A machine room AI value keeper decision method, system, device and medium

Abstract

The application belongs to the field of data processing, and specifically discloses a machine room AI value keeper decision-making method, system, device and medium. In the aspect of data collection, in order to obtain accurate data, RC low-pass filtering and TVS diode are used for electrostatic protection, in the aspect of fault detection, a Bayesian-generative adversarial network is used to dynamically generate a small sample fault model, a dynamic threshold is used, the traditional static threshold rule is solved, historical fault data is compared through an LSTM model to obtain a confidence degree, double verification is used to reduce fault false alarm problems, in the aspect of fault reporting, AI multi-modal interaction and decision-making are used, alarm->disposal->verification closed-loop logic is formed, an AI alarm instruction is converted into a DTMF dialing signal through an FPGA controller, an improved RJ11 interface is used to realize external dialing of maintenance personnel, AI multi-modal interaction and decision-making are realized, and voiceprint detection and current analysis are used for the handled results to realize verification of problem solving.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to a decision-making method, system, equipment, and medium for AI duty officers in computer rooms. Background Technology

[0002] Due to the development of digital technology in recent years, AI-driven informatization and digitalization have gradually become strategic goals. The foundation and network supporting AI information technology consists of various data centers and server rooms. In recent years, AI-assisted detection systems have been gradually introduced into the field of intelligent operation and maintenance technology for server rooms. However, in unattended server room scenarios of Class C and below, real-time fault response and false alarm rate control have become core pain points.

[0003] The existing alarm method, a data center detection alarm and alarm cancellation method, includes the following steps: (1) Encoding the data center equipment and detecting it in real time. When an abnormality is detected in the data center equipment, alarm information of different levels is generated; (2) Encapsulating the alarm information into a class object through encoding; (3) Converting the alarm information encapsulated in step (2) into template alarm information corresponding to the template message fixed format according to the template message fixed format of WeChat public account; (4) Sending the template alarm information to the web service server. The web service server sends the received template alarm information to the WeChat cloud server. The WeChat server pushes the received template alarm information to the following users according to the alarm information level; (5) After receiving the push message notification, the customer can send an instruction to the web service server to cancel the alarm.

[0004] The prior art disclosed in CN117975646A is a method for using AI to manage duty personnel, comprising: S1: using a general pre-trained framework GLM, based on a novel autoregressive blank-filling objective, GLM formalizes the NLU task into a blank-filling problem containing a task description, and the blank-filling problem containing the task description generates an answer through autoregression; S2: the pre-trained framework GLM is trained by optimizing the autoregressive blank-filling objective; S3: after training the pre-trained framework GLM, multi-task pre-training is introduced; S4: the pre-trained framework GLM is fine-tuned to obtain the target instruction;

[0005] In real-world data centers, electromagnetic interference is strong, making collected fault information susceptible to interference. Furthermore, the current method of using static thresholds to determine fault status, without establishing a verification model based on historical data and adjusting the thresholds in real-time, suffers from a high false alarm rate. Fault reporting typically relies on proactive push notifications or SMS alerts, resulting in slow alarm response times. Once the notification is sent, the fault event is considered over, with no further attention paid to its resolution, indicating a lack of a closed-loop strategy for problem-solving and no measures to mitigate errors. Moreover, while existing technologies feature multi-turn AI interaction, they lack the ability to independently address problems, merely providing simple communication and alerts without self-learning capabilities. This leads to the use of fixed models for problem identification, resulting in a high false alarm rate and the absence of a closed-loop strategy encompassing problem discovery, reporting, resolution, and verification. Summary of the Invention

