Multi-scene-oriented computing power infrastructure construction process parameter dynamic optimization method

Abstract

The invention discloses a multi-scene-oriented computing power infrastructure construction process parameter dynamic optimization method, and belongs to the technical field of intelligent construction. Comprising the following steps: collecting computing power infrastructure construction scene data in real time, and identifying scene categories; in combination with a random forest-gradient boosting tree fusion algorithm and the structured matrix, constructing and training a construction process parameter model, inputting a scene type, and outputting an initial construction process parameter combination; the initial construction process parameter combination is executed, and deviation judgment is conducted on the construction effect; when the deviation is greater than a deviation threshold value, retraining the construction process parameter model, and updating the structured matrix; the method is high in multi-scene adaptability, and the problem that in the prior art, the efficiency loss is large due to the fact that the general parameter adaptability is poor is solved; through the combination of the random forest-gradient boosting tree fusion algorithm and the structured matrix, the non-linear relation fitting is more accurate, the cabinet installation levelness is ensured to meet the operational stability requirement of the computing power equipment, and the quality standard-reaching rate is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mobile communication network, in particular to a multi-scene-oriented computing power infrastructure construction process parameter dynamic optimization method. BACKGROUND

[0002] Under the impetus of digital economy, the construction demand of computing power infrastructure such as data center and edge node increases by 20% per year, and the scene is diverse, outdoor, indoor, large, small, and the construction needs to meet high stability and high precision; in the prior art, for example, CN202111448116.8, only based on multi-construction party constraints (period, resources) optimization, without considering the scene difference of computing power infrastructure in outdoor soil bearing capacity, indoor electromagnetic interference, etc., the general parameter adaptability is poor, and there is no scene distinction; focusing on resource balance and construction period, without covering computing power infrastructure exclusive parameters such as cabinet precision and air duct efficiency, which cannot meet the operation demand of computing power equipment, and the parameter dimension is single.

[0003] In view of the limitations of the prior art, a method of scene quantization, exclusive modeling, real-time optimization and closed-loop verification is proposed, which solves the problems of multi-scene adaptation, parameter precision, adjustment timeliness through computing power infrastructure exclusive scene classification, parameter model and monitoring index, and finally improves the construction efficiency and quality. SUMMARY

[0004] The purpose of the present application is to provide a multi-scene-oriented computing power infrastructure construction process parameter dynamic optimization method to solve the problems raised in the background art.

[0005] In order to solve the above technical problems, the present application provides the following technical scheme: The multi-scene-oriented computing power infrastructure construction process parameter dynamic optimization method comprises: Real-time collection of computing power infrastructure construction scene data and identification of scene category; Combining random forest-gradient boosting tree fusion algorithm and structured matrix, constructing and training construction process parameter model, inputting scene type, and outputting initial construction process parameter combination; Execute the initial construction process parameter combination, and make deviation judgment on the construction effect; when the deviation is greater than the deviation threshold, retrain the construction process parameter model and update the structured matrix.

[0006] As an preferred scheme of the multi-scene-oriented computing power infrastructure construction process parameter dynamic optimization method, the specific implementation process of real-time collection of computing power infrastructure construction data and identification of scene category comprises: Before the construction of computing power infrastructure, the computing power infrastructure construction scene data is collected by sensors and database query, and the computing power infrastructure construction scene data includes environmental data and computing power infrastructure exclusive information; The collected computing infrastructure construction scene data is matched with each index of each preset scene to determine the index score of each index; the analytic hierarchy process is used to set the index weight, and the index scores of each preset scene are weighted and summed to obtain the total score of each preset scene; The total score of each preset scene is compared with the score threshold of each preset scene to determine the scene category.

