Data center energy saving control method, system and storage medium

By integrating quantum computing, GAN and multimodal perception technologies, real-time analysis and optimization of data center energy management, the existing systems are solved inadequate energy consumption prediction and responsiveness, and more efficient energy management and cost reduction are achieved.

CN118819269BActive Publication Date: 2024-11-29GUANGZHOU CHUANGBO MECH & ELECTRICAL EQUIP INSTALLATION
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411312379.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-20
Publication Date
2024-11-29
Estimated Expiration
2044-09-20

AI Technical Summary

Technical Problem

The existing data center energy management system lacks in-depth analysis and flexible response capabilities to complex data, and cannot effectively predict future energy consumption demand and environmental changes, resulting in limited energy efficiency improvements and failure to use artificial intelligence technology to optimize energy management efficiency.

Method used

Integrate quantum computing optimization algorithms, generative adversarial networks (GANs) and multimodal perception and adaptive control technologies to collect data through multimodal sensors, monitor and analyze in real time, use machine learning and deep learning models to predict and optimize energy configurations, Combined with quantum encryption technology to ensure data security.

Benefits of technology

It realizes accurate analysis and prediction of energy demand in data centers, dynamically adjusts energy allocation, improves energy efficiency and reduces operating costs, solves the shortcomings of existing technology in energy consumption prediction and adaptive response, and improves the self-optimization of the system. ability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118819269B_ABST
    Figure CN118819269B_ABST
Patent Text Reader

Abstract

The present invention proposes a data center energy-saving control method, system and storage medium, the method includes: arranging sensors to collect multimodal data and preprocessing; integrating multimodal data, building an intermediate monitoring platform to monitor and analyze data in real time; designing a generator and discriminator architecture to generate fault training data and fault test data; applying machine learning algorithms to conduct in-depth analysis of multimodal data; using quantum computing algorithms to process abnormal data of abnormal patterns; building a data integration framework and designing an adaptive control algorithm based on model predictive control to automatically adjust network configuration; building a prediction model based on deep learning, using the latest collected data to train and optimize the prediction model; using quantum encryption technology to encrypt the communication and data of the above steps to prevent data leakage. The present invention not only improves the data security and privacy protection capabilities of the communication network, but also enhances the network's adaptability and response speed to new threats.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of data center energy saving, and in particular, relates to a data center energy saving control method, system and storage medium. Background Art

[0002] According to statistics, the proportion of energy consumption in data centers to the global total energy consumption has been increasing year by year. High energy consumption not only increases operating costs, but also has a significant impact on the environment. Existing data center energy management systems mainly rely on traditional centralized control methods and simple feedback adjustment mechanisms. These systems often perform energy consumption optimization based on preset operating modes and rules, and lack the ability to conduct in-depth analysis and flexible response to real-time data. For example, traditional systems may only adjust air conditioning equipment based on temperature and power usage, but fail to fully consider multiple factors such as server operating status and external environmental changes. In addition, these systems usually do not have learning and adaptive capabilities, and cannot automatically optimize strategies based on historical data and operating results, resulting in limited energy efficiency improvements.

[0003] Currently, although some methods such as PID control and simple machine learning algorithms are used to try to improve the energy efficiency of data centers, these methods still have shortcomings in processing complex data, predicting future energy consumption, and automatically adjusting operation strategies. For example, these systems often cannot effectively predict future energy demand and environmental changes, and therefore cannot achieve predictive maintenance or dynamic energy allocation. In addition, most existing technologies fail to effectively utilize the latest artificial intelligence technologies, such as deep learning and reinforcement learning, which have the potential to significantly optimize the energy management efficiency of data centers.

[0004] Therefore, there is an urgent need to develop a new data center energy management system that can analyze and predict the energy needs of the data center in real time and accurately, and dynamically adjust energy allocation based on complex data inputs to achieve higher energy efficiency and lower operating costs. Summary of the invention

[0005] The purpose of the present invention is to design a data center energy-saving control method, system and storage medium, and to manage the energy efficiency of a data center by integrating quantum computing optimization algorithm, generative adversarial network (GAN) and multimodal perception and adaptive control technology.

