Storage, calculation and training integrated cloud computing server optimization method based on human brain neural network

Abstract

The invention discloses a storage, calculation and training integrated cloud computing server optimization method based on a human brain neural network. The method comprises the following steps: S1, designing a self-adaptive learning model based on the human brain neural network; s2, forming an integrated framework; s3, adopting a novel processor design; s4, optimizing an adaptive learning model architecture of the human brain neural network through a deep learning technology; s5, dynamically adjusting distribution of computing resources based on the real-time data flow and the computing requirements; s6, realizing data transmission path optimization by designing an adaptive data routing algorithm; s7, self-optimization is realized in the cloud computing server by using a synaptic plasticity mechanism of the adaptive learning model, and the adaptive learning model can autonomously adjust parameters and optimize performance according to historical data and a current state; and S8, integrating an intelligent fault detection and repair module in the adaptive learning model. The system state is monitored in real time, faults are detected and repaired, and the reliability and stability of the system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of cloud computing servers, and in particular to a storage, computing and training integrated cloud computing server optimization method based on a human brain neural network. Background Art

[0002] With the rapid development of information technology and the popularization of cloud computing, cloud computing servers play an important role in data storage, computing, and training. However, the existing cloud computing server architecture has some significant defects, which limits its potential in high-performance computing and data processing.

[0003] Traditional cloud computing server architectures usually separate storage, computing, and training functions. This design results in frequent data transmission between modules, increasing the overall energy consumption of the system. The frequent transmission of data between storage modules, computing modules, and training modules not only consumes a lot of energy, but also leads to a decrease in the system processing speed. Especially when processing large-scale data, this frequent data transmission greatly reduces the energy efficiency of the system and cannot meet the needs of efficient computing. In addition, the transmission of data between different modules also increases the risk of data loss or damage, affecting the security and reliability of data.

[0004] The cloud computing servers in the existing technology also have significant deficiencies in energy consumption management. Traditional energy consumption optimization methods mainly rely on the upgrade of hardware equipment and the improvement of power consumption management strategies, but these methods often cannot fundamentally solve the energy consumption problem. For example, by increasing the computing power of the processor and the storage capacity of the memory, although the computing efficiency can be improved to a certain extent, it also brings higher energy consumption. In addition, the existing energy consumption optimization strategies often lack flexibility and cannot be dynamically adjusted according to the actual workload and environment, resulting in inefficient energy consumption management.

[0005] Existing technologies also face challenges in data transmission and security. The data transmission path in traditional cloud computing architecture is complex, and data is easily disturbed during transmission, resulting in data loss or damage. Although redundant storage and data verification mechanisms have been introduced in existing technologies, these methods increase the complexity and cost of the system and fail to effectively solve data security issues. Especially in large-scale distributed computing environments, the complexity and risks of data transmission are more prominent, affecting the stability and reliability of the system.

[0006] In addition, existing cloud computing servers also have significant deficiencies in fault detection and repair. Traditional fault detection methods mainly rely on predefined fault models and rules, lack self-learning and adaptive capabilities, and cannot detect and repair complex faults in a timely manner. When a system fails, manual intervention is often required for troubleshooting and repair, resulting in a long fault repair time and affecting the normal operation of the system. Existing fault detection methods cannot achieve real-time monitoring and automatic repair, which limits the self-optimization and self-repair capabilities of cloud computing servers. Summary of the invention

[0007] One purpose of the present invention is to propose a storage, computing and training integrated cloud computing server optimization method based on human brain neural network. The present invention utilizes the adaptive learning ability of human brain neural network, and through the integrated integration of storage, computing and training functions, reduces data transmission, reduces energy consumption and improves computing efficiency. At the same time, combined with intelligent fault detection and repair module, by real-time monitoring of system status, detection and repair of faults, the reliability and stability of the system are improved.

[0008] According to an embodiment of the present invention, a storage, computing and training integrated cloud computing server optimization method based on a human brain neural network includes the following steps:

[0009] S1. Design an adaptive learning model based on human brain neural network. The adaptive learning model contains multiple levels of neurons, each of which has the functions of storage, calculation and training;

[0010] S2, integrating storage, computing and training functions in cloud computing servers to form an integrated architecture;

[0011] S3, adopts a new processor design, the processor includes multi-functional cores, each core can independently complete storage, computing and training tasks, and optimizes the data channel design;

[0012] S4. Optimize the adaptive learning model architecture of the human brain neural network through deep learning technology;

[0013] S5, dynamically adjust the allocation of computing resources based on real-time data flow and computing needs;

[0014] S6. Optimize the data transmission path by designing an adaptive data routing algorithm;

[0015] S7. Using the synaptic plasticity mechanism of the adaptive learning model, self-optimization is achieved in the cloud computing server. The adaptive learning model can autonomously adjust parameters and optimize performance based on historical data and current status.

[0016] S8. Integrate intelligent fault detection and repair modules into the adaptive learning model, and use the self-learning ability of the human brain neural network to monitor the system status in real time, detect and repair faults.

