Computer room energy consumption control system based on machine vision

Abstract

The invention discloses a computer room energy consumption control system based on machine vision, which relates to the field of energy consumption control and comprises a regulation and control index determination module, an energy efficiency index acquisition module, an index preliminary optimization module, an environment coincidence analysis module and an index re-optimization module. Generating an initial energy consumption regulation index of a target temperature control domain through the basic energy efficiency characteristic parameters, identifying a dynamic energy consumption hot spot region in the machine room, extracting a corresponding multi-dimensional energy efficiency influence feature set, and performing adaptive optimization on the initial energy consumption regulation index based on a time sequence change feature of the multi-dimensional energy efficiency influence feature set to obtain a dynamic energy consumption regulation result; generating a preliminary optimization energy consumption regulation index, combining with a historical normal state data set, generating an environmental parameter conformity index and a multi-dimensional environmental stability coefficient, and combining with the multi-dimensional environmental parameter conformity index to optimize the preliminary optimization energy consumption regulation index again, and generating a final energy consumption regulation index of the target temperature control domain; and the waste of resources is greatly reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of energy consumption control, in particular to a computer room energy consumption control system based on machine vision. BACKGROUND

[0002] The computer room energy consumption control method in the prior art has the following deficiencies: The computer room energy consumption control method in the prior art relies heavily on limited temperature, humidity and other basic environmental sensor data, or only on the power consumption readings of IT equipment for judgment. This single data source cannot comprehensively and stereoscopically consider the real running state in the computer room. Moreover, the simple rule control based on fixed thresholds cannot understand the time sequence correlation and dynamic change trend between data. It cannot predict the heat load change according to historical data and real-time trend, nor can it dynamically adjust the strategy according to the importance of parameters. The existing method usually controls the computer room as a whole or performs simple physical zoning, lacks the ability to accurately identify and dynamically divide the cabinet clusters with similar thermodynamic characteristics, and cannot locate the precise energy consumption hotspot area and its boundary through thermodynamic zoning algorithm and machine vision, resulting in that the cooling resources cannot be accurately distributed on demand. The existing technology rarely considers the mutual coupling relationship between multiple environmental parameters and its comprehensive influence on heat dissipation efficiency. Its control strategy lacks a dynamic baseline based on historical normal operation data. The decision is often based on the comparison of absolute values and fixed thresholds, rather than the relative deviation from the historical health state. At the same time, the existing technology often adopts a conservative strategy, prioritizing equipment safety at the expense of energy efficiency, resulting in high power usage efficiency index, difficulty in achieving two-way intelligent adjustment, and serious resource waste.

[0003] In order to solve the above-mentioned defects, a technical solution is provided. SUMMARY

[0004] In view of the defects of the prior art, the present application provides a computer room energy consumption control system based on machine vision.

[0005] The technical solution adopted by the present application to solve the above technical problems is: A computer room energy consumption control system based on machine vision, comprising: A control index determination module acquires the basic energy efficiency characteristic parameters of the target temperature control domain, multi-modal machine vision data, multi-dimensional environmental parameter time series data set, and generates the initial energy consumption control index of the target temperature control domain based on the basic energy efficiency characteristic parameters; An energy efficiency index acquisition module identifies the dynamic energy consumption hotspot area in the computer room based on the multi-modal machine vision data, and extracts the corresponding multi-dimensional energy efficiency influence feature set; An index preliminary optimization module, based on the time sequence variation characteristics of the multi-dimensional energy efficiency influence feature set, performs adaptive optimization on the initial energy consumption regulation index to generate a preliminary optimized energy consumption regulation index; An environment compliance analysis module, based on real-time environment parameter data of the target temperature control domain, in combination with a historical normal state data set, generates an environment parameter compliance index; An index re-optimization module, based on the time sequence variation characteristics of the environment parameters, generates a multi-dimensional environment stability coefficient, and performs re-optimization on the preliminary optimized energy consumption regulation index in combination with the multi-dimensional environment parameter compliance index to generate a final energy consumption regulation index of the target temperature control domain.

[0006] Further, the regulation index determination module comprises: A target determination unit performs division analysis on the computer room according to a region division basis to obtain a target temperature control domain, and the region division basis comprises cabinet power density, heat dissipation characteristics and air flow organization correlation degree; A data acquisition unit acquires basic energy efficiency characteristic parameters, multi-modal machine vision data and multi-dimensional environment parameter time sequence data set of the target temperature control domain, and the basic energy efficiency characteristic parameters comprise device inherent parameters, cabinet structure parameters and historical operation parameters.

[0007] Further, the regulation index determination module further comprises: A coefficient generation unit performs multi-dimensional thermal efficiency evaluation on the device inherent parameters to obtain a thermal inertia load factor, performs fluid dynamics characteristic analysis on the cabinet structure parameters to obtain an aerodynamic thermodynamic impedance factor, and quantifies running steady state characteristics of the historical operation parameters to obtain a running entropy coefficient; An initial index unit performs dynamic weight distribution and multi-source data fusion on the thermal inertia load factor, the aerodynamic thermodynamic impedance factor and the running entropy coefficient to obtain an initial energy consumption regulation index.

[0008] Further, the energy efficiency index acquisition module comprises: A graph generation unit extracts device state visual features, temperature distribution features and personnel behavior features by analyzing multi-modal machine vision data of each target temperature control domain, and obtains a thermodynamic situation graph by analyzing abnormal area determination mechanism and color labeling preset rules; A boundary positioning unit performs boundary evolution on the thermodynamic situation graph through semantic segmentation and semantic weighted active contour model, and performs rigid constraint alignment on the device physical contour during the boundary evolution process to obtain a multi-modal fusion thermal domain boundary; A feature set extraction unit delimits energy consumption hot spot sub-regions of each target temperature control domain based on the multi-modal fusion thermal domain boundary, and extracts a corresponding multi-dimensional energy efficiency influence feature set, including a calculation load density index, a heat dissipation efficiency decay coefficient and a personnel activity thermal disturbance index.

[0009] Further, the index preliminary optimization module comprises: a feature extraction unit, which establishes a time sequence variation model of a multi-dimensional energy efficiency influence feature set, extracts a variation rate, a fluctuation amplitude and a trend feature of a calculation load density index, a heat dissipation performance attenuation coefficient and a personnel activity heat disturbance index in a time dimension, and generates a feature set dynamic feature vector; a feature weight allocation unit which calculates a dynamic weight of each feature parameter in a current time sequence window by using an improved entropy weight method.

