Remote monitoring method and remote monitoring system thereof

Abstract

The invention relates to the field of monitoring methods and monitoring systems, in particular to a remote monitoring method and a remote monitoring system thereof, and the method comprises the following steps: obtaining original data collected by a multi-modal sensor array; preprocessing the original data to obtain preprocessed data; based on the preprocessed data, adaptive multi-mode compression and intelligent transmission are executed, and compressed data are obtained; the compressed data is decompressed, and the decompressed data is obtained; fusing and analyzing the decompressed data to obtain an analysis result; executing anomaly detection and early warning based on the analysis result to obtain an anomaly detection result; on the basis of the anomaly detection result, feedback and optimization are carried out, system parameters are updated, and the compression rate of 92% is achieved through a self-adaptive multi-mode compression algorithm; and the data transmission delay is reduced to 45ms from 280ms in the traditional method, so that the real-time performance of the system is improved.

Description

Technical Field

[0001] The present invention relates to the field of monitoring methods and monitoring systems, and more specifically, to a remote monitoring method and a remote monitoring system thereof. Background Art

[0002] With the rapid development of Industry 4.0 and intelligent manufacturing, remote monitoring technology plays an increasingly important role in modern industrial production. Especially in the monitoring of key facilities such as large-scale industrial equipment such as smelting blast furnaces and chemical reactors, the performance of the remote monitoring system is directly related to production efficiency, safety and economic benefits.

[0003] In recent years, remote monitoring technology has evolved from simple data collection to intelligent analysis. Early remote monitoring systems mainly relied on fixed-frequency data collection and simple threshold alarms. With the advancement of technology, more complex data analysis algorithms such as machine learning and deep learning have been introduced to improve the accuracy of anomaly detection. At the same time, in order to cope with the growing amount of data, various data compression and transmission optimization technologies have also been widely used.

[0004] However, despite the progress made in existing technologies, there are still many challenges and shortcomings in the face of the complexity and dynamism of modern industrial production:

[0005] 1. Insufficient adaptability of data collection: Most existing systems use a fixed sampling frequency and cannot be dynamically adjusted according to device status and environmental changes. This may result in insufficient data being collected at critical moments, or a large amount of redundant data being generated under normal conditions, resulting in a waste of resources.

[0006] 2. Low data compression efficiency: Traditional data compression methods often use general algorithms, such as ZIP or GZIP, which cannot fully utilize the characteristics of industrial data. This results in unsatisfactory compression rates or loss of important information when pursuing high compression rates.

[0007] 3. Insufficient network transmission optimization: Existing systems usually use fixed network transmission strategies, which cannot cope with complex and changing industrial network environments. This may lead to high data transmission delays and high packet loss rates, affecting the effect of real-time monitoring.

[0008] 4. Limited anomaly detection accuracy: Although machine learning algorithms have been introduced, most systems still have difficulty in effectively processing multimodal, high-dimensional industrial data. This results in insufficient anomaly detection accuracy and a high risk of false positives and false negatives.

[0009] 5. Slow system response: Due to the inefficiency of each link of data processing, transmission and analysis, the response time of many existing systems is long and cannot meet the needs of time-sensitive industrial scenarios.

[0010] 6. Low energy efficiency: Existing systems often ignore the optimization of energy efficiency, resulting in increased energy consumption while improving monitoring effects.

[0011] Take the industrial remote monitoring system of a well-known company as an example. Although the system has been widely used in large steel plants, it still has significant problems. It uses a fixed sampling interval of 5 seconds, which generates a large amount of redundant data when the blast furnace is operating normally, and may miss key information in emergencies. The system uses the traditional JPEG2000 algorithm for image compression. Although the compression rate reaches 70%, important details are often lost. In terms of anomaly detection, the system uses a rule-based approach combined with a simple statistical model, and has limited processing capabilities for complex multi-source data, resulting in anomaly detection accuracy of only about 85%. In addition, the system's average response time exceeds 5 seconds, which seems to be incapable of handling emergencies.

[0012] In view of the above problems, there is an urgent need for a new remote monitoring method and its remote monitoring system, which can comprehensively improve the efficiency and accuracy of data collection, compression, transmission, analysis and response, while taking into account the optimization of energy efficiency. Summary of the invention

[0013] The purpose of the present invention is to provide a remote monitoring method and a remote monitoring system thereof, which are used to solve the problems of low data transmission efficiency, poor real-time performance, inaccurate abnormality detection, etc. existing in the remote monitoring of existing industrial equipment.

[0014] To solve the above technical problems, the present invention adopts the following technical solution: a remote monitoring method, comprising the following steps:

[0015] Obtaining raw data collected by a multimodal sensor array;

[0016] Preprocessing the original data to obtain preprocessed data;

[0017] Based on the preprocessed data, performing adaptive multimodal compression and intelligent transmission to obtain compressed data;

[0018] Decompressing the compressed data to obtain decompressed data;

[0019] fusing and analyzing the decompressed data to obtain analysis results;

[0020] Based on the analysis results, perform anomaly detection and early warning to obtain anomaly detection results;

[0021] Based on the abnormal detection results, feedback and optimization are performed to update system parameters.