[0006] In order to achieve the above object, a first aspect of the present invention provides a data center energy saving control method, comprising the following steps:

[0007] Arranging multimodal sensors to collect multimodal data and preprocessing the multimodal data; wherein the multimodal sensors include temperature sensors, optical fiber vibration sensors, flow monitors and signal strength sensors;

[0008] Integrate multimodal data and build an intermediate monitoring platform to monitor and analyze multimodal data in real time;

[0009] Design the generator and discriminator architecture to simulate normal and various fault conditions, and generate fault training data and fault test data;

[0010] Apply machine learning algorithms to conduct in-depth analysis of multimodal data and identify abnormal patterns;

[0011] Use quantum computing algorithms to process abnormal data with abnormal patterns and optimize fault handling decisions;

[0012] Build a data integration framework and design an adaptive control algorithm based on model predictive control to automatically adjust the network configuration;

[0013] Build a deep learning-based prediction model and continuously train and optimize the prediction model using the latest collected data;

[0014] Use quantum encryption technology to encrypt communications and data to prevent data leakage;

[0015] Among them, in the process of using quantum computing algorithm to process abnormal data of abnormal mode, in optimizing fault handling decision, the optimized fault handling decision is converted into Hamiltonian expression, and the ground state with the lowest energy is calculated, which is expressed as follows:

[0016] ;

[0017] in, represents the edge set in the network, It is the edge The weight of Is a node status, is the state of node j.

[0018] Furthermore, the preprocessing includes: performing sliding average filtering on the data output by each sensor to perform calculation, which is expressed as follows:

[0019] ;

[0020] in, is the half width of the sliding window, Represents the data output at time t after sliding average filtering.

[0021] Furthermore, the integration of multimodal data and construction of an intermediate monitoring platform to monitor and analyze the multimodal data in real time specifically includes the following steps:

[0022] The data received by the multimodal sensor is synchronized with the data time, as shown below:

[0023] ;

[0024] in, Represents the original timestamp, Represents a reference timestamp, Indicates the correction value;

[0025] A dynamic data stream fusion algorithm is designed to integrate multimodal data. The data fusion calculation is expressed as follows:

[0026] ;

[0027] in, represents fused data, Indicates Sensor data, represents the weight assigned according to the importance and reliability of the data, Indicates the number of data categories;

[0028] A real-time data analysis mechanism is designed to instantly identify abnormal patterns and potential faults from the fused data. At the same time, the adaptive threshold technology is used to analyze the fused data in real time and dynamically adjust the threshold of fault detection based on historical data. The formula for fault detection is as follows:

[0029] ;

[0030] in, represents the current data point, and represent the moving average and standard deviation respectively, represents the sensitivity coefficient of adaptive adjustment, Indicates the fault threshold.

[0031] Furthermore, the application of machine learning algorithms to conduct in-depth analysis of multimodal data and identify abnormal patterns specifically includes the following steps:

[0032] Normalize the input data;

[0033] Use principal component analysis to reduce the dimension of the normalized data and extract features, as shown below:

[0034] ;

[0035] in, is the normalized data matrix, is the weight matrix composed of eigenvectors, is the feature matrix after dimensionality reduction;

[0036] The dynamic time warping algorithm is used to perform pattern matching on the extracted time series features to identify abnormal behaviors or failure modes in the data, as shown below:

[0037] ;

[0038] in, and are two time series, and These are and Elements of a time series

[0039] The identified anomalies or failure modes are fed back in real time, and a linear regression model is used to predict the development trend of network failures.

[0040] Furthermore, the design generator and discriminator architecture simulates normal and various fault conditions, including:

[0041] Design Builder, including:

[0042] Input 100-dimensional random noise vector , build a multi-layer fully connected network, use the ReLU activation function, and the last layer uses the tanh activation function to output simulated fault data with the same dimension as the real network fault data;

[0043] Design the discriminator, including:

[0044] Input simulated fault data of the same dimension as the output of the generator, build a multi-layer fully connected network, use the LeakyReLU activation function, and output a single scalar;

[0045] The loss function is designed to train the generator and the discriminator alternately, and the Adam optimizer is used; wherein the loss function of the generator is It is expressed as follows:

[0046] ;

[0047] in, represents a random noise vector, is with The corresponding conditional variable is expressed as the network state or the preset failure mode, m represents the number of samples, represents the discriminator, G represents the generator;

[0048] The loss function of the discriminator It is expressed as follows:

[0049] ;

[0050] in, is the classification loss, is a hyperparameter that controls the impact of classification loss, express Elements of a time series.

