Method and system for adjusting energy consumption of machine room based on AI

Abstract

The invention discloses a method and system for adjusting energy consumption of a machine room based on AI, and relates to the technical field of energy consumption adjusting.The method comprises the steps that all energy consumption devices in the machine room are collected, the importance degree of each energy consumption device to work of the machine room is calculated, and main devices and secondary devices are screened out; training the preprocessed data by using deep learning, capturing the change trend of the energy consumption data along with the time sequence, and constructing an energy consumption prediction model; training the preprocessed data by using a convolutional neural network, performing feature learning on the equipment state data, and constructing a fault early warning model; calculating the correlation between the equipment fault and the energy consumption change, and analyzing the influence degree of the fault on the energy consumption to obtain a fault energy consumption function; constructing a fault propagation map in the machine room; and the energy consumption prediction model and the fault early warning model are combined to monitor the energy consumption data in the machine room in real time, and energy adjustment strategies are formulated for different monitoring results.

Description

Technical Field

[0001] The present invention relates to the technical field of energy consumption regulation, and in particular to a method and system for regulating energy consumption of a computer room based on AI. Background Art

[0002] With the rapid development of AI technologies such as deep learning, more advanced algorithms and models have been introduced into the field of computer room energy consumption management. Enterprises have begun to use algorithms such as neural networks and reinforcement learning to optimize and control the cooling system, power supply system, etc. of the computer room, and automatically adjust the equipment operating parameters by real-time monitoring and analyzing a large amount of operating data to achieve energy-saving goals. Many data centers have begun to carry out AI-based energy-saving pilot projects and have achieved certain energy-saving effects. For example, the data centers of some large Internet companies have realized the intelligent control of the air-conditioning system and reduced the cooling energy consumption by applying AI technology. AI-based computer room energy consumption regulation technology has been more widely used and promoted. More and more companies have realized that reducing the energy consumption of computer rooms can not only meet the requirements of energy saving and low carbon, but also reduce operating costs and improve the competitiveness of enterprises.

[0003] However, when adjusting and predicting energy consumption in the computer room today, the impact of equipment failure on energy consumption is often ignored. When a device fails, the normally predicted energy consumption cannot adapt to the equipment in the computer room, and the normal operation of the computer room cannot be guaranteed. When the equipment fails, the faulty equipment may cause fault propagation. Blindly adjusting energy consumption will not only fail to reduce energy consumption, but also cause greater losses in the computer room. This may cause losses in the computer room due to untimely energy supply. How to use AI to adjust the energy consumption of equipment under normal operation and failure, not only to reduce energy consumption but also to ensure the normal operation of equipment, is crucial. Summary of the invention

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

[0005] To achieve the above object, the present invention provides the following technical solutions: A method for adjusting energy consumption of a computer room based on AI, the method comprising the following steps: S100, collecting all energy-consuming equipment in the computer room, extracting the working characteristics of the energy-consuming equipment in the computer room, and classifying the energy-consuming equipment using the working characteristics; calculating the importance of each energy-consuming equipment to the work of the computer room, and screening out the main equipment and the secondary equipment; Furthermore, the specific steps for screening out major equipment and minor equipment are as follows: S101, collect all energy-consuming equipment in the computer room, collect the working data of all energy-consuming equipment, and organize and generate a list; the list includes the working time, working method, working purpose and equipment introduction in the working data of each energy-consuming equipment, and all the data in the list are uniformly converted into text data; generate a list for each energy-consuming equipment, clean the text data in the list, and remove special characters, punctuation marks and stop words in the list; then split the text data in the list into individual words, calculate the word frequency and inverse document frequency of each word after the split, and use the calculated word frequency and inverse document frequency to obtain the contribution value of each word; compare the contribution values ​​of all words in the energy-consuming equipment list, and select the word with the largest contribution value as the working feature of each energy-consuming equipment; S102, converting the extracted working characteristics of each energy-consuming device into a feature vector, where each row represents a sample energy-consuming device and each column represents a working characteristic; dividing all feature vectors into a training set and a test set; inputting the training set into a support vector machine for training to obtain a classification model, and after the training is completed, inputting the test set into the classification model to obtain a category label; using the classification model to classify the extracted working characteristics of the energy-consuming devices and marking them with category labels; arranging the energy-consuming devices with the same category labels to construct a set; S103. Extract equipment failure maintenance time, proportion of equipment energy consumption in total energy consumption, equipment usage frequency and equipment working time from each energy-consuming equipment list; construct a hierarchical model, which includes a target layer, a criterion layer and a scheme layer; compare each element in the scheme layer with each other for each criterion in the criterion layer to construct a judgment matrix, calculate the eigenvector and eigenvalue of the judgment matrix, and obtain the weight of each element after consistency verification; the elements in the scheme layer correspond to the four types of data in the extracted energy-consuming equipment list; multiply the four data values ​​of the energy-consuming equipment by the corresponding weights and add them together to obtain the importance of each energy-consuming equipment; then sort the importance from small to large, calculate the difference between adjacent importances, select the maximum importance difference as the classification node, and divide different types of energy-consuming equipment into primary equipment and secondary equipment; primary equipment is energy-consuming equipment with an importance greater than the classification node, and secondary equipment is energy-consuming equipment with an importance less than the classification node.

[0006] After classifying the energy-consuming equipment in the computer room and further dividing it into major equipment and minor equipment, managers can quickly identify the key maintenance objects and concentrate the main manpower, material resources and time on the maintenance of major equipment to ensure the stable operation of these key equipment, thereby protecting the core functions of the entire computer room. Different maintenance plans and strategies are formulated according to the different characteristics and importance of major equipment and minor equipment.