[0017] Optionally, the S1 includes:

[0018] S11. Design an adaptive learning model based on the human brain neural network. The adaptive learning model adopts a heterogeneous computing architecture and contains multiple levels of neurons. Each neuron has storage, computing and training functions:

[0019]

[0020] Among them, A ij represents the activation value of the i-th neuron in the j-th layer, W ik represents the weight from the i-th neuron to the k-th neuron, X kj represents input data, B i represents the bias, T ij represents the temporal dependency, reflecting the temporal effect of dynamically adjusting weights and biases;

[0021] S12. In a heterogeneous computing architecture, each neuron is composed of different types of computing units, including general-purpose processors, graphics processors, and field-programmable gate arrays, to achieve optimized processing of different types of computing tasks:

[0022] P total =α·P CPU +β·P GPU +γ·P FPGA +λ·C comm ;

[0023] Among them, P total Represents the total computing power, P CPU , P GPU and P FPGA They represent the computing power of CPU, GPU and FPGA respectively, α, β, γ are the weight coefficients of each computing unit, and C comm represents the communication overhead, λ is the communication overhead coefficient;

[0024] S13, the storage unit uses a three-dimensional stacked memory to store the input and output data of the neuron;

[0025]

[0026] Among them, M represents the state of the storage unit, S i represents the capacity of the storage unit, V represents the input voltage, V i represents the operating voltage of the storage unit, σ represents the standard deviation of the voltage, and δ is the offset value of the storage unit;

[0027] S14. The computing unit implements parallel computing and accelerated computing through different types of processors. General-purpose processors are used to handle control and simple computing tasks, graphics processors are used to handle large-scale parallel computing tasks, and field programmable gate arrays are used to accelerate specific algorithms:

[0028]

[0029] Among them, Q represents the total computing power, v jk and k Represent weights and input data respectively, c j represents the bias, σ 1 represents the activation function, τ 1 represents the time constant, u i Indicates the time point, u 0 Indicates the current time;

[0030] S15, the training unit uses an adaptive synaptic plasticity mechanism to adjust the weights and biases of neurons through the error back propagation algorithm:

[0031]

[0032] Where Δv i represents the change in weight, η 1 represents the learning rate, z j Indicates the actual output, Represents the predicted output, y ij represents the input data, σ 2 represents the activation function, b j represents the net input of the neuron, φ 1 (t) represents the time-dependent regulation function;

[0033] S16. In a heterogeneous computing architecture, each computing unit transmits and shares data through an interconnected network:

[0034]

[0035] Among them, U represents the data transmission time, E represents the data volume, F represents the bandwidth, and P loss2 represents the probability of data loss, L 1 represents the basic transmission delay, d 1 Indicates the transmission distance, d 1,0 Indicates the reference distance.

[0036] Optionally, the S4 includes:

[0037] S41. In the storage-computing-training integrated architecture, a hybrid design of quantum computing units and traditional computing units is adopted. Quantum computing units are used to process highly complex training tasks, and traditional computing units are used for daily data processing and storage:

[0038]

[0039] Among them, total represents the total state of the hybrid quantum-classical computing system, α 1 and β 1 denote the contribution coefficients of quantum computing unit and traditional computing unit respectively, |ψ i > and |φ j > respectively represent the states of quantum bits and classical bits, n and m represent the number of quantum bits and classical bits, respectively;

[0040] S42. Design an adaptive learning model based on a quantum neural network. The quantum neural network contains multiple qubits, each of which represents a neuron state:

[0041]

[0042] Among them, Ψ represents the total state of the quantum neural network, α i and β j denote the weights of quantum bits and classical bits respectively, |χ j > and |η j > represents the entangled state of the jth quantum bit and the classical bit, n and m represent the number of quantum bits and the number of classical bits respectively;

[0043] S43. In the quantum computing unit, quantum entanglement and quantum superposition are used to achieve efficient parallel computing, and the weights and biases of the neural network are updated through quantum gate operations:

[0044]

[0045] in, represents the quantum gate operation matrix, θ 1 , λ 1 represents the angle parameter that controls the rotation of the quantum state;

[0046] S44. During each training process, the quantum tunneling effect is used to accelerate weight updates:

[0047]

[0048] Where, ΔW ij represents the weight change, η 2 represents the learning rate, represents the partial derivative of the loss function with respect to the weight, d represents the tunneling distance, represents Planck's constant, ω represents frequency, and t represents time;

[0049] S45. Use the ultra-high computing speed of quantum computing units to quickly process and train large-scale data, while using traditional computing units to verify and store results:

[0050]

[0051] Among them, T process represents the total processing time, Q speed and C speed They represent the processing speeds of the quantum computing unit and the traditional computing unit respectively, N represents the data scale, and M represents the amount of data that the traditional computing unit needs to process.

[0052] S46. Introducing the time crystal effect into quantum neural networks, using the periodic oscillation characteristics of time crystals to enhance the dynamic adjustment capability of the adaptive learning model:

[0053] H(t)=H 0 +Vcos(ωt)+γsin 2 (ω 1 t / 2);

[0054] Where H(t) represents the time-dependent Hamiltonian, H 0 represents the initial Hamiltonian, V represents the disturbance term, γ represents the oscillation frequency, ω 1 represents the nonlinear effect coefficient of the time crystal;

[0055] S47. Through the hybrid quantum-classical architecture, fast data storage, computing and training can be achieved, improving the overall performance and energy efficiency of cloud computing servers:

[0056]

[0057] Among them, P total represents the total computing power, α i1 and β j1 They represent the contribution coefficients of quantum computing unit and traditional computing unit, Q i and C j Represent the computing power of quantum computing unit and traditional computing unit respectively, λ 2 represents the data transmission coefficient, γ k Indicates the efficiency of the transmission channel, D k represents the data transmission volume, and n, m, and p represent the number of quantum computing units, traditional computing units, and data transmission channels, respectively.

