Data transmission method, device, equipment and storage medium based on Internet of Things

By using IoT sensors to generate device authentication codes and encrypted data in the micro-module computer room, combining multi-scale trend decomposition model and fault detection decision tree, the knowledge graph is built and reinforcement learning is solved, and the security of data transmission and operation and maintenance of micro-module computer room is solved. and real-time problems achieve efficient and reliable data transmission and fault diagnosis.

CN118900213BActive Publication Date: 2024-12-06SHENZHEN XUNHE IND CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411399514.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-10-09
Publication Date
2024-12-06
Estimated Expiration
2044-10-09

AI Technical Summary

Technical Problem

Micro-module computer rooms face data security and real-time challenges in data transmission and operation and maintenance. Traditional encryption and authentication mechanisms cannot meet the real-time and low latency requirements, and IoT devices increase the risk and complexity of network attacks.

Method used

Using the IoT-based data transmission method, data is collected through multiple IoT sensors, device authentication code is generated for identity authentication and encryption, and state prediction and fault diagnosis are predicted using multi-scale trend decomposition of long and short-term memory models and multi-level fault detection decision tree. , build a knowledge graph and perform reinforcement learning optimization to generate adaptive control strategies.

Benefits of technology

It improves the operating efficiency and reliability of the micro-module computer room, enhances the security of equipment authentication and data encryption, improves the reliability of data transmission and the sensitivity of fault detection, and realizes intelligent management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118900213B_ABST
    Figure CN118900213B_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of data transmission of the Internet of Things, and discloses a data transmission method, device, equipment and storage medium based on the Internet of Things. The method comprises: collecting original data packets of a micromodule computer room through multiple Internet of Things sensors, and generating an authentication data packet; performing identity authentication to generate a security data packet; performing data decryption and feature extraction, and obtaining a target feature set; predicting the operation state of the micromodule computer room through a multi-scale trend decomposition long short-term memory model to obtain a comprehensive prediction result; calculating multivariate statistics and performing fault diagnosis through a multi-level fault detection decision tree to obtain fault diagnosis information; constructing a knowledge graph of the micromodule computer room and performing reinforcement learning optimization according to the comprehensive prediction result and the fault diagnosis information, and generating an adaptive control strategy. The invention realizes the generation of an adaptive control strategy based on historical experience and real-time data, and improves the operation efficiency and reliability of the micromodule computer room.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of data transmission of the Internet of Things, and in particular to a data transmission method, device, equipment and storage medium based on the Internet of Things. Background Art

[0002] With the rapid development of data center technology, micro-module computer rooms, as a highly integrated and rapidly deployed solution, play an increasingly important role in the construction of modern data centers. However, the high degree of integration of micro-module computer rooms also brings complex operation and maintenance challenges, especially in terms of data security, real-time monitoring and intelligent management. Traditional data transmission and processing methods are difficult to meet the requirements of micro-module computer rooms for efficient, safe and intelligent operation and maintenance.

[0003] In terms of data transmission, micro-module computer rooms face challenges in data security and integrity. The transmission and storage of large amounts of sensitive data increases the risk of data leakage and tampering, while existing encryption and authentication mechanisms often fail to meet the real-time and low-latency requirements of micro-module computer rooms. In addition, the widespread use of IoT devices has improved the comprehensiveness of data collection, but has also increased the risk and complexity of cyber attacks. Summary of the invention

[0004] The present invention provides a data transmission method, device, equipment and storage medium based on the Internet of Things. The present invention realizes the generation of adaptive control strategies based on historical experience and real-time data, thereby improving the operating efficiency and reliability of a micromodule computer room.

[0005] In a first aspect, the present invention provides a data transmission method based on the Internet of Things, the data transmission method based on the Internet of Things comprising:

[0006] Collect the original data packets of the micromodule computer room through multiple IoT sensors, generate a device authentication code for each IoT sensor, add it to the original data packet, and obtain an authentication data packet;

[0007] Performing identity authentication based on the device authentication code, generating a session key, and encrypting the authentication data packet to obtain a security data packet, and at the same time, transmitting the security data packet to the cloud computing platform;

[0008] In the cloud computing platform, data decryption and feature extraction are performed on the security data packet, and a target feature set is obtained;

[0009] Inputting the target feature set into a multi-scale trend decomposition long short-term memory model to predict the operation status of a micro-module computer room, and obtaining a comprehensive prediction result;

[0010] Based on the comprehensive prediction results and actual observation data, multivariate statistics are calculated, and the multivariate statistics are input into a multi-level fault detection decision tree for fault diagnosis to obtain fault diagnosis information;

[0011] According to the comprehensive prediction results and the fault diagnosis information, a knowledge graph of the micro-module computer room is constructed and reinforcement learning optimization is performed to generate an adaptive control strategy.

[0012] In a second aspect, the present invention provides a data transmission device based on the Internet of Things, the data transmission device based on the Internet of Things comprising:

[0013] The acquisition module is used to collect the original data packets of the micromodule room through multiple IoT sensors, and generate a device authentication code for each IoT sensor, add it to the original data packet, and obtain an authentication data packet;

[0014] An authentication module, used to perform identity authentication based on the device authentication code, generate a session key, and encrypt the authentication data packet to obtain a security data packet, and at the same time, transmit the security data packet to the cloud computing platform;

[0015] An extraction module, used for performing data decryption and feature extraction on the security data packet, a target feature set, in the cloud computing platform;

[0016] A prediction module, used for inputting the target feature set into a multi-scale trend decomposition long short-term memory model to predict the operation status of the micro-module computer room and obtain a comprehensive prediction result;

[0017] A diagnosis module, used to calculate multivariate statistics based on the comprehensive prediction results and actual observation data, and input the multivariate statistics into a multi-level fault detection decision tree to perform fault diagnosis and obtain fault diagnosis information;

[0018] A generation module is used to construct a knowledge graph of a micro-module computer room and perform reinforcement learning optimization to generate an adaptive control strategy based on the comprehensive prediction results and the fault diagnosis information.

[0019] The third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned data transmission method based on the Internet of Things.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer executes the above-mentioned data transmission method based on the Internet of Things.

[0021] In the technical solution provided by the present invention, the security of device authentication and data encryption is enhanced through the physical unclonable function and multiple encryption mechanisms; the reliability and real-time performance of data transmission are improved by adopting Reed-Solomon error correction coding and MQTT protocol. By using technologies such as multi-scale decomposition and principal component analysis, efficient processing of complex data and extraction of key features are achieved, laying the foundation for subsequent analysis. The multi-scale trend decomposition long short-term memory model is adopted, combined with discrete wavelet transform and empirical mode decomposition, to achieve multi-scale and multi-dimensional prediction of the operating status of the micro-module computer room. By combining multivariate statistics and multi-level fault detection decision trees, the sensitivity and accuracy of fault detection are improved, and accurate diagnosis of fault types and causes is achieved. By constructing a knowledge graph of the micro-module computer room and combining it with reinforcement learning, the generation of adaptive control strategies based on historical experience and real-time data is achieved, which improves the operating efficiency and reliability of the micro-module computer room. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 Schematic diagram of the steps of the data transmission method based on the Internet of Things in an embodiment of the present invention;

[0024] Figure 2 Schematic diagram of the structure of a data transmission device based on the Internet of Things in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] Embodiments of the present invention provide a data transmission method, apparatus, device and storage medium based on the Internet of Things. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the data transmission method based on the Internet of Things in an embodiment of the present invention includes:

[0027] Step S1, collecting the original data packet of the micromodule computer room through multiple IoT sensors, and generating a device authentication code for each IoT sensor, adding it to the original data packet to obtain an authentication data packet;

