An AI-based intelligent energy consumption control system and method for computer rooms

Abstract

The present invention discloses an AI-based intelligent energy consumption control system and method for a computer room, which relates to the technical field of big data analysis. The present invention collects historical business load and energy consumption data by setting an energy consumption monitoring period, constructs a computer room energy consumption monitoring data set indexed by time series, and uses a neural network model to train the mapping relationship between business load and energy consumption, thereby solving the problem of lack of time series modeling in data association analysis; selects the total load of a reference period and the average load of subsequent periods as dual benchmarks, generates a benchmark business load through weighted fusion, combines the dual models of business load mapping and time series prediction to predict energy consumption, breaking through the limitations of traditional single benchmarks and single models; uses a normalized weight distribution method based on the inverse of historical errors to fuse the dual predicted energy consumption to generate a comprehensive total energy consumption, triggers a load balancing algorithm when the threshold is exceeded, regulates the business load, and reduces energy consumption through iterative adjustment, thereby achieving precise energy consumption control and intelligent resource scheduling.

Description

Technical Field

[0001] The present invention relates to the technical field of big data analysis, and specifically to an AI-based intelligent energy consumption control system and method for a computer room. Background Art

[0002] Computer rooms are the core infrastructure of the digital age, providing a secure and stable operating environment for servers and ensuring the continuity of data processing, storage, and transmission. From the real-time operation of enterprise business systems to the efficient delivery of cloud computing services, from the storage and analysis of massive amounts of data to disaster recovery and redundancy for critical business operations, all rely on the computer room's power supply, cooling system, network architecture, and security protection system. They are the physical cornerstone supporting the modern information and digital society, playing a decisive role in enterprise operational efficiency, data security, and service continuity.

[0003] Current energy consumption control of server rooms faces significant technical bottlenecks: as core energy consumption units, servers have energy consumption characteristics significantly affected by differences in hardware model, physical layout, and business load, resulting in strong heterogeneity in the energy consumption performance of different servers. Traditional control methods rely on fixed rules or simple statistical models, which are unable to adapt to the differentiated energy consumption baselines of server hardware parameters, and also have difficulty capturing the heat dissipation coupling effects between different servers. This leads to a lag between control strategies and actual energy consumption needs, and is generally subject to problems of insufficient control accuracy or delayed response. Therefore, there is an urgent need for a sophisticated intelligent control method for energy consumption in computer rooms that can achieve precise energy consumption control through in-depth data analysis and intelligent prediction. Summary of the Invention

[0004] The purpose of the present invention is to provide an AI-based intelligent energy consumption control system and method for a computer room to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solutions: an AI-based intelligent control method for energy consumption in a computer room, wherein the intelligent management method comprises the following steps:

[0006] Step S1: Set an energy consumption monitoring cycle, and obtain the historical business load data and energy consumption data of the computer room based on the energy consumption monitoring cycle, arrange and store the obtained business load data and energy consumption data according to the time series relationship, and construct a computer room energy consumption monitoring data set;

[0007] Step S1-1: Analyze the monitoring log files of the computer room energy consumption based on the energy consumption monitoring cycle to obtain business load data and energy consumption data in each historical energy consumption monitoring cycle of the computer room; the business load data is represented by a set of indicators of resource usage of each server in the computer room when processing business tasks during operation, including computing load, storage load, and network load; the energy consumption data is represented by the total power consumption of each server in the computer room during operation;

[0008] Step S1-2: Analyze the monitoring log file to obtain timestamp data, and store the business load data and energy consumption data in each historical energy consumption monitoring cycle in a relational database in chronological order using the timestamp data. Select the timestamp data as the primary key, and select the business load data and energy consumption data as field values. Construct a structured data set indexed by time series in the relational database, which is recorded as the computer room energy consumption monitoring data set.

[0009] By setting an energy consumption monitoring cycle, extracting business load and energy consumption data from monitoring log files, and storing them in a relational database with timestamps as indexes to build a structured data set, key data on computer room operations can be systematically and time-series retained, facilitating subsequent in-depth mining of the correlation between business load and energy consumption, and providing accurate and orderly basic data support for data-driven intelligent energy consumption regulation.

[0010] Step S2: Analyze the relationship between service load data and energy consumption data based on the data set of the computer room energy consumption monitoring data to obtain a mapping relationship between the service load and energy consumption of the computer room; and analyze the historical service load data in the data set of the computer room energy consumption monitoring data to select the energy consumption monitoring period with the highest total service load data and mark it as the reference energy consumption monitoring period;

[0011] Step S2-1: traverse and read the service load data and energy consumption data in the computer room energy consumption monitoring data set, normalize the service load data and energy consumption data, and use them as input data for the neural network model. After training the neural network model, a mapping relationship between the service load and energy consumption of the computer room is obtained;

[0012] The calculation formula for the mapping relationship between the business load and energy consumption of the computer room is as follows:

[0013] ;

[0014] Where E represents the energy consumption corresponding to the input business load L; g represents the model parameter set; L represents the business load of the input room business load and energy consumption mapping relationship; W （i） Represented as the weight matrix of the i-th layer; b （i） Represented as the bias vector of the i-th layer; z hidden It is represented as the hidden layer activation function; z output represents the output layer activation function; k represents the number of network layers, k=3, and the network layers include input layer, output layer and hidden layer;

[0015] Step S2-2: After normalizing the read service load data, the service load values ​​are added to obtain the total service load; based on the total service load of each energy consumption monitoring period in the data set of the computer room energy consumption monitoring data, the energy consumption monitoring period with the highest total service load data is selected and marked as the reference energy consumption monitoring period;

[0016] By normalizing the business load and energy consumption data in the computer room energy consumption monitoring data set and inputting them into the neural network model training, an accurate load-energy consumption mapping relationship can be constructed. At the same time, the reference period with the highest total business load can be screened. This not only realizes the deep modeling of dynamic correlations between data, but also provides a peak load reference for subsequent benchmark calculations, laying a dual foundation for data correlation analysis and benchmark anchoring for energy consumption prediction and regulation.