[0051] Furthermore, the adaptive control algorithm based on model predictive control is expressed as follows:

[0052] ;

[0053] in, It's in time The network state vector, is the control input, and is the parameter matrix of the system dynamics, is the process noise.

[0054] Furthermore, the prediction model based on deep learning is constructed, and the stochastic gradient descent method with adaptive learning rate is used to continuously train the model, and the optimization function is:

[0055] ;

[0056] in, are model parameters, is the time step The learning rate, is the loss function.

[0057] Furthermore, quantum encryption data is used to encrypt communications and data, including:

[0058] An adaptive encryption mechanism is designed based on quantum encryption technology. The encryption process uses the following form of adaptive adjustment algorithm:

[0059] ;

[0060] in, Represents the new key, represents the old key, represents the quantum link quality indicator, Indicates the time interval, Represents a key update function.

[0061] In a second aspect of the present invention, a data center energy-saving control system is provided, the system comprising:

[0062] A multimodal data collection module, used to arrange multimodal sensors to collect multimodal data and pre-process the multimodal data; wherein the multimodal sensors include temperature sensors, optical fiber vibration sensors, flow monitors and signal strength sensors;

[0063] Multimodal data integration module, used to integrate multimodal data and build an intermediate monitoring platform to monitor and analyze multimodal data in real time;

[0064] GAN simulation module, used to design the generator and discriminator architecture to simulate normal and various fault conditions, and generate fault training data and fault test data;

[0065] The abnormal pattern recognition module is used to apply machine learning algorithms to conduct in-depth analysis of multimodal data and identify abnormal patterns;

[0066] Optimize the fault handling decision module, which is used to process abnormal data in abnormal patterns using quantum computing algorithms and optimize fault handling decisions;

[0067] The network configuration automatic adjustment module is used to build a data integration framework and design an adaptive control algorithm based on model predictive control to automatically adjust the network configuration;

[0068] The prediction model optimization module is used to build a prediction model based on deep learning and continuously train and optimize the prediction model using the latest collected data;

[0069] Quantum encryption module, used to encrypt communications and data using quantum encryption technology to prevent data leakage;

[0070] Among them, in the process of using quantum computing algorithm to process abnormal data of abnormal mode, in optimizing fault handling decision, the optimized fault handling decision is converted into Hamiltonian expression, and the ground state with the lowest energy is calculated, which is expressed as follows:

[0071] ;

[0072] in, represents the edge set in the network, It is the edge The weight of Is a node status, is the state of node j.

[0073] The third aspect of the present invention further provides a storage medium storing non-transitory computer-readable instructions, which, when executed by a computer, can execute the steps in the aforementioned data center energy-saving control method.

[0074] The beneficial technical effects of the present invention are at least as follows:

[0075] (1) The present invention deploys multiple sensors in the data center to collect multiple data such as temperature, humidity, power consumption, and server operating status in real time. Using these multimodal data, the system can comprehensively monitor the operating environment and equipment status of the data center, thereby accurately grasping energy demand and potential energy saving points. This solves the problem of single data processing and inflexible response in the prior art.

[0076] The present invention uses GAN technology to simulate various normal and abnormal energy consumption scenarios to improve the accuracy of prediction and the adaptability of the system. The introduction of GAN enables the system to adjust strategies in advance without the actual occurrence of energy consumption peaks or equipment failures, effectively preventing potential energy consumption problems, which directly addresses the deficiencies of existing technologies in energy consumption prediction and dynamic response.

[0077] The present invention uses quantum computing to optimize energy allocation and scheduling, and uses its super computing power to handle complex optimization problems, achieving fast and efficient energy management. This provides a new solution to the limitations of traditional methods in dealing with the dynamic optimization of complex systems.

[0078] The present invention combines adaptive control technology with a continuous machine learning process to continuously optimize the energy management strategy based on actual operating data. This enables the system to self-optimize and continuously improve energy efficiency, fundamentally solving the problem of the lack of self-learning and optimization capabilities in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0080] Figure 1 This is a flow chart of the data center energy saving control method of the present invention.

[0081] Figure 2 This is a framework diagram of the data center energy-saving control system of the present invention. DETAILED DESCRIPTION

[0082] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0083] In one or more embodiments, Figure 1As shown, the data center energy saving control method of the present invention is disclosed, and the method includes the following steps S1-S8:

[0084] S1. Arrange multimodal sensors to collect multimodal data and preprocess the multimodal data; wherein the multimodal sensors include temperature sensors, optical fiber vibration sensors, flow monitors and signal strength sensors.

