Prefabricated quick-plug power supply interface system suitable for cabinet and power distribution method

Abstract

The invention belongs to the technical field of cabinet power distribution, and provides a preassembled fast-plug power interface system suitable for a cabinet and a power distribution method.The method comprises the steps that the load power distribution quantity output by an AI in a historical database and the actual demand quantity collected by a sensor and a BMC are called, a power distribution deviation value and a deviation ratio are calculated, the deviation value is clustered through a DBSCAN algorithm, and a power distribution result is obtained; a time distribution dynamic range of the power distribution is output, a deviation rate classification scene is combined, strong correlation features are mined through an Apriori-Temporal algorithm, a quantile regression LSTM prediction model is constructed, the model outputs a current deviation range, an AI scene matching degree is judged, if the AI scene matching degree is not matched, distribution quantity is corrected, a hardware threshold value is verified, and then power distribution is executed, so that scene-based accurate power distribution is realized, and power distribution safety and efficiency are guaranteed.

Description

Technical Field

[0001] This invention belongs to the field of cabinet power distribution technology, specifically a pre-installed quick-connect power interface system and power distribution method applicable to cabinets. Background Technology

[0002] The development of the digital economy is driving the evolution of data centers towards high density. As the core power distribution unit, the stability of the server rack directly determines the operational safety of loads such as servers and AI computing equipment, and has become the mainstream power distribution solution for server racks.

[0003] Artificial intelligence (AI) technology supports the intelligent upgrade of quick-connect interface systems. Traditional power distribution relies on manual pre-setting of capacity, which easily leads to energy efficiency and safety conflicts under dynamic loads, such as idle capacity during off-peak hours and overload during peak hours. AI, on the other hand, can predict power demand and dynamically allocate capacity by learning historical load, business characteristics, and equipment power consumption models. Theoretically, this can significantly improve resource utilization and has already been initially applied in high-density AI cabinets and cloud data centers.

[0004] However, in practical applications, the bottleneck of mismatch between AI prediction of power shortage and actual current demand becomes prominent: on the load side, there are instantaneous current surges in AI training and large model inference, and new services or hardware upgrades cause power consumption characteristics to deviate from historical data, resulting in inaccurate model predictions. In summary, the AI-driven integration of rack-mount quick-connect interface systems is an inevitable trend in high-density load management; however, the overload problem caused by the mismatch between predicted and actual needs urgently needs to be addressed. To this end, this invention provides a pre-installed quick-connect power interface system and power distribution method suitable for racks. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0006] The technical solution adopted by this invention to solve its technical problem is: In one aspect, the present invention provides a pre-installed quick-connect power interface power distribution method suitable for server racks, comprising: Step S10: Based on the historical power distribution database, obtain the load power distribution amount and the actual power load demand, and calculate the power distribution deviation value and the power distribution deviation rate; Step S20: Perform cluster analysis on the power distribution deviation values ​​and output the dynamic range of the power distribution deviation values ​​over time; Step S30: Based on the dynamic range of the power distribution deviation value over time and the power distribution deviation rate, classify the load scenarios, and obtain a strong correlation feature combination through the time-series association rule algorithm based on the classification results. Based on the strong correlation feature combination, construct a power distribution deviation prediction range model. Step S40: Based on the power distribution deviation prediction range model and the current time interval, output the current power distribution deviation value range. Based on the current power distribution deviation value range, determine whether the artificial intelligence (AI) matches the current scene features. If they do not match, then perform power distribution according to the power distribution deviation prediction range model.

[0007] As a further improvement of the present invention, the specific process for calculating the power distribution deviation value and the power distribution deviation rate is as follows: The load power allocation corresponding to the target load and the actual power load demand are retrieved. The power allocation deviation value corresponding to the load is obtained by subtracting the actual power load demand from the load power allocation. The power distribution deviation rate is obtained by dividing the power distribution deviation value by the load power distribution amount.