[0017] Step S3: Select the total service load of the reference energy consumption monitoring period as the first benchmark service load; select the arithmetic average of the service loads from the energy consumption monitoring period next to the reference energy consumption monitoring period to the energy consumption monitoring period before the current energy consumption monitoring period as the second benchmark service load; and calculate the benchmark service load of the current energy consumption monitoring period through weighted fusion based on the first benchmark service load and the second benchmark service load.

[0018] Step S3-1: Select the total service load of the reference energy consumption monitoring period as the first benchmark service load of the computer room in the current energy consumption monitoring period; and analyze and read the end timestamp data of the reference energy consumption monitoring period. Based on the analyzed end timestamp data, sequentially obtain the total load data of each energy consumption monitoring period, up to the energy consumption monitoring period before the current energy consumption monitoring period;

[0019] Step S3-2: Calculate the arithmetic mean of the total load of each energy consumption monitoring period to obtain a second benchmark service load of the computer room in the current energy consumption monitoring period;

[0020] Step S3-3, assigning weight coefficients to the first reference service load and the second reference service load respectively, and the weight assignments satisfy that the sum of the weight coefficient of the first reference service load and the weight coefficient of the second reference service load is equal to one;

[0021] Step S3-4: multiplying the first benchmark service load by the first weight coefficient to obtain a first weighted value of the benchmark service load; multiplying the second benchmark service load by the second weight coefficient to obtain a second weighted value of the benchmark service load; and adding the first weighted value and the second weighted value to obtain the benchmark service load of the computer room in the current energy consumption monitoring period;

[0022] The weighted fusion calculation formula is as follows:

[0023] ;

[0024] Where, L level It represents the benchmark service load of the computer room in the current energy consumption monitoring cycle; L1 represents the first benchmark service load; q1 represents the weight coefficient of the first benchmark service load; L2 represents the second benchmark service load; q2 represents the weight coefficient of the second benchmark service load;

[0025] By selecting the total load of the reference period as the first benchmark and the average load of subsequent periods as the second benchmark, and using weighted fusion to calculate the benchmark business load of the current period, we not only anchor the historical peak load characteristics, but also combine the recent load change trends. Through the dynamic weighting of the two benchmarks, the benchmark load can more flexibly adapt to business fluctuations, providing energy consumption forecasts with load benchmark parameters that are more in line with actual business scenarios.

[0026] Step S4: Based on the benchmark business load, the total energy consumption of the computer room in the current energy consumption monitoring period is obtained by mapping the business load of the computer room to the energy consumption, which is recorded as the business load mapped total energy consumption; based on the computer room energy consumption monitoring data set, the total energy consumption in the current energy consumption monitoring period is predicted by the time series prediction model, which is recorded as the time series predicted total energy consumption;

[0027] Step S4-1: Input the benchmark service load obtained in step S3 into the trained mapping relationship between the service load and energy consumption of the computer room, and output the total energy consumption of the computer room corresponding to the benchmark service load through the mapping relationship, which is recorded as the service load mapped total energy consumption;

[0028] Step S4-2: Clean the data in the computer room energy consumption monitoring data set, normalize it, and input it into the time series prediction model to predict the total energy consumption in the current energy consumption monitoring cycle, which is recorded as the time series predicted total energy consumption;

[0029] The calculation formula for the total energy consumption predicted by the time series prediction model is as follows:

[0030] ;

[0031] Where y(t) represents the total predicted energy consumption in the next energy consumption monitoring period t, which is recorded as the total time series predicted energy consumption; h represents the basic growth rate; m represents the offset; r a Expressed as trend mutation adjustment amount; e a (t) represents the trend mutation indicator function; N represents the order of the Fourier series; P represents the time length of the energy consumption monitoring cycle; e n It is represented by the fluctuation amplitude generated by Fourier through the data learning cycle in the data set of the computer room energy consumption monitoring data; u nIt is represented by the phase generated by Fourier through the data learning cycle in the data set of the computer room energy consumption monitoring data; v t Represented as random noise;

[0032] The total energy consumption of the business load mapping is calculated by combining the baseline business load with the load-energy consumption mapping relationship. At the same time, a time series model is used to predict historical energy consumption data, forming a two-dimensional energy consumption prediction mechanism. This mechanism deduces from the essential relationship between business load and energy consumption, and also predicts based on the time series patterns of energy consumption data itself. The two prediction methods complement each other and provide multi-perspective prediction data support for comprehensive energy consumption assessment.

