional layer of ResNet 18-1D, including: Initial convolutional block: Convolution kernel size 7, stride 2, 64-channel output, combined with batch normalization (BatchNorm) and ReLU activation function for preliminary feature extraction. Wide kernel convolution captures multi-scale low-level features and expands the initial receptive field.

[0046] Residual module: Contains 8 residual blocks, each consisting of two convolutional layers (kernel size 5, stride 1) and residual connections. The number of channels is increased to 256 to enhance feature extraction capabilities. This module significantly improves the feature capture capability of the non-stationary one-dimensional electrical signal generated by the full-current nanogenerator (TENG) through the residual learning mechanism. The specific process is as follows: (1) Gradient optimization of residual connection: Each residual block adds the input directly to the output of the convolution layer through a skip connection to form ,in, is the convolution transformation, The residual connections alleviate the vanishing gradient problem in deep networks, ensuring stable convergence of the 18-layer network during training. To address the transient pulse characteristics of TENG signals (such as the 10 ms current spike generated by acid rain droplet impact), the residual connections preserve low-level features of the original signal (such as peak amplitude) while allowing high-level convolutional layers to capture complex patterns (such as pH-related signal decay).

[0047] (2) Multi-scale feature extraction: The two convolutional layers (kernel size 5) of each residual block expand the receptive field through continuous convolution operations, capturing the local temporal patterns (such as spike width and rise / fall time) and global periodic features (such as the signal interval caused by multiple droplets) of the TENG signal. In this example, the difference between the spike width of the pH 3 acid rain signal (about 8 ms) and the pH 7 pure water signal (about 5 ms) is effectively extracted. Increasing the number of channels from 64 to 256 further enhances the model's ability to express multi-scale features.

[0048] (3) Batch Normalization and Activation Enhancement: Each convolutional layer is followed by batch normalization (BatchNorm) and ReLU activation functions. BatchNorm standardizes feature distribution to accelerate training and improve the model's robustness to different droplet volumes or impact velocities. ReLU enhances nonlinear expression capabilities, enabling the model to better distinguish the electrical signal characteristics corresponding to the pH value of acid rain. For example, low-pH acid rain (pH 3) has a low charge transfer efficiency due to the high H⁺ concentration, resulting in a small signal amplitude. The residual module amplifies the discrimination of such weak features through nonlinear transformation.

[0049] (4) Compatibility with TENG signals: TENG signals have high noise and non-stationary characteristics (such as the randomness of acid rain droplet impact). The residual module separates noise from pH-related signals through deep feature fusion. For example, the low-frequency component (0-2 kHz) of the pH 3 signal and the high-frequency component (2-5 kHz) of the pH 7 signal are effectively distinguished, significantly improving the classification accuracy (98.5% on the test set).

[0050] In summary, the residual module achieves efficient feature capture and robust classification of acid rain electrical signals with different pH values ​​through gradient optimization, multi-scale feature extraction and nonlinear enhancement, combined with the transient and non-stationary characteristics of TENG signals, providing core technical support for intelligent acid rain monitoring.

[0051] Global average pooling: Reduce the dimension of the feature map to a 256-dimensional vector.

[0052] Fully connected layer (FC): Outputs the probability distribution of 7 pH value categories (pH 1-7, step size 1) through two layers of fully connected networks (256→128→7).

[0053] Exemplarily, in step S2, the feature extraction process is implemented through the residual module and convolution layer of ResNet 18-1D, and efficient feature capture and pH value classification are performed on the one-dimensional time-series electrical signal generated by the full-current nanogenerator (TENG), specifically including: feature extraction and classification.

[0054] The feature extraction process includes: (1) Time domain feature extraction: The residual block captures the local temporal patterns and global patterns of the TENG signal through multi-layer convolution (kernel size 5, stride 1). The local pattern includes the current spike characteristics caused by the impact of acid rain droplets, such as the peak width (approximately 8 ms for pH = 3 and approximately 5 ms for pH = 7), rise time, and decay time; the global pattern includes the periodic regularity of multiple droplet impacts (such as a signal interval of 1 drop / second). These features reflect the effect of acid rain with different pH values ​​on the solid-liquid contact electrification efficiency. For example, low pH acid rain (pH = 3) has a high H⁺ concentration, which inhibits charge transfer and results in a low signal amplitude.

[0055] (2) Multi-scale feature fusion: The eight residual blocks of ResNet 18-1D achieve multi-scale feature fusion by gradually increasing the number of channels (64 → 128 → 256). Low-level convolutions retain signal details (such as spike shapes), while high-level convolutions capture abstract patterns (such as the periodic modulation of pH on the signal). Residual connections (skip connections) ensure that the deep network retains the original signal information during the fusion process, avoiding information loss. For example, the amplitude difference between pH 3 and pH 7 signals (100 nA vs. 600 nA) is amplified through multi-scale fusion, enhancing classification discrimination.