[0008] As a further improvement of the present invention, the specific process of the dynamic range of the output power distribution deviation value over time is as follows: The power allocation deviation value and the corresponding timestamp are aligned in time sequence. The power allocation deviation value and the corresponding timestamp are used as the core feature dimension. Density clustering algorithm is used for cluster analysis to divide the power allocation deviation value into multiple feature clusters and mark the continuous time interval corresponding to each feature cluster. Each feature cluster corresponds to a type of power distribution deviation value with similar fluctuation patterns, that is, the dynamic distribution range of power distribution deviation values ​​within the corresponding time interval.

[0009] As a further improvement of the present invention, the specific process of classifying load scenarios is as follows: From the feature clusters corresponding to the dynamic range of the distribution of power allocation deviation over time, extract the time intervals associated with the feature clusters, obtain the corresponding power allocation deviation rate dataset within the time interval, and calculate the mean of the power allocation deviation rate dataset. If the mean of the power distribution deviation rate dataset is greater than or equal to the first preset percentage threshold within a time interval, it is classified as an AI high-load training scenario. If, within a time interval, the mean of the power distribution deviation rate dataset is less than the first preset percentage threshold but greater than or equal to the second preset percentage threshold, it is classified as a business peak scenario. If the mean of the power distribution deviation rate dataset is less than the second preset percentage threshold within a time interval, it is classified as a business off-peak scenario.

[0010] As a further improvement of the present invention, the specific process of obtaining the strongly correlated feature combination is as follows: The mean of the time interval, classification results, and power distribution deviation rate datasets is integrated as the input dataset for the time series association rule algorithm. The Apriori-Temporal algorithm is selected, and the core parameters are set as follows: Time window constraint: The minimum time unit is 30 minutes, matching the actual operating cycle of the load; Support threshold: set to 15%; Confidence threshold: set to 80%; A preliminary rule set is generated, and weakly associated rules with support <15% or confidence <80% are filtered out to finally obtain a precise set of strongly associated feature combinations.

[0011] As a further improvement of the present invention, the specific process of constructing the power distribution deviation prediction range model is as follows: The time intervals, classification results, and mean values ​​of the power allocation deviation rate dataset in the strongly correlated feature combination are standardized and encoded, and integrated into the model input feature set; the training set, validation set, and test set are divided in a 7:2:1 ratio. The quantile regression LSTM model is selected. The structure of the quantile regression LSTM model includes: Input layer: Receives encoded strongly correlated feature vectors; LSTM layer: captures the temporal correlation patterns of features; Quantile Output Layer: Outputs the biquantile of the power distribution deviation value. With the goal of minimizing the quantile loss function, it is trained using the training set and the parameters are iteratively adjusted using the validation set to finally construct the power distribution deviation prediction range model.

[0012] As a further improvement of the present invention, the specific process of outputting the current power distribution deviation value range is as follows: Get the current time and the current classification result; Based on the time interval corresponding to the current time matching feature cluster, and combined with the current classification result, the power allocation deviation prediction range model is input. Through the quantile output layer of the power allocation deviation prediction range model, the confidence interval of the current power allocation deviation is obtained, that is, the current power allocation deviation value range.

[0013] As a further improvement of the present invention, the specific process of determining whether the artificial intelligence (AI) matches the features of the current scene is as follows: Collect the actual power distribution deviation value within the current time interval, and compare the actual power distribution deviation value within the current time interval with the current power distribution deviation value range output by the power distribution deviation prediction range model; Calculate the coverage of the actual power distribution deviation value within the current time interval with the range of the current power distribution deviation value output by the power distribution deviation prediction range model; If the coverage is less than the coverage threshold, it is considered a match; If the error rate is greater than the coverage threshold, a mismatch is determined.