[0028] It is understandable that the execution subject of the present invention may be a data transmission device based on the Internet of Things, or may be a terminal or a server, which is not specifically limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0029] Specifically, data is collected from the temperature sensor, humidity sensor, power consumption sensor, airflow sensor and device status sensor in the micromodule room. The data packet collected by each sensor contains the unique identification (ID) of the sensor, the type of data, the actual measured value and the timestamp of the collection. These information together constitute the original data packet. The physical unclonable function module integrated in each IoT sensor is used to respond to the random challenge value and generate a unique device fingerprint. The device fingerprint is the unique identification of each sensor, which is non-copyable and unpredictable and has a high degree of security. Based on the device fingerprint, a fixed-length device authentication code is generated. The device authentication code is associated with the original data packet of the sensor and added to the original data packet to form a first data packet. In order to ensure the consistency and operability of the data during transmission and processing, the first data packet is formatted. The data from different types of sensors are unified into a standardized format, usually the JSON format. Through formatting, the second data packet is obtained, which maintains good compatibility between different platforms and systems. The data in the second data packet is segmented, and the sliding window method is used. Each window contains 100 data points, and the sliding step is set to 10 data points to obtain the initial segmented data. Perform wavelet transform on the initial segmented data and denoise the data to remove the noise and interference that may be introduced during the acquisition process, so as to obtain more accurate segmented data. Repackage the denoised segmented data with the corresponding device authentication code to obtain an authentication data packet. Considering that there may be problems such as network interruption in the IoT environment, the authentication data packet is stored in the local cache. This strategy can effectively deal with network interruption situations, ensuring that data will not be lost even when the network is unavailable, and network recovery data can continue to be transmitted to the target cloud computing platform, ensuring the continuity and integrity of data transmission.

[0030] Step S2: Perform identity authentication based on the device authentication code, generate a session key, and encrypt the authentication data packet to obtain a security data packet. At the same time, the security data packet is transmitted to the cloud computing platform;

[0031] Specifically, the device authentication code generated by each device is verified by applying the elliptic curve digital signature algorithm to ensure that the device's authentication result is credible. Based on the authentication result, a shared key is generated using the elliptic curve Diffie-Hellman key exchange protocol. The shared key is usually 256 bits, has a high degree of security and confidentiality, and can ensure the encryption strength of the subsequent data transmission process. The HMAC key derivation function is applied to the shared key, and SHA-256 is used as the hash function to obtain the final session key. The authentication data packet is encrypted using the session key, and a 12-byte random initialization vector is set at the same time to increase the randomness and security of the encryption process to obtain the encrypted data. The HMAC-SHA256 message authentication code of the encrypted data is calculated. The key derived from the session key is used as the key of HMAC to generate a data integrity check value to ensure that the data in the transmission process will not be tampered with. Even if some data packets are damaged during the transmission process, they can be detected and corrected in time. The encrypted data, initialization vector and data integrity check value are packaged to construct a secure data packet. In order to improve the reliability of data transmission, the secure data packet is processed in blocks, and the size of each block of data is set to 1024 bytes. The Reed-Solomon error correction coding algorithm is used for encoding, and the parameters are set to RS (255, 223). Through this step, the anti-interference ability of the data during transmission is enhanced, ensuring that even if some data packets are lost or damaged during transmission, they can be restored through the error correction mechanism. At the transmission level, the MQTT protocol is used for data transmission, and the QoS level is set to 1. Each data packet will be transmitted at least once and will be confirmed to ensure the reliability of data transmission. In order to ensure the security of the transmission process, the entire transmission process is carried out through the TLS 1.3 encrypted channel. This encrypted channel provides additional security to ensure that the data will not be intercepted or tampered with when it is transmitted to the pre-configured cloud computing platform address.

[0032] Step S3: In the cloud computing platform, decrypt and extract features of the security data packet to obtain the target feature set;

[0033] Specifically, the cloud computing platform performs Reed-Solomon decoding on the received encoded data packets and uses RS (255, 223) parameters for error correction to ensure that even if some data packets are erroneous or damaged during transmission, the original secure data packets can still be restored. The redundancy check mechanism of Reed-Solomon decoding can effectively repair possible transmission errors and ensure the integrity and accuracy of the data. Protocol Buffers is used to deserialize the original secure data packets and convert the serialized binary data back into an operable data structure. Through deserialization, the platform extracts the encrypted data, initialization vector, and data integrity check value from the secure data packet to obtain the deconstructed data components. Based on the pre-shared key material, the HKDF key derivation process is re-executed to generate a session key consistent with the sender. The key derivation process uses the same hash function and parameter settings as the sender to ensure that the session key generated at the receiving end is consistent with the sender. The cloud computing platform uses the decryption key and initialization vector to perform AES-256-GCM decryption on the encrypted data to obtain the original authentication data packet and restore the original state when it was sent. The original authentication data packet is parsed by JSON, and the content of the data packet is converted into a readable JSON format to facilitate subsequent data extraction and analysis. Various sensor data, including temperature, humidity, power consumption, airflow, and equipment status information, are extracted from it to form a structured data set that represents the operating status of the micromodule room at a specific time point. The time series of the structured data set is decomposed into trend, seasonal, and residual components. The multi-dimensional decomposition method helps to reveal long-term trends, periodic changes, and random fluctuations in the data, providing a richer analysis perspective. Based on the decomposed data, statistical features, time domain features, and frequency domain features are calculated to form a preliminary feature set that can reflect the core features and potential problems of system operation. Principal component analysis is performed on the preliminary feature set, and the feature vectors are projected into a new coordinate system through linear transformation, and the principal components with a cumulative contribution rate of 95% are selected. The principal component analysis effectively reduces the dimension of the data, retains the most important information in the data, and obtains the target feature set.

[0034] Step S4, inputting the target feature set into the multi-scale trend decomposition long short-term memory model to predict the operation status of the micro-module computer room to obtain a comprehensive prediction result;

[0035] Specifically, the target feature set is decomposed by discrete wavelet transform, and the data is decomposed into components of different frequencies to obtain low-frequency approximate coefficients and high-frequency detail coefficients of three scales. The low-frequency approximate coefficients usually contain the overall trend information of the data, while the high-frequency detail coefficients reflect the fluctuation characteristics of the data. The low-frequency approximate coefficients are input into the long-term prediction LSTM network in the multi-scale trend decomposition long short-term memory model. The long-term prediction LSTM network consists of three LSTM layers, each containing 128 neurons, and uses the tanh activation function to handle nonlinear relationships. Through the long-term prediction LSTM network, the long-term trend of the data is captured and the long-term trend prediction results are generated. At the same time, the high-frequency detail coefficients of the three scales are respectively input into the three parallel medium-term prediction LSTM networks in the multi-scale trend decomposition long short-term memory model. Each medium-term prediction network consists of two LSTM layers, each containing 64 neurons, and uses the ReLU activation function. The medium-term prediction LSTM network is mainly used to capture the medium-term fluctuation characteristics in the data and generate medium-term fluctuation prediction results of three scales. The target feature set is subjected to empirical mode decomposition (EMD). EMD is an adaptive signal decomposition method that decomposes the time series into five intrinsic mode functions (IMFs) and a residual term. The five intrinsic mode functions represent different frequency components in the data, while the residual term reflects the part of the data that cannot be captured by the mode function. The five intrinsic mode functions are respectively input into five parallel short-term prediction LSTM networks in the multi-scale trend decomposition long short-term memory model. Each short-term prediction network consists of one LSTM layer, each layer contains 32 neurons, and uses the Sigmoid activation function. These networks are mainly used to capture the short-term fluctuation characteristics in the data and generate five short-term pattern prediction results. The residual term is input into the fully connected neural network in the multi-scale trend decomposition long short-term memory model. The fully connected neural network contains three hidden layers, each layer contains 16 neurons, and uses the LeakyReLU activation function for nonlinear processing to generate residual prediction results. The attention mechanism is used to perform weighted fusion on the long-term trend prediction results, three medium-term fluctuation prediction results, five short-term pattern prediction results, and residual prediction results. The attention mechanism enables the system to dynamically adjust the weights of different prediction results according to their importance, and obtain a more accurate and reliable comprehensive prediction result.

