Intelligent perception driven real-time data monitoring and safety evaluation system and method

Abstract

The invention discloses an intelligent perception driven real-time data monitoring and safety evaluation system and method in the technical field of intelligent perception, and the method carries out the monitoring and evaluation after the data fusion of a sensor, and the data fusion comprises the steps: obtaining the real-time data of the sensor; constructing a virtual sensor model in the multi-level neural network by using real-time data of the sensor; performing data prediction according to the relevance of the multi-source data through the virtual sensor model to obtain the output of the virtual sensor model; performing error correction on the output of the virtual sensor model according to an error correction mechanism; and performing data fusion based on the output of the virtual sensor model after error correction. According to the method, a virtual sensor model is generated on the basis of a multi-level neural network in combination with real-time data of different sensors, and nodes which may have data drift or errors are predicted and calibrated in real time on the basis of relevance of multi-source data, so that highly consistent and high-precision data fusion is realized.

Description

Technical Field

[0001] The present invention relates to an intelligent perception driven real-time data monitoring and safety assessment system and method, belonging to the field of intelligent perception technology. Background Art

[0002] Intelligent perception technology refers to mapping the signals of the physical world to the digital world through cameras, microphones or other sensor hardware devices, with the help of cutting-edge technologies such as voice recognition and image recognition, and then further elevating these digital information to a cognitive level. With the rapid development of the Internet of Things, artificial intelligence and big data technologies, real-time data monitoring and security assessment systems based on intelligent perception have been widely used in industrial production, smart cities, environmental monitoring and other fields. These systems collect multi-dimensional and high-frequency data in real time through a large number of sensor nodes, providing an important basis for decision support and security management.

[0003] In the existing technology, most multi-sensor data fusion methods rely on fixed rules or simple weighted averages, which are difficult to adapt to real-time dynamically changing environments, resulting in insufficient accuracy and reliability of data fusion, and limiting the system's ability to adapt to complex scenarios. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a real-time data monitoring and safety assessment system and method driven by intelligent perception, which combines the real-time data of different sensors to generate a virtual sensor model, and based on the correlation of multi-source data, predicts and calibrates nodes that may have data drift or errors in real time, thereby achieving highly consistent and high-precision data fusion.

[0005] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0006] In a first aspect, the present invention provides a method for real-time data monitoring and security assessment driven by intelligent perception, wherein the method monitors and assesses after fusing sensor data, wherein the data fusion comprises:

[0007] Get real-time data from sensors;

[0008] Use the real-time data from sensors to build a virtual sensor model in a multi-level neural network;

[0009] The virtual sensor model is used to predict data according to the correlation of multi-source data to obtain the virtual sensor model output;

[0010] Performing error correction on the virtual sensor model output according to the error correction mechanism;

[0011] Data fusion is performed based on the error-corrected virtual sensor model output.

[0012] Furthermore, the virtual sensor model output is:

[0013]

[0014] in: is the output of the virtual sensor model, f is the nonlinear mapping function learned by the deep neural network, and w i (t) is the weight coefficient of the th sensor, x i (t) is the output of sensor node i;

[0015] The error correction formula of the virtual sensor model is:

[0016]

[0017] in: is the corrected virtual sensor model output, is the virtual sensor model output, γ is the error correction coefficient, and z(t) is the actual observation value or ground truth.

[0018] Furthermore, the output data of the sensor is evaluated for sensor accuracy using a deep reinforcement learning calibration model, and the deep reinforcement learning calibration model is calibrated through multiple experimental learning and optimization, and preprocessed using a model-based adaptive filtering algorithm. At the same time, the latest sensor status is fed back into the filter training process through model adaptive updating.

[0019] Furthermore, the output data of the sensor adopts an adaptive data compression algorithm based on deep learning, and the data compression ratio is dynamically adjusted according to the sensor type, data content and current network bandwidth status, wherein the high-dimensional data in the output data is compressed after feature extraction through a convolutional neural network, and representative features related to the sensor monitoring content are extracted for encoding, and the low-dimensional data in the output data is compressed using differential coding and Huffman coding methods;

[0020] The output data of the sensor uses a bandwidth adaptive adaptive compression function to dynamically adjust the compression strategy according to bandwidth limitations and data characteristics. The bandwidth adaptive adaptive compression function formula includes:

[0021] C=CNN(D,w,λ)

[0022] Where: C is the compressed data, D is the original data, w is the learnable weight in the convolutional neural network, and λ is the dynamic bandwidth factor;

[0023] The bandwidth factor adjustment function is:

[0024]

[0025] Among them: B avail is the current available bandwidth, B req is the current transmission required bandwidth, and λ(t) is the bandwidth factor at time t.

[0026] Furthermore, the sensor network adopts a deep reinforcement learning routing optimization algorithm, and each node in the sensor network can make routing decisions based on the current network status, and predict energy consumption and load conditions based on a long short-term memory network model, and dynamically adjust task allocation according to the energy status and load conditions of the node through a hierarchical load balancing algorithm, and apply for and allocate tasks based on its current status and information in the blockchain, and make decisions based on its own status when participating in task scheduling and reach consensus through a consensus algorithm, and make decisions based on a utility function that comprehensively considers multiple factors of the node, and dynamically adjust the proportion of resource sharing according to the load and energy status.