[0058] Optionally, the S7 includes:

[0059] S71. In the adaptive learning model, design a parameter adjustment algorithm based on synaptic plasticity mechanisms, including long-term potentiation and long-term depression, to dynamically adjust the connection strength between neurons;

[0060] S72, using historical data and current status, calculating adjustment values ​​of synaptic weights, updating weights according to these adjustment values, and dynamically adapting to the current computing environment;

[0061] S73. In the synaptic plasticity mechanism, a time-dependent weight adjustment strategy is introduced to adjust the weight change through the time decay function;

[0062] S74, through the synaptic plasticity mechanism in the adaptive learning model, real-time monitoring and analysis of the operating status of the cloud computing server, and self-optimization based on the results of the optimization function;

[0063] S75. In the adaptive learning model, a feedback mechanism is used to adjust parameters according to the results of the optimization function;

[0064] S76. Through the dynamic adjustment of synaptic plasticity mechanism, cloud computing servers can be self-optimized under different workloads and environments.

[0065] S77. The adaptive learning model can autonomously adjust various parameters and optimize performance based on the historical data and current status of the cloud computing server.

[0066] Optionally, the S8 includes:

[0067] S81. Integrate intelligent fault detection module into the adaptive learning model, use the self-learning ability of human brain neural network, and detect faults in system operation through real-time monitoring of system status data:

[0068]

[0069] Among them, F d (t) represents the fault detection result at time t, N represents the number of neurons in the detection module, and α i is the weight coefficient of fault detection, f i (t) is the function of system state data, M is the number of input parameters, w ij is the weight, x j (t) is the jth input parameter, b i is the bias, σ is the activation function;

[0070] S82, the intelligent fault detection module collects and analyzes the status parameters of each component of the system, including processor utilization, memory usage, network latency, and memory read and write speed, to form a basic data set for fault detection:

[0071]

[0072] Among them, S(t) represents the set of system state parameters at time t, U cpu (t′) is the processor utilization at time t′, C cpu is the processor capacity, M mem (t′) is the memory usage at time t′, C mem is the memory capacity, D net (t′) is the network delay at time t′, L net is the network bandwidth, R stor (t′) is the memory read and write speed at time t′, S stor is the memory capacity;

[0073] S83. Utilize the pattern recognition capability of the human brain neural network to analyze the collected state parameters, identify abnormal patterns and determine whether there is a fault. The recognition process is based on the trained neural network model and can adaptively update and improve the detection algorithm:

[0074]

[0075] Among them, P fault (t) represents the failure probability at time t, Q is the number of layers of the neural network, β k is the weight coefficient of each layer, tanh is the hyperbolic tangent activation function, γ kl is the connection weight in each layer, S l (t) is the lth system state parameter;

[0076] S84. When a fault is detected, the intelligent fault detection module generates a fault report, indicating the specific location and type of the fault, and triggers the fault repair module to perform automatic repair;

[0077] S85, the fault repair module uses the self-learning ability of the human brain neural network to select an appropriate repair strategy based on the information in the fault report. The repair strategy includes reallocating computing resources, restarting faulty components, and adjusting system parameters.

[0078] S86. During the fault repair process, the fault repair module monitors the repair effect in real time and adjusts the repair strategy according to the actual situation;

[0079] S87. After the fault is repaired, the system records the fault detection and repair process, updates the parameters in the adaptive learning model, and optimizes future fault detection and repair capabilities.

[0080] The beneficial effects of the present invention are:

[0081] (1) The present invention integrates storage, computing and training functions into a unified architecture, effectively reducing the number and distance of data transmission between modules, and significantly reducing the overall energy consumption of the system. In traditional cloud computing architecture, the frequent transmission of data between different modules leads to high energy consumption. The present invention realizes on-site processing and training of data through the integrated storage, computing and training architecture design, reducing the need for data movement, thereby improving the energy efficiency of the system. In addition, the use of new processor design and optimized data channel design further improves computing efficiency and resource utilization, allowing the system to process more data and complete more complex computing tasks at the same power consumption.

[0082] (2) The present invention reduces the frequency of data transmission between storage, computing and training modules through adaptive data routing algorithms and optimized data transmission path design, thereby reducing the complexity and risk of data transmission. The adaptive data routing algorithm can dynamically adjust the data transmission path according to the actual workload and network status to ensure the security and reliability of data during transmission. At the same time, the integrated storage, computing and training architecture reduces the data transmission distance, reduces the possibility of data loss or damage, and further improves data security and system stability.

[0083] (3) The present invention integrates an intelligent fault detection and repair module into the adaptive learning model, which can monitor the system status in real time, detect and repair faults. Traditional fault detection methods rely on predefined fault models and rules and cannot adapt to complex and changing fault conditions. The present invention can dynamically adjust the detection and repair strategies through self-learning and adaptive algorithms, thereby improving the accuracy of fault detection and the timeliness of repair. By analyzing historical data and current status, the adaptive learning model can autonomously adjust various parameters to achieve self-optimization and performance improvement of the system, ensuring stable operation of the system under high load and complex environments.