[0056] (3) Nonlinear feature enhancement: Each convolutional layer is followed by a ReLU activation function and batch normalization (BatchNorm). ReLU improves the model's adaptability to the non-stationary characteristics of TENG signals by introducing nonlinear transformations, while BatchNorm ensures the model's robustness to different droplet volumes or impact velocities by standardizing feature distributions. These mechanisms enable the model to effectively distinguish weak signal differences, such as the current peak difference between pH 3 and pH 4 (approximately 20 nA).

[0057] Feature classification includes: mapping the FC layer output to a probability distribution through the softmax function, expressed as: ; Where, is the predicted probability, For the The logit values ​​of the pH categories, For the Logit values ​​for pH categories, summation in the denominator Normalize the index output of all categories to ensure the properties of probability distribution; The cross entropy loss function is used for optimization, and the expression is: ; Where, Cross-Entropy Loss is a function that measures the difference between the model's predicted probability distribution and the true label distribution. The smaller its value, the closer the prediction is to the true distribution. For the True labels of categories; The model is The predicted probability of each category is converted into a probability distribution through the Softmax function (all The sum is 1).

[0058] Exemplarily, after the ResNet 18-1D convolutional neural network feature extraction and classification, model training and verification are also performed to verify the feasibility of the ResNet 18-1D convolutional neural network.

[0059] The training dataset contains 7,000 acid rain electrical signal samples (pH 1-7, 1,000 samples per class). These samples were generated by testing TENGs with laboratory-synthesized acid rain (H₂SO₄ / HNO₃ mixture). The training and test sets were split 8:2. The Adam optimizer (with an initial learning rate of 0.001, decaying by 0.1 every 10 epochs) was used for 50 training epochs with a batch size of 32. The classification accuracy on the test set reached 98.5%.

[0060] Compared to traditional CNNs, ResNet 18-1D improves its feature extraction capabilities for non-stationary TENG signals through residual connections. Compared to traditional monitoring systems (latency > 1000ms), this system achieves end-to-end latency of less than 0.5s through NIMAX low-latency acquisition (latency < 5ms) and FuncAnimation fast rendering. The interface supports waveform scaling and, taking into account the transient characteristics of TENG signals (100ms pulses), achieves high-precision classification, thanks to the synergistic optimization of the unique time-frequency characteristics of TENG signals and deep learning, making it suitable for real-time, portable acid rain monitoring.

[0061] In Example 2, the intelligent acid rain sensing and monitoring system based on the full-current nanogenerator provided by the embodiment of the present invention includes: Full-current nanogenerator, used to monitor acid rain at different pH values; An electrical signal generating unit, used for generating electrical signals of acid rain with different pH values; The electrical signal acquisition, feature extraction, and classification unit is used to collect the electrical signals of acid rain with different pH values ​​generated by the full-current nanogenerator and preprocess them into standardized time series data. The preprocessed time series data is input into an algorithm processing module based on the ResNet 18-1D convolutional neural network. The network's residual block structure is used to extract multi-level time series features and classify the corresponding pH values. The algorithm processing module uses the ResNet 18-1D CNN to perform feature extraction and classification on the electrical signals of acid rain with different pH values. The interactive interface is used to display the waveform diagram, feature extraction results and pH value classification and identification results of the extracted and classified acid rain electrical signals in real time through the interactive interface, so as to enable users to dynamically query and analyze the monitoring data.

[0062] Example 3, as another possible implementation of the present invention, see Figure 2 , the preparation of full current nano power generation includes: Step 1: Using photo-curing 3D printing technology, design and print a resin outer model 16 for collecting the acid rain droplets 18 to be tested. A base substrate 15 (a fluoroethylene propylene copolymer (FEP) film (80 μm thick)) is attached to the lower surface of the resin outer model 16. Conductive tape (in this example, carbon tape) serves as the top electrode 12 and the bottom electrode 13. The top electrode 12 is attached with a conductive wire (in this example, a platinum wire) as a charge collection needle 14, while the bottom electrode 13 does not require a charge collection needle 14. The two electrodes are placed on the base substrate 15 (a fluoroethylene propylene copolymer (FEP) film) at a predetermined distance from each other and connected by wires (in this example, copper wire) as leads.

[0063] Step 2: Select a pH acid rain solution and drop it onto the inclined surface of the full-current nanogenerator, while ensuring that the acid rain droplets 18 to be tested can contact the charge collection needles 14 and the bottom electrode 13 respectively.