[0036] Step S5: Calculate multivariate statistics based on the comprehensive prediction results and actual observation data, and input the multivariate statistics into a multi-level fault detection decision tree for fault diagnosis to obtain fault diagnosis information;

[0037] Specifically, residuals are calculated for the comprehensive prediction results and actual observation data to obtain a residual sequence. The residual sequence reflects the deviation between the predicted value and the actual value, and is a key indicator for detecting anomalies and potential faults. The residual sequence is segmented to divide the data into smaller intervals, more accurately capture abnormal changes in the time series, and obtain segmented residual data. A series of multivariate statistical analyses are performed on the segmented residual data. Hotelling's T² statistic is calculated, and the covariance relationship of multiple variables is comprehensively considered. The calculated T² statistic sequence can help identify anomalies in the data. Q statistics are calculated. Q statistics are often used to detect the overall change characteristics of data and can capture abnormal patterns that cannot be identified by T² statistics. At the same time, the generalized variance (GV) statistic is calculated for the segmented residual data. The changes in data distribution are measured by the determinant of the sample covariance matrix to generate a GV statistic sequence. The Wishart statistic is calculated for the segmented residual data. The log-likelihood ratio method is used to generate the Wishart statistic sequence by comparing it with the normal distribution to enhance the ability to identify abnormal data. At the same time, the Kullback-Leibler (KL) divergence between the comprehensive prediction results and the actual observed data is calculated to measure the difference between the two probability distributions, help identify the significant deviation between the model prediction and the actual situation, and obtain the KL divergence sequence. The T² statistic sequence, Q statistic sequence, GV statistic sequence, Wishart statistic sequence and KL divergence sequence are input into the first layer of the multi-level fault detection decision tree. In the first layer, the C4.5 algorithm is used for preliminary classification. The algorithm can effectively classify the data according to the characteristics of the statistics, preliminarily judge the possible fault type, and obtain the preliminary fault type. Based on the preliminary fault type, combined with multivariate statistics, the second-level decision tree analysis is performed to obtain the refined fault type. The contribution diagram analysis of the refined fault type is performed to identify the most important contributing factors to the fault and determine which specific conditions or variables cause the abnormal situation. The detailed fault types and main contributing factors are input into the fault diagnosis knowledge graph for analysis. By comparing with historical data and known fault modes, the fault causes are confirmed and the final fault diagnosis information is generated.

[0038] Step S6: Based on the comprehensive prediction results and fault diagnosis information, a knowledge graph of the micro-module computer room is constructed and reinforcement learning optimization is performed to generate an adaptive control strategy.

[0039] Specifically, based on the comprehensive prediction results and fault diagnosis information, RDF triples are used to represent the relationship between entities. RDF triples can describe various entities in the computer room and their relationships, such as the interaction between devices, the association between operating status and environmental parameters, etc. The ontology model is defined using OWL (Web Ontology Language) to provide a clear semantic structure for the knowledge graph, so that different entities and relationships can be interpreted and operated under a unified framework. In this way, a comprehensive and structured knowledge graph of the micro-module computer room is constructed. After the knowledge graph is constructed, in order to adapt it to the needs of reinforcement learning, the graph embedding algorithm is applied to the knowledge graph. The graph embedding algorithm maps the entities and relationships in the knowledge graph to a low-dimensional vector space to obtain a graph embedding representation. Based on the graph embedding representation, a deep Q network (DQN) is constructed. The state space of the network contains the current environmental parameters, equipment status, and predicted future status, while the action space includes possible control instructions such as adjusting refrigeration equipment parameters, adjusting airflow distribution, and performing load balancing operations. The definition of state and action space ensures that the DQN model can comprehensively consider the operating status of the computer room and make appropriate adjustment strategies to obtain an initial DQN model. The initial DQN model is trained. During the training process, the ε-greedy strategy is used for exploration to balance the exploration and utilization of the model. In order to effectively store and utilize historical experience, the experience playback buffer size is set to 10,000, and the target network is updated every 50 time steps, so as to avoid the model from falling into the local optimum during the training process and ensure that it converges to a better strategy. After the training is completed, the trained DQN model is obtained. In order to optimize the decision-making process, the trained DQN model is combined with the Monte Carlo Tree Search (MCTS) algorithm to construct a reinforcement learning framework. In this framework, the MCTS algorithm is used for action selection, while the DQN model is used for state evaluation. Through this combination, the exploration efficiency and decision quality of the strategy are effectively improved. Through 500 iterations, the strategy network is gradually optimized to obtain a strategy model with excellent performance. Based on the optimized strategy network, combined with the comprehensive prediction results and fault diagnosis information of the micro-module computer room, an adaptive control strategy is generated. The strategy includes specific refrigeration equipment parameter adjustment instructions, airflow distribution optimization schemes, and load balancing strategies. These control strategies can dynamically adjust the operating status of the computer room, ensure that the equipment operates under optimal conditions, improve energy efficiency and reduce the possibility of failures, and ultimately achieve intelligent management of the micro-module computer room.

[0040] In the embodiment of the present invention, the security of device authentication and data encryption is enhanced through the physical unclonable function and multiple encryption mechanisms; the reliability and real-time performance of data transmission are improved by adopting Reed-Solomon error correction coding and MQTT protocol. By using technologies such as multi-scale decomposition and principal component analysis, efficient processing of complex data and extraction of key features are achieved, laying the foundation for subsequent analysis. The multi-scale trend decomposition long short-term memory model is adopted, combined with discrete wavelet transform and empirical mode decomposition, to achieve multi-scale and multi-dimensional prediction of the operating status of the micro-module computer room. By combining multivariate statistics and multi-level fault detection decision trees, the sensitivity and accuracy of fault detection are improved, and accurate diagnosis of fault types and causes is achieved. By constructing a knowledge graph of the micro-module computer room and combining it with reinforcement learning, the generation of adaptive control strategies based on historical experience and real-time data is achieved, which improves the operating efficiency and reliability of the micro-module computer room.

[0041] In a specific embodiment, the process of executing step S1 may specifically include the following steps:

[0042] Collect data from the temperature sensor, humidity sensor, power consumption sensor, airflow sensor, and equipment status sensor in the micromodule room to obtain the original data packet containing the sensor ID, data type, data value, and timestamp;

[0043] Utilize the physical unclonable function module of each IoT sensor to perform response calculations on the random challenge value to obtain a unique device fingerprint, and generate a fixed-length device authentication code based on the device fingerprint;

[0044] Add the device authentication code to the corresponding original data packet to form a first data packet, perform data formatting on the first data packet, unify different types of sensor data into JSON format, and obtain a second data packet;

[0045] The data in the second data packet is segmented, each window contains 100 data points, and the sliding step is 10 data points, to obtain initial segmented data;

[0046] Perform wavelet transform on the initial segmented data to obtain denoised segmented data, and repackage the denoised segmented data with the corresponding device authentication code to obtain an authentication data packet, which is stored in a local cache to cope with network interruption.

[0047] Specifically, data is collected from temperature sensors, humidity sensors, power consumption sensors, airflow sensors, and device status sensors. The data packet collected by each sensor contains four main elements: the sensor's unique identifier (ID), the type of data (such as temperature, humidity, etc.), the specific data value measured, and the timestamp of data collection. These elements constitute the original data packet. Utilize the physical unclonable functional module of each IoT sensor. The physical unclonable functional module can respond to a random challenge value and generate a unique device fingerprint. The device fingerprint is a hardware-based unique identifier that cannot be copied or predicted and can effectively prevent forgery and tampering. Assume that the random challenge value input to the PUF is , the physical unclonable function generates a unique response value based on the physical characteristics of the device , that is, the device fingerprint. Device fingerprint It can be expressed as:

[0048] ;