[0085] Specifically, the temperature sensor ( Use high-precision digital temperature sensor with a measurement range of to , accuracy . Fiber Optic Vibration Sensor ( The phase-sensitive optical time domain reflectometry technology can detect tiny vibrations along the line up to 40 kilometers. Using full-duplex high-precision traffic monitoring technology, it can analyze traffic size, packet type and flow direction in real time. Signal strength sensor Using high-sensitivity RF detectors, the measurement range can reach to , providing real-time data on signal attenuation and interference.

[0086] As an embodiment of the present invention, the installation location of the sensor is to install optical fiber vibration sensors along the main optical fiber channel, and install a node at a certain distance to build full-path monitoring. Further determine the hot spots of key network nodes such as routers, switches, core servers and other equipment to install temperature sensors. Install flow monitors at the main entrances and exits of the data center to monitor the flow in and out of the data center. Install signal strength sensors in key areas covered by the wireless network, especially areas where the signal may be interfered.

[0087] The data acquisition of the present invention is that all sensors are set to collect data once per second, that is, the acquisition frequency is At the same time, each sensor data is converted into a digital signal through an analog-to-digital converter (ADC) for further processing and analysis.

[0088] Specifically, the preprocessing includes: performing sliding average filtering on the data output by each sensor to reduce the influence of instantaneous noise, which is expressed as follows:

[0089] ;

[0090] in, is the half width of the sliding window, Represents the data output at time t after sliding average filtering, usually taken as .

[0091] Furthermore, an anomaly detection algorithm is used to preliminarily screen any abnormal or mutation data points that appear. For example, using a statistical threshold-based method, if a data point exceeds 3 standard deviations of the mean, it is marked as an anomaly.

[0092] Finally, the data is packaged and transmitted, with each data point accompanied by a timestamp and sensor ID, formatted as a JSON object, as follows:

[0093] "timestamp": , "sensor_id": ID, "value": ;

[0094] The TLS / SSL encryption protocol is used to ensure the security of data during transmission and prevent data from being stolen or tampered with. The MQTT protocol is used for data transmission. This protocol is suitable for IoT devices, has low power consumption and maintains a persistent connection state, and is suitable for real-time transmission of a large number of data points.

[0095] S2. Integrate multimodal data and build an intermediate monitoring platform to monitor and analyze multimodal data in real time, which specifically includes the following steps:

[0096] To ensure the timeliness and consistency of the data, the data collected by all sensors are first time-synchronized. Timestamp alignment technology is applied to ensure the alignment of various types of data in the time series, which is convenient for subsequent data analysis and processing. The data received by the multimodal sensor is time-synchronized, as shown below:

[0097] ;

[0098] in, is the original timestamp, is the reference timestamp (usually taken as zero o'clock on the day), is the correction value (network delay and processing time);

[0099] Design dynamic data stream fusion algorithms to integrate multimodal data, integrate data streams from different sensors in real time, use data buffers to manage data streams at different rates, and dynamically adjust algorithm parameters to adapt to real-time changes in data streams.

[0100] The data fusion calculation is expressed as follows:

[0101] ;

[0102] in, represents fused data, Indicates Sensor data, represents the weight assigned according to the importance and reliability of the data, Indicates the number of data categories;

[0103] A real-time data analysis mechanism is designed to instantly identify abnormal patterns and potential faults from the fused data. At the same time, the adaptive threshold technology is used to analyze the fused data in real time and dynamically adjust the threshold of fault detection based on historical data. The formula for fault detection is as follows:

[0104] ;

[0105] in, represents the current data point, and represent the moving average and standard deviation respectively, represents the sensitivity coefficient of adaptive adjustment, Indicates the fault threshold.

[0106] As a preferred embodiment of the present invention, an intuitive user interface is constructed to display network status and potential faults in real time. The interface includes multiple view modes, such as map view, table view and graph view, which support users to select different data display modes according to their needs. A responsive dashboard is designed to highlight key information and warnings according to the importance and urgency of the data.

[0107] S3. Design the generator and discriminator architecture to simulate normal and various fault conditions and generate fault training data and fault test data.