[0033] Step S5: Calculate the total energy consumption of the computer room in the current energy consumption monitoring cycle through weighted fusion based on the total energy consumption mapped by the business load and the total energy consumption predicted by the time series, and regulate the business load based on the total energy consumption of the computer room;

[0034] Step S5-1: Calculate the mean absolute error (MAE) between the total energy consumption of the business load mapping and the total energy consumption of the time series prediction in each historical energy consumption monitoring period and the actual total energy consumption of the computer room, and assign weight coefficients to the total energy consumption of the business load mapping and the total energy consumption of the time series prediction using a normalized weight allocation method based on the inverse of the MAE; the sum of the weight coefficients of the total energy consumption of the business load mapping and the total energy consumption of the time series prediction equals one;

[0035] The inverse of the mean absolute error is used for weight distribution using the following formula:

[0036] ;

[0037] ;

[0038] Where q map It is the weight coefficient of the total energy consumption of business load mapping; q time Expressed as the weight coefficient of the total energy consumption of time series prediction; MAE map Expressed as the mean absolute error between the total energy consumption of the business load mapping and the actual total energy consumption of the computer room; MAE time It is expressed as the average absolute error between the total energy consumption predicted by the time series and the actual total energy consumption of the computer room; q map With q time The result of the addition is equal to one;

[0039] The basic formula for calculating the mean absolute error is as follows:

[0040] ;

[0041] Where X represents the total number of historical energy consumption monitoring cycles; E actual，b It is expressed as the actual total energy consumption of the bth energy consumption monitoring cycle; Epred，b It is expressed as the total predicted energy consumption, including the total energy consumption of business load mapping and the total energy consumption of time series prediction. When calculating the mean absolute error, only one of the parameters is included in the calculation.

[0042] Step S5-2: Based on the assigned business load mapping total energy consumption weight coefficient and the time series predicted total energy consumption weight coefficient, the product of the business load mapping total energy consumption and the time series predicted total energy consumption multiplied by the corresponding weights is added together to obtain the total comprehensive energy consumption of the computer room in the current energy consumption monitoring cycle;

[0043] Step S5-3: Based on the calculated total energy consumption of the computer room in the current energy consumption monitoring cycle, energy consumption control is performed in combination with the threshold value set for the computer room energy consumption. The specific process of energy consumption control is as follows:

[0044] Step S5-3-1: When the total comprehensive energy consumption does not exceed the threshold set for the energy consumption of the computer room, it is determined that the energy consumption is normal and the energy consumption of the computer room is continuously monitored;

[0045] Step S5-3-2: When the total comprehensive energy consumption exceeds the threshold set for the energy consumption of the computer room, it is determined to be energy consumption abnormality. The load balancing algorithm is used to redistribute the service load according to the current number of connections, processing performance weights, and service priorities of each server in the computer room. The specific process of redistributing the service load is as follows:

[0046] Based on the priority of the server queue, low-priority business loads are migrated from high-energy-consuming servers to low-load servers. After the load migration is completed, the actual total energy consumption of the computer room is recalculated and, through an iterative adjustment mechanism, the total energy consumption of the computer room is brought back to the set threshold range.

[0047] By dynamically allocating weights based on historical errors and fusing the two predicted energy consumptions, a comprehensive total energy consumption assessment is formed, and then combined with thresholds for intelligent regulation. This not only avoids the limitations of a single prediction model, but also dynamically adjusts the business load according to the real-time energy consumption status, realizing closed-loop management from data collection, analysis, prediction to execution, and ultimately achieving precise control of the energy consumption of the computer room and optimal resource allocation, significantly improving energy utilization efficiency.

[0048] Furthermore, an AI-based intelligent energy consumption control system for a computer room includes a data acquisition module, a relationship analysis module, a benchmark calculation module, an energy consumption prediction module, and an energy consumption control module;

[0049] The data acquisition module is used to set the energy consumption monitoring cycle, obtain the historical business load data and energy consumption data of the computer room, and arrange and store them in a time series relationship to construct an energy consumption monitoring data set; the relationship analysis module is used to analyze the relationship between the business load data and the energy consumption data, generate a mapping relationship, and select the reference energy consumption monitoring cycle with the highest total business load; the benchmark calculation module is used to use the total load of the reference period as the first benchmark and the average load of subsequent periods as the second benchmark, and calculate the benchmark business load of the current period through weighted fusion; the energy consumption prediction module is used to calculate the total energy consumption of the business load mapping based on the benchmark business load and the mapping relationship, and at the same time predict the total energy consumption of the time series prediction through the time series model; the energy consumption control module is used to weightedly fuse the two predicted energy consumptions to obtain the comprehensive total energy consumption, and regulate the business load according to the threshold;

[0050] The data acquisition module includes a data acquisition unit and a time series storage unit; the data acquisition unit is used to extract the computing load, storage load, network load and total power consumption data of each historical period from the monitoring log file; the time series storage unit is used to store the business load and energy consumption data in a relational database based on the timestamp data, and construct a structured data set with time series as the index;

[0051] The relationship analysis module includes a mapping modeling unit and a cycle screening unit; the mapping modeling unit is used to normalize the business load and energy consumption data and generate a mapping relationship between load and energy consumption through neural network model training; the cycle screening unit is used to compare the total business load of each energy consumption monitoring period, screen and mark the reference period with the highest total load;