[0049] in, It is the response function of the physical unclonable functional module, which depends on the physical characteristics of the sensor and cannot be copied or simulated. Based on the device fingerprint, a fixed-length device authentication code is generated to ensure that the data of each sensor has a unique identity and can be used to verify the authenticity of the data in the subsequent data processing process. The device authentication code is added to the corresponding original data packet to form a first data packet that has undergone preliminary security processing. The first data packet is formatted. For example, the JSON format (JavaScript Object Notation) is selected for formatting. The JSON format has the advantages of being structured, easy to parse, and having strong cross-platform compatibility. After formatting, a second data packet is formed. The data in the second data packet is segmented. In order to ensure the temporal continuity of the data and the effectiveness of the processing, a sliding window segmentation method is adopted. Each window contains 100 data points, and the sliding step size is 10 data points. Through the segmentation method, the efficiency and accuracy of the processing are improved while ensuring the integrity of the data. Assume that the data sequence is , the sliding window function can be defined as:

[0050] ;

[0051] in, is the starting position of the window, the window size is 100, and the step length is 10. By sliding the window, multiple overlapping data segments are extracted from the data sequence to form initial segmented data. Wavelet transform is performed on the initial segmented data. Wavelet transform is an effective denoising tool. By decomposing the data into different frequency components, wavelet transform can separate the noise in the signal and obtain the denoised segmented data. Suppose the initial segmented data is , the wavelet transform can be expressed as:

[0052] ;

[0053] in, are the wavelet coefficients, is the scale parameter, is the translation parameter, is the mother wavelet function, is its conjugate complex number. By selecting the appropriate mother wavelet and scale parameter, the wavelet transform can effectively remove high-frequency noise in the signal, retain meaningful low-frequency information, and obtain smoother and more accurate segmented data. The denoised segmented data is repackaged with the corresponding device authentication code to form the final authentication data packet. The authentication data packet is stored in the local cache to cope with network interruptions.

[0054] In a specific embodiment, the process of executing step S2 may specifically include the following steps:

[0055] The device authentication code is verified using the elliptic curve digital signature algorithm to obtain an identity authentication result, and based on the identity authentication result, the elliptic curve Diffie-Hellman key exchange protocol is used to generate a shared key to obtain a 256-bit shared key;

[0056] Apply the HKDF key derivation function to the shared key, use SHA-256 as the hash function to obtain the session key, and use the session key to encrypt the authentication data packet, set a 12-byte random initialization vector, and obtain the encrypted data;

[0057] Calculate the HMAC-SHA256 message authentication code of the encrypted data, use the derived key of the session key as the HMAC key, and obtain the data integrity check value;

[0058] The encrypted data, initialization vector and data integrity check value are packaged to construct a secure data packet, and the secure data packet is divided into blocks, with each block size set to 1024 bytes. The Reed-Solomon error correction coding algorithm is used with the parameters set to RS (255, 223) to obtain a coded data packet.

[0059] The MQTT protocol is used, the QoS level is set to 1, and the encoded data packets are transmitted in batches to the pre-configured cloud computing platform address. During the transmission process, TLS 1.3 encrypted channels are used to ensure transmission security.

[0060] Specifically, the device authentication code is verified by the Elliptic Curve Digital Signature Algorithm (ECDSA) to obtain the identity authentication result. ECDSA is a digital signature algorithm based on the elliptic curve discrete logarithm problem, which has high security and small key size. Assume that the device authentication code is , the signature generated by the device is , the public key is , the verification process includes the following steps: Calculate the message hash SHA-256 , and then use the base point of the elliptic curve Calculate verification value .if The x-coordinate of If the signature verification is successful, the device's identity authentication is successful. Otherwise, the authentication fails. Based on the authentication result, the Elliptic Curve Diffie-Hellman (ECDH) key exchange protocol is used to generate a shared key. ECDH is a secure key exchange protocol that allows two devices to generate a shared key without directly transmitting the key. Assume that the private key of device A is , the public key of device B is , then device A can calculate the shared key:

[0061] ;

[0062] in, is the order of the elliptic curve, is the final shared key. The shared key is usually 256 bits and is used for subsequent encryption operations to ensure the confidentiality of data transmission. In order to enhance the security of the key, the HKDF key derivation function is applied to the shared key, and SHA-256 is used as the hash function. HKDF processes and expands the shared key through one or more steps to generate a session key. HKDF divides the input key material (shared key) into two stages: extraction and expansion. A pseudo-random key is generated by a salt value and the shared key, and the final session key is generated through an expansion function on this basis. The session key is used to encrypt the authentication data packet to ensure that the encryption key used for each communication is unique. When using the session key to encrypt the authentication data packet, a 12-byte random initialization vector is set. The initialization vector is part of the encryption algorithm, which is used to ensure that the generated ciphertext will be different even under the same plaintext and key. Assume that the authentication data packet is , the session key is , the initialization vector is , the encryption process can be expressed as:

[0063] ;

[0064] in, is the encrypted data. AES-256-GCM is a mode in the encryption standard with high efficiency and security. Calculate the HMAC-SHA256 message authentication code (MAC) of the encrypted data to verify the integrity of the data. HMAC-SHA256 is a hash-based message authentication code that uses the derived key of the session key as the HMAC key to perform a hash operation on the encrypted data to generate a fixed-length check value. Assume that the encrypted data is , the derived key is , then the HMAC calculation formula is:

[0065] ;

[0066] The HMAC value is attached to the encrypted data to ensure that the data has not been tampered with during transmission. The encrypted data, initialization vector and data integrity check value are packaged to construct a secure data packet. The secure data packet is processed in blocks, and the size of each block is set to 1024 bytes. The blocked data packet is processed by the Reed-Solomon error correction coding algorithm (RS). RS (255, 223) means that the data is encoded into blocks of 255 bytes, of which 223 bytes are the original data and 32 bytes are the error correction code. In this way, even if some bytes are erroneous during transmission, the data can still be recovered. The mathematical representation of RS coding is:

[0067] ;

[0068] in, is the encoded data block, is the original data block, is a generating polynomial. The MQTT protocol is used to transmit the encoded data packets in batches to the pre-configured cloud computing platform address. MQTT is a lightweight message transmission protocol suitable for low-bandwidth and unstable network connections in the IoT environment. Setting the QoS level to 1 means that each message will be transmitted and confirmed at least once to ensure the reliability of the data. At the same time, the system communicates through the TLS 1.3 encrypted channel during the transmission process. TLS (Transport Layer Security Protocol) provides encryption and integrity protection in data transmission. Through TLS1.3, the transmitted data is encrypted, and digital certificates can also be used for authentication to ensure that the data will not be tampered with or leaked midway.

[0069] In a specific embodiment, the process of executing step S3 may specifically include the following steps:

[0070] Perform Reed-Solomon decoding on the received coded data packet and use RS (255, 223) parameters for error correction to obtain the original secure data packet;

[0071] Use Protocol Buffers to deserialize the original secure data packet, extract the encrypted data, initialization vector, and data integrity check value, and obtain the deconstructed data components;

[0072] Based on the pre-shared key material, the HKDF key derivation process is re-executed to generate a session key consistent with the sender, and the key required for decryption is obtained;

[0073] Use the decryption key and initialization vector to perform AES-256-GCM decryption on the encrypted data to obtain the original authentication data packet;

[0074] Parse the original authentication data packet in JSON format and extract various sensor data, including temperature, humidity, power consumption, airflow, and device status information, to obtain a structured data set;

[0075] Decompose the time series of structured data sets into trend, seasonality and residual components to obtain multi-dimensional decomposition data, and calculate statistical features, time domain features and frequency domain features based on the multi-dimensional decomposition data to obtain a preliminary feature set;

[0076] The principal component analysis was performed on the preliminary feature set, and the principal components with a cumulative contribution rate of 95% were selected to obtain the target feature set.