[0022] Specifically, in the step of acquiring the raw data collected by the multimodal sensor array, an adaptive sampling frequency strategy is adopted:

[0023] f s (t) = f base +Δf·sigmoid(∑w i Δp i (t))

[0024] Among them, f s (t) is the real-time sampling frequency, f base is the basic sampling frequency, Δf is the adjustable frequency increment, w i is the weight of each parameter, Δp i (t) is the rate of change of parameter i.

[0025] Specifically, the step of preprocessing the original data includes:

[0026] Use the improved wavelet threshold denoising algorithm to remove noise:

[0027] X denoised =IDWT(T λ (DWT(X)))

[0028] Among them, X is the original data, DWT and IDWT are discrete wavelet transform and its inverse transform respectively, T λ is the adaptive threshold function:

[0029] T λ (x) = sign(x)(|x|-λ) + (1+exp(-α(|x|-β)))

[0030] λ, α, β are dynamically adjusted parameters.

[0031] The improved Z-score standardization method is used for data standardization:

[0032]

[0033] Among them, μ is the mean, σ is the standard deviation, ∈ and γ are dynamically adjusted parameters,

[0034] Specifically, the step of performing adaptive multimodal compression and intelligent transmission includes:

[0035] Use quantum-inspired neural networks to perform real-time data analysis and obtain data importance scores;

[0036] Based on the data importance score, performing adaptive multimodal compression to obtain compressed data;

[0037] Performing compression quality assessment on the compressed data to obtain a compression quality assessment result;

[0038] Based on the compression quality evaluation result, perform intelligent transmission scheduling and determine the data transmission strategy;

[0039] Based on the data transmission strategy, network self-optimization is performed to adjust network parameters.

[0040] Specifically, the calculation formula of the quantum inspired neural network is: QHNN(X)=σ(W2·Q Φ (W1·X+b1)+b2)

[0041] Among them, X is the input data, σ is the activation function, W1 and W2 are weight matrices, b1 and b2 are bias vectors, and Q Φ For quantum-inspired nonlinear transformations:

[0042]

[0043] Importance scoring function:

[0044] I(X)=∑w i QHNN(X i )+λH(X)

[0045] Among them, w i is the weight of each feature, H(X) is the entropy of the data, and λ is the balance parameter.

[0046] Specifically, the adaptive multimodal compression adopts a hybrid algorithm combining fractal compression and neural compression:

[0047] C(X)=αF c (X)+(1-α)N c (X)

[0048] Among them, X is the input data, C(X) is the compression result, and F c is the fractal compression function, N c is the neural compression function:

[0049] F c (X) = argmin∑D(R i (X), S j (X)

[0050] Among them, R i and S j are the range block and definition block respectively, and D is the distance function;

[0051] N c (X)=Decoder(Quantizer(Encoder(X)))

[0052] Among them, Encoder and Decoder use improved variational autoencoder structure, and Quantizer adopts soft quantization method.

[0053] Specifically, the step of decompressing the compressed data adopts a decompression algorithm corresponding to the compression:

[0054] For the fractal compression part:

[0055]

[0056] Among them, X′ is the decompression result, W is the iterated function system (IFS) defined in fractal coding, and X0 is the initial estimate;

[0057] For the nerve compression part:

[0058] X′=Decoder(Q- 1 (Z)

[0059] Among them, Q- 1 is the inverse quantization function, Decoder is the trained decoder network, and Z is the compressed data.

[0060] Specifically, the step of fusing and analyzing the decompressed data adopts the improved Dempster-Shafer evidence theory:

[0061]

[0062] Among them, m(A) is the basic probability distribution after fusion, m i (A) is the basic probability distribution of the i-th evidence source, K is the conflict degree, and w i The weight is adjusted dynamically.

[0063] Specifically, the execution anomaly detection and early warning step adopts an improved IsolationForest algorithm:

[0064] Anomaly Score:

[0065]

[0066] Where x is the data point to be detected, n is the number of samples, E(h(x)) is the average path length, and c(n) is the normalization factor;

[0067] Introducing a dynamic threshold mechanism:

[0068] τ(t)=μ(s)+k(t)σ(s)

[0069] k(t+1)=k(t)+δ(FPR(t)-FPR target)

[0070] Among them, τ(t) is the dynamic threshold, μ(s) and σ(s) are the mean and standard deviation of the anomaly score, k(t) is the dynamic adjustment factor, FPR is the false alarm rate, and δ is the learning rate.

[0071] A remote monitoring system for implementing the above method comprises:

[0072] Multimodal sensor arrays for acquiring raw data;

[0073] A preprocessing module, used for preprocessing the raw data to obtain preprocessed data;

[0074] An adaptive multi-modal compression and intelligent transmission module, used to perform compression and transmission based on the pre-processed data to obtain compressed data;

[0075] A decompression module, used for decompressing the compressed data to obtain decompressed data;

[0076] A data fusion and analysis module, used to fuse and analyze the decompressed data to obtain analysis results;

[0077] An anomaly detection and early warning module, used to perform anomaly detection and early warning based on the analysis result to obtain an anomaly detection result;

[0078] The feedback and optimization module is used to provide feedback and optimization based on the abnormality detection results and update system parameters.