[0007] S200, collecting energy consumption data, environmental data, equipment status data and fault history records in the historical work of the computer room, and pre-processing all the collected data; Furthermore, the specific steps for preprocessing all collected data and extracting data features are as follows: S201. Collect energy consumption data of each energy-consuming device through smart meters, use various environmental sensors to measure environmental data in the computer room, extract status data of each energy-consuming device from the equipment management system when the energy-consuming device is working, and extract records of historical working failures of energy-consuming devices from the system log.

[0008] S202. Collect energy consumption data and environmental data according to the time series, and use the timestamp in the time series to determine whether the collected data is repeated, and delete it if it is repeated; when there are missing values ​​in all types of collected data, calculate the mean of the adjacent data at the missing point, and use the mean to fill the missing data; after preprocessing the collected data, normalize all the collected data.

[0009] The collected data may contain various errors, such as data entry errors, erroneous data caused by sensor failure, etc. Through data cleaning, these errors can be identified and corrected, such as removing duplicate data and correcting obvious erroneous values, thereby ensuring the accuracy of the data. Preprocessed data can better meet the assumptions and requirements of the model, allowing the model to more accurately learn patterns and relationships in the data, thereby improving the model's prediction accuracy and generalization ability.

[0010] S300, using deep learning to train the pre-processed data, capturing the changing trend of energy consumption data over time series, and building an energy consumption prediction model; Furthermore, the specific steps of building an energy consumption prediction model are as follows: S301. Extract energy-consuming devices used to control environmental data in the computer room from the classified energy-consuming devices and define them as environmental devices, and define the remaining energy-consuming devices as non-environmental devices; extract energy consumption data of non-environmental devices in historical work from all preprocessed data, extract numerical features of energy consumption data values ​​of non-environmental devices based on time series, including mean value, maximum value and standard deviation, train data features using deep learning, and construct energy consumption change function of non-environmental devices that changes with time series within one year with a year as a cycle; S302. For the extracted environmental equipment, the energy consumption data and environmental data of the environmental equipment during historical operation are screened out from the preprocessed data, and a curve graph is drawn with the environmental data as an independent variable and the energy consumption data as a variable. The influence relationship of the environmental data on the energy consumption data of the environmental equipment is obtained by using the curve graph, and a polynomial function between the environmental data and the energy consumption data in each environmental equipment is obtained. Then, the polynomial functions of all environmental equipment are summed to generate an energy consumption change function of the environmental equipment in the computer room that changes with the environmental data; S303. Add the calculated energy consumption change function of non-environmental equipment that changes with time series and the energy consumption change function of environmental equipment that changes with environmental data to obtain a comprehensive energy consumption prediction function in the computer room; use deep learning to select historical data as a test set to train and test the energy consumption prediction function, and build an energy consumption prediction model.

[0011] An energy consumption change function is constructed for the classification of environmental equipment and non-environmental equipment, making the constructed energy consumption prediction model more accurate. It not only predicts energy consumption over time series, but also takes into account the impact of environmental data on energy consumption, greatly increasing the accuracy of energy consumption prediction.

[0012] S400, dividing the data into a training set and a test set, using a convolutional neural network to train the preprocessed data, performing feature learning on the equipment status data, and building a fault warning model; Furthermore, the specific steps of constructing the fault warning model are: S401, extracting equipment failure records and equipment status data from the preprocessed data, calculating the difference between all equipment status data before and after the failure occurred in the historical work for each energy-consuming equipment, determining the size of the difference between all equipment status data, and taking the equipment status data with the largest difference as the equipment failure feature; S402, extracting the fault characteristics of energy-consuming equipment when faults occurred in historical operation, calculating the average value and standard deviation of all fault characteristics, using the average value and standard deviation to calculate the fault threshold of the energy-consuming equipment, using the convolutional neural network algorithm to train the fault threshold and fault characteristics, and constructing a fault warning model for the energy-consuming equipment.

[0013] S500, collecting energy consumption data and equipment status data when equipment in the computer room fails during historical work, calculating the correlation between equipment failure and energy consumption change, analyzing the impact of the failure on energy consumption, and obtaining a failure energy consumption function; Furthermore, the specific steps to obtain the fault energy consumption function are: S501. Collect energy consumption data and equipment status data when a failure occurs in the historical operation of energy-consuming equipment, and use the Pearson correlation coefficient to calculate the correlation r between the energy consumption data and each equipment status data. When r=0, it is judged that the corresponding equipment status data and the energy consumption data are not correlated. When r≠0, it is judged that the corresponding equipment status data and the energy consumption data are correlated, and r>0 is positively correlated, and r<0 is negatively correlated. According to the correlation, all equipment status data related to the energy consumption data are obtained and defined as relevant status data; S502: Use a linear regression algorithm to fit the device status data related to the energy consumption data, draw a change curve, obtain a polynomial function of each related status data and energy consumption data, and sum them to obtain a fault energy consumption function.

[0014] When the equipment in the computer room is working, the specific energy consumption will change as the equipment fails. Some failures will increase energy consumption, while some failures will reduce energy consumption. Therefore, a specific analysis is performed to obtain the specific impact of relevant status data on energy consumption, which makes it more accurate to predict and adjust the energy consumption in the computer room, greatly reducing the impact of failures on the energy consumption in the computer room and causing greater losses.

[0015] S600, taking each energy-consuming device in the computer room as a node, collecting the types of devices that failed during the historical work of the computer room, extracting the fault associations between the energy-consuming devices, taking the fault associations as directed edges, and constructing a fault propagation graph in the computer room; Furthermore, the specific steps for constructing the fault propagation map in the computer room are as follows: S601, collect the records of failures in the historical work of the computer room, extract the type of faulty equipment and the time of the failure in the records; regard each energy-consuming equipment in the computer room as a node, and assign a unique identifier to each node; S602. Using the collected fault records from historical work, perform non-repetitive combinations of the faulty devices in the fault records, and calculate the probability of two devices in each faulty device combination failing at the same time. Specifically, divide the records of both devices failing in the combination by the total historical fault records. Set a probability threshold, and when the probability of the faulty device group is greater than the probability threshold, determine that the two devices have a fault association. Then, use the time series to determine the fault time sequence of the two devices with a fault association, obtain the fault causal relationship of the two devices, and use the fault causal relationship as a directed edge. Use the corresponding simultaneous failure probability as the weight of the directed edge. Use the nodes and directed edges to construct a fault propagation map in the computer room.