[0077] Specifically, the data packet is decoded by Reed-Solomon. This encoding method is an error correction coding technology used in data communication, which is suitable for combating burst errors in the channel. Assume that the received encoded data packet is , which is generated by Reed-Solomon coding, where is a polynomial with a length of 255 bytes, and its original data part The RS (255, 223) parameter is used for decoding, and the error correction algorithm can identify and repair errors in the data packet and restore the original secure data packet. . The decoding process includes calculating the comprehensive value of the received polynomial, solving the error location polynomial using the Euclidean algorithm, then applying the Chien search to locate the error location, and finally correcting the error values ​​at these locations to obtain the correct original security data packet. Use Protocol Buffers (Protobuf) to deserialize the original security data packet. Protocol Buffers is a flexible, language-neutral and extensible mechanism for serializing structured data. During the deserialization process, Protobuf parses the various fields in the security data packet and extracts the encrypted data, initialization vector (IN) and data integrity check value. The HKDF key derivation process is re-executed based on the pre-shared key material, and a session key consistent with the sender is generated from the shared key material to ensure that the decryption keys used by both parties are consistent. Assume that the shared key material is , generate a pseudo-random key (PRK) through the HKDF extraction step, and then generate the final session key through the expansion step

[0078] ;

[0079] Where info is the context information and length is the length of the required session key (usually 256 bits). This process ensures the correctness of the decryption key so that the data can be accurately decrypted. and the initialization vector (IV) extracted previously, perform AES-256-GCM decryption on the encrypted data. AES-256-GCM is an advanced encryption standard that combines the Galois / Counter Mode (GCM) mode to provide data encryption and data integrity verification. Assume that the encrypted data is , then the decryption process can be expressed as:

[0080] ;

[0081] in, is the original authentication data packet obtained by decryption. Perform JSON parsing on the original authentication data packet. JSON is a lightweight data exchange format. Through JSON parsing, various sensor data such as temperature, humidity, power consumption, airflow, and device status information are extracted from the original authentication data packet. After parsing, these data are stored as structured data sets, which facilitate subsequent data analysis and processing. Perform time series decomposition on the structured data set. Time series data usually contains multiple patterns, including long-term trends, seasonal fluctuations, and short-term residuals. By decomposing the time series, multi-dimensional decomposed data is obtained. Assume that the time series data is , decompose it into trend components , seasonal components and the residual components

[0082] ;

[0083] This decomposition helps identify different patterns in the data, which facilitates more accurate analysis and prediction. After completing the time series decomposition, the statistical features, time domain features, and frequency domain features are calculated for the multi-dimensional decomposition data. Statistical features include mean, variance, skewness, and kurtosis, which can describe the overall distribution of the data; time domain features include autocorrelation and delay, which can reflect the time correlation of the data; frequency domain features reveal the energy distribution of the data at different frequencies through methods such as Fourier transform. These features constitute the preliminary feature set. In order to reduce the feature dimension and retain the main information of the data, principal component analysis is performed on the preliminary feature set. Principal component analysis is a statistical technique that reduces the dimension of the data by linearly transforming the original variables into new uncorrelated variables (i.e., principal components). The selection of principal components is based on their explained variance contribution rate. Usually, principal components with a cumulative contribution rate of 95% are selected to obtain the final target feature set. Assume that the original feature matrix is , the process of principal component analysis can be expressed as:

[0084] ;

[0085] in, is the principal component matrix, is the eigenvector matrix. By reducing the dimension through principal component analysis, the complexity of the data is reduced while retaining most of the information, thus improving the computational efficiency of the subsequent algorithms.

[0086] In a specific embodiment, the process of executing step S4 may specifically include the following steps:

[0087] The target feature set is decomposed by discrete wavelet transform to obtain low-frequency approximate coefficients and high-frequency detail coefficients of three scales;

[0088] The low-frequency approximation coefficients are input into the long-term prediction LSTM network in the multi-scale trend decomposition long short-term memory model. The long-term prediction LSTM network contains 3 LSTM layers, each with 128 neurons, and uses the tanh activation function to obtain the long-term trend prediction results.

[0089] The high-frequency detail coefficients of the three scales are input into three parallel medium-term prediction LSTM networks in the multi-scale trend decomposition long short-term memory model. Each network contains two LSTM layers, each with 64 neurons, and uses the ReLU activation function to obtain medium-term fluctuation prediction results at three scales.

[0090] Perform empirical mode decomposition on the target feature set, extract five intrinsic mode functions and one residual term, and obtain multi-scale time series components;

[0091] The five intrinsic pattern functions are input into five parallel short-term prediction LSTM networks in the multi-scale trend decomposition long short-term memory model. Each network contains one LSTM layer and 32 neurons. The Sigmoid activation function is used to obtain five short-term pattern prediction results.

[0092] The residual term is input into the fully connected neural network in the multi-scale trend decomposition long short-term memory model, which contains 3 hidden layers, 16 neurons in each layer, and uses the LeakyReLU activation function to obtain the residual prediction result;

[0093] The attention mechanism is used to perform weighted fusion on the long-term trend prediction results, 3 medium-term fluctuation prediction results, 5 short-term pattern prediction results and residual prediction results to obtain a comprehensive prediction result.

[0094] Specifically, the target feature set is decomposed by discrete wavelet transform. Wavelet transform is a tool used for signal processing, which can decompose the original signal into components of different frequencies. For a given time series data , and apply discrete wavelet transform to decompose it into low-frequency approximate coefficients and high frequency detail factor ,in Indicates scale. Low frequency approximation coefficient The overall trend of the signal is captured, while the high-frequency detail coefficient It reflects the rapid changes in the signal. Through decomposition, low-frequency approximate coefficients and high-frequency detail coefficients of three different scales are obtained. These coefficients represent the characteristics of the signal at different time scales. The low-frequency approximate coefficients are input into the long-term prediction LSTM network in the multi-scale trend decomposition LSTM model. The LSTM network consists of three layers of LSTM units, each layer contains 128 neurons, and uses a tanh activation function. The LSTM (Long Short-Term Memory) network is a special recurrent neural network (RNN) that can effectively capture long-term dependencies in time series by introducing a gating mechanism. The long-term prediction LSTM network is designed to extract long-term trends in the data and generate long-term trend prediction results. Suppose the low-frequency approximate coefficient sequence is , the output of the LSTM network can be expressed as:

[0095] ;

[0096] in, is the hidden state of the long-term trend, i.e., the long-term trend prediction result. At the same time, the high-frequency detail coefficients of the three scales are respectively input into the three parallel medium-term prediction LSTM networks in the multi-scale trend decomposition LSTM model. Each medium-term prediction network consists of two layers of LSTM units, each layer contains 64 neurons, and uses the ReLU activation function. The ReLU activation function can effectively alleviate the gradient vanishing problem and maintain the training effect of the model in a deeper network. The goal of the medium-term prediction LSTM network is to capture the medium-term fluctuation characteristics in the data. Assume The high-frequency detail coefficient is , then the corresponding medium-term forecast output is:

[0097] ;

[0098] The high-frequency detail coefficients of these three scales correspond to three independent medium-term volatility prediction results. While processing the wavelet transform, the target feature set is subjected to empirical mode decomposition (EMD) to extract five intrinsic mode functions (IMFs) and a residual term. EMD is an adaptive signal decomposition method that iteratively decomposes a complex signal into several simple IMFs, each of which represents an independent oscillation mode in the signal. Suppose the target feature set is , the result of EMD decomposition can be expressed as:

[0099] ;

[0100] in, It is The intrinsic mode function, is the residual term. Through decomposition, multi-scale time series components are obtained. The five IMFs are respectively input into the five parallel short-term prediction LSTM networks in the multi-scale trend decomposition LSTM model. Each short-term prediction network contains only one layer of LSTM units, each layer contains 32 neurons, and uses the Sigmoid activation function. The Sigmoid function has good normalization characteristics and is suitable for modeling short-term changes. After each IMF passes through its corresponding short-term prediction network, a short-term pattern prediction result is generated. Assume IMF is IM , and its corresponding short-term prediction output is:

[0101] ;

[0102] The prediction results reflect the short-term changing trends of the data in different frequency components respectively. At the same time, the residual term obtained by EMD decomposition is input into the fully connected neural network in the multi-scale trend decomposition LSTM model. The fully connected network contains three hidden layers, 16 neurons in each layer, and uses the LeakyReLU activation function. LeakyReLU is an improved ReLU function that solves the "dead neuron" problem in ReLU by allowing the negative part of the input to pass with a smaller slope. The output of the fully connected network is the residual prediction result, which represents the changing trend of those signal components that are not captured by the IMF. The attention mechanism is used to weightedly fuse the long-term trend prediction results, three medium-term fluctuation prediction results, five short-term pattern prediction results and residual prediction results. The attention mechanism learns the importance weights of different prediction results, weights and combines each result to generate the final comprehensive prediction result. Suppose the long-term trend prediction result is , the medium-term volatility forecast result is , the short-term model prediction result is , the residual prediction result is , the attention weighted fusion can be expressed as:

[0103] ;

[0104] in, is the attention weight, It is the final comprehensive prediction result.