[0006] To address the aforementioned issues, this invention provides a decision-making method, system, equipment, and medium for AI-powered data center operators. In terms of data collection, to obtain accurate data, RC low-pass filtering and TVS diode electrostatic protection are used to suppress electromagnetic interference in the data center. For fault detection, a Bayesian generative adversarial network (B-GAN) is employed to dynamically generate small-sample fault models. Dynamic thresholding is used to overcome the limitations of traditional static threshold rules. Confidence levels are obtained by comparing historical fault data using an LSTM model. Double verification reduces false alarms. Fault reporting utilizes AI multimodal interaction and decision-making, forming a closed-loop logic of alarm → handling → verification. An FPGA controller converts AI alarm commands into DTMF dialing signals, which are then used to connect to maintenance personnel via a modified RJ11 interface, enabling AI multimodal interaction and decision-making. Voiceprint detection and current analysis are used to verify the resolved issues. Finally, fault voiceprints (12kHz pulses) are added to an edge sample library to enhance the safety identification capabilities of maintenance personnel. A Bayesian generative adversarial network (B-GAN) generator synthesizes similar spectra to update the discriminator weights.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A first aspect of the present invention provides a decision-making method for an AI-powered computer room operator, comprising the following steps:

[0009] Step S1, the step of multimodal perception and data acquisition, involves real-time collection of physical status data of equipment in the computer room, and edge computing of the collected pseudo-anomaly data;

[0010] Step S2, the edge computing step, establishes a small sample fault data identification model through the Bayesian Generative Adversarial Network (B-GAN) model, compares the historical fault patterns through the LSTM model, and determines whether the collected pseudo-abnormal data is abnormal based on the dynamic detection threshold. If the pseudo-abnormal data is determined to be abnormal, proceed to step S3; otherwise, the pseudo-abnormal data is considered normal data and no further processing is performed.

[0011] Step S3: Generate corresponding text and voice messages based on the abnormal data. The FPGA controller will generate AI alarm commands and convert them into DTMF dialing signals. By calling the RJ11 interface, the landline will be triggered to make an outbound call to contact maintenance personnel. At the same time, a fault work order will be created and the parameter information related to the fault will be entered.

[0012] Step S4: Based on the voice information of the maintenance personnel's response, the end-to-end ASR model is used to convert the voice information of the maintenance personnel into text, and keywords are identified. The validity of the voice is verified by the signal false trigger rate algorithm, fuzzy command processing is supported, and voiceprint detection and current analysis are used to confirm whether the problem has been resolved.

[0013] Step S5: Repeat steps S3 and S4. After multiple rounds of multimodal interactive decision-making, determine the validity based on instruction parsing and physical state verification, mark the fault work order status, and record the handling time. Generate an immutable evidence chain through the multimodal decision log of voiceprint + current + operation record, add the fault voiceprint to the edge sample library to enhance the security identification of maintenance personnel, and synthesize similar spectra through the Bayesian Generative Adversarial Network (B-GAN) generator to update the discriminator weights.

[0014] A second aspect of the present invention provides a computer room AI duty officer decision-making system, the system comprising: a multimodal perception module, an edge computing module, an active calling module, and an interactive decision-making module;

[0015] The multimodal sensing module includes a current sensor submodule, a voiceprint microphone submodule, a temperature probe submodule, and an anti-interference submodule. The anti-interference submodule employs an RC low-pass filter circuit and TVS diode electrostatic protection. The RC low-pass filter circuit includes a parallel 100pF capacitor and a 10kΩ resistor to suppress electromagnetic interference in the computer room. The anti-interference submodule ensures the accuracy of the physical status data of the computer room collected by the current sensor submodule, the voiceprint microphone submodule, and the temperature probe submodule.

[0016] The edge computing module includes: a Bayesian generative adversarial network autonomous learning submodule and an LSTM anomaly detection submodule; it generates small sample fault data through the Bayesian generative adversarial network, and dynamically adjusts the detection threshold through a discriminator, so that only 5 similar faults are needed to generate an identification model; the collected data is compared with historical data through the LSTM model to obtain the confidence level, and the fault information is judged based on the confidence level;

[0017] The active call module includes a speech synthesis submodule and an active dialing submodule. By integrating a text-to-speech submodule, text is converted into TTS speech for human-computer interaction. The active dialing submodule includes an FPGA dialing controller and a modified RJ11 interface. The FPGA dialing controller obtains the pre-set maintenance personnel's phone number, generates a DTMF signal to trigger the landline outbound dialing mechanism, and calls the modified RJ11 interface to implement the active dialing function. The modified RJ11 interface uses a shielded twisted-pair current sensor to reduce signal attenuation and calculates the signal false trigger rate. To achieve anti-interference design and reduce false trigger rate The calculation formula is:

[0018]

[0019] in For noise current, The actual signal current, when the false trigger rate A value less than 0.1% is not a false trigger;

[0020] The interactive decision-making module includes a speech parsing submodule, an AI multimodal interactive decision-making submodule, and an autonomous learning optimization submodule. It converts the speech of maintenance personnel into text using an end-to-end ASR model, identifies keywords, and then uses AI to perform multimodal interaction and decision-making based on these keywords. The decision-making process incorporates voiceprint detection and current physical state verification to determine whether an anomaly has been resolved. Voiceprint detection checks for the presence of a 12kHz peak voiceprint in the data center, while current physical state verification calculates the current during the recovery phase after handling the anomaly. Compared with the reference stage current correlation coefficient To determine whether the treatment is effective. The calculation formula is as follows:

[0021] ,

[0022] A value >0.98 indicates that the handling was effective; and through multimodal interaction and decision-making, an immutable evidence chain of voiceprint + current + operation record multimodal decision log is generated; through the self-learning optimization submodule, the voiceprint of this fault is added to the edge sample library to enhance the security identification of maintenance personnel; the B-GAN generator synthesizes similar spectra and updates the discriminator weights.

[0023] A third aspect of the present invention provides an electronic device including a memory 102, a processor 101, a display module 103, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps described in any of the above-described AI duty officer decision-making methods for computer rooms.

[0024] A fourth aspect of the present invention provides a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps described in any of the above-described AI duty officer decision-making methods for computer rooms.

[0025] The beneficial effects of this invention are as follows: By using multimodal perception anomaly data, a small-sample fault model is established through B-GAN, and the threshold is dynamically updated. A confidence level is obtained by comparing historical fault patterns using an LSTM model. The confidence level is used to determine whether the current fault is valid. An FPGA controller combined with a modified RJ11 enables proactive dialing to quickly contact maintenance personnel and resolve the problem as quickly as possible, rather than simply issuing a notification. AI multimodal interaction and decision-making further facilitate problem recording or resolution, generating a complete closed loop. Whether the problem is resolved is further determined based on voiceprint detection and current physical state verification. If the problem is resolved, an automatic learning function adds the current fault voiceprint (12kHz pulse) to the edge sample library, enhancing the safety identification capabilities of maintenance personnel. The B-GAN generator synthesizes similar spectra and updates the discriminator weights. Attached Figure Description

[0026] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a schematic flowchart of the method of the present invention;

[0028] Figure 2 This is a schematic diagram of the system of the present invention;

[0029] Figure 3 This is a schematic diagram of the device structure of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and systems according to various embodiments of this disclosure. It should be noted that each block in a flowchart or block diagram may represent a module, segment, or portion of code, which may include one or more executable instructions for implementing the logical functions specified in the various embodiments. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutively represented blocks may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, may be implemented using a dedicated hardware-based system that performs the specified functions or operations, or using a combination of dedicated hardware and computer instructions.

[0032] Example 1, as Figure 1 As shown, a decision-making method for an AI-powered data center operator is implemented using the following steps: Step S1: Through multimodal perception and data acquisition, real-time physical status data of data center equipment is collected, and the collected anomaly data is transmitted to the processing system for edge computing. The specific operation steps are as follows: Current data is obtained through a current sensor (sampling rate 1MHz), voiceprint data is obtained through a voiceprint microphone (20Hz–20kHz), and temperature data is obtained through a temperature probe. Electromagnetic interference in the data center is suppressed through an RC low-pass filter circuit and a TVS diode electrostatic protection to ensure the accuracy of the physical status data of the data center equipment. The RC low-pass filter circuit includes a 100pF capacitor and a 10kΩ resistor connected in parallel. The acquired anomaly data is then processed for edge computing.