[0016] S700 combines energy consumption prediction model and fault warning model to monitor energy consumption data in the computer room in real time, and formulates energy regulation strategies based on different monitoring results.

[0017] Furthermore, the specific steps for formulating energy regulation strategies based on different monitoring results are as follows: S701. When the computer room is working, real-time time and real-time environment data are collected, and the real-time time and real-time environment data are input into the energy consumption prediction model to obtain the real-time predicted energy consumption of the computer room; and the fault characteristics of each device in the computer room are collected. When the real-time fault characteristics are greater than the fault threshold, it is determined that the corresponding device has a fault risk and a fault warning is issued; S702, when there is a risk of equipment failure, collect relevant status data of the predicted faulty equipment and input it into the fault energy consumption function to calculate the real-time fault energy consumption, re-collect real-time environmental data to calculate the energy consumption of the equipment that has not failed, and then add it to the real-time fault energy consumption to obtain the energy consumption of the computer room after the failure, and judge the energy consumption of the computer room after the failure and the real-time predicted energy consumption. When the energy consumption of the computer room after the failure is greater than the real-time predicted energy consumption, classify and judge to formulate an adjustment strategy, specifically: When the faulty device is the primary device, the energy consumption of the secondary device is extracted and transmitted to the faulty device, so that the secondary device enters a dormant state; When the faulty device is a secondary device, the faulty device is directly put into a dormant state; When the energy consumption after a computer room failure is less than the real-time predicted energy consumption, no energy supply adjustment is performed.

[0018] By analyzing the impact of faults on energy consumption, we can accurately identify the abnormal increase in energy consumption caused by faults. When adjusting energy supply, we can specifically reduce or adjust this unnecessary energy supply to avoid energy waste, achieve refined management of computer room energy, and improve energy efficiency. Different faults have different impacts on the energy consumption of different equipment in the computer room. Combining the two models, we can dynamically adjust the energy allocation strategy according to the specific fault situation, give priority to supplying energy to equipment that is still operating normally and is critical to the business, ensure the energy needs of important equipment, and reasonably reduce the energy consumption of equipment that is more affected by the fault or non-critical equipment, so as to achieve adaptive energy allocation.

[0019] A system for regulating energy consumption in a computer room based on AI, the system for regulating energy consumption in a computer room based on AI comprising a data collection module, a classification module, a model building module, a correlation function module, an energy consumption prediction module, a graph building module and an energy consumption supply regulation module; The data collection module is used to collect energy consumption data, environmental data, equipment status data and fault history records in the historical work of the computer room, and pre-process all the collected data; The classification module is used to collect all energy-consuming equipment in the computer room and classify the energy-consuming equipment using working characteristics; calculate the importance of each energy-consuming equipment to the work of the computer room, and screen out the main equipment and the secondary equipment; The model building module is used to analyze the energy-consuming equipment in the computer room using an AI algorithm, and to build an energy consumption prediction model and a fault warning model respectively; The correlation function module is used to collect energy consumption data and equipment status data when equipment in the computer room fails in historical work, calculate the correlation between equipment failure and energy consumption change, analyze the impact of failure on energy consumption, and obtain a failure energy consumption function; The energy consumption prediction module is used to predict the energy consumption in the computer room using the energy consumption prediction model, and to judge the energy consumption after the computer room fails using the fault warning model; The graph construction module is used to collect the types of equipment that failed during the historical work of the computer room, extract the fault associations between energy-consuming equipment, use the fault associations as directed edges, and construct a fault propagation graph in the computer room; The energy consumption supply adjustment module is used to classify and discuss according to the judgment results and formulate corresponding energy consumption supply adjustment strategies.

[0020] The model building module includes an energy consumption prediction model building unit and a fault warning model building unit; The energy consumption prediction model building unit is used to analyze the energy consumption data of the environmental equipment and non-environmental equipment in the computer room respectively, obtain different energy consumption change functions, and then build an energy consumption prediction model; The fault warning model building unit is used to extract fault features in the equipment status data when faults occurred in historical work, calculate the working threshold value using the fault features and build a working warning model.

[0021] The energy consumption prediction module includes an energy consumption prediction unit and a fault energy consumption judgment unit; The energy consumption prediction unit is used to collect real-time time and real-time environmental data, input the real-time time and real-time environmental data into the energy consumption prediction model, and obtain the real-time predicted energy consumption of the computer room; The fault energy consumption judgment unit is used to collect relevant status data of the predicted fault equipment and input it into the fault energy consumption function to calculate the real-time fault energy consumption, and make a judgment with the real-time predicted energy consumption.

[0022] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention classifies the energy-consuming equipment in the computer room and further divides them into main equipment and secondary equipment, so that management personnel can quickly identify the key maintenance objects and concentrate the main manpower, material resources and time on the maintenance of the main equipment, ensuring the stable operation of these key equipment, thereby protecting the core functions of the entire computer room.