[0010] Further, the index preliminary optimization module further comprises: an energy consumption anomaly evaluation unit which, based on the feature set dynamic feature vector and the dynamic weight, performs parameter optimization on a basic Gaussian membership function to obtain an optimized Gaussian membership function; based on a fuzzy language value in the optimized Gaussian membership function, constructs a fuzzy rule base through historical data mining; based on the optimized Gaussian membership function and the fuzzy rule base, generates a fuzzy output set through weighted reasoning, and obtains a regional energy consumption state evaluation value through an area barycenter method; an index optimization unit which creates a preset control index optimization function, and optimizes an initial energy consumption control index of a target temperature control domain based on the regional energy consumption state evaluation value to generate a preliminary optimized energy consumption control index.

[0011] Further, the environment conformity analysis module comprises: a reference data extraction unit which retrieves data records with similar basic energy efficiency characteristic parameters and multi-modal machine vision features as the current target temperature control domain in a historical data set, filters out records in a normal operation state to constitute a reference data set; an environment parameter extraction unit which extracts various types of environment parameter data in the reference data set, calculates statistical distribution characteristics thereof, determines a normal fluctuation interval range of multi-dimensional environment parameters, and generates a multi-dimensional environment parameter reference interval set; the multi-dimensional environment parameters include air flow organization efficiency, cooling return temperature, supply and return air temperature difference, air age, and particulate matter concentration distribution; a conformity transformation unit which collects real-time multi-dimensional environment parameter data of the target temperature control domain, compares each parameter with the multi-dimensional environment parameter reference interval one by one, calculates a degree of deviation of each parameter from the reference interval, converts the deviation degree into a conformity score by using a fuzzy membership method, and finally generates a multi-dimensional environment parameter conformity index by weighting and comprehensively integrating the conformity scores of each parameter.

[0012] Further, the index re-optimization module comprises: an environment parameter time sequence analysis unit which, based on a multi-dimensional environment parameter time sequence data set, calculates a variation variance and a trend stability index of each parameter; The environment stability coefficient unit calculates the dynamic weight by using the entropy weight method based on the variance of each parameter and the trend stability index, and generates the multi-dimensional environment instability coefficient by weighting. The exponential final optimization unit processes the multi-dimensional environment parameter coincidence index and the multi-dimensional environment instability coefficient by using the environment situation evaluation function, generates the environment change index, and uses the index to perform final adjustment on the preliminary optimized energy consumption regulation index, thereby generating the final energy consumption regulation index of the target temperature control area.

[0013] Further, the regulation index optimization function comprises: wherein is the preliminary optimized energy consumption regulation index of the target temperature control area, is the initial energy consumption regulation index of the target temperature control area, E is the regional energy consumption state evaluation value, is the energy consumption state evaluation reference value, which represents the critical state of normality and abnormality, is the relative deviation amount based on the state evaluation value, which is used as an optimization factor, and K is the adjustment coefficient of the optimization strength, and K>0.

[0014] Further, the environment situation evaluation function comprises: wherein represents the environment change index, represents the gain coefficient, and epsilon is the smoothing factor, is the dynamic reference value, is the weight coefficient, and satisfies , C and S are the multi-dimensional environment parameter coincidence index and the multi-dimensional environment instability coefficient, respectively.

[0015] Compared with the prior art, the present application has the following beneficial effects: 1、The present application obtains the basic energy efficiency characteristic parameters of the target temperature control domain, multi-modal machine vision data, multi-dimensional environment parameter time series data set, and generates an initial energy consumption regulation index of the target temperature control domain based on the basic energy efficiency characteristic parameters; identifies the dynamic energy consumption hotspot area in the machine room based on the multi-modal machine vision data, and extracts the corresponding multi-dimensional energy efficiency influence feature set; performs adaptive optimization on the initial energy consumption regulation index based on the time series change characteristics of the multi-dimensional energy efficiency influence feature set, and generates a preliminary optimized energy consumption regulation index; generates an environment parameter compliance index based on the real-time environment parameter data of the target temperature control domain, combined with the historical normal state data set, which can not only consider the basic environment sensor data such as temperature and humidity, but also consider the power consumption readings of the IT equipment itself to judge the real running state in the machine room comprehensively and stereoscopically. The dynamic model can understand the time series correlation and dynamic change trend between the data, and can predict the heat load change according to the historical data and real-time trend, and can also dynamically adjust the strategy according to the importance of the parameters. The ability to accurately identify and dynamically divide the cabinet cluster with similar thermodynamic characteristics can accurately locate the energy consumption hotspot area and its boundary through thermodynamic partition algorithm and machine vision, and the cooling resources can be accurately distributed on demand.

[0016] 2、The present application generates a multi-dimensional environment stability coefficient based on the time series change characteristics of the environment parameters, and further optimizes the preliminary optimized energy consumption regulation index by combining the multi-dimensional environment parameter compliance index to generate the final energy consumption regulation index of the target temperature control domain, which can comprehensively consider the mutual coupling relationship between various environment parameters and its comprehensive influence on the heat dissipation efficiency, and the regulation strategy is based on the dynamic baseline of the historical normal operation data, and the decision can be compared with the relative deviation degree of the historical healthy state, and can be intelligently adjusted in both directions, greatly reducing the waste of resources. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The system application architecture diagram of the present application; Figure 2 The structure block diagram of the regulation index determination module in the system embodiment of the present application; Figure 3 The structure block diagram of the energy efficiency index acquisition module in the system embodiment of the present application; Figure 4 The structure block diagram of the index preliminary optimization module in the system embodiment of the present application; Figure 5 The structure block diagram of the environment compliance analysis module in the system embodiment of the present application; Figure 6 The structure block diagram of the index re-optimization module in the system embodiment of the present application. DETAILED DESCRIPTION

[0018] The technical solutions of the present application will be described clearly and completely below in combination with the drawings and specific embodiments.

[0019] Figure 1 An application architecture diagram of the system provided by the embodiment of the present application is shown.

[0020] In a preferred embodiment provided by the present application, a computer room energy consumption control system based on machine vision comprises: The regulation index determination module 100 acquires the basic energy efficiency characteristic parameters of the target temperature control domain, the multi-modal machine vision data, the multi-dimensional environmental parameter time series data set, and generates the initial energy consumption regulation index of the target temperature control domain based on the basic energy efficiency characteristic parameters.