[0027] Furthermore, the virtual sensor model uses few-sample learning technology and online update mechanism to perform model training, and uses real-time feedback and automatic model adjustment to perform model correction, and uses decision sharing and collaborative mechanism to perform task allocation execution and collaborative training, and uses model adaptation and dynamic adjustment to perform model adjustment, and uses data enhancement to perform model generalization, and uses modular integration and deployment configuration, and uses data encryption and model encryption technology.

[0028] Furthermore, the sensor performs real-time monitoring and prediction through an intelligent battery management system combined with machine learning and big data analysis. The intelligent battery management system uses the historical energy consumption data of the sensor and the current environmental conditions to establish an energy consumption prediction model, and the formula is:

[0029]

[0030] Where: E(t) is the energy state of the device, t is time, Δt is the time step, P i The energy consumption of each task, τ i is the execution time, H(t) is the energy obtained by recycling or charging, and the optimization objective function and constraints of the energy consumption prediction model are:

[0031]

[0032] Where: T i is the task set, U(E) is the energy state of the equipment, and w i For task T i The weight, S i For task T i Successful execution rate;

[0033] Energy constraint: E(t+△t)≥Emin, where Emin is the minimum threshold;

[0034] Time constraint: Σ i τ i ≤Δt.

[0035] In a second aspect, the present invention provides an intelligent perception-driven real-time data monitoring and security assessment system, comprising:

[0036] Data acquisition module: obtain real-time data from sensors;

[0037] Model building module: Use the real-time data of sensors to build a virtual sensor model in a multi-level neural network;

[0038] Data prediction module: performs data prediction based on the correlation of multi-source data through the virtual sensor model to obtain the virtual sensor model output;

[0039] Error correction module: performs error correction on the output of the virtual sensor model according to the error correction mechanism;

[0040] Data fusion module: performs data fusion based on the virtual sensor model output after error correction.

[0041] In a third aspect, the present invention provides an intelligent perception-driven real-time data monitoring and security assessment device, including a processor and a storage medium;

[0042] The storage medium is used to store instructions;

[0043] The processor is used to operate according to the instructions to execute the steps of any of the methods described above.

[0044] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of any of the above methods when executed by a processor.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. Based on a multi-level neural network, the present invention combines the real-time data of different sensors to generate a virtual sensor model node. The role of the virtual sensor model is to predict and calibrate nodes that may have data drift or errors in real time based on the correlation of multi-source data, thereby achieving highly consistent and high-precision data fusion;

[0047] Second, the present invention uses an intelligent adaptive filtering algorithm and a model-based dynamic adjustment method, combined with a deep learning model and reinforcement learning, so that the system can automatically adjust the filtering strategy according to different environmental conditions, sensor states and acquisition task requirements, and optimize the real-time and accuracy of data calibration;

[0048] 3. The present invention introduces an innovative bandwidth-adaptive adaptive compression function to dynamically adjust the compression strategy according to bandwidth limitations and data characteristics, thereby effectively improving data compression efficiency and minimizing transmission delay;

[0049] 4. The present invention further refines and improves the node dynamic scheduling and network optimization strategy through self-organizing network optimization, energy and load prediction, intelligent routing algorithm, decentralized task scheduling and other means to ensure the efficient operation of the sensor network in a changing environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0051] Figure 1 A flowchart of an intelligent perception-driven real-time data monitoring and security assessment method provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0052] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present invention can be combined with each other without conflict.

[0053] The following detailed description is an exemplary description, which is intended to provide further detailed description of the present invention. Unless otherwise specified, all technical terms used in the present invention have the same meaning as those generally understood by those skilled in the art to which the present invention belongs. The terms used in the present invention are only for describing specific embodiments, and are not intended to limit the exemplary embodiments according to the present invention.

[0054] Embodiment 1:

[0055] In an environment with multiple devices and heterogeneous sensors, how to ensure data consistency and synchronization is a complex issue. Most existing multi-sensor data fusion methods rely on fixed rules or simple weighted averages, which are difficult to adapt to real-time dynamically changing environments, resulting in insufficient accuracy and reliability of data fusion, and limiting the system's ability to adapt to complex scenarios. To this end, this embodiment proposes a real-time data monitoring and security assessment method driven by intelligent perception, which realizes cross-device data fusion through comprehensive modeling of multi-dimensional data sources based on virtual sensor model technology. The specific approach is to generate a virtual sensor model node based on multi-layer neural networks (Multilayer Neural Networks, MNN) combined with real-time data from different devices. The role of the virtual sensor model is to predict and calibrate nodes that may have data drift or errors in real time based on the correlation of multi-source data, thereby achieving highly consistent and high-precision data fusion.

[0056] Virtual sensor model building: The input-output relationship of multiple sensor nodes is trained through a deep neural network (DNN) to learn the environmental response model of each sensor. Then, the trained model is used to perform data prediction and error correction in the virtual sensor model to improve the accuracy of data fusion.

[0057] Virtual node coordination mechanism: Each device or sensor node not only relies on its own data to make decisions, but can also make data corrections based on the output of the virtual sensor model. The virtual sensor model can be dynamically adjusted based on the correlation and feedback information between multiple sensor data to reduce data deviation in a multi-source environment. The modeling formula of the virtual sensor model includes:

[0058] The data of each sensor node is processed and fused through a deep neural network. Let the output of sensor node i be x i (t), the trained virtual sensor model output for:

[0059]

[0060] Where: w i (t) is the weight coefficient of the th sensor, which is dynamically calculated based on the sensor quality; f is the nonlinear mapping function learned through a deep neural network, which is used to further improve the accuracy of data fusion.