[0033] Step S2, the edge computing step, establishes a small-sample fault data identification model using a Bayesian Generative Adversarial Network (B-GAN) model. It compares the collected suspected anomaly data with historical fault patterns using an LSTM model and determines whether the data is abnormal based on a dynamic detection threshold. If the suspected anomaly data is determined to be abnormal, proceed to step S3; otherwise, the suspected anomaly data is considered normal data and no further processing is performed. The specific operation steps are as follows:

[0034] Step S21: Generate a recognition model using the Bayesian Generative Adversarial Network (B-GAN) autonomous learning engine, with Gaussian noise input to the generator. and historical false alarm sample feature vector Synthesize small sample fault data; dynamically adjust the detection threshold using a discriminator. The threshold calculation formula is:

[0035]

[0036] in This represents the number of times the event was accidentally triggered. The total number of detections is 5; a recognition model can be generated once 5 similar faults are detected.

[0037] Step S22: Compare historical faults using an LSTM model to obtain fault information and confidence level. Determine whether the currently detected fault information is reliable if the confidence level is >95%.

[0038] Step S3: Generate corresponding text-to-speech (TTS) information based on the abnormal data. The FPGA controller generates an AI alarm command and converts it into a DTMF dialing signal. By calling the modified RJ11 interface, it triggers a landline outbound call to contact maintenance personnel. At the same time, a fault work order is created, and fault-related parameter information is entered. The specific operation steps are as follows:

[0039] Step S31: Based on the abnormal information obtained in step S2, the abnormal text information is converted into corresponding abnormal TTS speech through speech synthesis.

[0040] In step S32, the FPGA controller simultaneously reads the preset phone number, generates a DTMF dialing signal, and triggers an outbound call via the modified RJ11 interface to contact maintenance personnel. Simultaneously, a fault work order is created, and relevant fault parameters are entered. The RJ11 interface uses a shielded twisted-pair current sensor to reduce signal attenuation, and the signal false trigger rate is calculated. To achieve anti-interference design and reduce false trigger rate The calculation formula is:

[0041]

[0042] in For noise current, The actual signal current, when the false trigger rate If the percentage is less than 0.1%, the alarm is determined to be not a false alarm.

[0043] Step S33: After the call is connected, play the generated fault text (TTS) voice message.

[0044] Step S4: Based on the voice information provided by the maintenance personnel, an end-to-end ASR model is used to convert the personnel's voice into text, identify keywords, verify the validity of the voice using a signal false trigger rate algorithm, support fuzzy command processing, and confirm whether the problem has been resolved using voiceprint detection and current analysis algorithms. The specific operation steps are as follows:

[0045] Step S41: Based on the voice information replied by the maintenance personnel, the end-to-end ASR model is used to convert the voice of the maintenance personnel into text and identify keywords. The voice of the maintenance personnel is compared with the voiceprint feature adaptive filtering of the voiceprint data. Its dialect accent compatibility covers 6 major dialect areas, eliminating the influence of accent on recognition.

[0046] If step S42 identifies speech similar to that already processed, then proceed to step S43; otherwise, proceed to step S44.

[0047] Step S43: According to the preset processing flow, ask the staff whether it is 1. already processed 2. needs to be delayed 3. transferred to an expert, wait for the maintenance personnel to reply, execute S41, if the identification result is 1, then execute S45, if it is 2, then execute S44;

[0048] Step S44: Based on the information replied by the maintenance personnel, the AI ​​interacts and makes decisions with the maintenance personnel through multiple rounds of interaction based on the identified information.

[0049] Step S45: The system actively acquires the physical status data of the computer room, and uses a soundprint sensor to detect and confirm whether the 12kHz peak has disappeared (sound pressure < 40dB); it also performs physical status verification using current data and calculates the recovery current after handling. Compared with the reference stage current correlation coefficient , The calculation formula is as follows:

[0050]

[0051] in The mean value of the current data during the recovery phase. Current count during the reference phase The mean of the current data in the baseline phase, The standard deviation of the current data during the recovery phase. The standard deviation of the quasi-stage current data, if If the relevant formula is correct, then the treatment is deemed effective. If both voiceprint and voiceprint detection pass, proceed to step S5; otherwise, proceed to step S33.