[0079] Compared with the prior art, the present invention has the following beneficial effects:

[0080] 1. Efficient data compression: Through the adaptive multimodal compression algorithm, the present invention achieves a compression rate of 92%, which is much higher than the 70% of the traditional method, greatly reducing storage and transmission costs.

[0081] 2. Significantly reduced transmission delay: Thanks to intelligent transmission scheduling and network self-optimization, data transmission delay is reduced from 280ms in traditional methods to 45ms, improving the real-time performance of the system.

[0082] 3. Excellent anomaly detection accuracy: Using quantum-inspired neural networks and dynamic threshold mechanisms, the anomaly detection accuracy is increased to 97.5%, greatly reducing the risk of false positives and missed positives.

[0083] 4. Significant energy efficiency improvement: Through all-round optimization, this method achieves an 18.5% energy efficiency improvement, far exceeding the 5% of traditional methods, with significant economic and environmental benefits.

[0084] 5. Fast system response: The system response time is shortened to 1.2 seconds, nearly five times faster than the 5.8 seconds of the traditional method, greatly improving the ability to handle emergencies.

[0085] 6. Optimized network resource utilization: While improving data transmission efficiency, bandwidth utilization is controlled at 65%, reserving sufficient network resources for other applications.

[0086] In summary, the remote monitoring method provided by the present invention comprehensively solves the problems existing in the prior art and significantly improves the performance, reliability and efficiency of the industrial monitoring system. It not only provides more accurate and timely monitoring for industrial production, but also brings considerable economic benefits to enterprises by optimizing resource utilization. The application of this method will provide strong technical support for the development of Industry 4.0 and intelligent manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 A flow chart of a remote monitoring method of the present invention;

[0088] Figure 2 A flowchart for performing adaptive multi-modal compression and intelligent transmission of the present invention;

[0089] Figure 3 This is a framework flow chart of a remote monitoring system of the present invention. DETAILED DESCRIPTION

[0090] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation method, structure, features and effects of the intelligent management method based on the enterprise environment proposed by the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0091] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0092] A specific solution of a remote monitoring method provided by the present invention is described in detail below with reference to the accompanying drawings.

[0093] Please refer to Figure 1 The present invention provides a remote monitoring method, comprising the following steps:

[0094] Step 1: Acquire the raw data collected by the multimodal sensor array. In the specific implementation of the present invention, taking large-scale industrial smelting equipment as an example, the multimodal sensor array includes temperature sensors, pressure sensors, vibration sensors, gas concentration sensors, and high-definition cameras, etc. These sensors are distributed in key parts of the smelting equipment, such as furnaces, cooling systems, and tail gas treatment devices.

[0095] This step adopts an adaptive sampling frequency strategy:

[0096] f s (t) = f base +Δf·sigmoid(∑w i Δp i (t))

[0097] Among them, f s (t) is the real-time sampling frequency, f base is the base sampling frequency (for example, for a temperature sensor, f base may be 1 Hz), Δf is the adjustable frequency increment (e.g., 0.5 Hz), and w i is the weight of each parameter (for example, temperature change may be more important than pressure change), Δp i (t) is the rate of change of parameter i.

[0098] For example, when a rapid rise in furnace temperature is detected, the system automatically increases the sampling frequency of the temperature sensor to capture more detailed data.

[0099] Step 2: Preprocess the raw data. This step includes two main parts: noise removal and data normalization.

[0100] a) Noise removal, using improved wavelet threshold denoising algorithm:

[0101] X denoised =IDWT(T λ (DWT(X)))

[0102] Where X is the original data (e.g., temperature readings over a period of time), DWT and IDWT are discrete wavelet transform and its inverse transform, respectively, and T λ is the adaptive threshold function:

[0103] T λ (x) = sign(x)(|x|-λ) + (1+exp(-α(|x|-β)))

[0104] λ, α, and β are dynamically adjusted parameters. In practical applications, λ may be adjusted between 0.1 and 0.5, and α and β are optimized according to the characteristics of the signal.

[0105] b) Data standardization, using the improved Z-score standardization method:

[0106]

[0107] Among them, μ is the mean, σ is the standard deviation, ∈ and γ are dynamically adjusted parameters. Usually, ∈ may be between 0.001 and 0.1, and γ is adjusted between 0.1 and 1.

[0108] This improved normalization method is particularly suitable for dealing with extreme values ​​that may occur in smelting equipment, such as sudden temperature peaks.