[0108] Specifically, the design generator and discriminator architecture simulates normal and various fault conditions, including:

[0109] Design Builder, including:

[0110] Input 100-dimensional random noise vector , sampled from Gaussian distribution; construct a multi-layer fully connected network, use ReLU activation function, and use tanh activation function in the last layer to output simulated fault data with the same dimension as the real network fault data;

[0111] Design the discriminator, including:

[0112] Input simulated fault data of the same dimension as the output of the generator, build a multi-layer fully connected network, and use the LeakyReLU activation function to output a single scalar, indicating the probability that the input data is real data;

[0113] The loss function is designed to train the generator and the discriminator alternately, and the Adam optimizer is used; wherein the loss function of the generator is It is expressed as follows:

[0114] ;

[0115] in, represents a random noise vector, is with The corresponding conditional variable is expressed as the network state or the preset failure mode, m represents the number of samples, represents the discriminator, G represents the generator;

[0116] To improve the accuracy and robustness of the discriminator, an additional classification loss is added to ensure that the discriminator can not only judge the authenticity of the input data, but also identify the network state or fault type corresponding to the data. This can be achieved by adding an auxiliary classifier To achieve this, the classifier shares the feature extraction layer with the discriminator, and the loss function of the discriminator It is expressed as follows:

[0117] ;

[0118] in, is the classification loss, is a hyperparameter that controls the impact of classification loss, express Elements of a time series.

[0119] Specifically, use the Adam optimizer with the following parameters: learning rate First-order momentum decay factor .

[0120] As an embodiment of the present invention, after the GAN training is stable, the generator (G) is used to independently generate new fault data.

[0121] In terms of data usage, the generated fault data is used to train and refine the existing fault detection model to improve its ability to identify unknown fault types. Simulated fault data is regularly injected into the network system to test and evaluate the network's response strategy and recovery capabilities.

[0122] Furthermore, the generated fault data is integrated with real-time monitoring data to provide a more comprehensive view for network management. The training process of the generator is continuously adjusted based on the feedback from network administrators and system performance to make it simulate actual fault conditions more accurately.

[0123] S4. Apply machine learning algorithms to conduct in-depth analysis of multimodal data and identify abnormal patterns, including the following steps:

[0124] The input data is normalized; the input data includes multi-source data such as temperature, vibration, flow and signal strength; the normalization formula is:

[0125] ;

[0126] in, is the original data, and are the mean and standard deviation of the data type, respectively.

[0127] Use principal component analysis to reduce the dimension of the normalized data and extract features, as shown below:

[0128] ;

[0129] in, is the normalized data matrix, is the weight matrix composed of eigenvectors, is the feature matrix after dimensionality reduction;

[0130] The identified anomalies or failure modes are fed back to the network management system in real time so that corresponding measures can be taken quickly. The dynamic time warping algorithm is used to perform pattern matching on the extracted time series features to identify abnormal behaviors or failure modes in the data, as shown below:

[0131] ;

[0132] in, and are two time series, and These are and Elements of a time series

[0133] Feedback the identified anomalies or failure modes in real time, and then use the linear regression model to predict the development trend of network failures. Use a simple linear regression model to predict the development trend of network failures so that you can prepare in advance. The prediction model can be expressed as:

[0134] ;

[0135] in, is the predicted failure trend, is the network characteristic parameter, is the regression coefficient, is the error term.

[0136] S5. Use quantum computing algorithms to process abnormal data in abnormal patterns and optimize fault handling decisions.

[0137] Specifically, the network optimization problem is defined as finding the best network configuration, minimizing latency and maximizing data throughput. We formulate this problem as a combinatorial optimization problem. In the process of using quantum computing algorithms to process abnormal data in abnormal patterns, in optimizing fault handling decisions, the optimized fault handling decisions are converted into Hamiltonian expressions, and the ground state with the lowest energy is calculated, which is expressed as follows:

[0138] ;

[0139] in, represents the edge set in the network, It is the edge The weight of Is a node status, is the state of node j.

[0140] As an embodiment of the present invention, quantum annealing or quantum approximate optimization algorithm (QAOA) is selected to solve the problem. Quantum annealing finds the ground state of the system by slowly lowering the "temperature", while QAOA uses quantum circuits to approximate the ground state. The process of quantum annealing can be described as:

[0141] ;

[0142] in, is the initial state, is the time-varying annealing parameter, is the Hamiltonian of the system.

[0143] The QAOA process involves multiple quantum gate operations and measurements to optimize a specific objective function.