[0014] As a further improvement of the present invention, the specific process of power distribution is as follows: Based on the determination of mismatch, the confidence interval range of the power distribution deviation value range in this scenario is obtained. The load power distribution output of the AI ​​is then corrected based on this power distribution deviation value range. Using the load power distribution output by the AI ​​as a base, and combining the confidence interval range of the power distribution deviation value range output by the power distribution deviation prediction range model, the corrected load power distribution value is calculated. Corrected upper limit of load power allocation = Output of load power allocation by AI × (1 + Upper limit of the confidence interval of the power allocation deviation range output by the power allocation deviation prediction range model) The revised lower limit of load power allocation is calculated as follows: Load power allocation output by AI × (1 + lower limit of the confidence interval of the power allocation deviation range output by the power allocation deviation prediction range model). This expands the coverage range of the allocation and adapts to the actual demand fluctuations in the current scenario.

[0015] If the upper limit of the load power distribution exceeds the hardware threshold after correction, the upper limit will be adjusted to the maximum value corresponding to the hardware threshold. If the lower limit of the adjusted load power allocation is lower than the basic demand threshold of the load, then the lower limit will be adjusted to the basic demand threshold.

[0016] On the other hand, the present invention provides a pre-installed quick-connect power interface system suitable for server racks, comprising: Data acquisition and deviation calculation module: Based on the historical power distribution database, it acquires the load power distribution amount and the actual power load demand, and calculates the power distribution deviation value and the power distribution deviation rate; Deviation value clustering analysis module: performs clustering analysis on power distribution deviation values ​​and outputs the dynamic range of power distribution deviation value distribution over time; Load scenario classification and prediction model construction module: Based on the dynamic range of power distribution deviation value distribution over time and the power distribution deviation rate, load scenarios are classified, and based on the classification results, a strong correlation feature combination is obtained through the time series association rule algorithm. Based on the strong correlation feature combination, a power distribution deviation prediction range model is constructed. Deviation Range Prediction and Power Distribution Adaptation Module: Based on the power distribution deviation prediction range model and the current time interval, output the current power distribution deviation value range. Based on the current power distribution deviation value range, determine whether the artificial intelligence (AI) matches the current scene characteristics. If they do not match, then perform power distribution according to the power distribution deviation prediction range model.

[0017] The beneficial effects of this invention are as follows: 1. By calculating both the power distribution deviation value and the deviation rate, and combining DBSCAN clustering and scenario classification rules, it achieves accurate identification of different load scenarios such as high-load AI training and business peak and valley, solving the problem of insufficient feature matching of the original artificial intelligence scenario. Based on the deviation range output by the quantile regression LSTM prediction model, it corrects the power distribution amount and performs hardware load capacity threshold verification, effectively avoiding the risk of overload and underload, and ensuring the power distribution safety of the pre-installed quick-connect power interface in the cabinet. 2. By mining strong correlation features through the Apriori-Temporal algorithm, the power allocation is adapted to the fluctuations in actual load demand, reducing power redundancy or insufficiency and improving the utilization efficiency of cabinet power distribution resources. It is specifically designed for cabinet pre-installed quick-connect power interface systems, which can achieve intelligent power distribution without major modifications to existing cabinet hardware. It has good adaptability and implementation. By combining real-time data to iteratively adapt the prediction model and artificial intelligence, it can maintain the matching accuracy of load scenarios over a long period of time and ensure the stability of system power distribution performance. Attached Figure Description

[0018] The invention will now be further described with reference to the accompanying drawings.

[0019] Figure 1 This is a flowchart illustrating the steps of the pre-installed quick-connect power interface system and power distribution method applicable to server racks according to the present invention; Figure 2 This is a system module diagram of the edge device computing power resource scheduling system based on artificial intelligence, as described in this invention. Detailed Implementation

[0020] To facilitate understanding of the technical means, features, purpose, and effects of this invention, it should be clarified that: in this invention, both the load power allocation and the actual power load demand refer to current values, measured in amperes (A); the artificial intelligence (AI) specifically refers to the AI ​​adapted to the pre-installed quick-connect power interface system of this invention, whose function is to output the power allocation corresponding to the target load. The invention will be further described in detail below with reference to specific embodiments.