[0052] Step S5: Repeat steps S3 and S4. After multiple rounds of multimodal interactive decision-making, determine the validity based on voice command parsing and physical state verification, mark the fault work order status, and record the handling time. Generate an immutable evidence chain through the multimodal decision log of voiceprint + current + operation record, add the fault voiceprint to the edge sample library to enhance the security identification of maintenance personnel, and synthesize similar spectra through the Bayesian Generative Adversarial Network (B-GAN) generator to update the discriminator weights.

[0053] In summary, this invention utilizes multimodal sensing anomaly data to establish a small-sample fault model using B-GAN and dynamically updates thresholds. It then compares historical fault patterns using an LSTM model to obtain confidence levels, determining the validity of the current fault based on these confidence levels. An FPGA controller, combined with a modified RJ11 interface, enables proactive dialing to quickly contact maintenance personnel and resolve issues as rapidly as possible, rather than simply issuing a notification. AI multimodal interaction and decision-making further facilitate problem recording or resolution, creating a complete closed loop. Problem resolution is further determined by voiceprint detection and current physical state verification. If the problem is resolved, an automatic learning function adds the current fault voiceprint (12kHz pulse) to an edge sample library, enhancing safety identification by maintenance personnel. A B-GAN generator synthesizes similar spectra to update the discriminator weights. This invention presents a workflow from fault detection → outbound dialing → interaction → verification → learning. Through RJ11 interface modification and DTMF signal generation logic, it meets feasibility requirements and forms a closed-loop process encompassing sensing, calling, interaction, and verification.

[0054] This invention is applicable to large-scale cloud computing centers (such as Alibaba Cloud and AWS): real-time detection of faults such as current fluctuations and hard drive noises in more than 100,000 servers, triggering alarms within 5 seconds, and calling maintenance personnel via landline within 1 minute.

[0055] Edge computing nodes (such as 5G base station equipment rooms): When the network signal is poor in remote areas, calls can be made directly through telephone lines, avoiding fault response timeouts caused by network latency (traditional solutions > 30 minutes); Extended applications: Substations: Integrating temperature and current data, identifying transformer overheating risks (>120°C), and automatically calling on-duty personnel for emergency repairs; Communication base stations: Voiceprint recognition for burglary (sound pressure > 90dB), directly calling security personnel, 3 minutes faster than detection systems.

[0056] Example 2, as Figure 2 As shown, a computer room AI duty officer decision-making system includes: an integrated FPGA chip and ARM processor, a multimodal perception module, an edge computing module, an active call module, and an interactive decision-making module;

[0057] The multimodal sensing module includes a current sensor submodule, a voiceprint microphone submodule, a temperature probe submodule, and an anti-interference submodule. The anti-interference submodule employs an RC low-pass filter circuit and TVS diode electrostatic protection. The RC low-pass filter circuit includes a parallel 100pF capacitor and a 10kΩ resistor to suppress electromagnetic interference in the computer room. The anti-interference submodule ensures the accuracy of the physical status data of the computer room collected by the current sensor submodule, the voiceprint microphone submodule, and the temperature probe submodule.

[0058] The edge computing module includes: a Bayesian generative adversarial network autonomous learning submodule and an LSTM anomaly detection submodule; it generates small sample fault data through the Bayesian generative adversarial network, and dynamically adjusts the detection threshold through a discriminator, so that only 5 similar faults are needed to generate an identification model; the collected data is compared with historical data through the LSTM model to obtain the confidence level, and the fault information is judged based on the confidence level;

[0059] The active call module includes a speech synthesis submodule and an active dialing submodule. It integrates a text-to-speech submodule to convert text into TTS speech for human-computer interaction. The active dialing submodule includes an FPGA dialing controller and a modified RJ11 interface. The FPGA dialing controller obtains pre-set maintenance personnel phone numbers, generates DTMF signals to trigger the landline outbound dialing mechanism, and calls the modified RJ11 interface to implement the active dialing function. The modified RJ11 interface uses a shielded twisted-pair current sensor to reduce signal attenuation and calculates the signal false trigger rate. To achieve anti-interference design and reduce false trigger rate The calculation formula is:

[0060]

[0061] in For noise current, The actual signal current, when the false trigger rate A value less than 0.1% is not a false trigger.