[0061] There may be drift or error in the output of the virtual sensor model, so an error correction mechanism is introduced. Based on Kalman filtering and adaptive model adjustment, the virtual sensor model error correction formula is:

[0062]

[0063] Where: γ is the error correction coefficient, which is dynamically adjusted to ensure the accuracy of the output data, z(t) is the actual observation value or ground truth, is the virtual sensor model output, is the output of the modified virtual sensor model.

[0064] Kalman filtering and extended Kalman filtering (EKF) in the prior art can deal with the sensor drift problem to a certain extent, but these algorithms often face problems such as high computational complexity and poor adaptability when facing a large amount of high-dimensional data. Therefore, in view of the sensor data quality and calibration problems existing in the prior art, through the intelligent adaptive filtering algorithm and the model-based dynamic adjustment method, combined with the deep learning model and reinforcement learning, the system can automatically adjust the filtering strategy according to different environmental conditions, sensor status and acquisition task requirements, and optimize the real-time and accuracy of data calibration.

[0065] Deep reinforcement learning calibration model: Through reinforcement learning algorithms such as Deep Q-Learning or Proximal Policy Optimization (PPO), the accuracy of each sensor is evaluated in real time during system operation, and the calibration model is learned and optimized through multiple experiments. The system will dynamically determine whether the sensor has drift or data inconsistency, and adjust the data calibration strategy based on real-time feedback.

[0066] Dynamic filtering strategy: Combined with the model-based adaptive filtering algorithm, the data collected by the sensor is quickly preprocessed. At the same time, the latest sensor status is fed back to the filter training process through model adaptive update, thereby ensuring that the filter can dynamically adapt to environmental changes and sensor status changes.

[0067] In the case of limited network bandwidth, the real-time transmission of large-scale sensor data faces severe challenges. Traditional data transmission methods are prone to data loss and delay in low-bandwidth environments, which affects the real-time performance and reliability of the system and cannot meet the high requirements of data transmission for critical tasks. In order to optimize data transmission under limited bandwidth, this solution proposes a series of innovative technical solutions.

[0068] 1. Data compression and transmission optimization algorithm:

[0069] In a low-bandwidth environment, in order to minimize the amount of data transmission, an adaptive data compression algorithm based on deep learning is used. This algorithm can dynamically adjust the data compression ratio according to the sensor type, data content and current network bandwidth status. Specifically, high-dimensional data (such as video streams, image streams, etc.) are compressed after feature extraction through a convolutional neural network (CNN), and representative features related to the sensor monitoring content are extracted for encoding. Low-dimensional data (such as simple sensor data such as temperature and humidity) are compressed using differential coding and Huffman coding methods to ensure a balance between compression efficiency and data quality.

[0070] 1.1 Convolutional Neural Network (CNN) Compression:

[0071] Convolutional neural networks can extract spatial features of data through convolutional layers, reduce redundant information, and thus significantly improve the data compression ratio. This method is particularly suitable for high-dimensional data such as images and videos. It can significantly reduce the amount of data and improve transmission efficiency while ensuring data quality.

[0072] 1.2 Differential Coding and Huffman Coding:

[0073] For sensor data such as temperature and humidity, which usually changes slowly, differential coding is suitable for compression. Huffman coding can encode frequently occurring values ​​in a shorter format based on the probability of occurrence of the data, thereby further optimizing data transmission efficiency.

[0074] 2. Packet scheduling mechanism:

[0075] In order to effectively utilize bandwidth resources, reduce transmission delays, and ensure data real-time and integrity, this scheme proposes a dynamic scheduling algorithm that can flexibly adjust the transmission order of data packets according to network bandwidth, sensor data priority, and delay requirements.

[0076] 2.1 Priority Scheduling Mechanism:

[0077] According to the importance and real-time requirements of the data, key data (such as alarm information, system status, etc.) are transmitted first, and the transmission frequency of unimportant data (such as historical data, non-real-time data) is reduced. For urgent data or data with timeliness requirements, low-latency paths are used for priority transmission to ensure real-time performance.

[0078] 2.2 Bandwidth Adaptive Scheduling:

[0079] The algorithm dynamically evaluates the current bandwidth situation and adjusts the packet size according to the available bandwidth to avoid data loss or congestion caused by insufficient bandwidth.

[0080] 3. Error detection and recovery during transmission:

[0081] In a low-bandwidth environment, network instability may lead to problems such as data packet loss and transmission errors. Therefore, this solution introduces resumable coding (Reed-SolomonEncoding) technology to ensure the integrity of data during transmission.

[0082] 3.1 Recoverable Coding Technology:

[0083] Reed-Solomon coding is an efficient error correction coding method that adds redundant data for coding, so that even if some data packets are lost or damaged, the receiver can still recover some of the lost data. In this way, the system can still ensure the complete transmission of data in an unstable network environment, avoiding the impact of packet loss or transmission errors on system operation.