[0021] Example 1 like Figure 1 As shown in the embodiment of the present invention, the pre-installed quick-connect power interface power distribution method for server racks includes: Step S10: Based on the historical power distribution database, obtain the load power distribution amount and the actual power load demand, and calculate the power distribution deviation value and the power distribution deviation rate; In step S10, the specific process for calculating the power distribution deviation value and the power distribution deviation rate is as follows: The load power allocation and actual power load demand corresponding to the target load are retrieved from the historical power allocation database. The power allocation deviation value (absolute value) corresponding to the load is obtained by subtracting the actual power load demand from the load power allocation. Further calculate the power distribution deviation rate: Divide the power distribution deviation value by the load power distribution amount to obtain the ratio of the deviation value to the load power distribution amount (the result is expressed as a percentage). The power distribution deviation rate will serve as the core basis for subsequent load scenario classification.

[0022] Preferably, the process of obtaining the load power allocation is as follows: the artificial intelligence (AI) adapted to the rack pre-installed quick-connect power interface system of the present invention outputs the power allocation corresponding to the target load, specifically by retrieving the data records of the AI ​​output layer; The process of obtaining the actual power load demand is as follows: collect historical monitoring data from the high-precision current / power sensor integrated at the pre-installed quick-connect power interface of the cabinet, combine it with the real-time power consumption data reported by the load device BMC (Baseboard Management Controller), and obtain the actual power load demand of the load in the corresponding time period through data time sequence alignment, redundant data deduplication and 3σ criterion outlier filtering (removing jump data caused by instantaneous electromagnetic interference).

[0023] Example calculation of power distribution deviation: Taking a high-density AI server load as an example, the load power distribution amount output by its artificial intelligence AI is 15A (corresponding to a power of 3300W under 220V voltage). The actual power load demand obtained by integrating the data from the quick-connect interface sensor and the server BMC is 18A (corresponding to a power of 3960W). Then the power distribution deviation is 15A-18A=-3A. This negative deviation result indicates that the actual load demand exceeds the pre-allocated amount, and there is an overload risk at the corresponding load end.

[0024] It should be noted that the historical power distribution database is for the power distribution scenario of pre-installed quick-connect power interfaces in the cabinet, and stores the load power distribution and actual power load demand data for the entire operation cycle. Step S20: Perform cluster analysis on the power distribution deviation values ​​and output the dynamic range of the power distribution deviation values ​​over time; In step S20, the specific process of determining the dynamic range of the output power distribution deviation value over time is as follows: First, align the power distribution deviation value with the corresponding timestamp to ensure that the time dimension of the power distribution deviation value matches the actual fluctuation cycle of the load. Subsequently, the power distribution deviation data was screened using the 3σ criterion to eliminate extreme deviation values ​​caused by non-business load factors such as instantaneous electromagnetic interference and hardware failures, and to retain effective power distribution deviation values ​​that reflect the fluctuations in actual load demand.