[0064] Step 3: When the acid rain droplet 18 to be tested hits the friction layer surface 11 under the action of gravity, the solid-liquid interface triggers charge transfer due to the contact electrification effect, and the negative charge in the acid rain droplet 18 to be tested is injected into the friction layer, and the acid rain droplet 18 to be tested carries an equal amount of positive charge; subsequently, when the acid rain droplet 18 to be tested hits the top electrode 12 and the charge collection needle 14, the negative charge is transferred to the top electrode 12, and the remaining positive charge slides with the acid rain droplet 18 to be tested to the bottom electrode 13 to complete the transfer.

[0065] Example 4, as another possible implementation of the present invention, performs dynamic sensing and signal analysis of acid rain pH value based on full-current nanogenerator.

[0066] See Figure 3In order to demonstrate the sensing current signals of the full-current nanogenerator to acid rain with different pH values ​​and pure water, comprehensive tests were carried out on acid rain droplets with different pH values. The signal differences were significant. The device system has good sensitivity to acid rain with different pH values, which further verifies the superiority of this sensing system.

[0067] Step 1: dropping the acid rain droplet 18 to be tested onto the surface of the full-current nanogenerator; Step 2: Use a high-sensitivity data acquisition module to monitor in real time the electrical signal caused by the falling of the acid rain droplets 18 on the surface of the full-current nanogenerator.

[0068] Example 5, as another possible implementation of the present invention, performs acid rain pH intelligent detection based on ResNet18-1D.

[0069] The ResNet 18-1D convolutional learning network of the present invention extracts and deeply learns the current signals of acid rain with different pH values. By dropping acid rain droplets with different pH values, the interface displays the identification results of acid rain with different pH values ​​in real time. Specifically, the following steps are included: Step 1: Extract signal features through the full current pulse network and classify and output the identification results of different acid rain pH values; Step 2: The interactive interface displays signal waveforms, acid rain identification results, pictures and other information in real time.

[0070] See Figure 4 When switching from normal water droplets to acid rain, the output current signal drops rapidly, indirectly demonstrating the device's excellent response speed. Notably, the device's triboelectric layer surface exhibits excellent recoverability. When normal water droplets are reintroduced, the output signal returns to its initial level within just two drops of solution. This rapidly reversible response demonstrates: 1) the hydrogen ion adsorption process in acid rain is fully reversible; 2) the surface state of the triboelectric material is highly stable; and 3) the double-layer structure has a strong self-healing ability.

[0071] See Figure 5 Through experiments, the electrical signal data of solutions with different pH values ​​were collected. After deep learning with the ResNet 18-1D model, the recognition accuracy of solutions with different pH values ​​reached 99.86%, which shows high-precision recognition and reversible response characteristics.

[0072] The above description is only a preferred specific implementation method of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. An intelligent acid rain sensing and monitoring method based on full-current nanogenerator, characterized in that: The method comprises the following steps: S1, preparing a full-current nanogenerator to generate electrical signals of acid rain with different pH values, using a process of photocuring 3D printing to achieve biomimetic structure reproduction to prepare the full-current nanogenerator, so that it generates electrical signals with different characteristics when exposed to acid rain with different pH values; S2 collects the electrical signals of acid rain with different pH values ​​generated by the full-current nanogenerator and preprocesses them into standardized time series data. The preprocessed time series data is input into an algorithm processing module based on a ResNet 18-1D convolutional neural network. The network's residual block structure extracts multi-level time series features and classifies and identifies the corresponding pH values. The algorithm processing module uses the ResNet 18-1D CNN to extract and classify the electrical signals of acid rain with different pH values. S3, the waveform diagram, feature extraction results and pH value classification and identification results of the extracted and classified acid rain electrical signals are displayed in real time through the interactive interface, so as to enable users to dynamically query and analyze the monitoring data.

2. The intelligent acid rain sensing and monitoring method based on full-current nanogenerator according to claim 1 is characterized in that: In step S1, a full-current nanogenerator is prepared, including: Using photo-curing 3D printing technology, a resin external model (16) is designed and printed for collecting the acid rain droplets to be tested (18); A base substrate (15) is provided at the lower portion of the resin outer model (16), and a friction layer surface (11) is provided on the base substrate (15); a top electrode (12) and a bottom electrode (13) are respectively installed on the top and bottom of the friction layer surface (11), and a charge collecting needle (14) is installed on the top electrode (12).