[0084] (4) The present invention has made many innovations in the design of neural network architecture and optimization algorithm. An adaptive learning model based on the human brain neural network is designed, and a multi-level neuron structure is adopted to realize the organic combination of storage, computing and training functions. Through the introduction of deep learning technology and quantum computing units, the adaptive learning model architecture of the neural network is optimized, and the computing speed and processing power are improved. In addition, a time-dependent weight adjustment strategy is introduced into the synaptic plasticity mechanism, and the weight change is adjusted by the time decay function to ensure the stability and convergence speed of the model training. Finally, through the intelligent fault detection and repair module, the pattern recognition and self-learning ability of the human brain neural network are utilized to realize real-time monitoring and automatic repair of system faults, thereby improving the reliability and self-optimization ability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0086] Figure 1 This is a flow chart of a storage, computing and training integrated cloud computing server optimization method based on human brain neural network proposed by the present invention;

[0087] Figure 2 This is a schematic diagram of the hierarchical structure of an adaptive learning model for a storage, computing and training integrated cloud computing server optimization method based on a human brain neural network proposed in the present invention. DETAILED DESCRIPTION

[0088] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0089] refer to Figure 1-2 , a storage, computing and training integrated cloud computing server optimization method based on human brain neural network, comprising the following steps:

[0090] S1. Design an adaptive learning model based on human brain neural network. The adaptive learning model contains multiple levels of neurons, each of which has the functions of storage, calculation and training;

[0091] S2, integrating storage, computing and training functions in cloud computing servers to form an integrated architecture;

[0092] S3, adopts a new processor design, the processor includes multi-functional cores, each core can independently complete storage, computing and training tasks, and optimizes the data channel design;

[0093] S4. Optimize the adaptive learning model architecture of the human brain neural network through deep learning technology;

[0094] S5, dynamically adjust the allocation of computing resources based on real-time data flow and computing needs;

[0095] S6. Optimize the data transmission path by designing an adaptive data routing algorithm;

[0096] S7. Using the synaptic plasticity mechanism of the adaptive learning model, self-optimization is achieved in the cloud computing server. The adaptive learning model can autonomously adjust parameters and optimize performance based on historical data and current status.

[0097] S8. Integrate intelligent fault detection and repair modules into the adaptive learning model, and use the self-learning ability of the human brain neural network to monitor the system status in real time, detect and repair faults.

[0098] In this implementation, S1 includes:

[0099] S11. Design an adaptive learning model based on the human brain neural network. The adaptive learning model adopts a heterogeneous computing architecture and contains multiple levels of neurons. Each neuron has storage, computing and training functions:

[0100]

[0101] Among them, A ij represents the activation value of the i-th neuron in the j-th layer, W ik represents the weight from the i-th neuron to the k-th neuron, X kj represents input data, B i represents the bias, T ij represents the temporal dependency, reflecting the temporal effect of dynamically adjusting weights and biases;

[0102] S12. In a heterogeneous computing architecture, each neuron is composed of different types of computing units, including general-purpose processors, graphics processors, and field-programmable gate arrays, to achieve optimized processing of different types of computing tasks:

[0103] P total =α·P CPU +β·P GPU +γ·P FPGA +λ·C comm ;

[0104] Among them, P total Represents the total computing power, P CPU , P GPU and P FPGA They represent the computing power of CPU, GPU and FPGA respectively, α, β, γ are the weight coefficients of each computing unit, and C comm represents the communication overhead, λ is the communication overhead coefficient;

[0105] S13, the storage unit uses a three-dimensional stacked memory to store the input and output data of the neuron;

[0106]

[0107] Among them, M represents the state of the storage unit, S i represents the capacity of the storage unit, V represents the input voltage, V i represents the operating voltage of the storage unit, σ represents the standard deviation of the voltage, and δ is the offset value of the storage unit;

[0108] S14. The computing unit implements parallel computing and accelerated computing through different types of processors. General-purpose processors are used to handle control and simple computing tasks, graphics processors are used to handle large-scale parallel computing tasks, and field programmable gate arrays are used to accelerate specific algorithms:

[0109]

[0110] Among them, Q represents the total computing power, v jk and k Represent weights and input data respectively, c j represents the bias, σ 1 represents the activation function, τ 1 represents the time constant, u i Indicates the time point, u 0 Indicates the current time;

[0111] S15, the training unit uses an adaptive synaptic plasticity mechanism to adjust the weights and biases of neurons through the error back propagation algorithm:

[0112]

[0113] Where Δv i represents the change in weight, η 1 represents the learning rate, z j Indicates the actual output, Represents the predicted output, y ij represents the input data, σ 2 represents the activation function, b j represents the net input of the neuron, φ 1 (t) represents the time-dependent regulation function;

[0114] S16. In a heterogeneous computing architecture, each computing unit transmits and shares data through an interconnected network:

[0115]

[0116] Among them, U represents the data transmission time, E represents the data volume, F represents the bandwidth, and P loss2 represents the probability of data loss, L 1 represents the basic transmission delay, d 1 Indicates the transmission distance, d 1,0 Indicates the reference distance.

[0117] In this implementation, S4 includes:

[0118] S41. In the storage-computing-training integrated architecture, a hybrid design of quantum computing units and traditional computing units is adopted. Quantum computing units are used to process highly complex training tasks, and traditional computing units are used for daily data processing and storage:

[0119]

[0120] Among them, total represents the total state of the hybrid quantum-classical computing system, α 1 and β 1 denote the contribution coefficients of quantum computing unit and traditional computing unit respectively, |ψ i 〉 and |φ j > respectively represent the states of quantum bits and classical bits, n and m represent the number of quantum bits and classical bits, respectively;

[0121] S42. Design an adaptive learning model based on a quantum neural network. The quantum neural network contains multiple qubits, each of which represents a neuron state:

[0122]

[0123] Among them, Ψ represents the total state of the quantum neural network, α i and β j denote the weights of quantum bits and classical bits respectively, |χ j > and |η j > represents the entangled state of the jth quantum bit and the classical bit, n and m represent the number of quantum bits and the number of classical bits respectively;

[0124] S43. In the quantum computing unit, quantum entanglement and quantum superposition are used to achieve efficient parallel computing, and the weights and biases of the neural network are updated through quantum gate operations:

[0125]

[0126] in, represents the quantum gate operation matrix, θ 1 , λ 1 represents the angle parameter that controls the rotation of the quantum state;

[0127] S44. During each training process, the quantum tunneling effect is used to accelerate weight updates:

[0128]

[0129] Where, ΔW ij represents the weight change, η 2 represents the learning rate, represents the partial derivative of the loss function with respect to the weight, d represents the tunneling distance, represents Planck's constant, ω represents frequency, and t represents time;

[0130] S45. Use the ultra-high computing speed of quantum computing units to quickly process and train large-scale data, while using traditional computing units to verify and store results:

[0131]

[0132] Among them, T process represents the total processing time, Q speed and C speed They represent the processing speeds of the quantum computing unit and the traditional computing unit respectively, N represents the data scale, and M represents the amount of data that the traditional computing unit needs to process.

[0133] S46. Introducing the time crystal effect into quantum neural networks, using the periodic oscillation characteristics of time crystals to enhance the dynamic adjustment capability of the adaptive learning model:

[0134] H(t)=H 0 +Vcos(ωt)+γsin 2 (ω 1 t / 2);

[0135] Where H(t) represents the time-dependent Hamiltonian, H 0 represents the initial Hamiltonian, V represents the disturbance term, γ represents the oscillation frequency, ω 1 represents the nonlinear effect coefficient of the time crystal;

[0136] S47. Through the hybrid quantum-classical architecture, fast data storage, computing and training can be achieved, improving the overall performance and energy efficiency of cloud computing servers:

[0137]

[0138] Among them, P total represents the total computing power, α i1 and β j1 They represent the contribution coefficients of quantum computing unit and traditional computing unit, Q i and C j Represent the computing power of quantum computing unit and traditional computing unit respectively, λ 2 represents the data transmission coefficient, γ k Indicates the efficiency of the transmission channel, D k represents the data transmission volume, and n, m, and p represent the number of quantum computing units, traditional computing units, and data transmission channels, respectively.

[0139] In this implementation, S7 includes:

[0140] S71. In the adaptive learning model, design a parameter adjustment algorithm based on synaptic plasticity mechanisms, including long-term potentiation and long-term depression, to dynamically adjust the connection strength between neurons;

[0141] S72, using historical data and current status, calculating adjustment values ​​of synaptic weights, updating weights according to these adjustment values, and dynamically adapting to the current computing environment;

[0142] S73. In the synaptic plasticity mechanism, a time-dependent weight adjustment strategy is introduced to adjust the weight change through the time decay function;

[0143] S74, through the synaptic plasticity mechanism in the adaptive learning model, real-time monitoring and analysis of the operating status of the cloud computing server, and self-optimization based on the results of the optimization function;

[0144] S75. In the adaptive learning model, a feedback mechanism is used to adjust parameters according to the results of the optimization function;

[0145] S76. Through the dynamic adjustment of synaptic plasticity mechanism, cloud computing servers can be self-optimized under different workloads and environments.

[0146] S77. The adaptive learning model can autonomously adjust various parameters and optimize performance based on the historical data and current status of the cloud computing server.

[0147] In this implementation, S8 includes:

[0148] S81. Integrate intelligent fault detection module into the adaptive learning model, use the self-learning ability of human brain neural network, and detect faults in system operation through real-time monitoring of system status data:

[0149]

[0150] Among them, F d (t) represents the fault detection result at time t, N represents the number of neurons in the detection module, and α i is the weight coefficient of fault detection, f i (t) is the function of system state data, M is the number of input parameters, w ij is the weight, x j (t) is the jth input parameter, b i is the bias, σ is the activation function;

[0151] S82, the intelligent fault detection module collects and analyzes the status parameters of each component of the system, including processor utilization, memory usage, network latency, and memory read and write speed, to form a basic data set for fault detection:

[0152]

[0153] Among them, S(t) represents the set of system state parameters at time t, U cpu (t′) is the processor utilization at time t′, C cpu is the processor capacity, M mem (t′) is the memory usage at time t′, C mem is the memory capacity, D net (t′) is the network delay at time t′, L net is the network bandwidth, R stor (t′) is the memory read and write speed at time t′, S stor is the memory capacity;

[0154] S83. Utilize the pattern recognition capability of the human brain neural network to analyze the collected state parameters, identify abnormal patterns and determine whether there is a fault. The recognition process is based on the trained neural network model and can adaptively update and improve the detection algorithm:

[0155]

[0156] Among them, P fault (t) represents the failure probability at time t, Q is the number of layers of the neural network, β k is the weight coefficient of each layer, tanh is the hyperbolic tangent activation function, γ kl is the connection weight in each layer, S l (t) is the lth system state parameter;

[0157] S84. When a fault is detected, the intelligent fault detection module generates a fault report, indicating the specific location and type of the fault, and triggers the fault repair module to perform automatic repair;

[0158] S85, the fault repair module uses the self-learning ability of the human brain neural network to select an appropriate repair strategy based on the information in the fault report. The repair strategy includes reallocating computing resources, restarting faulty components, and adjusting system parameters.

[0159] S86. During the fault repair process, the fault repair module monitors the repair effect in real time and adjusts the repair strategy according to the actual situation;

[0160] S87. After the fault is repaired, the system records the fault detection and repair process, updates the parameters in the adaptive learning model, and optimizes future fault detection and repair capabilities.