[0025] Using the power distribution deviation value and the corresponding timestamp as the core feature dimension, a density clustering algorithm (such as DBSCAN) is selected to carry out clustering analysis; Density clustering algorithm can adapt to the non-uniform distribution characteristics of power distribution deviation values, avoiding the rigid division of deviation clusters by traditional clustering algorithms; the density clustering algorithm divides the power distribution deviation values ​​into multiple feature clusters and marks the continuous time interval corresponding to each feature cluster; Each feature cluster corresponds to a type of power distribution deviation value with similar fluctuation patterns, that is, the dynamic distribution range of power distribution deviation value within the corresponding time interval; Step S30: Based on the dynamic range of the power distribution deviation value over time and the power distribution deviation rate, classify the load scenarios, and obtain a strong correlation feature combination through the time-series association rule algorithm based on the classification results. Based on the strong correlation feature combination, construct a power distribution deviation prediction range model. In step S30, the specific process of classifying load scenarios is as follows: First, extract the time interval associated with the feature cluster from the feature cluster corresponding to the dynamic range of the power distribution deviation over time, and simultaneously obtain the power distribution deviation rate dataset (composed of the power distribution deviation rate calculated in step S10) within the time interval. Then, the mean of the power distribution deviation rate dataset is calculated, and combined with the actual operating characteristics of the cabinet load, the load scenarios are accurately classified: If the mean of the power distribution deviation rate dataset is greater than or equal to the first preset percentage threshold (e.g., 10%) within the target time interval, the corresponding load scenario will be classified as an AI high load training scenario. If, within the target time interval, the mean of the power distribution deviation rate dataset is less than the first preset percentage threshold (e.g., 10%) but greater than or equal to the second preset percentage threshold (e.g., 5%), then the corresponding load scenario will be classified as a business peak scenario. If the mean of the power distribution deviation rate dataset is less than the second preset percentage threshold (e.g., 5%) within the target time interval, the corresponding load scenario will be classified as a business off-peak scenario. In step S30, based on the classification results, the specific process of obtaining strongly correlated feature combinations through the temporal association rule algorithm is as follows: The mean of the power distribution deviation rate dataset (for AI high-load training scenarios, peak business scenarios, and off-peak business scenarios) is integrated into the input dataset for the time-series association rule algorithm. The Apriori-Temporal algorithm, which is adapted to the time-series characteristics of power distribution scenarios, is selected, and the core parameters are set as follows: Time window constraint: The minimum time unit is 30 minutes, matching the actual operating cycle of the load; Support threshold: set to 15% (to filter feature combinations that appear in at least 15% of the time interval); Confidence threshold: set to 80% (to ensure the reliability of the association between feature combinations); Based on the above parameters, the dataset is mined to generate a preliminary rule set (e.g., weekdays 8:00-8:30, AI high-load training scenario, mean deviation +25%).

[0026] Weak association rules with support <15% or confidence <80% are filtered out, while redundant combinations with no business value (such as minor deviation combinations in nighttime standby scenarios) are removed, ultimately yielding a precise set of strong association feature combinations. In step S30, the specific process of constructing the power distribution deviation prediction range model based on the combination of strongly correlated features is as follows: The time intervals (discretized into time period identifiers), classification results (converted into embedding vectors), and the mean of the power distribution deviation rate dataset in the strongly correlated feature combination are standardized and encoded, and integrated into the model input feature set; the training set, validation set, and test set are divided in a 7:2:1 ratio to ensure that each set contains feature data of all types of load scenarios. The quantile regression LSTM model is selected (to meet the interval prediction needs of time series data). The structure of the quantile regression LSTM model includes: Input layer: Receives encoded strongly correlated feature vectors; LSTM layer: Captures the temporal correlation patterns of features (sets 2 hidden layers, with 64 nodes); Quantile output layer: Outputs the double quantiles of the power distribution deviation value (such as the upper and lower limits of the 95% confidence interval), corresponding to the boundaries of the prediction range; The model was trained using the training set with the goal of minimizing the quantile loss function; parameters were iteratively adjusted using the validation set (e.g., setting the time window length to 60 minutes and the quantile thresholds to 0.025 and 0.975) to ensure that the prediction range covered ≥95% of the actual deviation. If the percentage of actual deviations falling within the predicted range is less than 95%, then we backtrack the strongly correlated feature combinations, supplement the historical data of the corresponding load scenarios, and retrain; finally, we construct the power distribution deviation prediction range model. Step S40: Based on the power distribution deviation prediction range model and the current time interval, output the current power distribution deviation value range. Based on the current power distribution deviation value range, determine whether the artificial intelligence (AI) matches the current scene features. If they do not match, then perform power distribution according to the power distribution deviation prediction range model. In step S40, the specific process of outputting the current power distribution deviation value range based on the power distribution deviation prediction range model and the current time interval is as follows: Get the current time (accurate to the minute) and the current classification result; Based on the time interval corresponding to the current time matching feature cluster, and combined with the current classification result, the power distribution deviation prediction range model is input. Through the quantile output layer of the power distribution deviation prediction range model, the confidence interval of the current power distribution deviation (such as the upper and lower limits under 95% confidence) is obtained, that is, the current power distribution deviation value range. Based on the current power distribution deviation range and considering the hardware capacity of the current quick-connect interface (such as the deviation safety threshold corresponding to the interface's rated current), the current power distribution deviation range output by the power distribution deviation prediction model is verified: If the upper limit of the range exceeds the hardware safety threshold, the upper limit will be corrected to the deviation value corresponding to the safety threshold to ensure that the output range meets the safety constraints of the power distribution system. In step S40, the specific process of determining whether the current scenario features match based on the current power distribution deviation range is as follows: Collect the actual power distribution deviation value within the current time interval (the difference between the load power distribution output by artificial intelligence prediction and the actual power load demand), and compare the actual power distribution deviation value within the current time interval with the current power distribution deviation value range output by the power distribution deviation prediction range model; Calculate the coverage of the actual power distribution deviation value within the current time interval with the range of the current power distribution deviation value output by the power distribution deviation prediction range model; If the coverage is less than the coverage threshold (e.g., 90%), it is determined that the AI ​​is matching the features of the current scene. If the error rate is greater than the coverage threshold (e.g., 90%), it is determined that the artificial intelligence (AI) does not match the features of the current scene. For example, taking the weekday 8:00-8:30 time interval (30 minutes, collecting 1 actual deviation value per minute) as an example, there are 30 actual power distribution deviation values ​​in this interval, specifically distributed as follows: 6 values ​​are +18%, 12 values ​​are +22%, 8 values ​​are +25%, and 4 values ​​are +32%. The power distribution deviation prediction range model outputs a deviation value range of +15% to +28% for this interval. The number of actual deviation values ​​falling within the range is 6 + 12 + 8 = 26. The calculated coverage is 26 ÷ 30 ≈ 86.7% (below the 90% threshold, indicating insufficient matching of current scene features by the original artificial intelligence). In step S40, the specific process of power distribution is as follows: Based on the determination of mismatch, the specific process of power allocation according to the power allocation deviation prediction range model is as follows: Obtain the confidence interval range (e.g., +15% to +28%) of the power distribution deviation value in this scenario.