[0007] As a preferred scheme of the multi-scene-oriented computing infrastructure construction process parameter dynamic optimization method, the specific implementation process of constructing and training the construction process parameter model, inputting the scene type, and outputting the initial construction process parameter combination includes: Based on the scene category, historical construction case data under the scene category is called from the historical construction database, including construction process parameters and corresponding construction effect data, and a training set and a test set are constructed; Based on the training set, a fusion model capable of predicting construction effect is constructed by using a random forest-gradient boosting tree fusion algorithm: The scene features and construction process parameters corresponding to the scene category in the training set are taken as the input feature vector, and the key performance indicators of the construction effect data are taken as the target vector to construct and train a first random forest model; the importance score of each input feature is calculated through the first random forest model, and the input features are filtered according to the importance score to obtain a key feature subset with an importance higher than a preset threshold; The key feature subset is taken as the input, and the key performance indicators of the construction effect data are taken as the training target to construct and train a second gradient boosting tree model; in the training process, the second gradient boosting tree model iteratively optimizes the residual error of the previous round of prediction to fit the nonlinear relationship between the construction process parameters and the construction effect; The trained fusion model is verified using the test set to evaluate its prediction accuracy; when the prediction accuracy meets the preset requirement, the final fusion model is output; Based on the trained fusion model, a structured matrix including scene features, construction process parameters, and optimization targets is constructed; The scene category is input into the structured matrix to retrieve and output an initial recommended construction process parameter set; The initial recommended construction process parameters are reasonably judged, and when the judgment fails, the n groups of cases with the smallest deviation are called from the historical qualified cases, each construction process parameter is weighted and calculated to obtain a new recommended construction process parameter set, and the reasonableness is judged; until the judgment is successful, a final group of construction process parameters is output as the initial construction process parameter combination.

[0008] The rationality judgment includes a preset constraint condition of a construction process parameter. When the initial recommended construction process parameter does not satisfy the constraint condition, the judgment fails; and when the initial recommended construction process parameter satisfies the constraint condition, the judgment succeeds.

[0009] As a preferred scheme of the multi-scene-oriented dynamic optimization method of construction process parameters of computing power infrastructure, the specific implementation process of the step of executing the initial construction process parameter combination and judging the deviation of the construction effect includes: The initial construction process parameter combination is executed, and computing power infrastructure construction data in the construction process are collected in real time through sensors and on-site monitoring equipment installed on the construction equipment; Based on the initial construction process parameter combination, optimization target data are extracted from the structured matrix; The computing power infrastructure construction data are compared with the optimization target data. When the deviation is less than or equal to a deviation threshold, the current construction process parameter is maintained; and when the deviation is greater than the deviation threshold, the construction process parameter model is retrained, and the structured matrix is updated As a preferred scheme of the multi-scene-oriented dynamic optimization method of construction process parameters of computing power infrastructure, the importance score of each input feature is calculated through the first random forest model. Specifically, the method of Gini impurity reduction or out-of-bag data error reduction is used for measurement.

[0010] As a preferred scheme of the multi-scene-oriented dynamic optimization method of construction process parameters of computing power infrastructure, the residual error of the previous round of prediction is taken as the focus for iterative optimization. Specifically, the second gradient boosting tree model minimizes the negative gradient of the loss function with respect to the current model prediction value by constructing a series of regression trees. That is, the residual error between the overall prediction result of the last round and the true label is approximated.

[0011] A computer device includes at least one processor and a memory. The memory stores a computer program. When the computer program is executed by the at least one processor, the computer device implements the multi-scene-oriented dynamic optimization method of construction process parameters of computing power infrastructure.

[0012] A computer readable storage medium stores a computer program. When the computer program is executed by a processor, the multi-scene-oriented dynamic optimization method of construction process parameters of computing power infrastructure is implemented.

[0013] Compared with the prior art, the beneficial effects achieved by the present application are: in the multi-scene oriented computing power infrastructure construction process parameter dynamic optimization method provided by the present application, the computing power infrastructure construction scene data is collected in real time, and the scene category is identified; the construction process parameter model is constructed and trained in combination with the random forest-gradient boosting tree fusion algorithm and the structured matrix, the scene type is input, and the initial construction process parameter combination is output; the initial construction process parameter combination is executed, and the deviation of the construction effect is determined; when the deviation is greater than the deviation threshold, the construction process parameter model is retrained, and the structured matrix is updated; the present application has strong multi-scene adaptability, and solves the problem of large efficiency loss caused by poor adaptability of general parameters in the prior art; through the combination of the random forest-gradient boosting tree fusion algorithm and the structured matrix, the nonlinear relationship fitting is more accurate, ensuring that the cabinet installation level meets the stable operation requirement of computing power equipment, and the quality standard rate is improved. BRIEF DESCRIPTION OF DRAWINGS

[0014] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, which together with the embodiments of the present application, serve to explain the present application, and do not constitute a limitation on the present application.