[0109] Please refer to Figure 2 , Step 3: Perform adaptive multi-modal compression and intelligent transmission. This step is the core of the present invention and includes the following parts:

[0110] a) Use quantum-inspired neural networks for real-time data analysis, where X is the input data (e.g., preprocessed multimodal sensor data), σ is the activation function (e.g., ReLU), W1 and W2 are weight matrices, b1 and b2 are bias vectors, and Q Φ For quantum-inspired nonlinear transformations:

[0111] QHNN(X)=σ(W2·Q Φ (W1·X+b1)+b2)

[0112]

[0113] Importance scoring function:

[0114] I(X)=∑w i QHNN(X i )+λH(X)

[0115] Among them, w i is the weight of each feature (for example, temperature data may be more important than vibration data), H(X) is the entropy of the data, and A is the balance parameter (usually between 0.1 and 0.5). This method can quickly identify key data in the smelting process, such as abnormal temperature fluctuations or sudden changes in gas concentration.

[0116] b) Perform adaptive multimodal compression

[0117] C(X)=αF c (X)+(1-α)N c (X)

[0118] Among them, X is the input data, C(X) is the compression result, and F c is the fractal compression function, N cis the neural compression function, and α is the dynamically adjusted mixing coefficient (usually between 0.3 and 0.7).

[0119] F c (X) = argmin∑D(R i (X), S j (X)

[0120] Among them, R i and S j are range blocks and definition blocks, respectively, and D is the distance function. This method is particularly suitable for compressing data with self-similarity, such as the temperature curve of smelting equipment.

[0121] N c (X)=Decoder(Quantizer(Encoder(X)))

[0122] The Encoder and Decoder use an improved variational autoencoder structure, and the Quantizer uses a soft quantization method, which is suitable for compressing complex multimodal data such as device vibration features and video streams.

[0123] c) Compression quality evaluation

[0124] Q(C(X))=w1R(X,X′)+w2S(X,X′)+w3T(X′)

[0125] Among them, R is the information retention rate, S is the structural similarity, T is the computational complexity, and W1, W2, and W3 are dynamically adjusted weights (the sum is 1). This evaluation method ensures that the compressed data still retains key information in the governance process.

[0126] d) Intelligent transmission scheduling, using the improved AntCo1onyOptimization (ACO) algorithm to optimize the transmission path:

[0127]

[0128] Among them, τ ij is the pheromone concentration, ρ is the evaporation rate (usually between 0.1 and 0.3), and are the pheromone increments of the current best path and the global best path respectively. This method can find the optimal data transmission path in a complex industrial network environment and ensure the timely transmission of key data.

[0129] e) Network self-optimization, using a hybrid optimization method that combines quantum genetic algorithm with deep reinforcement learning:

[0130] Quantum chromosome representation: |ψ>=∑α i|i>

[0131] Quantum revolving door:

[0132] Policy network: π(a|s)=softmax(f θ (s))

[0133] Value Network: V(s) = g φ (s)

[0134] Hybrid optimization objective function: J = w1J QGA +w2J PPO +w3R

[0135] Among them, J QGA is the fitness function of the quantum genetic algorithm, J PPO is the objective function of deep reinforcement learning, R is the regularization term. w1, W2, and W3 are weights (the sum is 1). This method can dynamically adjust network parameters to adapt to the complex and changeable network environment of industrial sites.

[0136] Step 4: Decompress the compressed data. This step uses the decompression algorithm corresponding to the compression:

[0137] For the fractal compression part:

[0138]

[0139] Where X' is the decompression result, W is the iterated function system (IFS) defined in fractal coding, and X0 is the initial estimate. Usually, 10-20 iterations can get good results.

[0140] For the nerve compression part:

[0141] X′=Decoder(Q - 1(Z))

[0142] Among them, Q -1 is the inverse quantization function, Decoder is the trained decoder network, and Z is the compressed data.

[0143] This decompression method can effectively restore various monitoring data of smelting equipment, including temperature curves, pressure changes, gas concentrations, etc.

[0144] Step 5: Fusion and analysis of the decompressed data. This step uses the improved Dempster-Shafer evidence theory:

[0145]

[0146] Among them, m(A) is the basic probability distribution after fusion, mi (A) is the basic probability distribution of the i-th evidence source, K is the conflict degree, and w i The weight is adjusted dynamically.

[0147] This method is particularly suitable for integrating multi-source heterogeneous data of smelting equipment, such as comprehensive analysis of temperature, pressure, gas concentration and other data to obtain a more accurate assessment of the equipment's operating status.

[0148] Step 7: Provide feedback and optimization, and use meta-learning methods for continuous optimization:

[0149]

[0150] Among them, θ is the model parameter, α is the learning rate, and L τ is the meta-learning objective function:

[0151]

[0152] Among them, τ is the task set, D i is the data of the i-th task, and R(θ) is the regularization term.

[0153] Based on the anomaly detection results, the system will automatically adjust the parameters of each module, such as adjusting the sensor sampling frequency, updating the weight of the compression algorithm, optimizing the network transmission path, etc., forming a closed-loop optimization system.