[0021] Specifically, Figure 2 A structural block diagram of the regulation index determination module 100 in the system provided by the embodiment of the present application is shown.

[0022] In a preferred embodiment provided by the present application, the regulation index determination module 100 specifically comprises: The target determination unit 101 performs division analysis on the computer room according to the division basis to obtain the target temperature control domain, and the division basis includes cabinet power density, heat dissipation characteristics and air flow organization correlation degree; The data acquisition unit 102 acquires the basic energy efficiency characteristic parameters of the target temperature control domain, the multi-modal machine vision data, and the multi-dimensional environmental parameter time series data set, and the basic energy efficiency characteristic parameters include device inherent parameters, cabinet structure parameters and historical operation parameters; The coefficient generation unit 103 performs multi-dimensional thermal efficiency evaluation on the device inherent parameters to obtain a thermal inertia load factor, performs fluid dynamics characteristic analysis on the cabinet structure parameters to obtain an aerodynamic thermodynamic impedance factor, and quantifies the running entropy value coefficient by running steady state characteristic of the historical operation parameters; The initial index unit 104 performs dynamic weight distribution and multi-source data fusion on the thermal inertia load factor, the aerodynamic thermodynamic impedance factor and the running entropy value coefficient to obtain the initial energy consumption regulation index.

[0023] In the embodiment of the present application, the target temperature control domain in the target determination unit 101 refers to the computer room being divided into a cabinet cluster area with similar thermal characteristics and needing unified regulation by a thermodynamic partition algorithm. Each target temperature control domain contains 5 physically adjacent cabinet units. The area division is based on cabinet power density, heat dissipation characteristics, and air flow organization correlation. Specifically, the cabinet power density refers to the real-time power consumption value of the unit area cabinet, reflecting the heat intensity of the area. The heat dissipation characteristics are quantified by the heat dissipation efficiency coefficient, indicating the cooling air volume required for unit power consumption. The air flow organization correlation uses the correlation coefficient to represent the mutual influence degree of the air flow path between different cabinets, and the correlation coefficient takes a value of 0-1. The algorithm standardizes all the above parameters of the cabinets in the computer room. Standardization refers to the Min-Max standardization method, which normalizes the values to the interval [0, 1] and eliminates the dimension effect of the data. The thermal similarity between cabinets is calculated by the Euclidean distance. When calculating the thermal similarity between cabinets, the standardized power density, heat dissipation characteristics, and air flow organization correlation are substituted into the weighted Euclidean distance formula to calculate the distance value of the thermal difference between cabinets. Then, the inverse proportional function is used to convert the distance value to a similarity value in the range of 0-1. The closer the value is to 1, the more similar the thermal characteristics of the two cabinets are. The K-means clustering algorithm is used to merge the cabinets with high similarity and physical adjacency into the same target temperature control domain. Cabinets with high similarity and physical adjacency refer to cabinets with similarity greater than a set threshold and physical adjacency less than a set threshold, such as less than 5 cabinet units. The significance of this division method is to break the simple partition of physical location in the prior art, and to dynamically optimize and regulate based on the thermal characteristics, which can avoid local overheating and reduce cold waste, and improve the efficiency of the cooling system.

[0024] The multi-modal machine vision data in the data acquisition unit 102 includes equipment running state visual feature data of the target temperature control domain collected by a multi-spectral camera group deployed in the computer room. Specifically, it includes visible light vision data, device panel indicator light state, screen display state, fan physical rotation state collected by a high-definition camera group; infrared thermal imaging data, device surface temperature distribution data, thermal channel temperature gradient data collected by an infrared thermal imager array; dynamic behavior recognition data, personnel activity trajectory, stay time, operation behavior data in the computer room collected by a motion capture camera group.

[0025] The multi-dimensional environment parameter time series data set is micro-environment time series data of a target temperature control domain collected by an environment monitoring sensor. A multi-source environment sensor array including a temperature and humidity sensor, a wind speed sensor, and a particle counter is deployed in the target temperature control domain to collect original environment parameter data at a fixed sampling frequency within a complete monitoring period to form a multi-dimensional environment parameter time series data set. The environment parameters include air flow organization efficiency, cooling return temperature, supply and return air temperature difference, air age, and particulate matter concentration distribution.

[0026] The basic energy efficiency characteristic parameters include device inherent parameters, server rated power, processor model quantity, and heat dissipation design power obtained through a device management system; cabinet structure parameters, cabinet size, arrangement density, and vent hole rate obtained through a computer room infrastructure database; and historical operation parameters, historical power consumption of the region obtained through an energy management system.

[0027] The thermal inertia load factor in the coefficient generation unit 103 is calculated by a multiple linear regression model. Specifically, inherent parameters of all servers in the target temperature control domain are collected, including but not limited to rated power P rat , processor core quantity N cor , and heat dissipation design power TDP, to form a feature vector. A regression model trained in advance based on a large amount of computer room device data is used, which has the form: where a is the intercept and β1, β2, β3... are the regression coefficients of each feature variable. The feature vector of the temperature control zone is input into the model to obtain an original heat load prediction value. Through the Min-Max normalization method, the original prediction value of all temperature control zones is normalized to the interval [0, 1] to obtain a thermal inertia load factor. The higher the factor value, the greater the theoretical heat load of the equipment in the region. The aerodynamic thermodynamic impedance factor is generated by a computational fluid dynamics (CFD) simulation model. Specifically, a digital three-dimensional geometric model of the target temperature control zone is constructed according to the physical structure parameters of the cabinet, such as size, arrangement density, and ventilation hole rate. Under the set boundary conditions, such as air conditioning supply air velocity and temperature, the CFD steady-state simulation is run to simulate the airflow organization, pressure field, and velocity field in the computer room. The aerodynamic thermodynamic impedance factor is calculated from the simulation results, for example, defined as (target region pressure difference / reference region pressure difference) x (1-average wind speed / nominal wind speed). The calculation result is normalized to the interval [0, 1], and the higher the coefficient value, the greater the resistance to airflow in this region and the lower the heat dissipation efficiency. The operation entropy coefficient is calculated by time series analysis. The system calls the historical power consumption time series data of the target temperature control zone of the energy management system in the past period, calculates the statistical characteristics of the time series to measure the volatility, including the standard deviation and the coefficient of variation, where the coefficient of variation is the ratio of the standard deviation to the mean, which is used to eliminate the influence of different mean levels on the comparison of volatility. The running fluctuation state value is calculated by integrating the above statistical characteristics, for example, running fluctuation state value = standard deviation x coefficient of variation, and normalized to the range [0, 1] by percentile sorting or range method to obtain the running entropy coefficient. The higher the coefficient value, the more intense the fluctuation of the historical power consumption in the region, and the more unstable the running state.