[0161] Embodiment 1:

[0162] On a world-renowned large-scale e-commerce platform, users visit hundreds of millions of users every day, and the platform needs to process a large amount of user data, including browsing history, purchase history, and evaluation data. This data requires fast storage, real-time calculation, and frequent training of machine learning models to predict user behavior, optimize recommendation algorithms, and improve user experience. However, the traditional cloud computing server architecture has the problem of separation of storage, computing, and training functions, resulting in frequent data transmission between modules, increasing system energy consumption, reducing processing speed, and posing the risk of data loss.

[0163] In order to verify the feasibility and effectiveness of the present invention, we conducted a simulation test in the actual operating environment of the e-commerce platform, demonstrating the advantages of the storage, computing and training integrated cloud computing server based on the human brain neural network in processing large-scale data.

[0164] We collected browsing, clicking, and purchasing records of 10 million users on the platform from June 15, 2024 to June 30, 2024. The data is directly stored in the server through the storage unit of the adaptive learning model, avoiding the data transmission between the storage and computing modules in the traditional method.

[0165] While the data is being stored, the computing unit begins to extract features from the data. Taking the user's browsing history as an example, the system extracts features such as the user's click sequence, page dwell time, and browsed page category. These features are quickly processed and modeled through an adaptive learning model. In Example 1, a user with the user ID "U123456" browsed 30 pages on June 18, 2024, with an average dwell time of 45 seconds, and the categories of clicked pages were mainly concentrated in electronic products and household appliances.

[0166] Using historical data from the past two weeks, the system trained the user behavior prediction model. During the training process, the system automatically adjusted the weights and biases of the neural network, and used the synaptic plasticity mechanism to achieve dynamic adjustment of parameters. In Example 1, the system detected that the click behavior of the user with user ID "U789012" was highly correlated with past purchase behavior during browsing, so the model weights were adjusted to improve prediction accuracy. Through the integrated storage, computing and training architecture, the model training time is shortened from 5 hours of traditional methods to 2 hours.

[0167] After the model training is completed, the platform begins to predict and recommend user behavior in real time. In Example 1, the user ID "U345678" browsed the digital product page at 8 pm on June 29, 2024, and the system immediately recommended related accessories and discount information based on his browsing history and purchase history. The user response time is shortened from 200 milliseconds in traditional methods to 100 milliseconds, greatly improving the user experience.

[0168] During this process, the intelligent fault detection module monitors the system status in real time. At 2 pm on June 20, 2024, the system detected that the CPU utilization of the server with the source IP address "192.168.1.101" increased abnormally, reaching more than 95%, and the memory utilization rate also exceeded 90%. The intelligent fault detection module analyzed parameters such as CPU utilization, memory usage, and network latency, and identified that the server was at risk of resource overload. The system generates a detailed fault report, recording the time of the fault, server IP, utilization data, and risk assessment results. The fault report is sent to the system administrator in real time, and the administrator quickly takes resource reallocation measures to reduce the load on the server and avoid system crashes.

[0169] Number of data transmissions: From June 15, 2024 to June 30, 2024, the number of data transmissions using traditional methods is 50,000 times, while the method of the present invention only requires 10,000 times, significantly reducing the transmission of data between storage, computing and training modules.

[0170] System energy consumption: By reducing the number of data transmissions, the system energy consumption of the method of the present invention is reduced to 80kWh, which is 60% lower than the 200kWh of the traditional method.

[0171] Training time: During the model training process, the traditional method takes 5 hours to complete the training, while the method of the present invention only takes 2 hours, which greatly improves the training efficiency.

[0172] Data loss rate: The data loss rate of the traditional method is 0.05%, while the data loss rate of the method of the present invention is reduced to 0.01%, which effectively ensures data security.

[0173] User response time: In the real-time prediction and recommendation process, the user response time of the traditional method is 200ms, while the method of the present invention shortens the response time to 100ms, significantly improving the user experience.

[0174] Through the above specific embodiments and detailed data comparison, the method of the present invention has shown significant technical advantages and effects in the actual application of large-scale e-commerce platforms. By reducing the number of data transmissions, reducing system energy consumption, shortening training time, reducing data loss rate and improving user response time, the method of the present invention effectively solves the problems of high energy consumption, complex data transmission and data loss risk existing in traditional cloud computing servers, and realizes the comprehensive optimization of cloud computing servers in high-performance computing and data processing.

[0175] Example 1 demonstrates the feasibility and effectiveness of the storage, computing and training integrated cloud computing server optimization method based on human brain neural network in practical applications, and provides an important reference and reference for the future development of cloud computing technology.

[0176] The present invention integrates storage, computing and training functions into a unified architecture, effectively reducing the number and distance of data transmission between modules, and significantly reducing the overall energy consumption of the system. In traditional cloud computing architecture, the frequent transmission of data between different modules leads to high energy consumption. The present invention realizes on-site processing and training of data through the integrated storage, computing and training architecture design, reduces the need for data movement, and thus improves the energy efficiency of the system. In addition, the use of new processor design and optimized data channel design further improves computing efficiency and resource utilization, allowing the system to process more data and complete more complex computing tasks at the same power consumption.

[0177] The present invention reduces the frequency of data transmission between storage, computing and training modules through adaptive data routing algorithms and optimized data transmission path design, and reduces the complexity and risk of data transmission. The adaptive data routing algorithm can dynamically adjust the data transmission path according to the actual workload and network status to ensure the security and reliability of data during transmission. At the same time, the storage, computing and training integrated architecture reduces the data transmission distance, reduces the possibility of data loss or damage, and further improves the data security and system stability.