[0154] Please refer to Figure 3 The present invention also provides a remote monitoring system for implementing the above method, comprising:

[0155] Multimodal sensor array 1, installed at key locations of smelting equipment to obtain raw data;

[0156] Preprocessing module 2, removing noise and standardizing the original data;

[0157] Adaptive multimodal compression and intelligent transmission module 3, which performs data analysis, compression, quality assessment and transmission scheduling;

[0158] Decompression module 4, restoring the compressed data;

[0159] Data fusion and analysis module 5, comprehensive analysis of multi-source data;

[0160] Anomaly detection and early warning module 6, identifies abnormal conditions and issues an alarm;

[0161] The feedback and optimization module 7 dynamically adjusts the parameters of each module according to the system operation status.

[0162] The hardware architecture of the remote monitoring system of the present invention mainly includes the following parts:

[0163] 1. Multimodal sensor array:

[0164] Temperature sensor: PT100 platinum resistance temperature sensor with a measuring range of -200°C to 650°C and an accuracy of ±0.1°C.

[0165] Pressure sensor: Use MEMS pressure sensor with a measuring range of 0-100MPa and an accuracy of 0.1%FS.

[0166] Vibration sensor: Three-axis MEMS accelerometer with a measurement range of ±16g and a bandwidth of DC-3200Hz.

[0167] Gas concentration sensor: Using electrochemical gas sensor, it can detect harmful gases such as CO, H2S, SO2, etc. with an accuracy of ±2%.

[0168] HD Camera: 4K resolution, 60fps, with low-light enhancement.

[0169] These sensors are connected to field data acquisition units via industrial Ethernet or industrial wireless networks such as WirelessHART.

[0170] 2. Field data acquisition unit: Use an industrial-grade embedded computer, such as Advantech ARK-3520L, equipped with an Intel Core i7 processor, 32 GB RAM, and 1 TB SSD. This unit runs a real-time operating system (such as VxWorks) and is responsible for data acquisition, preprocessing, and preliminary compression.

[0171] 3. Edge computing server: Use high-performance workstations such as Dell Precision 7920 Tower, equipped with dual-core Intel Xeon Gold processors, 128GB RAM, NVIDIA Quadro RTX 6000 GPU and 10TB NVMe SSD array. Responsible for performing complex data processing, compression and analysis tasks.

[0172] 4. Industrial Gateway: Use industrial IoT gateways such as Moxa UC-8580, which supports 5G, Wi-Fi 6, and wired Ethernet, and is responsible for secure data transmission.

[0173] 5. Cloud server cluster: Use Alibaba Cloud ECS server cluster, configure elastic scaling, for data storage, in-depth analysis and visualization.

[0174] The software architecture of this system adopts microservice design, which mainly includes the following modules:

[0175] 1. Data acquisition and preprocessing module, developed in C++ language and Qt framework, realizes the acquisition, filtering and preliminary standardization of sensor data. The key algorithms include adaptive sampling frequency adjustment (step 1), wavelet denoising and improved Z-score standardization (step 2).

[0176] 2. Adaptive multimodal compression module, developed in Python and PyTorc framework, mainly realizes intelligent data compression. Key algorithms include quantum inspired neural network (QHNN), fractal compression, and neural compression (steps 3a-b).

[0177] 3. Intelligent transmission scheduling module, which adopts the Go development language and a self-developed distributed scheduling framework. Its main function is to optimize data transmission paths and strategies. Key algorithms include the improved Ant Colony Optimization (ACO) algorithm (step 3d).

[0178] 4. The network self-optimization module uses the Python development language and the Ray framework. Its main function is to dynamically adjust network parameters. The key algorithms include quantum genetic algorithm and deep reinforcement learning hybrid optimization (step 3e).

[0179] 5. Data decompression and fusion analysis module, using Python development language, NumPy, SciPy framework, the main function is data decompression, multi-source data fusion, key algorithms include iterated function system (IFS) decompression, improved Dempster-Shafer evidence theory (steps 4-5)

[0180] 6. The anomaly detection and early warning module adopts the Python development language and the Scikit-learn (custom extension) framework. Its main function is to identify system anomalies and issue alarms. The key algorithms include the improved IsolationForest algorithm and the dynamic threshold mechanism (step 6).

[0181] 7. Feedback and optimization module, using Python development language, self-developed closed-loop optimization framework, the main function is to dynamically adjust the parameters of each module based on the system operation status, the key algorithm includes multi-objective optimization algorithm.

[0182] 8. The visualization and user interaction module adopts JavaScript development language, Vue.js, D3.js framework, and its main functions are data visualization, real-time monitoring, and alarm display.

[0183] The remote monitoring system of the present invention works as follows:

[0184] 1. Data collection: Multimodal sensors continuously collect various parameters of the smelting equipment. The field data collection unit adjusts the sampling frequency according to the adaptive sampling frequency strategy (step 1), especially increasing the sampling frequency of relevant sensors when an anomaly is detected.

[0185] 2. Preprocessing: The field data acquisition unit performs preliminary filtering and standardization on the raw data (step 2). This step effectively reduces the noise of the data and unifies different types of data to the same scale.