[0028] The weight proportions of the thermal inertia load factor, the aerodynamic thermodynamic impedance factor and the operation entropy value coefficient in the initial index unit 104 are determined by the entropy weight method. The method automatically calculates the weight by analyzing the discrete degree of each coefficient value. The coefficient with greater numerical difference is considered to contain more information, and the weight allocated is higher. After determining the weight of each coefficient, the initial energy consumption regulation index is obtained by using a weighted fusion algorithm. Specifically, each coefficient is multiplied by its corresponding weight and summed. The initial energy consumption regulation index is a value between 0 and 1. The higher the initial energy consumption regulation index, the worse the comprehensive performance of the target temperature control domain in the three dimensions of device heat generation, heat dissipation difficulty and operation instability, and therefore the more urgent need for energy efficiency regulation measures. In the subsequent cooling resource allocation, a higher priority should be given. However, the initial index only reflects the static basic characteristics of the temperature control domain and does not take into account dynamic factors such as real-time operating state and environmental parameters. Therefore, if energy consumption control is only based on the initial index, accurate regulation cannot be achieved. To solve this problem, the initial energy consumption regulation index is set as a baseline reference value. In subsequent steps, the initial index is dynamically corrected and optimized based on multi-modal machine vision data and real-time environmental parameter data, so that the final regulation strategy can accurately respond to the real-time state of the computer room and achieve precise control of the energy consumption of the computer room.

[0029] Further, the computer room energy consumption control system based on machine vision further comprises: An energy efficiency index acquisition module 200 identifies dynamic energy consumption hot spot areas in the computer room based on multi-modal machine vision data and extracts corresponding multi-dimensional energy efficiency influence feature sets.

[0030] Specifically, Figure 3 The structure block diagram of the energy efficiency index acquisition module 200 in the system provided by the embodiment of the application is shown.

[0031] In the preferred embodiment provided by the application, the energy efficiency index acquisition module 200 specifically comprises: A graph generation unit 201 extracts device state visual features, temperature distribution features and personnel behavior features by analyzing multi-modal machine vision data of each target temperature control domain, analyzes through an abnormal area determination mechanism and color labeling preset rules, and obtains a thermodynamic situation graph; A boundary positioning unit 202 performs boundary evolution on the thermodynamic situation graph through semantic segmentation and a semantic weighted active contour model, and performs rigid constraint alignment on the device physical contour during the boundary evolution process to obtain a multi-modal fusion heat domain boundary. The feature set extraction unit 203 demarcates the energy consumption hot spot sub-regions based on the multi-modal fusion thermal domain boundary for each target temperature control domain, and extracts the corresponding multi-dimensional energy efficiency influence feature set, including calculating the load density index, the heat dissipation performance attenuation coefficient and the personnel activity thermal disturbance index.

[0032] In the embodiment of the present application, three special convolutional neural network branches are used to process visible light, infrared thermal imaging and dynamic behavior data respectively, to extract device state visual features, temperature distribution features and personnel behavior features, wherein the device state visual features are indicator light flicker frequency, screen brightness value and fan speed, the temperature distribution features are surface temperature gradient and thermal spot morphology, and the personnel behavior features are personnel density and moving track. The abnormal sub-region is determined through an abnormal region determination mechanism, specifically: 1) device state abnormality determination, comparing the extracted visual features with the device normal state database, if the indicator light flicker frequency exceeds the type-specific threshold value, the screen brightness value is continuously lower than the set threshold value, the fan angular velocity deviates from the rated speed by more than 20%, and any two of the above three conditions are met, an abnormality flag is triggered, and the three-dimensional space coordinates are recorded; 2) temperature abnormality determination, using an adaptive threshold segmentation algorithm to calculate the Z-score value of the temperature distribution in the temperature control domain, marking the region with continuous temperature Z-score>3 as a thermal abnormal region, and obtaining its spatial position information, wherein the Z-score value is a standard score, which is not described here in the prior art; 3) personnel gathering determination, using a density clustering algorithm to calculate the spatial distribution density of personnel, identifying the region with personnel density>2 people per square meter and lasting>5 minutes, and labeling the spatial range. The spatial coordinate information of the three types of abnormal regions is projected into the three-dimensional space model of the machine room, and color labeling is performed according to the preset rules, red represents double abnormal sub-regions of device state abnormality and temperature abnormality, yellow represents single abnormal sub-regions, and blue represents personnel gathering sub-regions, to generate a thermodynamic situation map. The map not only contains abnormal type information, but also accurately records the spatial coordinates and geometric morphology of each abnormal region, providing a spatial information basis for subsequent multi-modal fusion thermal domain boundary generation.

[0033] The red, yellow and blue regions in the thermodynamic state map are semantically segmented, and the pixel clusters of each color region are extracted; based on the abnormal severity level represented by different colors, a semantic weighted active contour model is used for boundary evolution, specifically, the red region is assigned the highest weight such as 0.6, the yellow region is the second such as 0.3, and the blue region is the lowest such as 0.1, so that the initial boundary expands preferentially to the high-weight severe abnormal region. The boundary evolution process is rigidly constrained to align with the physical outline of the equipment identified in the high-resolution visible light image, such as the metal frame of the cabinet and the precise edge of the server shell. This constraint ensures that the generated thermal domain boundary will never cut the equipment entity, but strictly follows the physical boundary of the equipment, so that the delineated multi-modal fusion thermal domain boundary not only contains the severity information of the thermodynamic anomaly, but also has accurate and locatable physical space coordinates. This method combines abnormal semantic weight and equipment physical geometry constraint, and differentiates the boundary evolution by color-coded semantic weight, which not only ensures the highest priority of precise definition of the red severe abnormal region, but also retains the complete spatial information of the yellow and blue regions of different levels of abnormal regions and potential influencing factors, providing a structured spatial framework for subsequent extraction of multi-dimensional energy efficiency influence feature set.