[0015] Figure 1 is a step schematic diagram of the multi-scene oriented computing power infrastructure construction process parameter dynamic optimization method of the present application. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0017] Please refer to Figure 1 In the first embodiment of the present application: a multi-scene oriented computing power infrastructure construction process parameter dynamic optimization method is provided, which comprises the following steps: Step 1: collecting computing power infrastructure construction scene data in real time, and identifying the scene category; Specifically, the specific implementation process of collecting computing power infrastructure construction data in real time and identifying the scene category comprises: Before the computing power infrastructure construction, the computing power infrastructure construction scene data is collected by means of sensors and database queries, and the computing power infrastructure construction scene data includes environmental data and computing power infrastructure information; The collected computing infrastructure construction scene data is matched with each index of each preset scene to determine the index score of each index; the analytic hierarchy process is used to set the index weight, and the index scores of each preset scene are weighted and summed to obtain the total score of each preset scene; The total score of each preset scene is compared with the score threshold of each preset scene to determine the scene category.

[0018] The purpose of this step is scene information collection and classification; through sensors, database queries, and other means, the environmental information and computing infrastructure construction exclusive information of the computing infrastructure construction scene are collected, the environmental information includes but is not limited to geological conditions such as soil type, rock hardness, climate conditions such as temperature, humidity, wind power, construction site size and shape, and surrounding environment restrictions such as whether it is close to a residential area, whether there are special buildings, etc. The computing infrastructure construction exclusive information includes cabinet deployment density (units / m2), design deviation allowance of heat dissipation air duct, power access capacity, and surrounding electromagnetic interference intensity.

[0019] The computer device classifies the collected scene information according to the classification rule to determine the category to which the construction scene belongs, such as outdoor large-scale computing center construction scene, indoor small-scale computing machine room reconstruction scene, etc. The role of this step is to provide a targeted scene basis for subsequent process parameter optimization, ensuring that the optimization scheme can adapt to the characteristics of different scenes.

[0020] The classification rule first uses the analytic hierarchy process (AHP) to set the index weight: geology 25%, climate 20%, site 15%, computing exclusive 30%, and surrounding environment 10%, and preset scene thresholds, for example: outdoor large-scale computing center scene: soil bearing capacity ≥200kPa, cabinet deployment density ≥8 units / m2, power access capacity ≥1000kW; indoor small-scale computing machine room reconstruction scene: site flatness deviation ≤2mm, air duct deviation allowance ≤3mm, electromagnetic interference ≤40dB; secondly, the scene is determined through index matching-weight accumulation-threshold comparison, specifically: index matching includes matching the actual value with the quantitative standard to determine the index score; weight accumulation includes index contribution value=index score×index layer weight×criterion layer weight, and the total score is obtained by summing all index contribution values; threshold comparison includes presetting the score threshold of each scene, and the total score corresponding to the threshold interval is the final scene category. For example, a project "outdoor scene score 82%, indoor 18%", is classified as an outdoor large-scale scene. The "scene accurate portrait" is provided for subsequent optimization, avoiding the adaptation failure of existing patent general parameters in computing infrastructure construction multiple scenes.

[0021] Step 2: Combine the random forest-gradient boosting tree fusion algorithm and the structured matrix to build and train the construction process parameter model, input the scene type, and output the initial construction process parameter combination; Specifically, the specific implementation process of the construction and training of the construction process parameter model, inputting a scene type, and outputting an initial construction process parameter combination includes: Based on the scene category, historical construction case data under the scene category is retrieved from the historical construction database, including construction process parameters and corresponding construction effect data, and a training set and a test set are constructed and trained; wherein the construction process parameters include cabinet installation level, heat dissipation air duct welding qualification rate, and special hoisting equipment power; the construction effect data includes construction period, cost, and cabinet installation qualification rate; Based on the training set, a fusion model capable of predicting construction effect is constructed by using a random forest-gradient boosting tree fusion algorithm: The scene features and construction process parameters corresponding to the scene category in the training set are taken as the input feature vector, and the key performance indicators of the construction effect data are taken as the target vector, to construct and train a first random forest model; the importance score of each input feature is calculated through the first random forest model, and the input features are filtered according to the importance score, to obtain a key feature subset with an importance higher than a preset threshold; The key feature subset is taken as the input, and the key performance indicators of the construction effect data are taken as the training target, to construct and train a second gradient boosting tree model; in the training process, the second gradient boosting tree model iteratively optimizes the residual error of the previous round of prediction as the focus, to fit the nonlinear relationship between the construction process parameters and the construction effect, such as the nonlinear relationship fitting of the air duct size-heat dissipation efficiency and other computing infrastructure; The trained fusion model is verified using the test set to evaluate its prediction accuracy; when the prediction accuracy meets the preset requirement, the final fusion model is output; Based on the trained fusion model, a structured matrix including scene features, construction process parameters, and optimization targets is constructed; The scene category is input into the structured matrix, and an initial recommended construction process parameter set is retrieved and output; The initial recommended construction process parameters are reasonably judged, when the judgment fails, the n groups of cases with the smallest deviation are retrieved from the historical qualified cases, the weighted calculation of each construction process parameter is performed, a new recommended construction process parameter set is obtained, and the reasonable judgment is performed; until the judgment is successful, a final construction process parameter set is output as the initial construction process parameter combination.