[0105] In a specific embodiment, the process of executing step S5 may specifically include the following steps:

[0106] Perform residual calculation on the comprehensive prediction results and actual observation data to obtain a residual sequence, and then segment the residual sequence to obtain segmented residual data;

[0107] The Hotelling's T² statistic is calculated on the segmented residual data to obtain the T² statistic sequence, and the Q statistic is calculated on the segmented residual data to obtain the Q statistic sequence;

[0108] The generalized variance statistics are calculated for the segmented residual data, and the determinant of the sample covariance matrix is ​​used to obtain the GV statistic series;

[0109] The Wishart statistic is calculated for the segmented residual data, and the log-likelihood ratio method is used to obtain the Wishart statistic sequence. The KL divergence between the comprehensive prediction results and the actual observed data is calculated to obtain the KL divergence sequence.

[0110] The T² statistic sequence, Q statistic sequence, GV statistic sequence, Wishart statistic sequence and KL divergence sequence are input into the first layer of the multi-level fault detection decision tree, and the C4.5 algorithm is used for preliminary classification to obtain the preliminary fault type;

[0111] Based on the preliminary fault type, combined with multivariate statistics, the second-level decision tree analysis is performed to obtain the detailed fault type;

[0112] A contribution graph analysis is performed on the refined fault type to obtain the main contributing factors of the fault, and the refined fault type and the main contributing factors of the fault are input into the fault diagnosis knowledge graph for analysis to obtain fault diagnosis information.

[0113] Specifically, the comprehensive prediction results are compared with the actual observed data, and the residual between them is calculated. The residual refers to the difference between the predicted value and the actual value, and the formula is expressed as:

[0114] ;

[0115] in, It's at the time The residual of is the actual observed value, is the comprehensive forecast result. By calculating the residual of the entire time series, the residual sequence is obtained. The residual sequence is segmented and the long time series is divided into several shorter intervals. Each interval corresponds to a different time window, which helps to capture the characteristic changes of the residual in different time periods. Assume that the residual sequence is , through the sliding window method, the size of each window is , the step length is , and obtain multiple segmented residual data. For each segment of residual data, perform Hotelling's Statistical calculation. Hotelling's is a multivariate statistic used to detect anomalies in multidimensional data. Assume that the mean vector of the residual data is , the covariance matrix is , then Hotelling's The statistic can be expressed as:

[0116] ;

[0117] in, is the residual vector, is the inverse matrix of the covariance matrix. The value constitutes A sequence of statistics that can be used to identify abnormal patterns in the residual data. The Q statistic is calculated for the same segmented residual data. The Q statistic, also known as the squared prediction error statistic, measures the degree of fit between the residual and the model. For each segment of data, the Q statistic sequence is obtained by calculating the sum of the squares of the residual vector. The Q statistic can effectively detect the overall deviation of the system and identify abnormal patterns that have not been captured in the system. The generalized variance (GV) statistic is calculated for the segmented residual data. The generalized variance uses the determinant of the sample covariance matrix to measure the overall variability of the data set. Let the sample covariance matrix be , generalized variance It can be expressed as

[0118] ;

[0119] in, is the determinant of the sample covariance matrix. The generalized variance can capture the overall trend of change in a multivariate data set. The Wishart statistic is calculated for the piecewise residual data. The Wishart distribution is an important distribution in multivariate statistics, which is used to estimate the covariance matrix of the multidimensional normal distribution. Using the log-likelihood ratio method, the Wishart statistic sequence is obtained, which is used to evaluate the degree of deviation of the residual data from the multivariate normal distribution. The difference between the prediction and the actual data is identified by calculating the KullbackLeibler (KL) divergence between the comprehensive prediction results and the actual observed data. The KL divergence is an asymmetric indicator that measures the difference between two probability distributions. Suppose the predicted probability distribution is , the actual probability distribution is , then the KL divergence can be expressed as:

[0120] ;

[0121] The larger the KL divergence, the more significant the difference between the two distributions. This indicator helps capture significant deviations between the model and the actual situation. The statistic sequence, Q statistic sequence, GV statistic sequence, Wishart statistic sequence and KL divergence sequence are input into the first layer of the multi-level fault detection decision tree. The first layer of the decision tree model uses the C4.5 algorithm for preliminary classification. C4.5 is a classic classification algorithm that selects the optimal features for splitting through the information gain ratio to achieve the best classification effect. The results of the preliminary classification provide the preliminary types of potential faults in the system. Based on the preliminary fault types, the second-level decision tree analysis is combined with multivariate statistics to obtain more refined fault types. Through further analysis, more specific fault modes in the system are distinguished, thereby improving the accuracy of fault diagnosis. Contribution graph analysis is performed on the refined fault types to identify which specific statistics or features play a key role in the occurrence of the fault. By analyzing the main contributing factors, the root cause of the fault is better understood, and a basis for fault prevention and handling is provided. The refined fault types and the main contributing factors of the fault are input into the fault diagnosis knowledge graph for analysis. The knowledge graph is a semantic network that associates fault types, contributing factors and existing fault knowledge to generate a structured knowledge base.

[0122] In a specific embodiment, the process of executing step S6 may specifically include the following steps:

[0123] Based on the comprehensive prediction results and fault diagnosis information, RDF triples are used to represent the relationship between entities, and OWL is used to define the ontology model to obtain the knowledge graph of the micro-module computer room.

[0124] Apply graph embedding algorithm to the knowledge graph of micro-module computer room, map entities and relationships in the knowledge graph of micro-module computer room to low-dimensional vector space, and obtain graph embedding representation;

[0125] Based on the graph embedding representation, a deep Q network is constructed, in which the state space includes the current environmental parameters, equipment status and predicted future status, and the action space includes adjusting the parameters of the refrigeration equipment, adjusting the air flow distribution and load balancing operations, and the initial DQN model is obtained;

[0126] The initial DQN model is trained, the ε-greedy strategy is used for exploration, the experience replay buffer size is set to 10000, and the target network is updated every 50 time steps to obtain the trained DQN model;

[0127] The trained DQN model is combined with the MCTS algorithm to build a reinforcement learning framework, where the MCTS algorithm is used for action selection and the DQN model is used for state evaluation. After 500 iterations, the optimized policy network is obtained.

[0128] Based on the optimized strategy network, combined with the comprehensive prediction results and fault diagnosis information of the micro-module computer room, an adaptive control strategy is generated. The adaptive control strategy includes refrigeration equipment parameter adjustment instructions, airflow distribution optimization scheme and load balancing strategy.