[0178] The present invention integrates an intelligent fault detection and repair module in the adaptive learning model, which can monitor the system status in real time, detect and repair faults. Traditional fault detection methods rely on predefined fault models and rules and cannot adapt to complex and changeable fault conditions. The present invention can dynamically adjust the detection and repair strategies through self-learning and adaptive algorithms, thereby improving the accuracy of fault detection and the timeliness of repair. Through the analysis of historical data and current status, the adaptive learning model can autonomously adjust various parameters to achieve self-optimization and performance improvement of the system, ensuring the stable operation of the system under high load and complex environment.

[0179] The present invention has made many innovations in the design of neural network architecture and optimization algorithm. An adaptive learning model based on human brain neural network is designed, and a multi-level neuron structure is adopted to realize the organic combination of storage, computing and training functions. Through the introduction of deep learning technology and quantum computing unit, the adaptive learning model architecture of neural network is optimized, and the computing speed and processing power are improved. In addition, a time-dependent weight adjustment strategy is introduced in the synaptic plasticity mechanism, and the weight change is adjusted by the time decay function to ensure the stability and convergence speed of model training. Finally, through the intelligent fault detection and repair module, the pattern recognition and self-learning ability of human brain neural network are utilized to realize real-time monitoring and automatic repair of system faults, thereby improving the reliability and self-optimization ability of the system.

[0180] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A storage, computing and training integrated cloud computing server optimization method based on human brain neural network, characterized in that: The steps include: S1. Design an adaptive learning model based on human brain neural network. The adaptive learning model contains multiple levels of neurons, each of which has the functions of storage, calculation and training; S2, integrating storage, computing and training functions in cloud computing servers to form an integrated architecture; S3, adopts a new processor design, the processor includes multi-functional cores, each core can independently complete storage, computing and training tasks, and optimizes the data channel design; S4. Optimize the adaptive learning model architecture of the human brain neural network through deep learning technology; S5, dynamically adjust the allocation of computing resources based on real-time data flow and computing needs; S6. Optimize the data transmission path by designing an adaptive data routing algorithm; S7. Using the synaptic plasticity mechanism of the adaptive learning model, self-optimization is achieved in the cloud computing server. The adaptive learning model can autonomously adjust parameters and optimize performance based on historical data and current status. S8. Integrate intelligent fault detection and repair modules into the adaptive learning model, and use the self-learning ability of the human brain neural network to monitor the system status in real time, detect and repair faults.

2. According to claim 1, a storage, computing and training integrated cloud computing server optimization method based on human brain neural network is characterized in that: The S1 includes: S11. Design an adaptive learning model based on the human brain neural network. The adaptive learning model adopts a heterogeneous computing architecture and contains multiple levels of neurons. Each neuron has storage, computing and training functions: Among them, A ij represents the activation value of the i-th neuron in the j-th layer, W ik represents the weight from the i-th neuron to the k-th neuron, X kj represents the input data, B i represents the bias, T ij represents the temporal dependency, reflecting the temporal effect of dynamically adjusting weights and biases; S12. In a heterogeneous computing architecture, each neuron is composed of different types of computing units, including general-purpose processors, graphics processors, and field-programmable gate arrays, to achieve optimized processing of different types of computing tasks: P total =α·P CPU +β·P GPU +γ·P FPGA +λ·C comm ; Among them, P total Represents the total computing power, P CPU , P GPU and P FPGA They represent the computing power of CPU, GPU and FPGA respectively, α, β, γ are the weight coefficients of each computing unit, and C comm represents the communication overhead, λ is the communication overhead coefficient; S13, the storage unit uses a three-dimensional stacked memory to store the input and output data of the neuron; Among them, M represents the state of the storage unit, S i represents the capacity of the storage unit, V represents the input voltage, V i represents the operating voltage of the storage unit, σ represents the standard deviation of the voltage, and δ is the offset value of the storage unit; S14. The computing unit implements parallel computing and accelerated computing through different types of processors. General-purpose processors are used to handle control and simple computing tasks, graphics processors are used to handle large-scale parallel computing tasks, and field programmable gate arrays are used to accelerate specific algorithms: Among them, Q represents the total computing power, v jk and k Represent weights and input data respectively, c j represents bias, σ1 represents activation function, τ1 represents time constant, u i Indicates a time point, u0 indicates the current time; S15, the training unit uses an adaptive synaptic plasticity mechanism to adjust the weights and biases of neurons through the error back propagation algorithm: Where Δv i represents the change in weight, η1 represents the learning rate, z j Indicates the actual output, Represents the predicted output, y ij represents input data, σ2 represents activation function, b j represents the net input of the neuron, φ1(t) represents the time-dependent regulation function; S16. In a heterogeneous computing architecture, each computing unit transmits and shares data through an interconnected network: Among them, U represents the data transmission time, E represents the data volume, F represents the bandwidth, and P loss2 represents the probability of data loss, L1 represents the basic transmission delay, d1 represents the transmission distance, and d 1,0 Indicates the reference distance.