[0052] The benchmark calculation module includes a dual-benchmark determination unit and a weighted fusion unit; the dual-benchmark determination unit is used to select the total load of the reference period as the first benchmark and calculate the arithmetic mean of the total load of subsequent periods as the second benchmark; the weighted fusion unit is used to assign weights to the dual-benchmark loads and generate the current period benchmark service load through weighted calculation;

[0053] The energy consumption prediction module includes a mapping calculation unit and a time series prediction unit; the mapping calculation unit is used to input the benchmark business load into the trained mapping relationship model and output the corresponding business load mapping energy consumption; the time series prediction unit is used to clean and normalize the energy consumption monitoring data set and predict the total energy consumption of the current period through the time series model;

[0054] The energy consumption control module includes a comprehensive evaluation unit and a load adjustment unit; the comprehensive evaluation unit is used to calculate the historical error, assign weights by the error inverse normalization method, and generate the total comprehensive energy consumption; the load adjustment unit is used to migrate low-priority loads according to the number of server connections, performance weights and business priorities through a load balancing algorithm when the comprehensive energy consumption exceeds the threshold, so that the energy consumption falls back to the threshold range;

[0055] The output end of the data acquisition module is electrically connected to the input end of the relationship analysis module; the output end of the relationship analysis module is electrically connected to the input end of the benchmark calculation module; the output end of the benchmark calculation module is electrically connected to the input end of the energy consumption prediction module; the output end of the energy consumption prediction module is electrically connected to the input end of the energy consumption control module.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. The present invention collects historical business load and energy consumption data by setting an energy consumption monitoring cycle, constructs a computer room energy consumption monitoring data set indexed by time series, and uses a neural network model to train the mapping relationship between business load and energy consumption. It solves the problems of lack of time series modeling and inaccurate mapping relationship capture in data association analysis in the existing technology, and provides structured data support and dynamic association models for energy consumption regulation.

[0058] 2. The present invention selects the total load of the reference period as the first benchmark and the average load of subsequent periods as the second benchmark, generates the current period benchmark business load through weighted fusion, and combines the business load mapping and time series prediction dual models to perform energy consumption prediction, breaking through the limitations of traditional single benchmark and single prediction model, realizing dynamic adaptation to business fluctuations, and improving the accuracy and timeliness of energy consumption prediction.

[0059] 3. The present invention adopts a normalized weight distribution method based on the inverse of historical errors to weightedly fuse the total energy consumption of business load mapping and the total energy consumption of time series prediction to generate a comprehensive total energy consumption. When the comprehensive energy consumption exceeds the threshold, the load balancing algorithm is triggered to migrate low-priority businesses from high-energy-consuming servers to low-load servers according to the number of server connections, processing performance weights and business priorities, and the energy consumption is brought back to the threshold range through an iterative adjustment mechanism, thereby solving the energy waste problem caused by the single prediction model and extensive load distribution in traditional regulation, and realizing precise control of energy consumption and intelligent scheduling of server resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flow chart of an AI-based intelligent energy consumption control method for a computer room according to the present invention;

[0061] Figure 2 This is a structural diagram of an AI-based intelligent energy consumption control system for a computer room according to the present invention. DETAILED DESCRIPTION

[0062] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0063] Example 1: Figure 1 As shown, the present invention provides a technical solution, an AI-based intelligent control method for energy consumption in a computer room, and the intelligent management method includes the following steps:

[0064] Step S1: Set an energy consumption monitoring cycle, and obtain the historical business load data and energy consumption data of the computer room based on the energy consumption monitoring cycle, arrange and store the obtained business load data and energy consumption data according to the time series relationship, and construct a computer room energy consumption monitoring data set;

[0065] Step S1-1: Analyze the monitoring log files of the computer room energy consumption based on the energy consumption monitoring cycle to obtain business load data and energy consumption data in each historical energy consumption monitoring cycle of the computer room; the business load data is represented by a set of indicators of resource usage of each server in the computer room when processing business tasks during operation, including computing load, storage load, and network load; the energy consumption data is represented by the total power consumption of each server in the computer room during operation;

[0066] Step S1-2: Analyze the monitoring log file to obtain timestamp data, and store the business load data and energy consumption data in each historical energy consumption monitoring cycle in a relational database in chronological order using the timestamp data. Select the timestamp data as the primary key, and select the business load data and energy consumption data as field values. Construct a structured data set indexed by time series in the relational database, which is recorded as the computer room energy consumption monitoring data set.

[0067] In the specific implementation, the energy consumption monitoring cycle is set to 1 day. Taking a large cloud data center as an example, by parsing the hourly monitoring log files, the computing load (such as CPU utilization), storage load (such as disk IOPS), network load (such as bandwidth usage) and total power consumption data of each server are extracted and stored in the MySQL database in timestamp order. The table structure is established with the timestamp as the primary key to ensure that the data is arranged in minute-level time series to avoid data dislocation due to log parsing errors. At the same time, historical data loss is prevented by regularly backing up the database.