[0027] The load power distribution output is corrected based on the power distribution deviation range. Using the load power distribution output from the AI ​​as a base, and combining the confidence interval of the power distribution deviation range output by the power distribution deviation prediction range model, the corrected load power distribution is calculated. Corrected upper limit of load power allocation = Output of load power allocation by AI × (1 + Upper limit of the confidence interval of the power allocation deviation range output by the power allocation deviation prediction range model) The revised lower limit of load power allocation is calculated as follows: Load power allocation output by AI × (1 + lower limit of the confidence interval of the power allocation deviation range output by the power allocation deviation prediction range model). This expands the coverage range of the allocation and adapts to the actual demand fluctuations in the current scenario.

[0028] If the upper limit of the load power distribution exceeds the hardware threshold after correction, the upper limit will be adjusted to the maximum value corresponding to the hardware threshold. If the lower limit of the revised load power allocation is lower than the basic demand threshold of the load, the lower limit will be adjusted to the basic demand threshold to ensure that the allocation meets safety and functional constraints.

[0029] Example 2 like Figure 2 As shown, based on Embodiment 1, the present invention provides a pre-installed quick-connect power interface system suitable for server racks, comprising: Data acquisition and deviation calculation module: Based on the historical power distribution database, it acquires the load power distribution amount and the actual power load demand, and calculates the power distribution deviation value and the power distribution deviation rate; Deviation value clustering analysis module: performs clustering analysis on power distribution deviation values ​​and outputs the dynamic range of power distribution deviation value distribution over time; Load scenario classification and prediction model construction module: Based on the dynamic range of power distribution deviation value distribution over time and the power distribution deviation rate, load scenarios are classified, and based on the classification results, a strong correlation feature combination is obtained through the time series association rule algorithm. Based on the strong correlation feature combination, a power distribution deviation prediction range model is constructed. Deviation Range Prediction and Power Distribution Adaptation Module: Based on the power distribution deviation prediction range model and the current time interval, output the current power distribution deviation value range. Based on the current power distribution deviation value range, determine whether the artificial intelligence (AI) matches the current scene characteristics. If they do not match, then perform power distribution according to the power distribution deviation prediction range model.