[0023] 2. The present invention combines the energy consumption prediction model with the fault warning model. When predicting energy consumption, it analyzes the impact of faults on energy consumption and formulates a reasonable adjustment strategy to avoid excessive energy consumption caused by faults, directly reducing the energy cost expenditure of the computer room. At the same time, it reduces the maintenance and replacement costs caused by further damage to the equipment due to faults, helps to optimize the overall operating costs of the computer room and improve economic benefits. Rapidly responding to faults and adjusting energy consumption can minimize the impact of faults on the business of the computer room and ensure business continuity. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a module distribution diagram of a system for adjusting energy consumption of a computer room based on AI according to the present invention; Figure 2 The present invention is a schematic diagram of the steps of a method for adjusting energy consumption of a computer room based on AI. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0026] Example: Figure 1-Figure 2 As shown, the present invention provides a technical solution. A method for adjusting energy consumption of a computer room based on AI, the method comprising the following steps: S100, collecting all energy-consuming equipment in the computer room, extracting the working characteristics of the energy-consuming equipment in the computer room, and classifying the energy-consuming equipment using the working characteristics; calculating the importance of each energy-consuming equipment to the work of the computer room, and screening out the main equipment and the secondary equipment; The specific steps to screen out major and minor equipment are: S101. Collect all energy-consuming equipment in the computer room, collect the working data of all energy-consuming equipment, and sort out and generate a list; the list includes the working time, working method, working purpose and equipment introduction in the working data of each energy-consuming equipment, and all the data in the list are uniformly converted into text data; generate a list for each energy-consuming equipment, clean the text data in the list, and remove special characters, punctuation marks and stop words in the list; then split the text data in the list into individual words, calculate the word frequency and inverse document frequency of each word after the split, and use the calculated word frequency and inverse document frequency to obtain the contribution value of each word; the formula is: ; In the formula, Con represents the contribution value of each word, nt represents the number of times word t appears in a single document, k represents the number of words in a single document, N represents the number of documents in the list, and m t Indicates the number of documents containing word t in the list; Compare the contribution values ​​of all words in the list of energy-consuming equipment and select the word with the largest contribution value as the working characteristic of each energy-consuming equipment; S102, converting the extracted working characteristics of each energy-consuming device into a feature vector, where each row represents a sample energy-consuming device and each column represents a working characteristic; dividing all feature vectors into a training set and a test set; inputting the training set into a support vector machine for training to obtain a classification model, and after the training is completed, inputting the test set into the classification model to obtain a category label; using the classification model to classify the extracted working characteristics of the energy-consuming devices and marking them with category labels; arranging the energy-consuming devices with the same category labels to construct a set; S103, extracting equipment failure maintenance time, equipment energy consumption ratio in total energy consumption, equipment usage frequency and equipment working time from each energy-consuming equipment list; constructing a hierarchical model, the hierarchical model includes a target layer, a criterion layer and a scheme layer; constructing a judgment matrix by comparing each element in the scheme layer with each criterion in the criterion layer, calculating the eigenvector and eigenvalue of the judgment matrix, and obtaining the weight of each element after consistency verification; wherein the elements in the scheme layer correspond to the four types of data in the extracted energy-consuming equipment list; multiplying the four data values ​​of the energy-consuming equipment by the corresponding weights and adding them together to obtain the importance of each energy-consuming equipment; the formula is: ; In the formula, Im represents the importance of each energy-consuming device, S represents the four data values ​​of the energy-consuming device, β represents the weight corresponding to each data value, and i represents the data value in the i-th place; Then, the importance is sorted from small to large, the difference between adjacent importance is calculated, and the maximum importance difference is selected as the classification node, and different types of energy-consuming equipment are divided into primary equipment and secondary equipment; the primary equipment is the energy-consuming equipment with an importance greater than the classification node, and the secondary equipment is the energy-consuming equipment with an importance less than the classification node.

[0027] After classifying the energy-consuming equipment in the computer room and further dividing it into major equipment and minor equipment, managers can quickly identify the key maintenance objects and concentrate the main manpower, material resources and time on the maintenance of major equipment to ensure the stable operation of these key equipment, thereby protecting the core functions of the entire computer room. Different maintenance plans and strategies are formulated according to the different characteristics and importance of major equipment and minor equipment.

[0028] S200, collecting energy consumption data, environmental data, equipment status data and fault history records in the historical work of the computer room, and pre-processing all the collected data; The specific steps for preprocessing all collected data and extracting data features are: S201. Collect energy consumption data of each energy-consuming device through smart meters, use various environmental sensors to measure environmental data in the computer room, extract status data of each energy-consuming device from the equipment management system when the energy-consuming device is working, and extract records of historical working failures of energy-consuming devices from the system log.

[0029] S202. Collect energy consumption data and environmental data according to the time series, and use the timestamp in the time series to determine whether the collected data is repeated, and delete it if it is repeated; when there are missing values ​​in all types of collected data, calculate the mean of the adjacent data at the missing point, and use the mean to fill the missing data; after preprocessing the collected data, normalize all the collected data.

[0030] The collected data may contain various errors, such as data entry errors, erroneous data caused by sensor failure, etc. Through data cleaning, these errors can be identified and corrected, such as removing duplicate data and correcting obvious erroneous values, thereby ensuring the accuracy of the data. Preprocessed data can better meet the assumptions and requirements of the model, allowing the model to more accurately learn patterns and relationships in the data, thereby improving the model's prediction accuracy and generalization ability.

[0031] S300, using deep learning to train the pre-processed data, capturing the changing trend of energy consumption data over time series, and building an energy consumption prediction model; The specific steps to build an energy consumption prediction model are: S301. Extract energy-consuming devices used to control environmental data in the computer room from the classified energy-consuming devices and define them as environmental devices, and define the remaining energy-consuming devices as non-environmental devices; extract energy consumption data of non-environmental devices in historical work from all preprocessed data, extract numerical features of energy consumption data values ​​of non-environmental devices based on time series, including average value, maximum value and standard deviation, train data features using deep learning, and construct energy consumption change function E1=f(T) of non-environmental devices changing with time series within one year with a year as a cycle; E1 represents energy consumption changing with time series, and T represents time; S302. For the extracted environmental equipment, the energy consumption data and environmental data of the environmental equipment during historical operation are screened out from the preprocessed data. The environmental data is used as an independent variable and the energy consumption data is used as a variable to draw a curve graph. The influence relationship of the environmental data on the energy consumption data of the environmental equipment is obtained by using the curve graph. The polynomial function between the environmental data and the energy consumption data in each environmental equipment is obtained. Then, the polynomial functions of all environmental equipment are summed to generate the energy consumption change function of the environmental equipment in the computer room as the environmental data changes. ; E2 represents the energy consumption that varies with environmental data, H j represents the jth type of environmental data; p represents the total number of types of environmental data; S303. Add the calculated energy consumption change function of non-environmental equipment that changes with time series and the energy consumption change function of environmental equipment that changes with environmental data to obtain a comprehensive energy consumption prediction function in the computer room; use deep learning to select historical data as a test set to train and test the energy consumption prediction function, and construct an energy consumption prediction model E=E1+E2, where E represents the comprehensive energy consumption after prediction.