[0068] Step S2: Analyze the relationship between service load data and energy consumption data based on the data set of the computer room energy consumption monitoring data to obtain a mapping relationship between the service load and energy consumption of the computer room; and analyze the historical service load data in the data set of the computer room energy consumption monitoring data to select the energy consumption monitoring period with the highest total service load data and mark it as the reference energy consumption monitoring period;

[0069] Step S2-1: traverse and read the service load data and energy consumption data in the computer room energy consumption monitoring data set, normalize the service load data and energy consumption data, and use them as input data for the neural network model. After training the neural network model, a mapping relationship between the service load and energy consumption of the computer room is obtained;

[0070] Step S2-2: After normalizing the read service load data, the service load values ​​are added to obtain the total service load; based on the total service load of each energy consumption monitoring period in the data set of the computer room energy consumption monitoring data, the energy consumption monitoring period with the highest total service load data is selected and marked as the reference energy consumption monitoring period;

[0071] In the specific implementation, Min-Max normalization is used to map the load and energy consumption data to the interval [0, 1]. Taking the BP neural network as an example, three hidden layers are set, and the number of nodes in each layer is 1.5 times the number of input features. The mapping relationship model is obtained by training with the Adam optimizer for 1000 rounds. When screening the reference period, the total business load of each quarter is traversed. For example, the load peak period of an e-commerce platform during the promotion season should be considered. It is necessary to exclude abnormal load fluctuations caused by temporary tasks to ensure that the selected reference period is the normal business peak period.

[0072] Step S3: Select the total service load of the reference energy consumption monitoring period as the first benchmark service load; select the arithmetic average of the service loads from the energy consumption monitoring period next to the reference energy consumption monitoring period to the energy consumption monitoring period before the current energy consumption monitoring period as the second benchmark service load; and calculate the benchmark service load of the current energy consumption monitoring period through weighted fusion based on the first benchmark service load and the second benchmark service load.

[0073] Step S3-1: Select the total service load of the reference energy consumption monitoring period as the first benchmark service load of the computer room in the current energy consumption monitoring period; and analyze and read the end timestamp data of the reference energy consumption monitoring period. Based on the analyzed end timestamp data, sequentially obtain the total load data of each energy consumption monitoring period, up to the energy consumption monitoring period before the current energy consumption monitoring period;

[0074] Step S3-2: Calculate the arithmetic mean of the total load of each energy consumption monitoring period to obtain a second benchmark service load of the computer room in the current energy consumption monitoring period;

[0075] Step S3-3, assigning weight coefficients to the first reference service load and the second reference service load respectively, and the weight assignments satisfy that the sum of the weight coefficient of the first reference service load and the weight coefficient of the second reference service load is equal to one;

[0076] Step S3-4: multiplying the first benchmark service load by the first weight coefficient to obtain a first weighted value of the benchmark service load; multiplying the second benchmark service load by the second weight coefficient to obtain a second weighted value of the benchmark service load; and adding the first weighted value and the second weighted value to obtain the benchmark service load of the computer room in the current energy consumption monitoring period;

[0077] In specific implementation, if the reference period is a historical high-load period, extract its total business load as the first benchmark, obtain all the total loads after the period and before the current period in timestamp order, and calculate the arithmetic mean as the second benchmark; the initial weight can be allocated according to the business characteristics, such as increasing the first benchmark weight when it is biased towards peak load, and the weight sum must meet the normalization conditions; it is necessary to ensure that the time range is clearly defined to avoid overlapping of previous and subsequent cycle data. The period interval can be strictly divided by the timestamp index. For example, starting from the last timestamp of the reference period, the subsequent period data is extracted in sequence until the time point before the current period to ensure the timing accuracy of the benchmark calculation.

[0078] Step S4: Based on the benchmark business load, the total energy consumption of the computer room in the current energy consumption monitoring period is obtained by mapping the business load of the computer room to the energy consumption, which is recorded as the business load mapped total energy consumption; based on the computer room energy consumption monitoring data set, the total energy consumption in the current energy consumption monitoring period is predicted by the time series prediction model, which is recorded as the time series predicted total energy consumption;

[0079] Step S4-1: Input the benchmark service load obtained in step S3 into the trained mapping relationship between the service load and energy consumption of the computer room, and output the total energy consumption of the computer room corresponding to the benchmark service load through the mapping relationship, which is recorded as the service load mapped total energy consumption;

[0080] Step S4-2: Clean the data in the computer room energy consumption monitoring data set, normalize it, and input it into the time series prediction model to predict the total energy consumption in the current energy consumption monitoring cycle, which is recorded as the time series predicted total energy consumption;

[0081] During implementation, the benchmark load is input into the trained LSTM neural network mapping model, the number of memory units is set to 128, and the business load mapping energy consumption is output. The 3σ principle is used to clean outliers in historical energy consumption data, and the current cycle energy consumption is predicted using the Prophet time series model, with annual and weekly seasonal parameters set. Data normalization must be consistent with that during training. For example, energy consumption data must be normalized to the range [0, 1000] to avoid prediction bias due to scale differences.