[0030] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A power distribution method for pre-installed quick-connect power interfaces suitable for server racks, characterized in that: include: Step S10: Based on the historical power distribution database, obtain the load power distribution amount and the actual power load demand, and calculate the power distribution deviation value and the power distribution deviation rate; Step S20: Perform cluster analysis on the power distribution deviation values ​​and output the dynamic range of the power distribution deviation values ​​over time; Step S30: Based on the dynamic range of the power distribution deviation value over time and the power distribution deviation rate, classify the load scenarios, and obtain a strong correlation feature combination through the time-series association rule algorithm based on the classification results. Based on the strong correlation feature combination, construct a power distribution deviation prediction range model. Step S40: Based on the power distribution deviation prediction range model and the current time interval, output the current power distribution deviation value range. Based on the current power distribution deviation value range, determine whether the artificial intelligence (AI) matches the current scene features. If they do not match, then perform power distribution according to the power distribution deviation prediction range model.

2. The pre-installed quick-connect power interface power distribution method for server racks according to claim 1, characterized in that: The specific process for calculating the power distribution deviation value and the power distribution deviation rate is as follows: The load power allocation corresponding to the target load and the actual power load demand are retrieved. The power allocation deviation value corresponding to the load is obtained by subtracting the actual power load demand from the load power allocation. The power distribution deviation rate is obtained by dividing the power distribution deviation value by the load power distribution amount.

3. The pre-installed quick-connect power interface power distribution method for server racks according to claim 1, characterized in that: The specific process for determining the dynamic range of the output power distribution deviation value over time is as follows: The power allocation deviation value and the corresponding timestamp are aligned in time sequence. The power allocation deviation value and the corresponding timestamp are used as the core feature dimension. Density clustering algorithm is used for cluster analysis to divide the power allocation deviation value into multiple feature clusters and mark the continuous time interval corresponding to each feature cluster. Each feature cluster corresponds to a type of power distribution deviation value with similar fluctuation patterns, that is, the dynamic distribution range of power distribution deviation values ​​within the corresponding time interval.

4. The pre-installed quick-connect power interface power distribution method for server racks according to claim 3, characterized in that: The specific process for classifying load scenarios is as follows: From the feature clusters corresponding to the dynamic range of the distribution of power allocation deviation over time, extract the time intervals associated with the feature clusters, obtain the corresponding power allocation deviation rate dataset within the time interval, and calculate the mean of the power allocation deviation rate dataset. If the mean of the power distribution deviation rate dataset is greater than or equal to the first preset percentage threshold within a time interval, it is classified as an AI high-load training scenario. If, within a time interval, the mean of the power distribution deviation rate dataset is less than the first preset percentage threshold but greater than or equal to the second preset percentage threshold, it is classified as a business peak scenario. If the mean of the power distribution deviation rate dataset is less than the second preset percentage threshold within a time interval, it is classified as a business off-peak scenario.

5. The pre-installed quick-connect power interface power distribution method for server racks according to claim 1, characterized in that: The specific process for obtaining the strongly correlated feature combination is as follows: The mean of the time interval, classification results, and power distribution deviation rate datasets is integrated as the input dataset for the time series association rule algorithm. The Apriori-Temporal algorithm is selected, and the core parameters are set as follows: Time window constraint: The minimum time unit is 30 minutes, matching the actual operating cycle of the load; Support threshold: set to 15%; Confidence threshold: set to 80%; A preliminary rule set is generated, and weakly associated rules with support <15% or confidence <80% are filtered out to finally obtain a precise set of strongly associated feature combinations.