[0034] Based on the multi-modal fusion thermal domain boundary, the energy consumption hotspot sub-region of each target temperature control domain is delineated, and its corresponding multi-dimensional energy efficiency influence feature set is extracted, including the calculation of load density index, heat dissipation efficiency attenuation coefficient and personnel activity heat disturbance index. Specifically, based on the real-time power consumption data of the servers in the energy consumption hotspot sub-region and the processor utilization rate, the load density index is calculated by using the weighted average algorithm, which is equal to (∑(single server power consumption x CPU utilization rate)) / area, directly reflecting the spatial concentration of the operation load; by comparing the actual measured temperature in the energy consumption hotspot sub-region with the expected temperature of the ideal heat dissipation model based on CFD simulation, the heat dissipation efficiency attenuation coefficient is calculated, which is equal to (measured maximum temperature-expected temperature) / (device allowed maximum temperature-expected temperature), quantifying the attenuation degree of heat dissipation efficiency; by analyzing the personnel activity data in the energy consumption hotspot sub-region, the number of personnel, the length of stay and the activity intensity per unit time are counted, among which the activity intensity is calculated based on the motion amplitude of the key points of the skeleton, and the value is obtained by calculating the displacement amplitude change of these key points per unit time, and the data is processed by the Min-Max standardization method and then calculated by the regression model to obtain the personnel activity heat disturbance index, wherein the model obtains the weight coefficients ρ1, ρ2, ρ3 and the intercept term ρ0 of each parameter by least squares fitting, forming the regression equation heat disturbance index = ρ0 + ρ1X1 + ρ2X2 + ρ3X3, where X1, X2 and X3 are the number of personnel, the length of stay and the activity intensity per unit time after standardization, and the index reflects the heat load contribution of human body heat dissipation to the microenvironment of the computer room.

[0035] Further, the computer room energy consumption control system based on machine vision further comprises: An index preliminary optimization module 300 performs self-adaptive optimization on an initial energy consumption regulation index based on time sequence variation characteristics of a multi-dimensional energy efficiency influence feature set, and generates a preliminary optimized energy consumption regulation index.

[0036] Specifically, Figure 4 The structure block diagram of the index preliminary optimization module 300 in the system provided by the embodiment of the present application is shown.

[0037] In the preferred embodiment provided by the present application, the index preliminary optimization module 300 specifically comprises: A feature extraction unit 301 establishes a time sequence variation model of the multi-dimensional energy efficiency influence feature set, extracts the variation rate, fluctuation amplitude and trend characteristics of the computational load density index, the heat dissipation efficiency attenuation coefficient and the personnel activity heat disturbance index in the time dimension, and generates a feature set dynamic feature vector; A feature weight distribution unit 302 calculates the dynamic weight of each feature parameter in the current time sequence window by using an improved entropy weight method; An energy consumption anomaly evaluation unit 303 performs parameter optimization on a basic Gaussian membership function based on the feature set dynamic feature vector and the dynamic weight, and obtains an optimized Gaussian membership function; based on the fuzzy language values in the optimized Gaussian membership function, a fuzzy rule base is constructed through historical data mining; based on the optimized Gaussian membership function and the fuzzy rule base, a fuzzy output set is generated through weighted reasoning, and a regional energy consumption state evaluation value is obtained through the area barycenter method; An index optimization unit 304 creates a preset regulation index optimization function, and optimizes the initial energy consumption regulation index of the target temperature control domain based on the regional energy consumption state evaluation value, generates a preliminary optimized energy consumption regulation index, and uses the index for the generation of the final optimized energy consumption regulation index.

[0038] The regulation index optimization function is wherein is the preliminary optimized energy consumption regulation index of the target temperature control domain, is the initial energy consumption regulation index of the target temperature control domain, and E is the regional energy consumption state evaluation value, is an energy consumption state evaluation reference value, which is an experience value 0.5, representing the critical state of normal and abnormal, is a relative deviation amount based on the state evaluation value, which is an optimization factor, and K is an adjustment coefficient of the optimization strength, and K>0; In the embodiment of the present application, the feature extraction unit 301 performs deep time sequence analysis on the calculation load density index, heat dissipation performance attenuation coefficient and personnel activity heat disturbance index by constructing a time sequence change model of the multi-dimensional energy efficiency influence feature set. The specific time sequence change model uses a sliding window mechanism to obtain continuous sampling values of each index within a preset time window, and extracts three types of key time sequence features, which are change rate, fluctuation amplitude and trend feature. The change rate is obtained by calculating the first-order difference of the values of adjacent time points and normalizing, the fluctuation amplitude is obtained by calculating the ratio of the standard deviation to the mean of the data in the window, i.e. the coefficient of variation, and the trend feature is defined by the change directionality, which is the monotonicity trend of the feature value over time, and the number of change slopes per unit time is obtained. Finally, the three types of time sequence features of each index are combined in a fixed order to obtain a feature set dynamic feature vector, which contains 9 dimensions, i.e. 3 indexes x 3 types of features. This vector represents the evolution law of the energy efficiency state of the target temperature control area, and provides a data basis for subsequent intelligent optimization.

[0039] The feature weight allocation unit 302 uses an improved entropy weight method to realize dynamic allocation of feature weights. Specifically, the time sequence feature values of the 9 dimensions in the parameter dynamic feature vector are standardized to eliminate the dimension influence; the information entropy of each feature is calculated, wherein the faster the feature value change rate or the greater the fluctuation amplitude, the smaller the information entropy, indicating that the feature has lower uncertainty and higher information contribution in evaluation; the weight of each feature is obtained by normalizing the reciprocal of its information entropy, and the feature with smaller information entropy obtains higher weight; the dynamic weight allocation scheme of each feature in the current time sequence window is output, wherein the normalization processing is that the weight of each feature is equal to dividing 1 by the information entropy of the feature, and then dividing by the sum of the reciprocals of the information entropies of all features.