[0032] An energy consumption change function is constructed for the classification of environmental equipment and non-environmental equipment, making the constructed energy consumption prediction model more accurate. It not only predicts energy consumption over time series, but also takes into account the impact of environmental data on energy consumption, greatly increasing the accuracy of energy consumption prediction.

[0033] S400, dividing the data into a training set and a test set, using a convolutional neural network to train the preprocessed data, performing feature learning on the equipment status data, and building a fault warning model; The specific steps to build a fault warning model are: S401, extracting equipment failure records and equipment status data from the preprocessed data, calculating the difference between all equipment status data before and after the failure occurred in the historical work for each energy-consuming equipment, determining the size of the difference between all equipment status data, and taking the equipment status data with the largest difference as the equipment failure feature; S402, extracting the fault characteristics of the energy-consuming equipment when the fault occurred in the historical operation, calculating the average value and standard deviation of all fault characteristics, and using the average value and standard deviation to calculate the fault threshold of the energy-consuming equipment, the formula is: ,In the formula, Gy represents the fault threshold, Gp represents the mean value of the fault feature, and Gb represents the standard deviation of the fault feature; The convolutional neural network algorithm is used to train the fault threshold and fault characteristics, and a fault warning model for energy-consuming equipment is constructed.

[0034] S500, collecting energy consumption data and equipment status data when equipment in the computer room fails during historical work, calculating the correlation between equipment failure and energy consumption change, analyzing the impact of the failure on energy consumption, and obtaining a failure energy consumption function; The specific steps to obtain the fault energy consumption function are: S501. Collect energy consumption data and equipment status data when a failure occurs in the historical operation of energy-consuming equipment, and use the Pearson correlation coefficient to calculate the correlation r between the energy consumption data and each equipment status data. When r=0, it is judged that the corresponding equipment status data and the energy consumption data are not correlated. When r≠0, it is judged that the corresponding equipment status data and the energy consumption data are correlated, and r>0 is positively correlated, and r<0 is negatively correlated. According to the correlation, all equipment status data related to the energy consumption data are obtained and defined as relevant status data; S502: Use a linear regression algorithm to fit the equipment status data related to the energy consumption data, draw a change curve, obtain a polynomial function of each related status data and energy consumption data, and sum them to obtain a fault energy consumption function. In the formula, Eg represents energy consumption data, G e represents the polynomial function of the e-th relevant state data, and q represents the total number of relevant state data; When the equipment in the computer room is working, the specific energy consumption will change as the equipment fails. Some failures will increase energy consumption, while some failures will reduce energy consumption. Therefore, a specific analysis is performed to obtain the specific impact of relevant status data on energy consumption, which makes it more accurate to predict and adjust the energy consumption in the computer room, greatly reducing the impact of failures on the energy consumption in the computer room and causing greater losses.

[0035] S600, taking each energy-consuming device in the computer room as a node, collecting the types of devices that failed during the historical work of the computer room, extracting the fault associations between the energy-consuming devices, taking the fault associations as directed edges, and constructing a fault propagation graph in the computer room; The specific steps to construct the fault propagation map in the computer room are: S601, collect the records of failures in the historical work of the computer room, extract the type of faulty equipment and the time of the failure in the records; regard each energy-consuming equipment in the computer room as a node, and assign a unique identifier to each node; S602. Using the collected fault records from historical work, the faulty devices in the fault records are combined in pairs without duplication, and the probability of two devices in each faulty device combination failing at the same time is calculated. Specifically, the record of both devices failing in the combination is divided by the total number of historical fault records; the formula is: ; In the formula, Q(A, B) represents the probability that the faulty equipment groups A and B fail at the same time, Z represents the total number of fault records, and C(A∪B) represents the number of records when equipment A and B fail at the same time; A probability threshold is set. When the probability of a faulty device group is greater than the probability threshold, it is determined that there is a fault correlation between the two devices. Then, the time series is used to determine the failure time sequence of the two devices with a fault correlation, and the failure causal relationship of the two devices is obtained. The failure causal relationship is used as a directed edge. The corresponding simultaneous failure probability is used as the weight of the directed edge. The nodes and directed edges are used to construct a fault propagation map in the computer room.

[0036] S700 combines energy consumption prediction model and fault warning model to monitor energy consumption data in the computer room in real time, and formulates energy regulation strategies based on different monitoring results.

[0037] The specific steps for formulating energy regulation strategies based on different monitoring results are as follows: S701. When the computer room is working, real-time time and real-time environment data are collected, and the real-time time and real-time environment data are input into the energy consumption prediction model to obtain the real-time predicted energy consumption of the computer room; and the fault characteristics of each device in the computer room are collected. When the real-time fault characteristics are greater than the fault threshold, it is determined that the corresponding device has a fault risk and a fault warning is issued; S702, when there is a risk of equipment failure, collect relevant status data of the predicted faulty equipment and input it into the fault energy consumption function to calculate the real-time fault energy consumption, re-collect real-time environmental data to calculate the energy consumption of the equipment that has not failed, and then add it to the real-time fault energy consumption to obtain the energy consumption of the computer room after the failure, and judge the energy consumption of the computer room after the failure and the real-time predicted energy consumption. When the energy consumption of the computer room after the failure is greater than the real-time predicted energy consumption, classify and judge to formulate an adjustment strategy, specifically: When the faulty device is the primary device, the energy consumption of the secondary device is extracted and transmitted to the faulty device, so that the secondary device enters a dormant state; When the faulty device is a secondary device, the faulty device is directly put into a dormant state; When the energy consumption after a computer room failure is less than the real-time predicted energy consumption, no energy supply adjustment is performed.