6. The pre-installed quick-connect power interface power distribution method for server racks according to claim 1, characterized in that: The specific process for constructing the power distribution deviation prediction range model is as follows: The time intervals, classification results, and mean values ​​of the power allocation deviation rate dataset in the strongly correlated feature combination are standardized and encoded, and integrated into the model input feature set; the training set, validation set, and test set are divided in a 7:2:1 ratio. The quantile regression LSTM model is selected. The structure of the quantile regression LSTM model includes: Input layer: Receives encoded strongly correlated feature vectors; LSTM layer: captures the temporal correlation patterns of features; Quantile Output Layer: Outputs the biquantile of the power distribution deviation value. With the goal of minimizing the quantile loss function, it is trained using the training set and the parameters are iteratively adjusted using the validation set to finally construct the power distribution deviation prediction range model.

7. The pre-installed quick-connect power interface power distribution method for server racks according to claim 1, characterized in that: The specific process for outputting the current power distribution deviation range is as follows: Get the current time and the current classification result; Based on the time interval corresponding to the current time matching feature cluster, and combined with the current classification result, the power allocation deviation prediction range model is input. Through the quantile output layer of the power allocation deviation prediction range model, the confidence interval of the current power allocation deviation is obtained, that is, the current power allocation deviation value range.

8. The pre-installed quick-connect power interface power distribution method for server racks according to claim 1, characterized in that: The specific process for determining whether the artificial intelligence (AI) matches the features of the current scene is as follows: Collect the actual power distribution deviation value within the current time interval, and compare the actual power distribution deviation value within the current time interval with the current power distribution deviation value range output by the power distribution deviation prediction range model; Calculate the coverage of the actual power distribution deviation value within the current time interval with the range of the current power distribution deviation value output by the power distribution deviation prediction range model; If the coverage is less than the coverage threshold, it is considered a match; If the error rate is greater than the coverage threshold, a mismatch is determined.

9. The pre-installed quick-connect power interface power distribution method for server racks according to claim 1, characterized in that: The specific process for power distribution is as follows: Based on the determination of mismatch, the confidence interval range of the power distribution deviation value range in this scenario is obtained. The load power distribution output of the AI ​​is then corrected based on this power distribution deviation value range. Using the load power distribution output by the AI ​​as a base, and combining the confidence interval range of the power distribution deviation value range output by the power distribution deviation prediction range model, the corrected load power distribution value is calculated. Corrected upper limit of load power allocation = Output of load power allocation by AI × (1 + Upper limit of the confidence interval of the power allocation deviation range output by the power allocation deviation prediction range model) The revised lower limit of load power allocation is calculated as follows: Load power allocation output by AI × (1 + lower limit of the confidence interval of the power allocation deviation range output by the power allocation deviation prediction range model). This expands the coverage range of the allocation and adapts to the actual demand fluctuations in the current scenario. If the upper limit of the load power distribution exceeds the hardware threshold after correction, the upper limit will be adjusted to the maximum value corresponding to the hardware threshold. If the lower limit of the adjusted load power allocation is lower than the basic demand threshold of the load, then the lower limit will be adjusted to the basic demand threshold.

10. A pre-installed quick-connect power interface system suitable for server racks, used to execute the power distribution method according to any one of claims 1-9, characterized in that: include: Data acquisition and deviation calculation module: Based on the historical power distribution database, it acquires the load power distribution amount and the actual power load demand, and calculates the power distribution deviation value and the power distribution deviation rate; Deviation value clustering analysis module: performs clustering analysis on power distribution deviation values ​​and outputs the dynamic range of power distribution deviation value distribution over time; Load scenario classification and prediction model construction module: Based on the dynamic range of power distribution deviation value distribution over time and the power distribution deviation rate, load scenarios are classified, and based on the classification results, a strong correlation feature combination is obtained through the time series association rule algorithm. Based on the strong correlation feature combination, a power distribution deviation prediction range model is constructed. Deviation Range Prediction and Power Distribution Adaptation Module: Based on the power distribution deviation prediction range model and the current time interval, output the current power distribution deviation value range. Based on the current power distribution deviation value range, determine whether the artificial intelligence (AI) matches the current scene characteristics. If they do not match, then perform power distribution according to the power distribution deviation prediction range model.

Images