[0062] The interactive decision-making module includes a speech parsing submodule, an AI multimodal interactive decision-making submodule, and an autonomous learning optimization submodule. It converts the speech of maintenance personnel into text using an end-to-end ASR model, identifies keywords, and then uses AI to perform multimodal interaction and decision-making based on these keywords. The decision-making process incorporates voiceprint detection and current physical state verification to determine whether an anomaly has been resolved. Voiceprint detection checks for the presence of a 12kHz peak voiceprint in the data center, while current physical state verification calculates the current during the recovery phase after handling the anomaly. Compared with the reference stage current correlation coefficient To determine whether the treatment is effective. The calculation formula is as follows:

[0063] ,

[0064] A value >0.98 indicates that the handling was effective; and through multimodal interaction and decision-making, an immutable evidence chain of voiceprint + current + operation record multimodal decision log is generated; through the self-learning optimization submodule, the voiceprint of this fault is added to the edge sample library to enhance the security identification of maintenance personnel; the B-GAN generator synthesizes similar spectra and updates the discriminator weights.

[0065] Example 3, as Figure 3 As shown, a computer device includes a processor 101, a memory 102, a display module 103, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the computer room AI duty officer decision-making method described in Embodiment 1.

[0066] Example 4: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the computer room AI duty officer decision-making method described in Example 1.

[0067] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0068] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0069] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0070] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0071] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0072] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A decision-making method for AI-powered computer room operators, characterized in that, Includes the following steps: Step S1, the step of multimodal perception and data acquisition, involves real-time collection of physical status data of equipment in the computer room, and edge computing of the collected pseudo-anomaly data; Step S2, the edge computing step, establishes a small sample fault data identification model through the Bayesian Generative Adversarial Network (B-GAN) model, compares the historical fault patterns through the LSTM model, and determines whether the collected pseudo-abnormal data is abnormal based on the dynamic detection threshold. If the pseudo-abnormal data is determined to be abnormal, proceed to step S3; otherwise, the pseudo-abnormal data is considered normal data and no further processing is performed. Step S3: Generate corresponding text and voice information based on the abnormal data. The FPGA controller generates an AI alarm command and converts it into a DTMF dialing signal. By calling the RJ11 interface, it triggers the landline to make an outbound call to contact maintenance personnel. At the same time, a fault work order is created and the parameter information related to the fault is entered. Step S4: Based on the voice information of the maintenance personnel's response, the end-to-end ASR model is used to convert the voice information of the maintenance personnel into text, and keywords are identified. The validity of the voice is verified by the signal false trigger rate algorithm, fuzzy command processing is supported, and voiceprint detection and current analysis are used to confirm whether the problem has been resolved. Step S5: Repeat steps S3 and S4. After multiple rounds of multimodal interactive decision-making, determine the validity based on voice command parsing and physical state verification, mark the fault work order status, and record the handling time. Generate an immutable evidence chain through the multimodal decision log of voiceprint + current + operation record, add the fault voiceprint to the edge sample library to enhance the security identification of maintenance personnel, and synthesize similar spectra through the Bayesian Generative Adversarial Network (B-GAN) generator to update the discriminator weights.

2. The decision-making method for AI-powered computer room operators according to claim 1, characterized in that: The specific operation steps of step S1 are as follows: obtain current data through a current sensor, obtain voiceprint data through a voiceprint microphone, obtain temperature data through a temperature probe, and suppress electromagnetic interference in the computer room through an RC low-pass filter circuit and a TVS diode electrostatic protection to ensure the accuracy of the physical status data of the equipment in the computer room. The RC low-pass filter circuit includes a 100pF capacitor and a 10kΩ resistor connected in parallel.