[0144] Furthermore, according to the optimal configuration output by the quantum algorithm, the working status of network devices is adjusted in real time, such as the route selection of routers and the bandwidth allocation of switches. The optimization results are fed back to the network operation center for subsequent monitoring and adjustment decisions. At the same time, the performance of the quantum optimization algorithm in the actual network environment is regularly evaluated, including its effect in fault response and network load balancing. The parameters and strategies of the quantum algorithm are adjusted according to the evaluation results to improve the accuracy of decision-making and response speed.

[0145] S6. Build a data integration framework and design an adaptive control algorithm based on model predictive control to automatically adjust the network configuration.

[0146] As an embodiment of the present invention, an integrated framework is constructed to collect data from various data sources (including sensor data, network performance logs, and user feedback). The real-time data stream passes through an advanced data processing pipeline, which includes data cleaning, feature extraction, and data normalization steps to standardize the input format. Then an adaptive control algorithm based on model predictive control (MPC) is designed. The algorithm uses current and predicted future data to calculate the control input.

[0147] Specifically, the adaptive control algorithm based on model predictive control is expressed as follows:

[0148] ;

[0149] in, It's in time The network state vector, is the control input, and is the parameter matrix of the system dynamics, is the process noise.

[0150] As a preferred embodiment of the present invention, a set of performance evaluation indicators are defined, such as delay, throughput and data packet loss rate, as well as network stability indicators. A feedback mechanism is designed, which adjusts the network performance based on real-time performance data and preset performance thresholds. and matrix for optimal network response.

[0151] According to the output of the control algorithm, the network device configuration is automatically adjusted, such as the routing strategy of the router, the traffic management settings of the switch, etc. After the control decision is implemented, the system will continuously monitor the implementation effect and adjust the control strategy parameters according to the results to form a closed-loop control system.

[0152] Furthermore, NAFCS is deployed at key network nodes to ensure that the system covers the core part of the network, and the system is iteratively updated regularly to introduce new control algorithm improvements and adapt to new network technologies and application requirements.

[0153] S7. Build a deep learning-based prediction model and continuously train and optimize the prediction model using the latest collected data.

[0154] Specifically, we build a prediction model based on deep learning, which can predict network performance indicators such as latency and throughput based on historical data. We use the stochastic gradient descent method with adaptive learning rate to continuously train the model, and the optimization function is:

[0155] ;

[0156] in, are model parameters, is the time step The learning rate, is the loss function.

[0157] As an embodiment of the present invention, the system collects real-time operation data and user feedback to monitor the performance of the current model, and then uses reinforcement learning technology, especially Learning, to adjust network configuration decisions. The update formula for learning is:

[0158] ;

[0159] in, and are the current and next states, respectively. is the action taken, It is the reward after taking action. is the learning rate, is the discount factor.

[0160] Finally, integrate the learning system into the existing network management platform, allowing the system to adjust the network configuration in real time based on the learning results. Implement a feedback control loop where network performance data and user satisfaction feedback are key inputs for adjusting network configuration. Regularly evaluate the effectiveness of the learning system and iteratively update the learning algorithm and model structure based on the performance evaluation results. Set up a performance monitoring dashboard that displays real-time data and forecast results, as well as the historical performance of the learning system and upcoming network adjustments.

[0161] S8. Use quantum encryption technology to encrypt the communications and data in the above steps to prevent data leakage.

[0162] Specifically, the quantum key distribution enhancement module: uses quantum key distribution (QKD) technology combined with traditional encryption technology to implement an adaptive encryption mechanism. This mechanism dynamically adjusts the encryption strength and key update frequency according to the network load and communication security level. According to the data type and sensitivity, the system automatically selects the appropriate encryption algorithm (such as AES-256, RSA2048, etc.) and uses the key provided by QKD for encryption.

[0163] The encryption process uses an adaptive tuning algorithm of the following form:

[0164] ;

[0165] in, Represents the new key, represents the old key, represents the quantum link quality indicator, Indicates the time interval, Represents a key update function.

[0166] As an embodiment of the present invention, machine learning technology (especially deep learning) is used to model and learn network user behavior. This module can identify normal and abnormal behavior patterns. The implementation of anomaly detection algorithms based on deep learning can automatically adapt to changes in network behavior. The anomaly detection algorithm may take the following forms:

[0167] ;

[0168] in It is behavioral characteristic data. is the standard of behavior (normal or abnormal), are model parameters, is a neural network, and the sigmoid function is used to output the anomaly probability.