[0038] By analyzing the impact of faults on energy consumption, we can accurately identify the abnormal increase in energy consumption caused by faults. When adjusting energy supply, we can specifically reduce or adjust this unnecessary energy supply to avoid energy waste, achieve refined management of computer room energy, and improve energy efficiency. Different faults have different impacts on the energy consumption of different equipment in the computer room. Combining the two models, we can dynamically adjust the energy allocation strategy according to the specific fault situation, give priority to supplying energy to equipment that is still operating normally and is critical to the business, ensure the energy needs of important equipment, and reasonably reduce the energy consumption of equipment that is more affected by the fault or non-critical equipment, so as to achieve adaptive energy allocation.

[0039] A system for regulating energy consumption in a computer room based on AI, the system for regulating energy consumption in a computer room based on AI comprising a data collection module, a classification module, a model building module, a correlation function module, an energy consumption prediction module, a graph building module and an energy consumption supply regulation module; The data collection module is used to collect energy consumption data, environmental data, equipment status data and fault history records in the historical work of the computer room, and pre-process all the collected data; The classification module is used to collect all energy-consuming equipment in the computer room and classify the energy-consuming equipment using working characteristics; calculate the importance of each energy-consuming equipment to the work of the computer room, and screen out the main equipment and the secondary equipment; The model building module is used to analyze the energy-consuming equipment in the computer room using an AI algorithm, and to build an energy consumption prediction model and a fault warning model respectively; The correlation function module is used to collect energy consumption data and equipment status data when equipment in the computer room fails in historical work, calculate the correlation between equipment failure and energy consumption change, analyze the impact of failure on energy consumption, and obtain a failure energy consumption function; The energy consumption prediction module is used to predict the energy consumption in the computer room using the energy consumption prediction model, and to judge the energy consumption after the computer room fails using the fault warning model; The graph construction module is used to collect the types of equipment that failed during the historical work of the computer room, extract the fault associations between energy-consuming equipment, use the fault associations as directed edges, and construct a fault propagation graph in the computer room; The energy consumption supply adjustment module is used to classify and discuss according to the judgment results and formulate corresponding energy consumption supply adjustment strategies.

[0040] The model building module includes an energy consumption prediction model building unit and a fault warning model building unit; The energy consumption prediction model building unit is used to analyze the energy consumption data of the environmental equipment and non-environmental equipment in the computer room respectively, obtain different energy consumption change functions, and then build an energy consumption prediction model; The fault warning model building unit is used to extract fault features in the equipment status data when faults occurred in historical work, calculate the working threshold value using the fault features and build a working warning model.

[0041] The energy consumption prediction module includes an energy consumption prediction unit and a fault energy consumption judgment unit; The energy consumption prediction unit is used to collect real-time time and real-time environmental data, input the real-time time and real-time environmental data into the energy consumption prediction model, and obtain the real-time predicted energy consumption of the computer room; The fault energy consumption judgment unit is used to collect relevant status data of the predicted fault equipment and input it into the fault energy consumption function to calculate the real-time fault energy consumption, and make a judgment with the real-time predicted energy consumption.

[0042] Embodiment: Now the energy consumption in a certain computer room is analyzed, and it is found that the equipment in the computer room includes Internet equipment, temperature control equipment and humidity control equipment; it is judged that the Internet equipment is the main equipment, and the temperature control equipment and humidity control equipment are the secondary equipment; The AI ​​algorithm is used to construct energy prediction models, fault warning models and fault energy consumption functions respectively; then a fault propagation map is constructed from Internet devices to temperature control devices; Collect real-time time and real-time environmental data, input them into the energy prediction model to get a real-time predicted energy consumption of 2300, collect real-time fault features for Internet devices 20, set the fault threshold to 25, determine that Internet devices have a fault risk, extract real-time device status data and input them into the fault energy consumption function to calculate the required energy consumption after the Internet fault is 1500, re-collect environmental data after the fault to get the required energy consumption of 3000; determine that it is greater than the real-time predicted energy consumption; Remove the temperature control equipment, put the humidity control equipment into sleep mode, and transfer the energy supply of the secondary equipment to the Internet equipment to maintain the normal operation of the main functions in the computer room. The maintenance personnel will then handle the work in a timely manner.

[0043] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A method for adjusting energy consumption of a computer room based on AI, characterized in that: The method comprises the following steps: S100, collecting all energy-consuming equipment in the computer room, extracting the working characteristics of the energy-consuming equipment in the computer room, and classifying the energy-consuming equipment using the working characteristics; calculating the importance of each energy-consuming equipment to the work of the computer room, and screening out the main equipment and the secondary equipment; S200, collecting energy consumption data, environmental data, equipment status data and fault history records in the historical work of the computer room, and pre-processing all the collected data; S300, using deep learning to train the pre-processed data, capturing the changing trend of energy consumption data over time series, and building an energy consumption prediction model; S400, dividing the data into a training set and a test set, using a convolutional neural network to train the preprocessed data, performing feature learning on the equipment status data, and building a fault warning model; S500, collecting energy consumption data and equipment status data when equipment in the computer room fails during historical work, calculating the correlation between equipment failure and energy consumption change, analyzing the impact of the failure on energy consumption, and obtaining a failure energy consumption function; S600, taking each energy-consuming device in the computer room as a node, collecting the types of devices that failed during the historical work of the computer room, extracting the fault associations between the energy-consuming devices, taking the fault associations as directed edges, and constructing a fault propagation graph in the computer room; S700 combines energy consumption prediction model and fault warning model to monitor energy consumption data in the computer room in real time, and formulates energy regulation strategies based on different monitoring results.