[0084] In order to improve the efficiency of redundant data utilization during data transmission, we introduced an innovative bandwidth-adaptive adaptive compression function to dynamically adjust the compression strategy according to bandwidth limitations and data characteristics. The formula includes:

[0085] C=CNN(D,w,λ)

[0086] Where: C is the compressed data, D is the original data, w is the learnable weight in the convolutional neural network (CNN), and λ is the dynamic bandwidth factor, which indicates the impact of the current network bandwidth on the data compression rate.

[0087] The bandwidth factor adjustment function is:

[0088]

[0089] Among them: B avail is the current available bandwidth, B req is the current transmission required bandwidth, and λ(t) is the bandwidth factor at time t.

[0090] Through this algorithm, the system can adjust the compression strategy according to the real-time bandwidth conditions, thereby effectively improving data compression efficiency while minimizing transmission delays.

[0091] Routing and scheduling issues in dynamic network environments: Sensor networks are highly dynamic, and the energy, load, and network topology of nodes may change at any time. Traditional static routing protocols cannot adapt to this dynamic environment, resulting in increased communication delays, uneven energy consumption, and even node failures, which reduces the robustness and stability of the network. In order to address the above issues, this solution proposes innovative technical solutions. Through self-organizing network optimization, energy and load prediction, intelligent routing algorithms, decentralized task scheduling, etc., it further refines and improves node dynamic scheduling and network optimization strategies to ensure the efficient operation of sensor networks in a changing environment.

[0092] 1. Self-organizing routing optimization algorithm

[0093] Traditional static routing protocols cannot meet the requirements of fast-changing node states and unstable network environments in dynamic sensor networks. Therefore, this solution adopts a deep reinforcement learning (DRL) routing optimization algorithm to enable the network to adaptively select routes based on environmental changes. Each node not only makes decisions based on local information, but also learns and optimizes global routing strategies through interaction with other nodes to ensure data transmission reliability and minimum latency.

[0094] 1.1 Dynamic Routing Decision

[0095] Each node can make routing decisions based on the current network status (such as neighbor node quality, remaining energy, bandwidth, delay, etc.). On this basis, routing decisions not only consider the transmission delay of the current path, but also include factors such as node energy consumption and data packet transmission priority. Through learning and dynamic adjustment, the system can always maintain the lowest delay and efficient data transmission.

[0096] 1.2 Adaptive Routing Path Adjustment

[0097] The system continuously evaluates the routing selection of each node through a deep reinforcement learning algorithm, and dynamically optimizes the routing path according to changes in network status. When a node is in a low energy state or has a high load, the system will adjust the data flow to nodes with lighter loads and sufficient energy to avoid excessive consumption of resources of individual nodes and ensure the stable operation of the entire network.

[0098] 2. Node load and energy efficiency optimization

[0099] In order to effectively deal with the energy consumption and load imbalance problems of sensor nodes, this scheme proposes a scheduling mechanism based on energy and load prediction. Each node predicts future resource requirements by analyzing historical load and energy consumption, and makes reasonable allocation during task scheduling, thereby ensuring the balance of node load and energy consumption.

[0100] 2.1 Energy consumption and load forecasting

[0101] Each node predicts energy consumption and load based on the long short-term memory network (LSTM) model. The model can predict the future energy demand and load of the node based on its past operating status. When the node's energy is about to be exhausted or the load is close to saturation, the system will adjust the task allocation in advance to avoid the node from failing due to insufficient resources.

[0102] 2.2 Dynamic load distribution

[0103] Through the hierarchical load balancing algorithm, the system can dynamically adjust task allocation according to the energy status and load of the nodes. High-load nodes will transfer some tasks to nodes with lighter loads to ensure reasonable allocation of resources within the network, thereby avoiding failure of some nodes due to overload and enhancing the robustness of the network.

[0104] 3. Decentralized task scheduling and blockchain technology

[0105] In order to improve the robustness of the network and prevent single point failures, this solution introduces blockchain technology to achieve decentralized task scheduling. Through the blockchain, the status information of each node (such as energy, load, health status, etc.) will be recorded and stored in a decentralized ledger to ensure the transparency, tamper-proof and fairness of task scheduling.

[0106] 3.1 Decentralized Scheduling Mechanism

[0107] Smart contracts in the blockchain automatically perform task scheduling without relying on centralized control nodes. Each node applies for and allocates tasks based on its current state and information in the blockchain. When the energy, load and other states of the node change, the smart contract automatically adjusts its task allocation to ensure that no node in the network fails due to excessive load or insufficient energy.

[0108] 3.2 Consensus Mechanism and Resource Sharing

[0109] The consensus mechanism in the blockchain ensures fairness and consistency in task scheduling. When participating in task scheduling, each node makes decisions based on its own energy, load and other states, and reaches consensus through the consensus algorithm, thereby ensuring the rational allocation of resources and efficient operation of the network. In addition, the blockchain can also record the execution of historical tasks to provide data support for subsequent optimization.

[0110] 4. Multi-node collaboration and optimization

[0111] In order to improve the utilization of network resources and enhance the stability of sensor networks, this scheme uses game theory to model the resource competition among multiple nodes. Through the game model, nodes can select appropriate tasks and resources according to their own utility functions, thereby optimizing the overall network benefits.

[0112] 4.1 Game Theory Resource Allocation

[0113] Each node makes decisions based on its own utility function, which takes into account factors such as the node's energy consumption, communication overhead, and load. The node uses a game decision algorithm to determine whether to accept tasks or forward data, avoiding excessive consumption of its own resources and ensuring network stability.