[0169] Preferably, a real-time monitoring dashboard is set up to monitor the security status of all network nodes. Use data visualization techniques, such as heat maps and time series graphs, to display encryption status and behavior analysis results. In addition to monitoring data, the system automatically adjusts security policies and behavior analysis parameters to optimize performance and response speed.

[0170] It should be noted that the system design complies with international data protection regulations (such as GDPR and CCPA) to achieve data processing compliance. Data access and management policies are implemented to ensure that sensitive data is protected throughout its life cycle, including data generation, storage, transmission and destruction.

[0171] In one or more embodiments, Figure 2 As shown, a data center energy-saving control system is disclosed, and the system includes:

[0172] The multimodal data collection module 101 is used to arrange multimodal sensors to collect multimodal data and pre-process the multimodal data; wherein the multimodal sensors include temperature sensors, optical fiber vibration sensors, flow monitors and signal strength sensors;

[0173] The multimodal data integration module 102 is used to integrate the multimodal data and build an intermediate monitoring platform to perform real-time monitoring and analysis on the multimodal data;

[0174] The GAN simulation module 103 is used to design the generator and discriminator architecture to simulate normal and various fault conditions, and generate fault training data and fault test data;

[0175] An abnormal pattern identification module 104 is used to apply a machine learning algorithm to conduct in-depth analysis on multimodal data and identify abnormal patterns;

[0176] An optimized fault handling decision module 105 is used to process abnormal data of abnormal patterns using a quantum computing algorithm to optimize fault handling decisions;

[0177] The network configuration automatic adjustment module 106 is used to build a data integration framework and design an adaptive control algorithm based on model predictive control to automatically adjust the network configuration;

[0178] The prediction model optimization module 107 is used to build a prediction model based on deep learning, and continuously train and optimize the prediction model using the latest collected data;

[0179] A quantum encryption module 108, used to encrypt the communication and data in the above steps using quantum encryption technology to prevent data leakage;

[0180] Among them, in the process of using quantum computing algorithm to process abnormal data of abnormal mode, in optimizing fault handling decision, the optimized fault handling decision is converted into Hamiltonian expression, and the ground state with the lowest energy is calculated, which is expressed as follows:

[0181] ;

[0182] in, represents the edge set in the network, It is the edge The weight of Is a node status, is the state of node j.

[0183] The present invention also provides a storage medium storing non-temporary computer-readable instructions. When the non-temporary computer-readable instructions are executed by a computer, the steps in the aforementioned data center energy-saving control method can be executed.

[0184] The above disclosures are only some preferred embodiments of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiments and equivalent changes made according to the claims of the present invention are still within the scope of the invention.

Claims

1. A data center energy saving control method, characterized in that: The following steps are involved: Arranging multimodal sensors to collect multimodal data and preprocessing the multimodal data; wherein the multimodal sensors include temperature sensors, optical fiber vibration sensors, flow monitors and signal strength sensors; Integrate multimodal data and build an intermediate monitoring platform to monitor and analyze multimodal data in real time; Design the generator and discriminator architecture to simulate normal and various fault conditions, and generate fault training data and fault test data; Apply machine learning algorithms to conduct in-depth analysis of multimodal data and identify abnormal patterns; Use quantum computing algorithms to process abnormal data with abnormal patterns and optimize fault handling decisions; Build a data integration framework and design an adaptive control algorithm based on model predictive control to automatically adjust the network configuration; Build deep learning-based predictive models and use the latest collected data to train and optimize predictive models; Use quantum encryption technology to encrypt the communication and data in the above steps to prevent data leakage; Among them, in the process of using quantum computing algorithm to process abnormal data of abnormal mode, in optimizing fault handling decision, the optimized fault handling decision is converted into Hamiltonian expression, and the ground state with the lowest energy is calculated, which is expressed as follows: ; in, represents the edge set in the network, It is the edge The weight of Is a node status, is the state of node j; The method of integrating multimodal data and building an intermediate monitoring platform to monitor and analyze the multimodal data in real time specifically includes the following steps: The data received by the multimodal sensor is synchronized with the data time, as shown below: ; in, Represents the original timestamp, Represents a reference timestamp, Indicates the correction value; A dynamic data stream fusion algorithm is designed to integrate multimodal data. The data fusion calculation is expressed as follows: ; in, represents fused data, Indicates Sensor data, represents the weight assigned according to the importance and reliability of the data, Indicates the number of data categories; A real-time data analysis mechanism is designed to instantly identify abnormal patterns and potential faults from the fused data. At the same time, the adaptive threshold technology is used to analyze the fused data in real time and dynamically adjust the threshold of fault detection based on historical data. The formula for fault detection is as follows: ; in, represents the current data point, and represent the moving average and standard deviation respectively, represents the sensitivity coefficient of adaptive adjustment, Indicates the fault threshold; The adaptive control algorithm based on model predictive control is expressed as follows: ; in, It's in time The network state vector, is the control input, and is the parameter matrix of the system dynamics, is the process noise; Use quantum cryptographic data to encrypt communications and data, including: An adaptive encryption mechanism is designed based on quantum encryption technology. The encryption process uses the following form of adaptive adjustment algorithm: ; in, Represents the new key, represents the old key, represents the quantum link quality indicator, Indicates the time interval, Represents a key update function.