[0186] 3. Edge computing: The preprocessed data is transferred to the edge computing server. Here, the system performs the following operations: Use QHNN to perform real-time data analysis and evaluate the data importance (step 3a); Based on the data importance, perform adaptive multimodal compression (step 3b); Evaluate the compression quality and adjust the compression parameters if necessary (step 3c)

[0187] 4. Intelligent transmission: The compressed data is transmitted through the industrial gateway. The intelligent transmission scheduling module uses the improved ACO algorithm to optimize the transmission path (step 3d). The network self-optimization module adjusts the network parameters in real time to adapt to the current network conditions (step 3e).

[0188] 5. Cloud processing: After the data reaches the cloud server, the system performs the following operations: data decompression (step 4), multi-source data fusion and analysis (step 5), anomaly detection and early warning (step 6).

[0189] 6. Feedback optimization: The system dynamically adjusts the parameters of each module according to the operating status and detection results (step 7). For example: adjusting the sensor sampling frequency; updating the weight of the compression algorithm; optimizing the network transmission strategy.

[0190] 7. Visualization and interaction: The processed data is presented to the user through the visualization module. The user can monitor the device status in real time, receive alarms, and perform necessary remote operations through the web interface or mobile app.

[0191] The key technical realization of the present invention is:

[0192] 1. Quantum-inspired neural network (QHNN): QHNN is implemented on the GPU of the edge computing server. It uses concepts from quantum computing, such as superposition and entanglement, to enhance the performance of traditional neural networks. Specifically, we use PyTorch's custom quantum layer to implement the QΦ transform, which greatly improves the network's ability to process complex multimodal data.

[0193] 2. Adaptive multimodal compression: The fractal compression algorithm is optimized for data with self-similarities (such as temperature curves), while neural compression is used to process more complex data (such as video streams). The system dynamically adjusts the weights α of the two methods to achieve the best compression effect.

[0194] 3. Intelligent transmission scheduling: The improved ACO algorithm is implemented in the Go language, using Goroutine to achieve high-concurrency path optimization. The algorithm takes into account multiple factors such as network latency, bandwidth, and data importance, and realizes intelligent routing in complex network environments.

[0195] 4. Anomaly detection: The improved IsolationForest algorithm is implemented through a custom extension on the Scikit-learn framework. The dynamic threshold mechanism can adaptively adjust the anomaly judgment criteria, greatly reducing false positives and false negatives.

[0196] In a blast furnace monitoring system of a large steel plant, the method and system of the present invention have been successfully applied:

[0197] 1. Data compression efficiency: The system compresses the original data volume from about 500MB per second to 50MB, with a compression rate of 90%, while maintaining the integrity of key information.

[0198] 2. Network optimization: Through intelligent transmission scheduling, data transmission delay is reduced from an average of 300ms to 50ms, greatly improving the real-time performance of the system.

[0199] 3. Abnormal detection: The system successfully warned of abnormal temperature fluctuations in blast furnaces multiple times, discovered problems in advance, and avoided possible equipment damage and production interruptions.

[0200] 4. Energy efficiency: Through precise monitoring and optimization of the production process, the factory's energy consumption has been reduced by about 15%.

[0201] 5. Maintenance cost: Improved predictive maintenance capabilities have reduced equipment maintenance costs by approximately 25%.

[0202] In summary, the remote monitoring method and system of the present invention realizes efficient, reliable and intelligent monitoring of industrial equipment through innovative algorithm design and optimized hardware configuration. It not only improves production efficiency and safety, but also provides important technical support for Industry 4.0 and intelligent manufacturing.

[0203] The remote monitoring method and system of the present invention can significantly improve data transmission efficiency, reduce network bandwidth usage, and enhance the accuracy and real-time performance of abnormality detection in the application of large-scale industrial smelting equipment. For example, in the actual application of a large steel smelting plant, the system successfully reduced the data transmission volume by 60%, and shortened the detection time of abnormal conditions from minutes to seconds, greatly improving production efficiency and safety.

[0204] In summary, the remote monitoring method and system provided by the present invention realize efficient, reliable and intelligent monitoring of industrial equipment through innovative algorithms and optimization strategies, and provide important technical support for Industry 4.0 and intelligent manufacturing.

[0205] In order to further verify the effectiveness of the present invention, the applicant provides the embodiments of the present invention and comparative examples of different conditions, and conducts detailed comparative analysis.

[0206] Embodiment 1: Remote monitoring method of the present invention

[0207] Comparative Example 1: Traditional fixed sampling rate and static compression method

[0208] Comparative Example 2: A method using a conventional neural network instead of a quantum-inspired neural network. Comparative Example 3: A method that does not include a network self-optimization module. Test environment: A blast furnace monitoring system in a large steel plant. The test period is 30 days.