[0022] The reasonable judgment includes the constraint conditions of the preset construction process parameters, when the initial recommended construction process parameters do not meet the constraint conditions, the judgment fails; when the initial recommended construction process parameters meet the constraint conditions, the judgment succeeds.

[0023] In addition to the cost deviation rate of the existing technology, the newly added computing infrastructure quality standard needs to meet the following conditions simultaneously: Process parameter deviation rate: deviation from the mean of the same scenario compliance case ≤10%; Exclusive quality indicators: cabinet installation level compliance rate ≥98%, air duct air speed uniformity deviation ≤10%; If not, retrieve the 3 groups of compliance cases with the smallest deviation, and correct them by weighting the construction effect score (quality 60%, efficiency 30%, cost 10%) until they meet the standard. Provide power infrastructure construction exclusive parameter benchmarks for the initial construction stage to reduce trial and error costs.

[0024] Step 3: Execute the initial construction process parameter combination and determine the deviation of the construction effect; when the deviation is greater than the deviation threshold, retrain the construction process parameter model and update the structured matrix.

[0025] Specifically, the specific implementation process of executing the initial construction process parameter combination and determining the deviation of the construction effect includes: Execute the initial construction process parameter combination and collect power infrastructure construction data in the construction process in real time through sensors installed on construction equipment and on-site monitoring equipment; including equipment operation data: speed, temperature, and energy consumption of special lifting equipment; process data: cabinet installation level, air duct air speed, and power connection resistance; real-time quality values: concrete foundation rebound value and welding flaw detection pass rate; transmitted to the data center in real time through the 5G industrial gateway, stored using time series database + scene type, facilitating subsequent tracing and optimization. Avoid the problem of not adjusting in time due to data lag in the prior art.

[0026] Based on the initial construction process parameter combination, extract optimization target data from the structured matrix; Compare the power infrastructure construction data with the optimization target data; when the deviation is less than or equal to the deviation threshold, maintain the current construction process parameters; when the deviation is greater than the deviation threshold, retrain the construction process parameter model and update the structured matrix.

[0027] Every 50 cases of the same scenario are accumulated, the construction process parameter model is automatically updated with new data, and the accuracy of the initial parameter rationality determination is improved. Achieve data closed loop of power infrastructure parameter optimization, and continuously improve model adaptability.

[0028] Specifically, the importance score of each input feature is calculated by the first random forest model, which is measured by the method of Gini impurity reduction or out-of-bag data error reduction.

[0029] Specifically, the residual error of the previous round of prediction is focused on iterative optimization, specifically including the second gradient boosting tree model by constructing a series of regression trees, and the construction target of each new tree is to minimize the negative gradient of the loss function with respect to the current model prediction value, that is, to approximate the residual error between the overall prediction result of the last round and the true label.

[0030] A computer device comprising at least one processor and a memory, the memory storing a computer program, when the computer program is executed by the at least one processor, the computer device implements a multi-scene-oriented computing power infrastructure construction process parameter dynamic optimization method.

[0031] The computer device can be a terminal, and the computer device comprises a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved by WIFI, operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0032] A computer-readable storage medium, the storage medium storing a computer program, the computer program being executed by a processor to implement a multi-scene-oriented computing power infrastructure construction process parameter dynamic optimization method.

[0033] It should be noted that, in this document, relational terms such as first and second and the like can only be used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that these entities or operations exist in any such actual relationship or order. Moreover, the terms "comprise", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a list of elements does not only include those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or apparatus.

[0034] Finally, it should be noted that the above only describes the preferred embodiments of the present application and is not intended to limit the present application, and although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent replacements to some technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for dynamic optimization of construction process parameters for computing infrastructure in multiple scenarios, characterized in that, The method includes: Real-time data collection of computing infrastructure construction scenarios, and identification of scenario categories; By combining the random forest-gradient boosting tree fusion algorithm and structured matrix, a construction process parameter model is constructed and trained. The input scene type outputs the initial combination of construction process parameters. The initial combination of construction process parameters is executed, and the deviation of the construction effect is judged. When the deviation is greater than the deviation threshold, the construction process parameter model is retrained and the structured matrix is ​​updated.