[0114] 4.2 Dynamic Resource Sharing Mechanism

[0115] In order to optimize resource usage, nodes can dynamically adjust the resource sharing ratio according to load and energy conditions. High-load nodes transfer part of the tasks to low-load nodes, and use smart contract mechanisms to ensure the rationality of task transfer, prevent resource waste and network instability.

[0116] 5. Comprehensive application of innovative solutions

[0117] By integrating advanced technologies such as deep reinforcement learning, LSTM energy load prediction, blockchain decentralized scheduling, and game theory, this solution can optimize network routing, assign node tasks, and balance load and energy consumption in a dynamically changing environment. The system can automatically adjust routing and task allocation strategies based on changes in network status, node load, and energy consumption, ensuring that the sensor network maintains efficient and stable operation in an unstable environment.

[0118] In the case of data scarcity and rapid environmental changes, traditional large-scale data training models are difficult to provide timely and accurate decision support. The model's generalization ability is insufficient in small samples and dynamic environments, and it cannot effectively respond to new situations, which affects the system's intelligence level and real-time response capabilities. Therefore, an intelligent decision-making and real-time feedback mechanism that can operate efficiently in small samples and dynamic environments is needed. This solution is deeply refined from multiple angles such as few-sample learning, real-time feedback, model adaptive adjustment, and collaborative computing to ensure that the system can still maintain high decision-making accuracy and real-time performance in complex environments.

[0119] 1. Few-shot learning and online update mechanism

[0120] 1.1 Few-shot Learning Technology

[0121] In scenarios where data is scarce, few-sample learning technology is used to enable the model to quickly learn and adapt to new environments with only a small number of samples.

[0122] Meta-Learning: Learn how to learn by training a meta-model on a large number of tasks. When faced with a new task, the model uses previous experience to learn at a faster speed and with fewer samples. For example, Model-Agnostic Meta-Learning (MAML) is used to enable the model to adapt to new tasks after a small number of gradient updates.

[0123] Transfer Learning: Using a model pre-trained in the source domain to transfer its knowledge to the target domain. Fine-tuning the high-level features of the model to adapt it to the new data distribution.

[0124] 1.2 Online Reinforcement Learning and Adaptive Adjustment

[0125] In order to enable the model to continuously adapt to environmental changes, online reinforcement learning (Online Reinforcement Learning) is introduced to enable the model to continuously learn and optimize during the interaction process.

[0126] Environmental interaction and reward mechanism: As an intelligent agent, the model interacts with the environment in real time and receives rewards or penalties based on the actions taken. By maximizing the cumulative rewards, the model continuously adjusts its strategy to achieve the optimal decision.

[0127] Strategy update and exploration mechanism: Use ε-greedy strategy or soft strategy to balance exploration (trying new strategies) and exploitation (using known best strategies) in the decision-making process to avoid falling into local optimality.

[0128] 2. Real-time feedback and automatic model adjustment

[0129] 2.1 Establishment of real-time feedback system

[0130] Build a complete real-time feedback system to ensure that the model can obtain environmental change information in a timely manner and make self-adjustments.

[0131] Data monitoring module: Continuously monitor sensor data, capture real-time changes, and identify anomalies. For example, use sliding window statistics or time series anomaly detection algorithms to detect data anomalies in real time.

[0132] User feedback collection: In scenarios with user interaction, collect user feedback information as a basis for model adjustment.

[0133] 2.2 Application of Incremental Learning Algorithm

[0134] In order to enable the model to learn new information without forgetting historical knowledge, an incremental learning algorithm is adopted.

[0135] Continuous learning: When the model receives new data, it gradually updates parameters to avoid full training and improve efficiency. It uses methods such as Elastic Weight Consolidation (EWC) to prevent catastrophic forgetting.

[0136] Model version management: Maintain multiple versions of the model, dynamically select or merge models based on environmental changes and performance indicators to ensure the accuracy of decision-making.

[0137] 3. Decision-making sharing and coordination mechanism

[0138] 3.1 Edge Node Collaborative Computing

[0139] Taking advantage of edge computing, each node works together to improve the overall efficiency of the system.

[0140] Distributed decision-making: Distribute decision-making tasks to multiple edge nodes to reduce the load on the central server and reduce latency.

[0141] Task decomposition and coordination: Complex decision-making tasks can be decomposed and completed by multiple nodes. The task decomposition algorithm is used to ensure the reasonable allocation and efficient execution of tasks.

[0142] 3.2 Federated Learning Framework

[0143] In order to achieve collaborative training of models while protecting data privacy, a federated learning framework is adopted.

[0144] Local model training: Each node independently trains the model on local data to avoid data transmission and protect privacy.

[0145] Model parameter aggregation: The central server or negotiation node aggregates the model parameters uploaded by each node and updates the global model. The Federated Averaging algorithm is used to achieve efficient model fusion.

[0146] 4. Model Adaptation and Dynamic Adjustment

[0147] 4.1 Dynamic Learning Rate and Regularization Strategy

[0148] In order to adapt to environmental changes, the model needs to dynamically adjust the learning rate and regularization parameters.

[0149] Adaptive learning rate: Adopt adaptive optimization algorithms (such as Adam and RMSprop) to automatically adjust the learning rate according to the gradient information to improve training efficiency and stability.