[0209] The testing indicators, standards and methods are as follows:

[0210] 1. Data compression rate

[0211] Standard: compressed data volume / original data volume*100%

[0212] Method: Calculate the average compression rate within 30 days

[0213] 2. Data transmission delay

[0214] Standard: Average time from data collection to reaching the cloud (ms)

[0215] Methodology: Use a network analysis tool to measure average transmission delay over a 30-day period

[0216] 3. Anomaly Detection Accuracy

[0217] Standard: (true positive examples + true negative examples) / (true positive examples + false positive examples + true negative examples + false negative examples) * 100% Method: Manually verify the system detection results and calculate the accuracy

[0218] 4. Improved energy efficiency

[0219] Standard: (original energy consumption - optimized energy consumption) / original energy consumption * 100%

[0220] Method: Compare the average energy consumption before and after the implementation of this method for 30 days

[0221] 5. System response time

[0222] Standard: Average time from data collection to abnormal alarm triggering (s)

[0223] Method: Calculate the average response time within 30 days

[0224] 6. Network bandwidth utilization

[0225] Standard: actual bandwidth used / total available bandwidth*100%

[0226] Method: Use a network monitoring tool to measure average bandwidth utilization over a 30-day period

[0227] The test results are shown in the following table:

[0228] Table 1: Performance comparison between examples and comparative examples

[0229] index Example 1 Comparative Example 1 Comparative Example 2 Comparative Example 3 Data compression rate 92％ 70％ 85％ 88％ Data transmission delay (ms) 45 280 120 95 Anomaly detection accuracy 97.5％ 85.0％ 92.0％ 94.5％ Energy efficiency improvement 18.5％ 5.0％ 12.0％ 15.0％ System response time(s) 1.2 5.8 2.5 2.0 Network bandwidth utilization 65％ 95％ 80％ 75％

[0230] The above results are analyzed and explained as follows:

[0231] 1. Data compression rate: Example 1 achieved a compression rate of 92%, which is much higher than other comparative examples. This is due to the adaptive multimodal compression algorithm of the present invention, which can dynamically adjust the compression strategy according to the importance and characteristics of the data. A high compression rate means lower storage costs and network transmission burden.

[0232] 2. Data transmission delay: The transmission delay of Example 1 is only 45ms, which is 50ms lower than the closest comparative example 3. This is mainly due to the intelligent transmission scheduling and network self-optimization modules of the present invention, which can optimize the transmission path and network parameters in real time. Low latency is crucial for real-time monitoring and fast response.

[0233] 3. Anomaly detection accuracy: The anomaly detection accuracy of Example 1 reaches 97.5%, which is significantly higher than other methods. This demonstrates the superiority of quantum-inspired neural networks in processing complex multimodal data, and the effectiveness of dynamic threshold mechanisms in reducing false positives and false negatives. High accuracy means more reliable monitoring and fewer false alarms.

[0234] 4. Improved energy efficiency: Example 1 achieved an 18.5% improvement in energy efficiency, far exceeding other methods. This shows that the present invention not only improves the monitoring effect, but also significantly reduces energy consumption through precise control and optimization, and has obvious economic and environmental benefits.

[0235] 5. System response time: The system response time of Example 1 is only 1.2 seconds, which is much faster than other methods. This is due to the end-to-end optimization of the present invention, including adaptive sampling, fast compression and efficient transmission. Fast response is crucial for preventing potential accidents and timely intervention.

[0236] 6. Network bandwidth utilization: The bandwidth utilization of Example 1 is 65%, which is lower than that of other methods. This may not seem optimal, but it actually reflects the high efficiency of the present invention. Lower bandwidth utilization means that less network resources are occupied while transmitting the same amount of effective information, leaving more bandwidth for other applications.

[0237] Based on the above results, Example 1 is determined to be the best example. It performs well in all key indicators, especially in data compression rate, transmission delay and anomaly detection accuracy.

[0238] Compared with traditional methods and some improved methods, the remote monitoring method of the present invention shows significant superiority in multiple key indicators. It not only greatly improves the efficiency of data processing and transmission, but also significantly improves the accuracy of anomaly detection and the system response speed. At the same time, while improving the monitoring effect, this method also achieves a significant improvement in energy efficiency, reflecting the combination of technological innovation and sustainable development.

[0239] These results fully demonstrate the effectiveness and advancement of the innovations of the invention, such as quantum-inspired neural networks, adaptive multimodal compression, intelligent transmission scheduling, and network self-optimization. Together, they form an efficient, intelligent, and reliable remote monitoring system, providing strong technical support for Industry 4.0 and smart manufacturing.

[0240] This comprehensive performance improvement not only improves production efficiency and safety, but also brings significant economic benefits to enterprises. For example, an 18.5% improvement in energy efficiency may save millions of yuan in energy costs for large factories each year. At the same time, an anomaly detection accuracy rate of up to 97.5% can effectively prevent equipment failures and production interruptions, further reducing operating costs and risks.

[0241] In general, this invention represents an important breakthrough in industrial remote monitoring technology and provides new possibilities and directions for the digital transformation of the manufacturing industry.

[0242] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A remote monitoring method, characterized in that: The following steps are involved: Obtaining raw data collected by a multimodal sensor array; Preprocessing the original data to obtain preprocessed data; Based on the preprocessed data, performing adaptive multimodal compression and intelligent transmission to obtain compressed data; Decompressing the compressed data to obtain decompressed data; fusing and analyzing the decompressed data to obtain analysis results; Based on the analysis results, perform anomaly detection and early warning to obtain anomaly detection results; Based on the abnormal detection results, feedback and optimization are performed to update system parameters.