[0129] Specifically, the knowledge graph of the micro-module computer room is constructed by defining the ontology model through RDF triples and OWL. RDF (Resource Description Framework) triples are used to represent the relationship between entities in the form of "entity-attribute-value" or "subject-predicate-object". RDF triples are used to describe various devices, states and relationships in the micro-module computer room. In order to better organize and reason about this information, OWL (Web Ontology Language) is used to define the ontology model. The ontology model provides a semantic structure for the knowledge graph, making the relationship between different entities clearer. For example, "equipment" and "state" can be defined as classes of the ontology, and the inheritance relationship, attributes and constraints between these classes can be described. In this way, a comprehensive and structured knowledge graph of the micro-module computer room is generated. A graph embedding algorithm is applied to the knowledge graph of the micro-module computer room to map the entities and relationships in the graph to a low-dimensional vector space. Graph embedding represents complex graph structure information in an easy-to-operate vector form, so that downstream tasks such as machine learning models can directly process these vectors. Assume that the entities in the knowledge graph are , the relationship is , through the graph embedding algorithm, the entity and relationship Mapped to vector and . The vectorized representation retains the semantic information in the graph structure and significantly reduces the complexity of the data. Based on the graph embedding representation, a deep Q network (DQN) model is constructed. DQN is a reinforcement learning algorithm suitable for dealing with complex decision-making problems with continuous states and discrete actions. In this model, the state space contains current environmental parameters, device states, and predicted future states. The state space can be represented by a vector, which contains multi-dimensional features such as temperature, humidity, device working status, etc. Assume that the state is The action space contains operations that can adjust the computer room environment, such as adjusting cooling equipment parameters, adjusting air flow distribution, and performing load balancing. The action space can be used Indicates that the goal of the DQN model is to learn The function is used to evaluate the value of each state-action pair and select the optimal action. In order to enable the DQN model to learn effectively, the model training is carried out using - Greedy strategy for exploration. The ε-greedy strategy combines exploration and exploitation, allowing the model to explore different actions more in the early stages of training to avoid falling into local optimality, and gradually increasing the exploitation of optimal actions as training progresses. During training, the size of the experience replay buffer is set to 10,000, and the model can sample data from the past 10,000 experiences to update the weights, which helps break the correlation of data and improve training stability. At the same time, the target network is updated every 50 time steps, and the stability of training is ensured and the model oscillation is prevented by fixing the target value for a certain period of time. After this training process, a DQN model that can well estimate the state-action value is obtained. In order to improve the optimization effect of the strategy, the trained DQN model is combined with the Monte Carlo Tree Search (MCTS) algorithm to construct a reinforcement learning framework. MCTS is a decision tree search algorithm based on random sampling, which gradually builds a search tree in a complex search space, thereby effectively evaluating the long-term value of different actions. In this framework, MCTS is used for action selection, while the DQN model is used for state evaluation. By combining the advantages of both, the optimal strategy can be explored more effectively in a large-scale state space. MCTS continuously selects and expands the current optimal subtree by simulating multiple game processes, and uses the valuation function of DQN to evaluate the state of leaf nodes, and updates the search tree in reverse according to the results. This combination allows the algorithm to gradually converge to an optimized policy network in 500 iterations. Based on the optimized policy network, the adaptive control strategy is generated in combination with the comprehensive prediction results and fault diagnosis information of the micro-module computer room. The adaptive control strategy ensures that the computer room always maintains the best state under different environmental conditions by dynamically adjusting the parameters of the refrigeration equipment, optimizing the airflow distribution, and performing load balancing. For example, if it is predicted that the temperature will rise in the future, the working parameters of the refrigeration equipment are adjusted in advance to prevent the equipment from malfunctioning due to excessive temperature. If the fault diagnosis information shows that the airflow distribution in a certain area is uneven, the cooling effect is improved by optimizing the airflow control strategy. The load balancing strategy is used to evenly distribute computing tasks among multiple devices to prevent individual devices from being overloaded. These adjustments are based on the joint evaluation results of DQN and MCTS to ensure the accuracy and real-time nature of the decision.

[0130] The above describes the data transmission method based on the Internet of Things in the embodiment of the present invention. The following describes the data transmission device based on the Internet of Things in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a data transmission device based on the Internet of Things includes:

[0131] The collection module is used to collect the original data packets of the micromodule room through multiple IoT sensors, and generate a device authentication code for each IoT sensor, add it to the original data packet, and obtain an authentication data packet;

[0132] The authentication module is used to perform identity authentication based on the device authentication code, generate a session key, and encrypt the authentication data packet to obtain a security data packet, and at the same time, transmit the security data packet to the cloud computing platform;

[0133] An extraction module, used for data decryption and feature extraction of security data packets, target feature sets in the cloud computing platform;

[0134] The prediction module is used to input the target feature set into the multi-scale trend decomposition long short-term memory model to predict the operation status of the micro-module computer room and obtain a comprehensive prediction result;

[0135] A diagnosis module is used to calculate multivariate statistics based on comprehensive prediction results and actual observation data, and input the multivariate statistics into a multi-level fault detection decision tree for fault diagnosis to obtain fault diagnosis information;

[0136] The generation module is used to build a knowledge graph of the micro-module computer room and perform reinforcement learning optimization to generate an adaptive control strategy based on the comprehensive prediction results and fault diagnosis information.

[0137] Through the collaboration of the above components, the physical unclonable function and multiple encryption mechanisms enhance the security of device authentication and data encryption; the use of Reed-Solomon error correction coding and MQTT protocol improves the reliability and real-time performance of data transmission. By using technologies such as multi-scale decomposition and principal component analysis, efficient processing of complex data and extraction of key features are achieved, laying the foundation for subsequent analysis. The multi-scale trend decomposition long-short-term memory model, combined with discrete wavelet transform and empirical mode decomposition, realizes the multi-scale and multi-dimensional prediction of the operating status of the micro-module computer room. By combining multivariate statistics and multi-level fault detection decision trees, the sensitivity and accuracy of fault detection are improved, and accurate diagnosis of fault types and causes is achieved. By constructing a knowledge graph of the micro-module computer room and combining it with reinforcement learning, the generation of adaptive control strategies based on historical experience and real-time data is realized, which improves the operating efficiency and reliability of the micro-module computer room.

[0138] The present invention also provides a computer device, which includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the data transmission method based on the Internet of Things in the above-mentioned embodiments.

[0139] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the data transmission method based on the Internet of Things.

[0140] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0141] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0142] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A data transmission method based on the Internet of Things, characterized in that: The method comprises: Collect the original data packets of the micromodule computer room through multiple IoT sensors, generate a device authentication code for each IoT sensor, add it to the original data packet, and obtain an authentication data packet; Performing identity authentication based on the device authentication code, generating a session key, and encrypting the authentication data packet to obtain a security data packet, and at the same time, transmitting the security data packet to the cloud computing platform; In the cloud computing platform, data decryption and feature extraction are performed on the security data packet, and a target feature set is obtained; The target feature set is input into the multi-scale trend decomposition long short-term memory model to predict the operation status of the micro-module computer room to obtain a comprehensive prediction result; specifically including: performing discrete wavelet transform decomposition on the target feature set to obtain low-frequency approximate coefficients and high-frequency detail coefficients of three scales; the low-frequency approximate coefficients are input into the long-term prediction LSTM network in the multi-scale trend decomposition long short-term memory model, the long-term prediction LSTM network includes three LSTM layers, each layer has 128 neurons, and a tanh activation function is used to obtain a long-term trend prediction result; the high-frequency detail coefficients of the three scales are respectively input into three parallel medium-term prediction LSTM networks in the multi-scale trend decomposition long short-term memory model, each network includes two LSTM layers, each layer has 64 neurons, and a ReLU activation function is used to obtain medium-term fluctuation prediction results of three scales. ; Perform empirical mode decomposition on the target feature set, extract 5 intrinsic mode functions and 1 residual term, and obtain multi-scale time series components; input the 5 intrinsic mode functions into 5 parallel short-term prediction LSTM networks in the multi-scale trend decomposition long short-term memory model, each network contains 1 LSTM layer, 32 neurons, and uses Sigmoid activation function to obtain 5 short-term mode prediction results; input the residual term into the fully connected neural network in the multi-scale trend decomposition long short-term memory model, which contains 3 hidden layers, 16 neurons in each layer, and uses LeakyReLU activation function to obtain residual prediction results; use the attention mechanism to perform weighted fusion on the long-term trend prediction results, 3 medium-term fluctuation prediction results, 5 short-term mode prediction results and residual prediction results to obtain a comprehensive prediction result; Based on the comprehensive prediction results and actual observation data, multivariate statistics are calculated, and the multivariate statistics are input into a multi-level fault detection decision tree for fault diagnosis to obtain fault diagnosis information; specifically including: performing residual calculation on the comprehensive prediction results and actual observation data to obtain a residual sequence, and segmenting the residual sequence to obtain segmented residual data; performing Hotelling's T² statistic calculation on the segmented residual data to obtain a T² statistic sequence, and performing Q statistic calculation on the segmented residual data to obtain a Q statistic sequence; performing generalized variance statistic calculation on the segmented residual data, using the determinant of the sample covariance matrix to obtain a GV statistic sequence; performing Wishart statistic calculation on the segmented residual data, using the log-likelihood ratio method to obtain a Wishart statistic sequence, and calculating the KL divergence of the comprehensive prediction results and the actual observation data to obtain a KL divergence sequence; and integrating the T² statistic sequence and the Q statistic sequence. The quantity sequence, the GV statistic sequence, the Wishart statistic sequence and the KL divergence sequence are input into the first layer of a multi-level fault detection decision tree, and a C4.5 algorithm is used for preliminary classification to obtain a preliminary fault type; based on the preliminary fault type, combined with the multivariate statistics, a second-level decision tree analysis is performed to obtain a refined fault type; a contribution graph analysis is performed on the refined fault type to obtain the main contributing factors of the fault, and the refined fault type and the main contributing factors of the fault are input into the fault diagnosis knowledge graph for analysis to obtain fault diagnosis information; According to the comprehensive prediction results and the fault diagnosis information, a knowledge graph of the micro-module computer room is constructed and reinforcement learning optimization is performed to generate an adaptive control strategy.