[0082] Step S5: Calculate the total energy consumption of the computer room in the current energy consumption monitoring cycle through weighted fusion based on the total energy consumption mapped by the business load and the total energy consumption predicted by the time series, and regulate the business load based on the total energy consumption of the computer room;

[0083] Step S5-1: Calculate the mean absolute error (MAE) between the total energy consumption of the business load mapping and the total energy consumption of the time series prediction in each historical energy consumption monitoring period and the actual total energy consumption of the computer room, and assign weight coefficients to the total energy consumption of the business load mapping and the total energy consumption of the time series prediction using a normalized weight allocation method based on the inverse of the MAE; the sum of the weight coefficients of the total energy consumption of the business load mapping and the total energy consumption of the time series prediction equals one;

[0084] Step S5-2: Based on the assigned business load mapping total energy consumption weight coefficient and the time series predicted total energy consumption weight coefficient, the product of the business load mapping total energy consumption and the time series predicted total energy consumption multiplied by the corresponding weights is added together to obtain the total comprehensive energy consumption of the computer room in the current energy consumption monitoring cycle;

[0085] Step S5-3: Based on the calculated total energy consumption of the computer room in the current energy consumption monitoring cycle, energy consumption control is performed in combination with the threshold value set for the computer room energy consumption. The specific process of energy consumption control is as follows:

[0086] Step S5-3-1: When the total comprehensive energy consumption does not exceed the threshold set for the energy consumption of the computer room, it is determined that the energy consumption is normal and the energy consumption of the computer room is continuously monitored;

[0087] Step S5-3-2: When the total comprehensive energy consumption exceeds the threshold set for the energy consumption of the computer room, it is determined to be energy consumption abnormality. The load balancing algorithm is used to redistribute the service load according to the current number of connections, processing performance weights, and service priorities of each server in the computer room. The specific process of redistributing the service load is as follows:

[0088] Based on the priority of the server queue, low-priority business loads are migrated from high-energy-consuming servers to low-load servers. After the load migration is completed, the actual total energy consumption of the computer room is recalculated and, through an iterative adjustment mechanism, the total energy consumption of the computer room is brought back to the set threshold range.

[0089] In specific implementation, based on the error calculation results between the two predicted energy consumptions and the actual energy consumption in each historical period, the inverse error normalization method is used to assign weights, and the prediction model with smaller error obtains a higher weight; when the integrated energy consumption after fusion is compared with the preset threshold, if it exceeds the threshold, the server’s current number of connections, processing performance weight and business priority are sorted, and low-priority businesses are migrated from high-energy consumption servers to low-load servers; after each load migration, a stable period is reserved and the energy consumption is re-evaluated. Through iterative adjustment, the energy consumption is gradually brought back to the threshold range to avoid fluctuations in server resources due to frequent migrations, and to ensure a balance between business continuity and energy consumption regulation.

[0090] Example 2, as Figure 2 As shown, the present invention provides an AI-based intelligent energy consumption control system for a computer room, which includes a data acquisition module, a relationship analysis module, a benchmark calculation module, an energy consumption prediction module, and an energy consumption control module;

[0091] The data acquisition module is used to set the energy consumption monitoring cycle, obtain the historical business load data and energy consumption data of the computer room, and arrange and store them in a time series relationship to construct an energy consumption monitoring data set; the relationship analysis module is used to analyze the relationship between the business load data and the energy consumption data, generate a mapping relationship, and select the reference energy consumption monitoring cycle with the highest total business load; the benchmark calculation module is used to use the total load of the reference period as the first benchmark and the average load of subsequent periods as the second benchmark, and calculate the benchmark business load of the current period through weighted fusion; the energy consumption prediction module is used to calculate the total energy consumption of the business load mapping based on the benchmark business load and the mapping relationship, and at the same time predict the total energy consumption of the time series prediction through the time series model; the energy consumption control module is used to weightedly fuse the two predicted energy consumptions to obtain the comprehensive total energy consumption, and regulate the business load according to the threshold;

[0092] The data acquisition module includes a data acquisition unit and a time series storage unit; the data acquisition unit is used to extract the computing load, storage load, network load and total power consumption data of each historical period from the monitoring log file; the time series storage unit is used to store the business load and energy consumption data in a relational database based on the timestamp data, and construct a structured data set with time series as the index;

[0093] The relationship analysis module includes a mapping modeling unit and a cycle screening unit; the mapping modeling unit is used to normalize the business load and energy consumption data and generate a mapping relationship between load and energy consumption through neural network model training; the cycle screening unit is used to compare the total business load of each energy consumption monitoring period, screen and mark the reference period with the highest total load;

[0094] The benchmark calculation module includes a dual-benchmark determination unit and a weighted fusion unit; the dual-benchmark determination unit is used to select the total load of the reference period as the first benchmark and calculate the arithmetic mean of the total load of subsequent periods as the second benchmark; the weighted fusion unit is used to assign weights to the dual-benchmark loads and generate the current period benchmark service load through weighted calculation;

[0095] The energy consumption prediction module includes a mapping calculation unit and a time series prediction unit; the mapping calculation unit is used to input the benchmark business load into the trained mapping relationship model and output the corresponding business load mapping energy consumption; the time series prediction unit is used to clean and normalize the energy consumption monitoring data set and predict the total energy consumption of the current period through the time series model;

[0096] The energy consumption control module includes a comprehensive evaluation unit and a load adjustment unit; the comprehensive evaluation unit is used to calculate the historical error, assign weights by the error inverse normalization method, and generate the total comprehensive energy consumption; the load adjustment unit is used to migrate low-priority loads according to the number of server connections, performance weights and business priorities through a load balancing algorithm when the comprehensive energy consumption exceeds the threshold, so that the energy consumption falls back to the threshold range;

[0097] The output end of the data acquisition module is electrically connected to the input end of the relationship analysis module; the output end of the relationship analysis module is electrically connected to the input end of the benchmark calculation module; the output end of the benchmark calculation module is electrically connected to the input end of the energy consumption prediction module; the output end of the energy consumption prediction module is electrically connected to the input end of the energy consumption control module.