2. The remote monitoring method according to claim 1, characterized in that: In the step of acquiring the raw data collected by the multimodal sensor array, an adaptive sampling frequency strategy is adopted: f s (t)=f base +Δf·sigmoid(∑w i Δp i (t)) Among them, f s (t) is the real-time sampling frequency, f base is the basic sampling frequency, Δf is the adjustable frequency increment, w i is the weight of each parameter, Δp i (t) is the rate of change of parameter i.

3. The remote monitoring method according to claim 1, characterized in that: The step of preprocessing the raw data comprises: Use the improved wavelet threshold denoising algorithm to remove noise: X denoised =IDWT(T λ (DWT(X))) Among them, X is the original data, DWT and IDWT are discrete wavelet transform and its inverse transform respectively, T λ is the adaptive threshold function: T λ (x)=sign(x)(|x|-λ) + ·(1+exp(-α(|x|-β))) λ, α, β are dynamically adjusted parameters. The improved Z-score standardization method is used for data standardization: Among them, μ is the mean, σ is the standard deviation, and ∈ and γ are dynamically adjusted parameters.

4. The remote monitoring method according to claim 1, characterized in that: The step of performing adaptive multi-modal compression and intelligent transmission includes: Use quantum-inspired neural networks to perform real-time data analysis and obtain data importance scores; Based on the data importance score, performing adaptive multimodal compression to obtain compressed data; Performing compression quality assessment on the compressed data to obtain a compression quality assessment result; Based on the compression quality evaluation result, perform intelligent transmission scheduling and determine the data transmission strategy; Based on the data transmission strategy, network self-optimization is performed to adjust network parameters.

5. The remote monitoring method according to claim 4, characterized in that: The calculation formula of the quantum inspired neural network is: QHNN(X)=σ(W2·Q Φ (W1·X+b1)+b2) Among them, X is the input data, σ is the activation function, W1 and W2 are weight matrices, b1 and b2 are bias vectors, and Q Φ For quantum-inspired nonlinear transformations: Importance scoring function: I(X)=∑w i QHNN(X i )+λH(X) Among them, w i is the weight of each feature, H(X) is the entropy of the data, and λ is the balance parameter.

6. The remote monitoring method according to claim 4, characterized in that: The adaptive multimodal compression adopts a hybrid algorithm combining fractal compression and neural compression: C(X)=αF c (X)+(1-α)N c (X) Among them, X is the input data, C(X) is the compression result, and F c is the fractal compression function, N c is the neural compression function: F c (X)=argmin∑D(R i (X),S j (X)) Among them, r i and S j are the range block and definition block respectively, and D is the distance function; N c (X)=Decoder(Quantizer(Encoder(X))) Among them, Encoder and Decoder use improved variational autoencoder structure, and Quantizer adopts soft quantization method.

7. The remote monitoring method according to claim 1, characterized in that: The step of decompressing the compressed data adopts a decompression algorithm corresponding to the compression: For the fractal compression part: Among them, X′ is the decompression result, W is the iterated function system (IFS) defined in fractal coding, and X0 is the initial estimate; For the nerve compression part: X′=Decoder(Q -1 (Z)) Among them, Q -1 is the inverse quantization function, Decoder is the trained decoder network, and Z is the compressed data.

8. The remote monitoring method according to claim 1, characterized in that: The step of fusing and analyzing the decompressed data adopts the improved Dempster-Shafer evidence theory: Among them, m(A) is the basic probability distribution after fusion, m i (A) is the basic probability distribution of the i-th evidence source, K is the conflict degree, and w i The weight is adjusted dynamically.

9. The remote monitoring method according to claim 1, characterized in that: The anomaly detection and early warning step adopts the improved IsolationForest algorithm: Anomaly Score: Where x is the data point to be detected, n is the number of samples, E(h(x)) is the average path length, and c(n) is the normalization factor; Introducing a dynamic threshold mechanism: τ(t)=μ(s)+k(t)σ(s) k(t+1)=k(t)+δ(FPR(t)-FPR target ) Among them, τ(t) is the dynamic threshold, μ(s) and σ(s) are the mean and standard deviation of the anomaly score, k(t) is the dynamic adjustment factor, FPR is the false alarm rate, and δ is the learning rate.

10. A remote monitoring system for implementing the method according to any one of claims 1 to 9, characterized in that: include: Multimodal sensor arrays for acquiring raw data; A preprocessing module, used for preprocessing the raw data to obtain preprocessed data; An adaptive multi-modal compression and intelligent transmission module, used to perform compression and transmission based on the pre-processed data to obtain compressed data; A decompression module, used for decompressing the compressed data to obtain decompressed data; A data fusion and analysis module, used to fuse and analyze the decompressed data to obtain analysis results; An anomaly detection and early warning module, used to perform anomaly detection and early warning based on the analysis result to obtain an anomaly detection result; The feedback and optimization module is used to provide feedback and optimization based on the abnormality detection results and update system parameters.

Images