2. According to claim 1, a method for adjusting energy consumption of a computer room based on AI is characterized in that: The specific steps of screening out the main equipment and the secondary equipment in S100 are: S101. Collect all energy-consuming equipment in the computer room, collect the working data of all energy-consuming equipment, and compile and generate a list; The list includes the working time, working method, working purpose and equipment introduction of each energy-consuming equipment, and all the data in the list are uniformly converted into text data; Generate a list for each energy-consuming device, clean the text data in the list, and remove special characters, punctuation marks, and stop words in the list; Then, the text data in the list is split into individual words, and the word frequency and inverse document frequency of each word after the split are calculated respectively. The contribution value of each word is obtained by using the calculated word frequency and inverse document frequency; Compare the contribution values ​​of all words in the list of energy-consuming equipment and select the word with the largest contribution value as the working characteristic of each energy-consuming equipment; S102, converting the extracted working characteristics of each energy-consuming device into a feature vector, where each row represents a sample energy-consuming device and each column represents a working characteristic; dividing all feature vectors into a training set and a test set; inputting the training set into a support vector machine for training to obtain a classification model, and after the training is completed, inputting the test set into the classification model to obtain a category label; using the classification model to classify the extracted working characteristics of the energy-consuming devices and marking them with category labels; arranging the energy-consuming devices with the same category labels to construct a set; S103, extracting equipment failure maintenance time, proportion of equipment energy consumption in total energy consumption, equipment usage frequency and equipment working time from each energy-consuming equipment list; By constructing a hierarchical model, the hierarchical model includes a target layer, a criterion layer, and a solution layer; for each criterion in the criterion layer, each element in the solution layer is compared pairwise to construct a judgment matrix, the eigenvector and eigenvalue of the judgment matrix are calculated, and the weight of each element is obtained after consistency verification; The elements in the scheme layer correspond to the four types of data in the extracted energy-consuming equipment list; the four data values ​​of the energy-consuming equipment are multiplied by the corresponding weights and then added to obtain the importance of each energy-consuming equipment; then the importance is sorted from small to large, the difference between adjacent importance values ​​is calculated, and the maximum importance difference value is selected as the classification node, and different types of energy-consuming equipment are divided into primary equipment and secondary equipment; the primary equipment is the energy-consuming equipment whose importance is greater than the classification node, and the secondary equipment is the energy-consuming equipment whose importance is less than the classification node.

3. According to claim 1, a method for adjusting energy consumption of a computer room based on AI is characterized in that: The specific steps of preprocessing all collected data and extracting data features in S200 are: S201, collecting energy consumption data of each energy-consuming device through smart meters, using various environmental sensors to measure environmental data in the computer room, extracting status data of each energy-consuming device from the device management system when the energy-consuming device is working, and extracting records of historical work failures of energy-consuming devices from the system log; S202, collecting energy consumption data and environmental data according to the time series, and using the timestamp in the time series to determine whether the collected data is repeated, and deleting the repeated data; when there are missing values ​​in all types of collected data, calculating the mean of the adjacent data at the missing location, and using the mean to fill the missing data; After preprocessing the collected data, all the collected data are normalized.

4. The method for adjusting energy consumption of a computer room based on AI according to claim 1, characterized in that: The specific steps of constructing the energy consumption prediction model in S300 are: S301. Extract energy-consuming devices used to control environmental data in the computer room from the classified energy-consuming devices and define them as environmental devices, and define the remaining energy-consuming devices as non-environmental devices; extract energy consumption data of non-environmental devices in historical work from all preprocessed data, extract numerical features of energy consumption data values ​​of non-environmental devices based on time series, including mean value, maximum value and standard deviation, train data features using deep learning, and construct energy consumption change function of non-environmental devices that changes with time series within one year with a year as a cycle; S302. For the extracted environmental equipment, the energy consumption data and environmental data of the environmental equipment during historical operation are screened out from the preprocessed data, and a curve graph is drawn with the environmental data as an independent variable and the energy consumption data as a variable. The influence relationship of the environmental data on the energy consumption data of the environmental equipment is obtained by using the curve graph, and a polynomial function between the environmental data and the energy consumption data in each environmental equipment is obtained. Then, the polynomial functions of all environmental equipment are summed to generate an energy consumption change function of the environmental equipment in the computer room that changes with the environmental data; S303, adding the calculated energy consumption change function of non-environmental equipment changing with time series and the energy consumption change function of environmental equipment changing with environmental data to obtain a comprehensive energy consumption prediction function in the computer room; using deep learning to select historical data as a test set to train and test the energy consumption prediction function, and construct an energy consumption prediction model; The specific steps of constructing the fault warning model in S400 are: S401, extracting equipment failure records and equipment status data from the preprocessed data, calculating the difference between all equipment status data before and after the failure occurred in the historical work for each energy-consuming equipment, determining the size of the difference between all equipment status data, and taking the equipment status data with the largest difference as the equipment failure feature; S402, extracting the fault characteristics of energy-consuming equipment when faults occurred in historical operation, calculating the average value and standard deviation of all fault characteristics, using the average value and standard deviation to calculate the fault threshold of the energy-consuming equipment, using the convolutional neural network algorithm to train the fault threshold and fault characteristics, and constructing a fault warning model for the energy-consuming equipment.