[0098] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. An AI-based intelligent energy consumption control method for a computer room, characterized by: The method for intelligently controlling energy consumption in a computer room comprises the following steps: Step S1: Set an energy consumption monitoring cycle, and obtain the historical business load data and energy consumption data of the computer room based on the energy consumption monitoring cycle, arrange and store the obtained business load data and energy consumption data according to the time series relationship, and construct a computer room energy consumption monitoring data set; Step S2: Analyze the relationship between service load data and energy consumption data based on the data set of the computer room energy consumption monitoring data to obtain a mapping relationship between the service load and energy consumption of the computer room; and analyze the historical service load data in the data set of the computer room energy consumption monitoring data to select the energy consumption monitoring period with the highest total service load data and mark it as the reference energy consumption monitoring period; Step S3: Select the total service load of the reference energy consumption monitoring period as the first benchmark service load; select the arithmetic average of the service loads from the energy consumption monitoring period next to the reference energy consumption monitoring period to the energy consumption monitoring period before the current energy consumption monitoring period as the second benchmark service load; and calculate the benchmark service load of the current energy consumption monitoring period through weighted fusion based on the first benchmark service load and the second benchmark service load. Step S4: Based on the benchmark business load, the total energy consumption of the computer room in the current energy consumption monitoring period is obtained by mapping the business load of the computer room to the energy consumption, which is recorded as the business load mapped total energy consumption; based on the computer room energy consumption monitoring data set, the total energy consumption in the current energy consumption monitoring period is predicted by the time series prediction model, which is recorded as the time series predicted total energy consumption; Step S5: Calculate the total energy consumption of the computer room in the current energy consumption monitoring cycle through weighted fusion based on the total energy consumption mapped by the business load and the total energy consumption predicted by the time series, and regulate the business load based on the total energy consumption of the computer room; Step S5-1: Calculate the mean absolute error (MAE) between the total energy consumption of the business load mapping and the total energy consumption of the time series prediction in each historical energy consumption monitoring period and the actual total energy consumption of the computer room, and assign weight coefficients to the total energy consumption of the business load mapping and the total energy consumption of the time series prediction using a normalized weight allocation method based on the inverse of the MAE; the sum of the weight coefficients of the total energy consumption of the business load mapping and the total energy consumption of the time series prediction equals one; Step S5-2: Based on the assigned business load mapping total energy consumption weight coefficient and the time series predicted total energy consumption weight coefficient, the product of the business load mapping total energy consumption and the time series predicted total energy consumption multiplied by the corresponding weights is added together to obtain the total comprehensive energy consumption of the computer room in the current energy consumption monitoring cycle; Step S5-3: Based on the calculated total energy consumption of the computer room in the current energy consumption monitoring cycle, energy consumption control is performed in combination with the threshold value set for the computer room energy consumption. The specific process of energy consumption control is as follows: Step S5-3-1: When the total comprehensive energy consumption does not exceed the threshold set for the energy consumption of the computer room, it is determined that the energy consumption is normal and the energy consumption of the computer room is continuously monitored; Step S5-3-2: When the total comprehensive energy consumption exceeds the threshold set for the energy consumption of the computer room, it is determined to be energy consumption abnormality. The load balancing algorithm is used to redistribute the service load according to the current number of connections, processing performance weights, and service priorities of each server in the computer room. The specific process of redistributing the service load is as follows: According to the priority of the server sorting queue, low-priority business loads are migrated from high-energy-consuming servers to low-load servers. After the load migration is performed, the actual total energy consumption of the computer room is recalculated, and through an iterative adjustment mechanism, the total energy consumption of the computer room is brought back to within the set threshold range.

2. The AI-based intelligent control method for energy consumption in a computer room according to claim 1 is characterized in that: The specific steps of step S1 are as follows: Step S1-1: Analyze the monitoring log files of the computer room energy consumption based on the energy consumption monitoring cycle to obtain business load data and energy consumption data in each historical energy consumption monitoring cycle of the computer room; the business load data is represented by a set of indicators of resource usage of each server in the computer room when processing business tasks during operation, including computing load, storage load, and network load; the energy consumption data is represented by the total power consumption of each server in the computer room during operation; Step S1-2: Analyze the monitoring log file to obtain timestamp data, and store the business load data and energy consumption data in each historical energy consumption monitoring cycle in a relational database in chronological order using the timestamp data. Select the timestamp data as the primary key, select the business load data and energy consumption data as the field values, and construct a structured data set indexed by time series in the relational database, which is recorded as the computer room energy consumption monitoring data set.

3. The AI-based intelligent control method for energy consumption in a computer room according to claim 2 is characterized in that: The specific steps of step S2 are as follows: Step S2-1: traverse and read the service load data and energy consumption data in the computer room energy consumption monitoring data set, normalize the service load data and energy consumption data, and use them as input data for the neural network model. After training the neural network model, a mapping relationship between the service load and energy consumption of the computer room is obtained; Step S2-2: normalize the read service load data and obtain the total service load by adding the service load values; According to the comparison of the total business load of each energy consumption monitoring period in the energy consumption monitoring data set of the computer room, the energy consumption monitoring period with the highest total business load data is selected and marked as the reference energy consumption monitoring period.