2. The data transmission method based on the Internet of Things according to claim 1, characterized in that: The original data packet of the micromodule computer room is collected by multiple IoT sensors, and a device authentication code of each IoT sensor is generated and added to the original data packet to obtain an authentication data packet, including: Collect data from the temperature sensor, humidity sensor, power consumption sensor, airflow sensor, and equipment status sensor in the micromodule room to obtain the original data packet containing the sensor ID, data type, data value, and timestamp; Using the physical unclonable function module of each IoT sensor, a response operation is performed on the random challenge value to obtain a unique device fingerprint, and based on the device fingerprint, a fixed-length device authentication code is generated; Adding the device authentication code to the corresponding original data packet to form a first data packet, performing data formatting processing on the first data packet, unifying different types of sensor data into a JSON format, and obtaining a second data packet; Segment the data in the second data packet, where each window contains 100 data points and the sliding step is 10 data points, to obtain initial segmented data; The initial segmented data is subjected to wavelet transformation to obtain denoised segmented data, and the denoised segmented data is repackaged with a corresponding device authentication code to obtain an authentication data packet, and the authentication data packet is stored in a local cache to cope with network interruption.

3. The data transmission method based on the Internet of Things according to claim 1, characterized in that: The identity authentication is performed based on the device authentication code, a session key is generated, and the authentication data packet is encrypted to obtain a security data packet, and at the same time, the security data packet is transmitted to the cloud computing platform, including: Performing elliptic curve digital signature algorithm verification on the device authentication code to obtain an identity authentication result, and based on the identity authentication result, using the elliptic curve Diffie-Hellman key exchange protocol to generate a shared key to obtain a 256-bit shared key; Applying the HKDF key derivation function to the shared key, using SHA-256 as the hash function, to obtain a session key, and using the session key to encrypt the authentication data packet, setting a 12-byte random initialization vector, to obtain encrypted data; Calculate the HMAC-SHA256 message authentication code of the encrypted data, use the derived key of the session key as the HMAC key, and obtain a data integrity check value; The encrypted data, the initialization vector and the data integrity check value are packaged to construct a secure data packet, and the secure data packet is processed in blocks, with each block size set to 1024 bytes, using the Reed-Solomon error correction coding algorithm, with the parameters set to RS (255, 223), to obtain a coded data packet; The MQTT protocol is adopted, the QoS level is set to 1, and the encoded data packets are transmitted in batches to a pre-configured cloud computing platform address. During the transmission process, the TLS 1.3 encrypted channel is used to ensure the transmission security.

4. The data transmission method based on the Internet of Things according to claim 3 is characterized in that: In the cloud computing platform, the security data packet is decrypted and features are extracted, and the target feature set includes: Perform Reed-Solomon decoding on the received coded data packet and use RS (255, 223) parameters for error correction to obtain the original secure data packet; Deserialize the original secure data packet using Protocol Buffers, extract the encrypted data, the initialization vector, and the data integrity check value, and obtain the deconstructed data components; Based on the pre-shared key material, the HKDF key derivation process is re-executed to generate a session key consistent with the sender, and the key required for decryption is obtained; Using the decryption key and the initialization vector, perform AES-256-GCM decryption on the encrypted data to obtain an original authentication data packet; Parsing the original authentication data packet in JSON format to extract various sensor data, including temperature, humidity, power consumption, airflow, and device status information, to obtain a structured data set; Decomposing the time series of the structured data set into trend, seasonality and residual components to obtain multi-dimensional decomposition data, and calculating statistical features, time domain features and frequency domain features based on the multi-dimensional decomposition data to obtain a preliminary feature set; The principal component analysis is performed on the preliminary feature set, and the principal components with a cumulative contribution rate of 95% are selected to obtain the target feature set.

5. The data transmission method based on the Internet of Things according to claim 1, characterized in that: According to the comprehensive prediction results and the fault diagnosis information, a knowledge graph of a micro-module computer room is constructed and reinforcement learning optimization is performed to generate an adaptive control strategy, including: Based on the comprehensive prediction results and the fault diagnosis information, RDF triples are used to represent the relationship between entities, and OWL is used to define the ontology model to obtain a knowledge graph of the micro-module computer room; Applying a graph embedding algorithm to the micro-module computer room knowledge graph, mapping entities and relationships in the micro-module computer room knowledge graph to a low-dimensional vector space, and obtaining a graph embedding representation; Based on the graph embedding representation, a deep Q network is constructed, wherein the state space includes current environmental parameters, equipment status and predicted future status, and the action space includes adjusting refrigeration equipment parameters, adjusting airflow distribution and load balancing operations, thereby obtaining an initial DQN model; The initial DQN model is trained, an ε-greedy strategy is used for exploration, the experience replay buffer size is set to 10000, and the target network is updated every 50 time steps to obtain a trained DQN model; The trained DQN model is combined with the MCTS algorithm to construct a reinforcement learning framework, wherein the MCTS algorithm is used for action selection and the DQN model is used for state evaluation, and the optimized policy network is obtained after 500 iterations; Based on the optimized strategy network, combined with the comprehensive prediction results and the fault diagnosis information of the micro-module computer room, an adaptive control strategy is generated, and the adaptive control strategy includes refrigeration equipment parameter adjustment instructions, airflow distribution optimization scheme and load balancing strategy.

6. A data transmission device based on the Internet of Things, characterized in that: Used to execute the data transmission method based on the Internet of Things as described in any one of claims 1 to 5, the device comprises: The acquisition module is used to collect the original data packets of the micromodule room through multiple IoT sensors, and generate a device authentication code for each IoT sensor, add it to the original data packet, and obtain an authentication data packet; An authentication module, used to perform identity authentication based on the device authentication code, generate a session key, and encrypt the authentication data packet to obtain a security data packet, and at the same time, transmit the security data packet to the cloud computing platform; An extraction module, used for performing data decryption and feature extraction on the security data packet, a target feature set, in the cloud computing platform; A prediction module, used for inputting the target feature set into a multi-scale trend decomposition long short-term memory model to predict the operation status of the micro-module computer room and obtain a comprehensive prediction result; A diagnosis module, used to calculate multivariate statistics based on the comprehensive prediction results and actual observation data, and input the multivariate statistics into a multi-level fault detection decision tree to perform fault diagnosis and obtain fault diagnosis information; A generation module is used to construct a knowledge graph of a micro-module computer room and perform reinforcement learning optimization to generate an adaptive control strategy based on the comprehensive prediction results and the fault diagnosis information.

7. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and is characterized in that when the processor executes the computer program, the data transmission method based on the Internet of Things described in any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor is enabled to execute the data transmission method based on the Internet of Things as claimed in any one of claims 1 to 5.