[0150] Dynamic regularization: Dynamically adjust regularization parameters according to model complexity and overfitting risk to prevent model overfitting or underfitting.

[0151] 4.2 Adaptation of model structure

[0152] In the case of large changes in the environment, it may be necessary to adjust the structure of the model.

[0153] Model pruning and expansion: Prune or expand the model based on its performance and resource constraints. Use Neural Architecture Search (NAS) to automatically find the optimal model structure.

[0154] Multi-model integration: When a single model is difficult to adapt to all situations, a multi-model integration strategy is adopted. The robustness of decision-making is improved through weighted average, voting mechanism and other methods.

[0155] 5. Data augmentation and simulation

[0156] In order to make up for the deficiency of small sample data, data enhancement and simulation techniques are used.

[0157] Data enhancement: Transform existing data (such as rotation, scaling, adding noise) to generate more training samples and improve the generalization ability of the model.

[0158] Simulation environment: Build a virtual simulation environment to generate synthetic data for model training. Generate realistic data samples using technologies such as Generative Adversarial Networks (GAN).

[0159] 6. System integration and deployment

[0160] 6.1 Modular Design

[0161] Intelligent decision-making and real-time feedback mechanisms are designed as pluggable modules to facilitate system integration and maintenance.

[0162] Interface standardization: clearly defined module interfaces and communication protocols ensure compatibility between components.

[0163] Flexible configuration: supports flexible configuration of models and algorithms according to different application scenarios and requirements to improve the adaptability of the system.

[0164] 6.2 Resource Management and Optimization

[0165] In an edge computing environment, resources are limited and need to be properly managed and optimized.

[0166] Resource scheduling: Dynamically schedule computing resources based on the urgency of the task and resource consumption. Use task scheduling algorithms to balance the load of each node.

[0167] Energy consumption optimization: Reduce system energy consumption through algorithm optimization and hardware acceleration. Improve efficiency by using low-power model design and hardware accelerators (such as FPGA, ASIC).

[0168] 7. Security and Privacy Protection

[0169] 7.1 Data encryption and transmission security

[0170] Ensure the security of data during transmission to prevent eavesdropping and tampering.

[0171] Transport Layer Security (TLS): uses encryption protocols to ensure the security of data transmission.

[0172] Data encryption: Encrypt sensitive data for storage and transmission to prevent unauthorized access.

[0173] 7.2 Model Security

[0174] Prevent models from being attacked or leaked, and protect intellectual property and system security.

[0175] Defense against adversarial attacks: Improve the model's resistance to adversarial sample attacks and enhance the robustness of the system.

[0176] Model encryption and watermarking: Encrypt the model or embed a watermark to prevent illegal copying and tampering.

[0177] In intelligent sensing systems, the energy consumption of sensors and devices directly affects the system's ability to continue operating. Especially in an environment where resources are limited and it is difficult to replace batteries or replenish energy, how to optimize energy consumption and extend the life of equipment has become a key challenge. To this end, this solution proposes innovative technical solutions from three aspects: energy adaptive mechanism, load-aware scheduling algorithm, and wireless charging and energy recovery, aiming to improve the energy efficiency of equipment and support the long-term stable operation of the system.

[0178] 1. Energy Adaptive Mechanism

[0179] 1.1 Intelligent Battery Management and Prediction Algorithm

[0180] By introducing an intelligent battery management system (BMS) combined with machine learning and big data analysis, the energy consumption of the sensor can be monitored and predicted in real time.

[0181] Historical data and environmental conditions analysis: Use the sensor's historical energy consumption data and current environmental conditions (such as temperature, humidity, and workload) to build an energy consumption prediction model. Use deep learning algorithms such as long short-term memory networks (LSTM) to predict future energy consumption trends.

[0182] Dynamically adjust the working mode: Based on the energy consumption prediction results, the system can automatically adjust the working mode of the device. For example, when it is predicted that the energy is insufficient, the device can enter low-power mode or sleep mode to maintain only the necessary functions; when the energy is sufficient or high-performance computing is required, the device can switch back to normal mode.

[0183] 1.2 Adaptive Energy Consumption Optimization Strategy

[0184] Task priority scheduling: Dynamically adjust the execution order and frequency of tasks according to their importance and urgency. Important tasks are executed first, and secondary tasks can be delayed or executed less frequently.

[0185] Energy-aware algorithms: Introduce energy-aware algorithm scheduling and select low-energy algorithm implementations at the algorithm level. For example, select algorithms with lower computational complexity or use approximate calculation methods to reduce computational energy consumption.

[0186] 2. Load-aware scheduling algorithm

[0187] 2.1 Adaptive Scheduling Algorithm

[0188] Comprehensive parameter evaluation: A comprehensive evaluation model is established by combining the system's real-time load, the sensor's battery power, computing power, network bandwidth and other parameters. A multi-objective optimization algorithm is used to achieve a balance between energy consumption, performance and response time.

[0189] Dynamically adjust the operating frequency: Dynamically adjust the processor's operating frequency (Dynamic Voltage and Frequency Scaling, DVFS) according to the load and energy status of the device. When the load is low or the energy is tight, the operating frequency is reduced to save energy; when the load is high or the energy is sufficient, the operating frequency is increased to meet performance requirements.