5. The method for adjusting energy consumption of a computer room based on AI according to claim 4 is characterized in that: The specific steps of obtaining the fault energy consumption function in S500 are: S501. Collect energy consumption data and equipment status data when a failure occurs in the historical operation of energy-consuming equipment, and use the Pearson correlation coefficient to calculate the correlation r between the energy consumption data and each equipment status data. When r=0, it is judged that the corresponding equipment status data and the energy consumption data are not correlated. When r≠0, it is judged that the corresponding equipment status data and the energy consumption data are correlated, and r>0 is positively correlated, and r<0 is negatively correlated. According to the correlation, all equipment status data related to the energy consumption data are obtained and defined as relevant status data; S502: Use a linear regression algorithm to fit the device status data related to the energy consumption data, draw a change curve, obtain a polynomial function of each related status data and energy consumption data, and sum them to obtain a fault energy consumption function.

6. The method for adjusting energy consumption of a computer room based on AI according to claim 5, characterized in that: The specific steps of constructing the fault propagation map in the computer room in S600 are: S601, collect the records of failures in the historical work of the computer room, extract the type of faulty equipment and the time of the failure in the records; regard each energy-consuming equipment in the computer room as a node, and assign a unique identifier to each node; S602, using the collected historical fault records, perform non-repetitive combinations of faulty devices in the fault records, and calculate the probability of two devices in each faulty device combination failing at the same time, specifically, dividing the record of both devices failing in the combination by the total historical fault records; A probability threshold is set. When the probability of a faulty device group is greater than the probability threshold, it is determined that there is a fault correlation between the two devices. Then, the time series is used to determine the failure time sequence of the two devices with a fault correlation, and the failure causal relationship of the two devices is obtained. The failure causal relationship is used as a directed edge. The corresponding simultaneous failure probability is used as the weight of the directed edge. The nodes and directed edges are used to construct a fault propagation map in the computer room.

7. The method for adjusting energy consumption of a computer room based on AI according to claim 6, characterized in that: The specific steps of formulating energy regulation strategies according to different monitoring results in S700 are: S701. When the computer room is working, real-time time and real-time environment data are collected, and the real-time time and real-time environment data are input into the energy consumption prediction model to obtain the real-time predicted energy consumption of the computer room; and the fault characteristics of each device in the computer room are collected. When the real-time fault characteristics are greater than the fault threshold, it is determined that the corresponding device has a fault risk and a fault warning is issued; S702: After determining that a device with a risk of failure is obtained, the fault propagation map in the computer room is used to analyze the faulty device, and all energy-consuming devices that have directed edges with the faulty device and are the source of the directed edges with the faulty device are extracted, and the extracted energy-consuming devices are regarded as suspicious devices; S602, when there is a risk of equipment failure, collect relevant status data of the predicted faulty equipment and input it into the fault energy consumption function to calculate the real-time fault energy consumption, re-collect real-time environmental data to calculate the energy consumption of the equipment that has not failed, and then add it to the real-time fault energy consumption to obtain the energy consumption of the computer room after the failure, judge the energy consumption of the computer room after the failure and the real-time predicted energy consumption, when the energy consumption of the computer room after the failure is greater than the real-time predicted energy consumption, classify and judge to formulate an adjustment strategy, mark all suspicious equipment as non-adjustment equipment when formulating the adjustment strategy, and do not need to adjust the energy supply of the non-adjustment equipment, and then specifically: When the faulty device is the main device, the energy consumption of the secondary device of the non-adjustable device is extracted and transmitted to the faulty device, so that the secondary device enters a dormant state; When the faulty device is a secondary device, the faulty device is directly put into a dormant state; When the energy consumption after a computer room failure is less than the real-time predicted energy consumption, no energy supply adjustment is performed.

8. A system for regulating energy consumption in a computer room based on AI, characterized in that: The AI-based system for regulating the energy consumption of the computer room includes a data collection module, a classification module, a model building module, a correlation function module, an energy consumption prediction module, a graph building module, and an energy consumption supply regulation module; The data collection module is used to collect energy consumption data, environmental data, equipment status data and fault history records in the historical work of the computer room, and pre-process all the collected data; The classification module is used to collect all energy-consuming equipment in the computer room and classify the energy-consuming equipment using working characteristics; calculate the importance of each energy-consuming equipment to the work of the computer room, and screen out the main equipment and the secondary equipment; The model building module is used to analyze the energy-consuming equipment in the computer room using an AI algorithm, and to build an energy consumption prediction model and a fault warning model respectively; The correlation function module is used to collect energy consumption data and equipment status data when equipment in the computer room fails in historical work, calculate the correlation between equipment failure and energy consumption change, analyze the impact of failure on energy consumption, and obtain a failure energy consumption function; The energy consumption prediction module is used to predict the energy consumption in the computer room using the energy consumption prediction model, and to judge the energy consumption after the computer room fails using the fault warning model; The graph construction module is used to collect the types of equipment that failed during the historical work of the computer room, extract the fault associations between energy-consuming equipment, use the fault associations as directed edges, and construct a fault propagation graph in the computer room; The energy consumption supply adjustment module is used to classify and discuss according to the judgment results and formulate corresponding energy consumption supply adjustment strategies.

9. The system for adjusting energy consumption of a computer room based on AI according to claim 8 is characterized in that: The model building module includes an energy consumption prediction model building unit and a fault warning model building unit; The energy consumption prediction model building unit is used to analyze the energy consumption data of the environmental equipment and non-environmental equipment in the computer room respectively, obtain different energy consumption change functions, and then build an energy consumption prediction model; The fault warning model building unit is used to extract fault features in the equipment status data when faults occurred in historical work, calculate the working threshold value using the fault features and build a working warning model.

10. The system for adjusting energy consumption of a computer room based on AI according to claim 8, characterized in that: The energy consumption prediction module includes an energy consumption prediction unit and a fault energy consumption judgment unit; The energy consumption prediction unit is used to collect real-time time and real-time environmental data, input the real-time time and real-time environmental data into the energy consumption prediction model, and obtain the real-time predicted energy consumption of the computer room; The fault energy consumption judgment unit is used to collect relevant status data of the predicted fault equipment and input it into the fault energy consumption function to calculate the real-time fault energy consumption, and make a judgment with the real-time predicted energy consumption.

Images