4. The AI-based intelligent control method for energy consumption in a computer room according to claim 3 is characterized by: The specific steps of step S3 are as follows: Step S3-1: Select the total service load of the reference energy consumption monitoring period as the first benchmark service load of the computer room in the current energy consumption monitoring period; and analyze and read the end timestamp data of the reference energy consumption monitoring period. Based on the analyzed end timestamp data, sequentially obtain the total load data of each energy consumption monitoring period, up to the energy consumption monitoring period before the current energy consumption monitoring period; Step S3-2: Calculate the arithmetic mean of the total load of each energy consumption monitoring period to obtain a second benchmark service load of the computer room in the current energy consumption monitoring period; Step S3-3, assigning weight coefficients to the first reference service load and the second reference service load respectively, and the weight assignments satisfy that the sum of the weight coefficient of the first reference service load and the weight coefficient of the second reference service load is equal to one; Step S3-4: multiplying the first reference service load by the first weight coefficient to obtain a first weighted value of the reference service load; The second reference business load is multiplied by the second weight coefficient to obtain a second weighted value of the reference business load; the first weighted value and the second weighted value are added to obtain the reference business load of the computer room in the current energy consumption monitoring cycle.

5. The AI-based intelligent control method for energy consumption in a computer room according to claim 4 is characterized in that: The specific steps of step S4 are as follows: Step S4-1: Input the benchmark service load obtained in step S3 into the trained mapping relationship between the service load and energy consumption of the computer room, and output the total energy consumption of the computer room corresponding to the benchmark service load through the mapping relationship, which is recorded as the service load mapped total energy consumption; Step S4-2: clean the data in the computer room energy consumption monitoring data set, normalize it, and input it into the time series prediction model to predict the total energy consumption in the current energy consumption monitoring cycle, which is recorded as the time series predicted total energy consumption.

6. An AI-based intelligent control system for energy consumption in a computer room, applied to the AI-based intelligent control method for energy consumption in a computer room according to any one of claims 1 to 5, characterized in that: The intelligent energy consumption control system for the computer room includes a data acquisition module, a relationship analysis module, a benchmark calculation module, an energy consumption prediction module and an energy consumption control module; The data acquisition module is used to set the energy consumption monitoring cycle, obtain the historical business load data and energy consumption data of the computer room, and arrange and store them in a time series relationship to construct an energy consumption monitoring data set; the relationship analysis module is used to analyze the relationship between the business load data and the energy consumption data, generate a mapping relationship, and select the reference energy consumption monitoring cycle with the highest total business load; the benchmark calculation module is used to use the total load of the reference cycle as the first benchmark and the average load of subsequent cycles as the second benchmark, and calculate the benchmark business load of the current cycle through weighted fusion; the energy consumption prediction module is used to calculate the total energy consumption of the business load mapping based on the benchmark business load and the mapping relationship, and at the same time predict the total energy consumption through the time series model; The energy consumption control module is used to weight and fuse the two predicted energy consumptions to obtain a total comprehensive energy consumption, and to control the service load according to the threshold; The output end of the data acquisition module is electrically connected to the input end of the relationship analysis module; the output end of the relationship analysis module is electrically connected to the input end of the reference calculation module; The output end of the reference calculation module is electrically connected to the input end of the energy consumption prediction module; the output end of the energy consumption prediction module is electrically connected to the input end of the energy consumption control module.

7. The AI-based intelligent energy consumption control system for a computer room according to claim 6 is characterized by: The data acquisition module includes a data acquisition unit and a time series storage unit; the data acquisition unit is used to extract the computing load, storage load, network load and total power consumption data of each historical period from the monitoring log file; the time series storage unit is used to store the business load and energy consumption data in a relational database based on the timestamp data, and construct a structured data set with time series as the index; The relationship analysis module includes a mapping modeling unit and a cycle screening unit; the mapping modeling unit is used to normalize the business load and energy consumption data and generate a mapping relationship between load and energy consumption through neural network model training; the cycle screening unit is used to compare the total business load of each energy consumption monitoring period, screen and mark the reference period with the highest total load; The benchmark calculation module includes a dual-benchmark determination unit and a weighted fusion unit; the dual-benchmark determination unit is used to select the total load of the reference period as the first benchmark, and calculate the arithmetic mean of the total load of subsequent periods as the second benchmark; the weighted fusion unit is used to assign weights to the dual-benchmark loads and generate the current period benchmark service load through weighted calculation.

8. The AI-based intelligent energy consumption control system for a computer room according to claim 6 is characterized by: The energy consumption prediction module includes a mapping calculation unit and a time series prediction unit; the mapping calculation unit is used to input the benchmark business load into the trained mapping relationship model and output the corresponding business load mapping energy consumption; the time series prediction unit is used to clean and normalize the energy consumption monitoring data set and predict the total energy consumption of the current period through the time series model; The energy consumption control module includes a comprehensive evaluation unit and a load adjustment unit; The comprehensive evaluation unit is used to calculate the historical errors, assign weights by using the inverse error normalization method, and fuse them to generate the total comprehensive energy consumption; The load adjustment unit is used to migrate low-priority loads according to the number of server connections, performance weights and business priorities through a load balancing algorithm when the comprehensive energy consumption exceeds a threshold, so that the energy consumption falls back to within the threshold range.

Images