3. The method for optimizing a storage, computing and training integrated cloud computing server based on a human brain neural network according to claim 1, characterized in that: The S4 includes: S41. In the storage-computing-training integrated architecture, a hybrid design of quantum computing units and traditional computing units is adopted. Quantum computing units are used to process highly complex training tasks, and traditional computing units are used for daily data processing and storage: Among them, total represents the total state of the hybrid quantum-classical computing system, α1 and β1 represent the contribution coefficients of the quantum computing unit and the traditional computing unit, respectively, |ψ i > and |φ j > respectively represent the states of quantum bits and classical bits, n and m represent the number of quantum bits and classical bits, respectively; S42. Design an adaptive learning model based on a quantum neural network. The quantum neural network contains multiple qubits, each of which represents a neuron state: Among them, Ψ represents the total state of the quantum neural network, α i and β j denote the weights of quantum bits and classical bits respectively, |χ j > and |η j > represents the entangled state of the jth quantum bit and the classical bit, n and m represent the number of quantum bits and the number of classical bits respectively; S43. In the quantum computing unit, quantum entanglement and quantum superposition are used to achieve efficient parallel computing, and the weights and biases of the neural network are updated through quantum gate operations: in, represents the quantum gate operation matrix, θ1, λ1 represents the angle parameter that controls the rotation of the quantum state; S44. During each training process, the quantum tunneling effect is used to accelerate weight updates: Where, ΔW ij represents the weight change, η2 represents the learning rate, represents the partial derivative of the loss function with respect to the weight, d represents the tunneling distance, represents Planck's constant, ω represents frequency, and t represents time; S45. Use the ultra-high computing speed of quantum computing units to quickly process and train large-scale data, while using traditional computing units to verify and store results: Among them, T process represents the total processing time, Q speed and C speed They represent the processing speeds of the quantum computing unit and the traditional computing unit respectively, N represents the data scale, and M represents the amount of data that the traditional computing unit needs to process. S46. Introducing the time crystal effect into quantum neural networks, using the periodic oscillation characteristics of time crystals to enhance the dynamic adjustment capability of the adaptive learning model: H(t)=H0+Vcos(ωt)+γsin 2 (ω1t / 2); Where H(t) represents the time-dependent Hamiltonian, H0 represents the initial Hamiltonian, V represents the perturbation term, γ represents the oscillation frequency, and ω1 represents the nonlinear effect coefficient of the time crystal; S47. Through the hybrid quantum-classical architecture, fast data storage, computing and training can be achieved, improving the overall performance and energy efficiency of cloud computing servers: Among them, P total represents the total computing power, α i1 and β j1 They represent the contribution coefficients of quantum computing unit and traditional computing unit, Q i and C j They represent the computing power of quantum computing unit and traditional computing unit respectively, λ2 represents the data transmission coefficient, and γ k Indicates the efficiency of the transmission channel, D k represents the data transmission volume, and n, m, and p represent the number of quantum computing units, traditional computing units, and data transmission channels, respectively.

4. The method for optimizing a cloud computing server based on a human brain neural network according to claim 1, characterized in that: The S7 includes: S71. In the adaptive learning model, design a parameter adjustment algorithm based on synaptic plasticity mechanisms, including long-term potentiation and long-term depression, to dynamically adjust the connection strength between neurons; S72, using historical data and current status, calculating adjustment values ​​of synaptic weights, updating weights according to these adjustment values, and dynamically adapting to the current computing environment; S73. In the synaptic plasticity mechanism, a time-dependent weight adjustment strategy is introduced to adjust the weight change through the time decay function; S74, through the synaptic plasticity mechanism in the adaptive learning model, real-time monitoring and analysis of the operating status of the cloud computing server, and self-optimization based on the results of the optimization function; S75. In the adaptive learning model, a feedback mechanism is used to adjust parameters according to the results of the optimization function; S76. Through the dynamic adjustment of synaptic plasticity mechanism, cloud computing servers can be self-optimized under different workloads and environments. S77. The adaptive learning model can autonomously adjust various parameters and optimize performance based on the historical data and current status of the cloud computing server.

5. The method for optimizing a cloud computing server based on a human brain neural network for storage, computing and training according to claim 1, characterized in that: The S8 includes: S81. Integrate intelligent fault detection module into the adaptive learning model, use the self-learning ability of human brain neural network, and detect faults in system operation through real-time monitoring of system status data: Among them, F d (t) represents the fault detection result at time t, N represents the number of neurons in the detection module, and α i is the weight coefficient of fault detection, f i (t) is the function of system state data, M is the number of input parameters, w ij is the weight, x j (t) is the jth input parameter, b i is the bias, σ is the activation function; S82, the intelligent fault detection module collects and analyzes the status parameters of each component of the system, including processor utilization, memory usage, network latency, and memory read and write speed, to form a basic data set for fault detection: Among them, S(t) represents the set of system state parameters at time t, U cpu (t′) is the processor utilization at time t′, C cpu is the processor capacity, M mem (t′) is the memory usage at time t′, C mem is the memory capacity, D net (t′) is the network delay at time t′, L net is the network bandwidth, R stor (t′) is the memory read and write speed at time t′, S stor is the memory capacity; S83. Utilize the pattern recognition capability of the human brain neural network to analyze the collected state parameters, identify abnormal patterns and determine whether there is a fault. The recognition process is based on the trained neural network model and can adaptively update and improve the detection algorithm: Among them, P fault (t) represents the failure probability at time t, Q is the number of layers of the neural network, β k is the weight coefficient of each layer, tanh is the hyperbolic tangent activation function, γ kl is the connection weight in each layer, S l (t) is the lth system state parameter; S84. When a fault is detected, the intelligent fault detection module generates a fault report, indicating the specific location and type of the fault, and triggers the fault repair module to perform automatic repair; S85, the fault repair module uses the self-learning ability of the human brain neural network to select an appropriate repair strategy based on the information in the fault report. The repair strategy includes reallocating computing resources, restarting faulty components, and adjusting system parameters. S86. During the fault repair process, the fault repair module monitors the repair effect in real time and adjusts the repair strategy according to the actual situation; S87. After the fault is repaired, the system records the fault detection and repair process, updates the parameters in the adaptive learning model, and optimizes future fault detection and repair capabilities.

Images