[0040] The energy consumption anomaly assessment unit 303 establishes a basic Gaussian membership function for each dimension of the dynamic feature vector of the feature set by combining expert experience initialization with historical data clustering analysis. Specifically, domain experts initially determine the center value c and standard deviation σ corresponding to the fuzzy linguistic values ​​of each feature based on prior knowledge, forming initial function parameters, where the fuzzy linguistic values ​​of each feature are low, medium, and high. The historical dataset is clustered using a fuzzy C-means clustering algorithm. The center value c set by the experts is optimized and adjusted by the cluster center, and the standard deviation σ is optimized and adjusted by the cluster variance to obtain the basic Gaussian membership function. If the feature weight is greater than a set threshold, the basic Gaussian membership function is reduced according to the weight ratio, such as multiplying the standard deviation σ of the Gaussian function by the weight decay coefficient to obtain the optimized Gaussian membership function, which can be adjusted according to the actual situation. Based on the fuzzy linguistic values ​​in the optimized Gaussian membership function, a fuzzy rule base is constructed through historical data mining. The rule premises combine linguistic values ​​from nine dimensions, such as IF (high feature in dimension 1) AND (medium feature in dimension 2) ... THEN... . The rule conclusions correspond to the semantic terms of the energy consumption anomaly level in the output variable, such as normal, slightly abnormal, or severely abnormal. The optimized Gaussian membership function is used to match the premises in the rule base. The prod-sum inference method is used to calculate the activation strength of each rule. The matching degree of the IF part of the rule is calculated by the algebraic product of the membership values ​​of each condition. The conclusion of the THEN part is obtained by performing an algebraic product operation between the activation strength and the output membership function to obtain the output fuzzy subset of each rule. All output subsets are aggregated using a sum operation to form the total output fuzzy set. The centroid position of this fuzzy set in the [0,1] universe is calculated using the area centroid method, with the formula y*=(Σμ i ×y i ) / (Σμ i ), where y i Discrete points in the universe of discourse refer to a pre-set sequence of discretized sampling points within the range [0,1] for the output variable's energy consumption anomaly level, such as y. i =[0,0.01,0.02,...,1.0],μ i The total membership degree refers to the membership degree at each specific discrete point y. i The above is the membership value calculated from the total output fuzzy set, which is the membership value corresponding to the point in the aggregated fuzzy set. y* is the regional energy consumption status assessment value, which represents the degree of energy consumption abnormality. 0-0.3 indicates normal, 0.3-0.7 indicates slight abnormality, and 0.7-1 indicates severe abnormality. It provides a precise basis for subsequent energy consumption regulation. The fuzzy output set is the output fuzzy subset and the total output fuzzy set.

[0041] When the regional energy consumption status assessment value E is higher than the benchmark value When the evaluation value E is higher than the reference value E0, it indicates that the energy consumption of the region is relatively serious, the correction factor is positive, the function will increase the initial energy consumption control index, thereby improving the energy efficiency control priority of the region; when the evaluation value E is lower than the reference value E0 , the correction factor is negative, and the function will correspondingly decrease the control index. The adjustment coefficient K is used to control the amplitude of optimization, so as to realize different degrees of control intensity.

[0042] Further, the computer room energy consumption control system based on machine vision further comprises: An environment conformity analysis module 400 generates an environment parameter conformity index based on real-time environment parameter data of the target temperature control domain and in combination with a historical normal state data set.

[0043] Specifically, Figure 5 The structure block diagram of the environment conformity analysis module 400 in the system provided by the embodiment of the application is shown.

[0044] In the preferred embodiment provided by the application, the environment conformity analysis module 400 specifically comprises: A reference data extraction unit 401 retrieves data records with similar basic energy efficiency characteristic parameters and multi-modal machine vision features of the current target temperature control domain in the historical data set, screens out records in a normal operation state, and constitutes a reference data set; An environment parameter extraction unit 402 extracts various environment parameter data in the reference data set, calculates statistical distribution characteristics thereof, determines a normal fluctuation interval range of multi-dimensional environment parameters, and generates a multi-dimensional environment parameter reference interval set; the multi-dimensional environment parameters include airflow organization efficiency, cooling return temperature, supply and return air temperature difference, air age, and particulate matter concentration distribution; A conformity transformation unit 403 collects real-time multi-dimensional environment parameter data of the target temperature control domain, compares each parameter with the multi-dimensional environment parameter reference interval, calculates the degree of deviation of each parameter from the reference interval, converts the deviation degree into a conformity score by using a fuzzy membership method, finally weights and integrates the conformity scores of all parameters, and generates a multi-dimensional environment parameter conformity index.

[0045] In the embodiment of the present application, retrieval is performed in the historical data set by an improved k-nearest neighbor algorithm, the feature vector of which is composed of the basic energy efficiency characteristic parameters of the target temperature control domain and multi-modal machine vision features, and the weighted Euclidean distance is used as the similarity measurement standard, wherein the weight coefficient of the machine vision features is set to 0.6, and the weight of the energy efficiency characteristic parameters is 0.4; in the screening process, the first k samples with the highest similarity are selected, for example, the value of k is determined to be 15 through cross-validation, and the records in the normal operating state are screened out based on the operating state labels of these samples, which need to meet the following three conditions at the same time: 1) the equipment load rate is continuously between 30% and 70% of the rated power; 2) the temperature fluctuation standard deviation is less than 0.5℃; 3) there is no any alarm event record, thereby forming a high-quality reference data set.

[0046] The normal fluctuation interval range refers to the confidence interval of the statistical distribution of the multi-dimensional environmental parameters in the reference data set under the normal operating state, and the multi-dimensional environmental parameters include air distribution efficiency, cooling return temperature, supply and return air temperature difference, air age, and particulate matter concentration distribution. The normal fluctuation range of each environmental parameter is determined by calculating the percentile of each environmental parameter, for example, the 5% and 95% percentiles. Specifically, the distribution curve of each parameter is fitted by using the kernel density estimation method, and then the 5% and 95% percentiles of the cumulative distribution function are taken as the upper and lower bounds of the interval. This method can effectively handle non-normal distribution data. The generated multi-dimensional environmental parameter reference interval set includes, for example, air distribution efficiency [0.35, 0.65], cooling return temperature [18℃, 24℃], supply and return air temperature difference [4℃, 8℃], air age [60s, 180s], and particulate matter concentration distribution [0.01mg / m³, 0.05mg / m³]. This set provides a dynamic reference for subsequent environmental state assessment.