2. The method for dynamic optimization of construction process parameters of computing infrastructure for multiple scenarios as described in claim 1, characterized in that, The specific implementation process of real-time acquisition of construction data for computing infrastructure and identification of scenario categories includes: Before the construction of computing infrastructure, data on the construction scenario of computing infrastructure is collected through sensors and database queries. The data on the construction scenario of computing infrastructure includes environmental data and computing infrastructure-specific information. The collected data on the construction scenarios of computing infrastructure are matched with the various indicators of each preset scenario to determine the score of each indicator; the weight of the indicators is set by the analytic hierarchy process, and the scores of the indicators of each preset scenario are weighted and summed to obtain the total score of each preset scenario. The total score for each preset scenario is compared with the score threshold for each preset scenario to determine the scenario category.

3. The method for dynamic optimization of construction process parameters of computing infrastructure for multiple scenarios as described in claim 2, characterized in that, The specific implementation process of constructing and training the construction process parameter model, inputting the scenario type, and outputting the initial combination of construction process parameters includes: Based on the scenario category, historical construction case data under the scenario category are retrieved from the historical construction database, including construction process parameters and corresponding construction effect data, and training set and test set are constructed. Based on the training set, a fusion model capable of predicting construction effects is constructed using a random forest-gradient boosting tree fusion algorithm. The scene features and construction process parameters corresponding to the scene categories in the training set are used as input feature vectors, and the key performance indicators of the construction effect data are used as target vectors to construct and train a first random forest model. The importance score of each input feature is calculated through the first random forest model, and the input features are filtered according to the importance score to obtain a subset of key features with an importance higher than a preset threshold. Using the key feature subset as input and the key performance indicators of construction effect data as training targets, a second gradient boosting tree model is constructed and trained. During the training process, the second gradient boosting tree model is iteratively optimized with the residual of the previous prediction as the focus, so as to fit the nonlinear relationship between construction process parameters and construction effect. The trained fusion model is validated using a test set to evaluate its prediction accuracy; when the prediction accuracy meets the preset requirements, the final fusion model is output. Based on the trained fusion model, a structured matrix including scene features, construction process parameters, and optimization objectives is constructed. Input the scene category into the structured matrix, retrieve and output a set of initially recommended construction process parameters; The rationality of the initially recommended construction process parameters is judged. If the judgment fails, n sets of compliant cases with the smallest deviation are selected from the historical compliant cases. The weighted calculation of each construction process parameter is performed to obtain a new set of recommended construction process parameters, and the rationality is judged. The process continues until the judgment is successful, and the final set of construction process parameters is output as the initial construction process parameter combination. The rationality determination includes preset constraints on construction process parameters. If the initially recommended construction process parameters do not meet the constraints, the determination fails; if the initially recommended construction process parameters meet the constraints, the determination succeeds.

4. The method for dynamic optimization of construction process parameters of computing infrastructure for multiple scenarios as described in claim 3, characterized in that, The specific implementation process of executing the initial construction process parameter combination and judging the deviation of the construction effect includes: The initial construction process parameter combination is executed, and the construction data of the computing infrastructure is collected in real time through sensors installed on the construction equipment and on-site monitoring equipment. Based on the initial combination of construction process parameters, optimization target data is extracted from the structured matrix; The construction data of the computing infrastructure is compared with the optimization target data. When the deviation is less than or equal to the deviation threshold, the current construction process parameters are maintained; when the deviation is greater than the deviation threshold, the construction process parameter model is retrained and the structured matrix is ​​updated.

5. The method for dynamic optimization of construction process parameters of computing infrastructure for multiple scenarios according to claim 4, characterized in that, The importance score of each input feature is calculated using the first random forest model, specifically measured by methods such as Gini impurity reduction or out-of-bag data error reduction.

6. The method for dynamic optimization of construction process parameters of computing infrastructure for multiple scenarios as described in claim 5, characterized in that, The iterative optimization, which focuses on the residuals of the previous prediction, specifically includes the second gradient boosting tree model constructing a series of regression trees. The goal of constructing each new tree is to minimize the negative gradient of the loss function with respect to the current model prediction value, that is, to approximate the residuals between the overall prediction results of the previous round and the true labels.

7. A computer device, characterized in that, It includes at least one processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the at least one processor, the computer device enables the computer device to implement the dynamic optimization method for construction process parameters of computing power infrastructure for multiple scenarios as described in any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method for dynamic optimization of construction process parameters of computing infrastructure for multiple scenarios as described in any one of claims 1 to 6.

Images