[0190] 2.2 Task Allocation and Collaboration

[0191] Load balancing strategy: In a multi-device environment, through task migration and load balancing, tasks of high-load devices are assigned to low-load devices to avoid overloading or energy exhaustion of individual devices.

[0192] Edge computing collaboration: Use edge computing nodes to assist in processing computing-intensive tasks and reduce the energy consumption of front-end devices. Through intelligent scheduling, some tasks can be offloaded to edge nodes with sufficient energy and strong computing capabilities.

[0193] 3. Wireless charging and energy recovery

[0194] 3.1 Wireless Charging Technology

[0195] Long-distance wireless power supply: Use electromagnetic induction, wireless radio frequency (RF) energy transmission and other technologies to achieve long-distance wireless charging of devices. Deploy wireless charging base stations to provide continuous energy supply for sensor devices that are difficult to replace batteries.

[0196] Intelligent charging management: The device can intelligently establish a connection with the wireless charging base station by monitoring its own power status. When the power level is below the threshold, it will automatically request charging; after charging is completed, it will disconnect to avoid energy waste.

[0197] 3.2 Energy recovery system

[0198] Environmental energy harvesting: Utilize tiny energies such as light energy, heat energy, vibration energy, and radio frequency energy in the environment to recycle energy through energy harvesting devices (such as solar cells, thermoelectric converters, and vibration energy harvesters).

[0199] Self-energy replenishment mechanism: The collected energy is stored in the device's battery or supercapacitor for use by the device. Combined with the energy adaptive mechanism, the energy use strategy is optimized to maximize energy utilization.

[0200] To further improve energy efficiency, this solution proposes a dynamic energy management algorithm to optimize the energy consumption of the device. Assume that the energy state of the device is E(t), and the task set is {T i}, the energy consumption of each task is P i , the execution time is τ i . Introducing energy consumption prediction model:

[0201]

[0202] Where: Δt is the time step, t is the time, and H(t) is the energy obtained by recycling or charging.

[0203] On this basis, the energy utility function U(E) is defined to reflect the energy utilization efficiency of the equipment. By optimizing the objective function:

[0204]

[0205] Where: w i For task T i The weight reflects the importance of the task, S i For task T i successful execution rate.

[0206] Constraints:

[0207] Energy constraint: E(t+△t)≥Emin, ensuring that the device power is not lower than the minimum threshold Emin;

[0208] Time constraint: Σ i τ i ≤Δt, the task execution time does not exceed the time step.

[0209] The above optimization problem is solved by Lagrange multiplier method or dynamic programming algorithm to obtain the optimal task scheduling solution, maximize energy utilization efficiency and extend equipment operation time.

[0210] 4. Comprehensive application of the solution

[0211] Intelligent scheduling and energy management: Through the combination of energy adaptive mechanism and load-aware scheduling algorithm, the equipment can intelligently adjust its own working state to meet performance requirements and save energy.

[0212] Energy acquisition and replenishment: Using wireless charging and energy recovery technology, devices can continuously obtain energy support and extend operating time in energy-constrained environments.

[0213] Synergy with other technical solutions: This solution cooperates with the previous solutions such as data quality improvement, self-organizing network optimization, intelligent decision-making and real-time feedback to jointly improve the overall performance and stability of the system.

[0214] Embodiment 2:

[0215] An intelligent perception-driven real-time data monitoring and security assessment system can implement an intelligent perception-driven real-time data monitoring and security assessment method described in Example 1, including:

[0216] Data acquisition module: obtain real-time data from sensors;

[0217] Model building module: Use the real-time data of sensors to build a virtual sensor model in a multi-level neural network;

[0218] Data prediction module: performs data prediction based on the correlation of multi-source data through the virtual sensor model to obtain the virtual sensor model output;

[0219] Error correction module: performs error correction on the output of the virtual sensor model according to the error correction mechanism;

[0220] Data fusion module: performs data fusion based on the virtual sensor model output after error correction.

[0221] Embodiment three:

[0222] The embodiment of the present invention further provides an intelligent perception driven real-time data monitoring and security assessment device, which can implement an intelligent perception driven real-time data monitoring and security assessment method described in the first embodiment, including a processor and a storage medium;

[0223] The storage medium is used to store instructions;

[0224] The processor is used to operate according to the instructions to perform the steps of the following method:

[0225] Get real-time data from sensors;

[0226] Use the real-time data from sensors to build a virtual sensor model in a multi-level neural network;

[0227] The virtual sensor model is used to predict data according to the correlation of multi-source data to obtain the virtual sensor model output;

[0228] Performing error correction on the virtual sensor model output according to the error correction mechanism;

[0229] Data fusion is performed based on the error-corrected virtual sensor model output.

[0230] Embodiment 4:

[0231] The embodiment of the present invention further provides a computer-readable storage medium, which can implement the intelligent perception-driven real-time data monitoring and security assessment method described in the first embodiment, and stores a computer program thereon, which implements the steps of the following method when executed by a processor:

[0232] Get real-time data from sensors;

[0233] Use the real-time data from sensors to build a virtual sensor model in a multi-level neural network;

[0234] The virtual sensor model is used to predict data according to the correlation of multi-source data to obtain the virtual sensor model output;

[0235] Performing error correction on the virtual sensor model output according to the error correction mechanism;

[0236] Data fusion is performed based on the error-corrected virtual sensor model output.