3. The intelligent acid rain sensing and monitoring method based on full-current nanogenerator according to claim 1 is characterized in that: In step S1, electrical signals of acid rain with different pH values ​​are generated, including: acid rain droplets (18) to be measured fall naturally through the acid rain circulation pipe (17), and electrical signals are generated by the acid rain droplets (18) to be measured hitting the surface (11) of the friction layer.

4. The intelligent acid rain sensing and monitoring method based on full-current nanogenerator according to claim 1 is characterized in that: In step S2, collecting electrical signals of acid rain with different pH values ​​generated by the full-current nanogenerator includes: The electrical signal generated by the all-current nanogenerator is collected through a NI acquisition card. The signal duration is 0.1-5 seconds and includes current pulses caused by the impact of acid rain droplets. The generated electrical signal is denoised, normalized, and subjected to time-frequency analysis. The time-frequency analysis includes applying short-time Fourier transform (STFT) to the signal to extract frequency components and time-domain variation characteristics to form a two-dimensional time-frequency spectrum.

5. The intelligent acid rain sensing and monitoring method based on full-current nanogenerator according to claim 1 is characterized in that: In step S2, the ResNet 18-1D convolutional neural network includes: The input layer receives a one-dimensional electrical signal vector; the input one-dimensional electrical signal is normalized to a fixed length and processed by sliding window segmentation; Initial convolution block, the convolution kernel is combined with batch normalization BatchNorm and ReLU activation function for preliminary feature extraction; the initial convolution block uses a one-dimensional convolution kernel to perform local feature scanning on the input signal; The residual module consists of 8 residual blocks, each of which consists of two convolutional layers and a residual connection; Global average pooling is used to reduce the dimensionality of feature maps; The fully connected layer FC outputs the probability distribution of multiple pH value categories through two layers of fully connected networks.

6. The intelligent acid rain sensing and monitoring method based on full-current nanogenerator according to claim 1 is characterized in that: In step S2, the feature extraction process is implemented through the residual module and convolutional layer of ResNet 18-1D, which efficiently captures features and classifies pH values ​​of the one-dimensional time-series electrical signals generated by the full-current nanogenerator. Specifically, it includes: Time domain features, using residual blocks to capture local and global patterns of signals; residual blocks capture local timing patterns and global patterns of TENG signals through multi-layer convolution; Multi-scale feature fusion: The eight residual blocks of ResNet 18-1D achieve multi-scale feature fusion by gradually increasing the number of channels. Low-level convolutions retain signal details, while high-level convolutions capture abstract patterns. Residual connections ensure that the deep network retains original signal information during the fusion process. Nonlinear feature enhancement: Each convolutional layer is followed by a ReLU activation function and batch normalization. ReLU improves the model's adaptability to the non-stationary characteristics of TENG signals by introducing nonlinear transformations. BatchNorm ensures the model's robustness to different droplet volumes or impact velocities by standardizing feature distributions.

7. The intelligent acid rain sensing and monitoring method based on full-current nanogenerator according to claim 1 is characterized in that: In step S2, feature classification includes mapping the FC layer output to a probability distribution through the softmax function, which is expressed as: ； Where, is the predicted probability, For the logit values ​​for each pH category; For the Logit values ​​for pH categories, summation in the denominator It is to normalize the index output of all categories; The cross entropy loss function is used for optimization, and the expression is: ； Where, is the cross entropy loss function, For the True labels of categories; The model is The predicted probability of each category.

8. The intelligent acid rain sensing and monitoring method based on full-current nanogenerator according to claim 1 is characterized in that: In step S2, after feature extraction and classification, model training and verification are also performed.

9. An intelligent acid rain sensing and monitoring system based on full-current nanogenerator, characterized in that: The system implements the intelligent acid rain sensing and monitoring method based on the full-current nanogenerator according to any one of claims 1 to 8, and the system comprises: Full-current nanogenerator, used to monitor acid rain at different pH values; An electrical signal generating unit, used for generating electrical signals of acid rain with different pH values; The electrical signal acquisition, feature extraction, and classification unit is used to collect the electrical signals of acid rain with different pH values ​​generated by the full-current nanogenerator and preprocess them into standardized time series data. The preprocessed time series data is input into an algorithm processing module based on the ResNet 18-1D convolutional neural network. The network's residual block structure is used to extract multi-level time series features and classify the corresponding pH values. The algorithm processing module uses the ResNet 18-1D CNN to perform feature extraction and classification on the electrical signals of acid rain with different pH values. The interactive interface is used to display the waveform diagram, feature extraction results and pH value classification and identification results of the extracted and classified acid rain electrical signals in real time through the interactive interface, so as to enable users to dynamically query and analyze the monitoring data.

Images