[0047] The multi-dimensional environmental parameter data of the target temperature control domain is collected by the real-time monitoring system, and the degree of deviation of each parameter from the reference interval is calculated using the relative deviation formula, Deviation = (measured value - median of reference interval) / (upper bound of reference interval - lower bound of reference interval) x 2. The deviation is converted into a compliance score by using a triangular membership function, wherein the score of the complete compliance interval is 1, the score of the critical state is 0.5, and the score of the exceeding critical state is 0, wherein the complete compliance interval is Deviation = 0, and the critical state is Deviation = ±0.5. The weights are assigned according to the importance of each parameter on energy consumption, specifically, the weight of air distribution efficiency is 0.3, the weight of cooling return temperature is 0.25, the weight of supply and return air temperature difference is 0.2, the weight of air age is 0.15, and the weight of particulate matter concentration is 0.1. The multi-dimensional environmental parameter compliance index is generated by weighted average calculation. The index takes a value in the range of 0-1, and the higher the value, the more the environmental parameters comply with the normal operating state.

[0048] Further, the machine vision-based computer room energy consumption control system further comprises: An index re-optimization module 500 generates a multi-dimensional environment stability coefficient based on the time sequence variation characteristics of the environment parameters, and re-optimizes the preliminary optimized energy consumption regulation index in combination with a multi-dimensional environment parameter compliance index to generate a final energy consumption regulation index of the target temperature control domain.

[0049] Specifically, Figure 6 The structural block diagram of the index re-optimization module 500 in the system provided by the embodiment of the application is shown.

[0050] In the preferred embodiment provided by the application, the index re-optimization module 500 specifically comprises: An environment parameter time sequence analysis unit 501 calculates the variation variance and trend stability index of each parameter based on the multi-dimensional environment parameter time sequence data set; An environment stability coefficient unit 502 calculates a dynamic weight by using an entropy weight method based on the variance and trend stability index of each parameter, and generates a multi-dimensional environment instability coefficient by weighting; An index final optimization unit 503 processes the multi-dimensional environment parameter compliance index and the multi-dimensional environment instability coefficient by using an environment situation assessment function to generate an environment abnormality index, and uses the index to finally adjust the preliminary optimized energy consumption regulation index to generate a final energy consumption regulation index of the target temperature control domain.

[0051] In the embodiment of the application, after each parameter time sequence data in the multi-dimensional environment parameter time sequence data set is preprocessed and stationarity tested, the time sequence variance σ² of each parameter time sequence data is calculated to quantify the fluctuation amplitude, and the trend stability index is calculated by using linear regression to fit the variation trend and calculating the residual standard deviation of the trend line. The smaller the index value is, the more stable the parameter variation trend is.

[0052] The variation variance and trend stability index of each parameter are normalized to eliminate the dimension influence, and then the dynamic weight of each parameter index is calculated by using an improved entropy weight method. The entropy weight method allocates the weight according to the discrete degree of each index value. The greater the discrete degree is, the higher the weight is. Finally, the normalized index value and the corresponding weight are weighted and summed to generate a comprehensive environment instability coefficient in the range of 0-1. The coefficient comprehensively reflects the fluctuation and unpredictability of the environment parameters in the time dimension. The greater the coefficient value is, the more unstable the environment state is.

[0053] The calculation process of the final energy consumption regulation index is as follows: the environment situation assessment function is wherein represents the environment abnormality index, represents a gain coefficient, a constant for controlling the overall adjustment sensitivity, usually set to 1 or less than 1, and ε is a smoothing factor, a very small positive number such as 0.001, for preventing the denominator from being zero and ensuring the calculation stability, is a dynamic reference value, obtained by calculating the moving average or median of the Q value in the historical data, for example, in the past week or month under normal operating conditions, where represents the typical problem level that the system is subjected to under normal conditions, is a weight coefficient, satisfying , C and S are the multi-dimensional environmental parameter conformity index and the multi-dimensional environmental instability coefficient, respectively; the environmental disturbance index is then applied to the preliminary energy consumption control index through an index optimization model to obtain the final energy consumption control index, and the index optimization model is , where is the final energy consumption control index, is the adjustment coefficient of the environmental disturbance index, and >0. The design of the environmental disturbance coefficient enables the energy consumption control index to have a two-way intelligent adjustment capability. If the current environmental problem level is the same as the historical normal level, the coefficient is zero, and the control index remains unchanged. When the environmental condition deteriorates and the problem level exceeds the historical normal level, the coefficient is positive, and the control index is correspondingly increased to enhance the control intensity. Conversely, when the environmental condition is better than the historical normal level, the coefficient is negative, and the control index is correspondingly reduced to reduce energy consumption, thereby achieving precise on-demand control. The method dynamically adjusts the control strategy through real-time comparison with the historical normal environment, ensuring the safety of equipment operation and optimizing the overall energy efficiency. The final energy consumption control index is the ultimate decision parameter for guiding the resource allocation of the computer room cooling system, and its value directly determines the priority and intensity of cooling measures to be taken in the region, aiming to achieve precise on-demand refrigeration, thereby achieving the dual goals of improving energy efficiency and ensuring equipment operation safety.

[0054] It should be understood that although each step in the flowchart of each embodiment of the present application is displayed in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in each embodiment can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or sub-steps or stages of other steps.

[0055] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer readable storage medium, and when the program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0056] The above is a description of the present application and should not be considered as a limitation. Those skilled in the art can make many modifications or replacements to the exemplary embodiments without departing from the essence of the present application, and all these modifications or replacements shall be included in the scope of the present application defined by the claims.

Claims

1. A machine vision based computer room energy consumption control system, characterized by, The method comprises the following steps: The control index determination module obtains the basic energy efficiency characteristic parameters, multi-modal machine vision data, and multi-dimensional environment parameter time series data set of the target temperature control domain, and generates an initial energy consumption control index of the target temperature control domain based on the basic energy efficiency characteristic parameters; The energy efficiency index acquisition module identifies the dynamic energy consumption hotspot area in the computer room based on the multi-modal machine vision data, and extracts the corresponding multi-dimensional energy efficiency influence feature set; The index preliminary optimization module performs adaptive optimization on the initial energy consumption control index based on the time series variation characteristics of the multi-dimensional energy efficiency influence feature set, and generates a preliminary optimized energy consumption control index; The environment compliance analysis module generates an environment parameter compliance index based on the real-time environment parameter data of the target temperature control domain and the historical normal state data set; The index re-optimization module generates a multi-dimensional environment stability coefficient based on the time series variation characteristics of the environment parameters, and re-optimizes the preliminary optimized energy consumption control index based on the multi-dimensional environment parameter compliance index to generate a final energy consumption control index of the target temperature control domain.