2. Data center energy-saving control system, characterized in that: The system comprises: A multimodal data collection module, used to arrange multimodal sensors to collect multimodal data and pre-process the multimodal data; wherein the multimodal sensors include temperature sensors, optical fiber vibration sensors, flow monitors and signal strength sensors; Multimodal data integration module, used to integrate multimodal data and build an intermediate monitoring platform to monitor and analyze multimodal data in real time; GAN simulation module, used to design the generator and discriminator architecture to simulate normal and various fault conditions, and generate fault training data and fault test data; The abnormal pattern recognition module is used to apply machine learning algorithms to conduct in-depth analysis of multimodal data and identify abnormal patterns; Optimize the fault handling decision module, which is used to process abnormal data in abnormal patterns using quantum computing algorithms and optimize fault handling decisions; The network configuration automatic adjustment module is used to build a data integration framework and design an adaptive control algorithm based on model predictive control to automatically adjust the network configuration; The prediction model optimization module is used to build a prediction model based on deep learning and use the latest collected data to train and optimize the prediction model; Quantum encryption module, used to encrypt communications and data using quantum encryption technology to prevent data leakage; Among them, in the process of using quantum computing algorithm to process abnormal data of abnormal mode, in optimizing fault handling decision, the optimized fault handling decision is converted into Hamiltonian expression, and the ground state with the lowest energy is calculated, which is expressed as follows: ; in, represents the edge set in the network, It is the edge The weight of Is a node status, is the state of node j; The method of integrating multimodal data and building an intermediate monitoring platform to monitor and analyze the multimodal data in real time specifically includes the following steps: The data received by the multimodal sensor is synchronized with the data time, as shown below: ; in, Represents the original timestamp, Represents a reference timestamp, Indicates the correction value; A dynamic data stream fusion algorithm is designed to integrate multimodal data. The data fusion calculation is expressed as follows: ; in, represents fused data, Indicates Sensor data, represents the weight assigned according to the importance and reliability of the data, Indicates the number of data categories; A real-time data analysis mechanism is designed to instantly identify abnormal patterns and potential faults from the fused data. At the same time, the adaptive threshold technology is used to analyze the fused data in real time and dynamically adjust the threshold of fault detection based on historical data. The formula for fault detection is as follows: ; in, represents the current data point, and represent the moving average and standard deviation respectively, represents the sensitivity coefficient of adaptive adjustment, Indicates the fault threshold; The adaptive control algorithm based on model predictive control is expressed as follows: ; in, It's in time The network state vector, is the control input, and is the parameter matrix of the system dynamics, is the process noise; Use quantum cryptographic data to encrypt communications and data, including: An adaptive encryption mechanism is designed based on quantum encryption technology. The encryption process uses the following form of adaptive adjustment algorithm: ; in, Represents the new key, represents the old key, represents the quantum link quality indicator, Indicates the time interval, Represents a key update function.

3. A storage medium, characterized in that: It stores non-transitory computer-readable instructions, and when the non-transitory computer-readable instructions are executed by a computer, the steps in the data center energy-saving control method as claimed in claim 1 can be executed.

Citation Information

Patent Citations

  • Power grid health assessment and analysis method based on multiple modes

    CN118657404A

  • Anomaly detection in pipelines and flowlines

    US20210116076A1