[0237] It is known from common technical knowledge that the present invention can be implemented by other embodiments that do not deviate from its spirit or essential features. Therefore, the above disclosed embodiments are only illustrative in all respects and are not exclusive. All changes within the scope of the present invention or within the scope equivalent to the present invention are included in the present invention.

[0238] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0239] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0240] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0241] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0242] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. An intelligent perception driven real-time data monitoring and security assessment method, characterized in that: The method monitors and evaluates sensor data after fusion, and the data fusion includes: Get real-time data from sensors; Use the real-time data from sensors to build a virtual sensor model in a multi-level neural network; The virtual sensor model is used to predict data according to the correlation of multi-source data to obtain the virtual sensor model output; Performing error correction on the virtual sensor model output according to the error correction mechanism; Data fusion is performed based on the error-corrected virtual sensor model output.

2. The intelligent perception driven real-time data monitoring and security assessment method according to claim 1 is characterized in that: The virtual sensor model output is: in: is the output of the virtual sensor model, f is the nonlinear mapping function learned by the deep neural network, and w i (t) is the weight coefficient of the th sensor, x i (t) is the output of sensor node i; The error correction formula of the virtual sensor model is: in: is the corrected virtual sensor model output, is the virtual sensor model output, γ is the error correction coefficient, and z(t) is the actual observation value or ground truth.

3. The intelligent perception driven real-time data monitoring and security assessment method according to claim 1 is characterized in that: The output data of the sensor is evaluated for sensor accuracy using a deep reinforcement learning calibration model. The deep reinforcement learning calibration model is calibrated through multiple experimental learning and optimization, and is preprocessed using a model-based adaptive filtering algorithm. At the same time, the latest sensor status is fed back into the filter training process through model adaptive updating.

4. The intelligent perception driven real-time data monitoring and security assessment method according to claim 1 is characterized in that: The output data of the sensor adopts an adaptive data compression algorithm based on deep learning, and dynamically adjusts the data compression ratio according to the sensor type, data content and current network bandwidth status. The high-dimensional data in the output data is compressed after feature extraction through a convolutional neural network, and representative features related to the sensor monitoring content are extracted for encoding. The low-dimensional data in the output data is compressed using differential coding and Huffman coding methods; The output data of the sensor uses a bandwidth adaptive adaptive compression function to dynamically adjust the compression strategy according to bandwidth limitations and data characteristics. The bandwidth adaptive adaptive compression function formula includes: C=CNN(D,w,λ) Where: C is the compressed data, D is the original data, w is the learnable weight in the convolutional neural network, and λ is the dynamic bandwidth factor; The bandwidth factor adjustment function is: Among them: B avail is the current available bandwidth, B req is the current transmission required bandwidth, and λ(t) is the bandwidth factor at time t.

5. The intelligent perception driven real-time data monitoring and security assessment method according to claim 1 is characterized in that: The sensor network adopts a deep reinforcement learning routing optimization algorithm, and each node in the sensor network can make routing decisions based on the current network status, and predict energy consumption and load conditions based on a long short-term memory network model, and dynamically adjust task allocation according to the energy status and load conditions of the node through a hierarchical load balancing algorithm, and apply for and allocate tasks based on its current status and information in the blockchain, and make decisions based on its own status when participating in task scheduling and reach consensus through a consensus algorithm, and make decisions based on a utility function that comprehensively considers multiple factors of the node, and dynamically adjust the proportion of resource sharing according to the load and energy status.

6. The intelligent perception driven real-time data monitoring and security assessment method according to claim 1 is characterized in that: The virtual sensor model uses few-sample learning technology and online update mechanism to perform model training, and uses real-time feedback and automatic model adjustment to perform model correction, and uses decision sharing and collaborative mechanism to perform task allocation execution and collaborative training, and uses model adaptation and dynamic adjustment to perform model adjustment, and uses data enhancement to generalize the model, and uses modular integration and deployment configuration, and uses data encryption and model encryption technology.

7. The intelligent perception driven real-time data monitoring and security assessment method according to claim 1 is characterized in that: The sensor performs real-time monitoring and prediction through an intelligent battery management system combined with machine learning and big data analysis. The intelligent battery management system uses the historical energy consumption data of the sensor and the current environmental conditions to establish an energy consumption prediction model, and the formula is: Where: E(t) is the energy state of the device, t is time, Δt is the time step, P i The energy consumption of each task, τ i is the execution time, H(t) is the energy obtained by recycling or charging, and the optimization objective function and constraints of the energy consumption prediction model are: Where: T i is the task set, U(E) is the energy state of the equipment, and w i For task T i The weight, S i For task T i Successful execution rate; Energy constraint: E(t+△t)≥Emin, where Emin is the minimum threshold; Time constraint: Σ i τ i ≤Δt.

8. An intelligent perception-driven real-time data monitoring and security assessment system, characterized in that: include: Data acquisition module: obtain real-time data from sensors; Model building module: Use the real-time data of sensors to build a virtual sensor model in a multi-level neural network; Data prediction module: performs data prediction based on the correlation of multi-source data through the virtual sensor model to obtain the virtual sensor model output; Error correction module: performs error correction on the output of the virtual sensor model according to the error correction mechanism; Data fusion module: performs data fusion based on the virtual sensor model output after error correction.

9. An intelligent perception-driven real-time data monitoring and security assessment device, characterized in that: including processor and storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images