2. The machine vision-based computer room energy consumption control system of claim 1, wherein, The control index determination module comprises: The target determination unit divides and analyzes the computer room based on the area division basis to obtain the target temperature control domain, and the area division basis includes cabinet power density, heat dissipation characteristics, and air flow organization correlation; The data acquisition unit obtains the basic energy efficiency characteristic parameters, multi-modal machine vision data, and multi-dimensional environment parameter time series data set of the target temperature control domain, and the basic energy efficiency characteristic parameters include device inherent parameters, cabinet structure parameters, and historical operation parameters.

3. The machine vision-based computer room energy consumption control system of claim 2, wherein, The control index determination module further comprises: The coefficient generation unit performs multi-dimensional thermal efficiency evaluation on the device inherent parameters to obtain a thermal inertia load factor, performs fluid dynamics characteristic analysis on the cabinet structure parameters to obtain an aerodynamic thermodynamic impedance factor, and quantifies the running steady state characteristics of the historical operation parameters to obtain a running entropy coefficient; The initial index unit performs dynamic weight distribution and multi-source data fusion on the thermal inertia load factor, the aerodynamic thermodynamic impedance factor, and the running entropy coefficient to obtain an initial energy consumption control index.

4. The machine vision-based computer room power consumption control system of claim 1, wherein, The energy efficiency index acquisition module comprises: The atlas generation unit extracts device state visual features, temperature distribution features, and personnel behavior features by analyzing the multi-modal machine vision data of each target temperature control domain, and obtains a thermodynamic situation atlas through abnormal area determination mechanism and color labeling preset rules; The boundary positioning unit performs boundary evolution on the thermodynamic situation atlas through semantic segmentation and semantic weighted active contour model, and aligns the device physical contour in the boundary evolution process to obtain a multi-modal fusion thermal domain boundary; The feature set extraction unit divides the energy consumption hotspot sub-area of each target temperature control domain based on the multi-modal fusion thermal domain boundary, and extracts the corresponding multi-dimensional energy efficiency influence feature set, including the calculation of load density index, heat dissipation efficiency decay coefficient, and personnel activity thermal disturbance index.

5. The machine vision-based computer room power consumption control system of claim 1, wherein, The index preliminary optimization module comprises: The feature extraction unit establishes a time sequence change model of the multi-dimensional energy efficiency influence feature set, extracts a change rate, a fluctuation amplitude and a trend feature of a calculation load density index, a heat dissipation performance attenuation coefficient and a personnel activity heat disturbance index in a time dimension, and generates a dynamic feature vector of the feature set; The feature weight allocation unit calculates dynamic weights of each feature parameter in a current time sequence window by using an improved entropy weight method.

6. The machine vision-based computer room power consumption control system of claim 5, wherein, The index preliminary optimization module further includes: The energy consumption anomaly evaluation unit optimizes a basic Gaussian membership function based on the dynamic feature vector of the feature set and the dynamic weights, to obtain an optimized Gaussian membership function; constructs a fuzzy rule base by historical data mining based on fuzzy language values in the optimized Gaussian membership function; generates a fuzzy output set by weighted reasoning based on the optimized Gaussian membership function and the fuzzy rule base, and obtains a regional energy consumption state evaluation value by an area barycenter method; The index optimization unit creates a preset control index optimization function, and optimizes an initial energy consumption control index of the target temperature control domain based on the regional energy consumption state evaluation value, to generate a preliminary optimized energy consumption control index.

7. The machine vision-based computer room power consumption control system of claim 1, wherein, The environment conformity analysis module includes: The reference data extraction unit searches data records with similar basic energy efficiency characteristic parameters and multi-modal machine vision features in the current target temperature control domain in the historical data set, filters out records in a normal operation state, and constitutes a reference data set; The environment parameter extraction unit extracts various environment parameter data in the reference data set, calculates statistical distribution characteristics thereof, determines a normal fluctuation interval range of multi-dimensional environment parameters, and generates a multi-dimensional environment parameter reference interval set; the multi-dimensional environment parameters include an air flow organization efficiency, a cooling return temperature, a supply-return air temperature difference, an air age, and a particulate matter concentration distribution; The conformity transformation unit collects real-time multi-dimensional environment parameter data of the target temperature control domain, compares the data with the multi-dimensional environment parameter reference interval item by item, calculates a degree of deviation of each parameter from the reference interval, transforms the degree of deviation into a conformity score by using a fuzzy membership method, and finally generates a multi-dimensional environment parameter conformity index by weightedly integrating conformity scores of all parameters.

8. The machine vision-based computer room power consumption control system of claim 1, wherein, The index re-optimization module includes: The environment parameter time sequence analysis unit calculates a change variance and a trend stability index of each parameter based on a multi-dimensional environment parameter time sequence data set; The environment stability coefficient unit calculates dynamic weights by using an entropy weight method based on the variance and the trend stability index of each parameter, and generates a multi-dimensional environment instability coefficient by weighting; The index final optimization unit generates an environment anomaly index by processing the multi-dimensional environment parameter conformity index and the multi-dimensional environment instability coefficient by using an environment situation evaluation function, and finally adjusts the preliminary optimized energy consumption control index by using the index, to generate a final energy consumption control index of the target temperature control domain.

9. The machine vision-based computer room power consumption control system of claim 6, wherein, The control index optimization function includes: wherein is a preliminary optimized energy consumption regulation index of the target temperature control area, is an initial energy consumption regulation index of the target temperature control area, E is a regional energy consumption state evaluation value, is an energy consumption state evaluation benchmark value, representing a critical state between normal and abnormal, is a relative deviation amount based on the state evaluation value, as an optimization factor, K is an adjustment coefficient of optimization intensity, and K>

0.

10. The machine vision-based computer room power consumption control system of claim 6, wherein, The environment situation evaluation function includes: wherein represents the environmental change index, represents the gain coefficient, and e is a smoothing factor, is a dynamic reference value, is a weight coefficient, satisfying C and S are a multi-dimensional environmental parameter conformity index and a multi-dimensional environmental instability coefficient, respectively.

Images