3. The decision-making method for AI-powered computer room operators according to claim 1, characterized in that... The specific steps of step S2 are as follows: Step S21: Generate a recognition model using the Bayesian Generative Adversarial Network (B-GAN) autonomous learning engine, with Gaussian noise input to the generator. and historical false alarm sample feature vector Synthesize small sample fault data; dynamically adjust the detection threshold using a discriminator. The threshold calculation formula is: in This represents the number of times the event was accidentally triggered. The total number of detections is used to generate an identification model based on five similar faults. Step S22: Compare historical faults using an LSTM model to obtain fault information and confidence level. Determine whether the currently detected fault information is reliable if the confidence level is >95%.

4. The decision-making method for AI-powered computer room operators according to claim 1, characterized in that: The specific steps of S3 are as follows: S31, based on the abnormal information obtained in step S2, the abnormal text information is converted into corresponding abnormal text speech information through speech synthesis. S32, at the same time, the FPGA controller reads the preset phone number, generates the phone number as a DTMF dialing signal, and triggers the landline to make an outbound call by calling the modified RJ11 interface to contact the maintenance personnel. At the same time, a fault work order is created and the parameter information related to the fault is entered. S33: After the call is connected, the generated fault text voice information is played.

5. The decision-making method for AI-powered computer room operators according to claim 4, characterized in that: The RJ11 interface uses a shielded twisted-pair current sensor to reduce signal attenuation and calculates the signal false trigger rate. To achieve anti-interference design and reduce false trigger rate The calculation formula is: in For noise current, The actual signal current, when the false trigger rate A value less than 0.1% is not a false trigger.

6. The decision-making method for AI-powered computer room operators according to claim 1, characterized in that... Step S4 is detailed below: S41, based on the maintenance personnel's voice, uses an end-to-end ASR model to convert the maintenance personnel's voice into text and identify keywords; If the recognized speech is similar to the processed speech, then execute S43; otherwise, execute S44. S43, according to the preset processing flow, ask the staff whether it is 1. already processed 2. needs to be delayed 3. transferred to an expert, wait for the operation and maintenance personnel to reply, execute S41, if the identification result is 1, then execute S45, if it is 2, then execute S44; S44. Based on the information replied by the maintenance personnel, the AI ​​interacts and makes decisions with the maintenance personnel through multiple rounds of interaction based on the identified information. S45, the system actively acquires the physical status data of the computer room, and uses adaptive filtering of the voiceprint features in the voiceprint data to confirm whether the 12kHz peak has disappeared; it also performs physical status verification using current data and calculates the recovery current after handling. Compared with the reference stage current correlation coefficient , The calculation formula is as follows: in The mean value of the current data during the recovery phase. Current count during the reference phase The mean of the current data in the baseline phase, The standard deviation of the current data during the recovery phase. The standard deviation of the quasi-stage current data, if If the relevant formula is correct, then the treatment is deemed effective. If both voiceprint and voiceprint detection pass, proceed to step S5; otherwise, proceed to step S33.

7. A decision-making system for AI-powered computer room duty officers, characterized in that: This system integrates an FPGA chip and an ARM processor, and includes: a multimodal perception module, an edge computing module, an active call module, and an interactive decision-making module; The multimodal sensing module includes a current sensor submodule, a voiceprint microphone submodule, a temperature probe submodule, and an anti-interference submodule. The anti-interference submodule employs an RC low-pass filter circuit and TVS diode electrostatic protection. The RC low-pass filter circuit includes a parallel 100pF capacitor and a 10kΩ resistor to suppress electromagnetic interference in the computer room. The anti-interference submodule ensures the accuracy of the physical status data of the computer room collected by the current sensor submodule, the voiceprint microphone submodule, and the temperature probe submodule. The edge computing module includes: a Bayesian generative adversarial network autonomous learning submodule and an LSTM anomaly detection submodule; it generates small sample fault data through the Bayesian generative adversarial network, and dynamically adjusts the detection threshold through a discriminator, so that only 5 similar faults are needed to generate an identification model; the collected data is compared with historical data through the LSTM model to obtain the confidence level, and the fault information is judged based on the confidence level; The active call module includes a speech synthesis submodule and an active dialing submodule. By integrating a text-to-speech submodule, text is converted into text-to-speech information for human-computer interaction. The active dialing submodule includes an FPGA dialing controller and a modified RJ11 interface. The FPGA dialing controller obtains the pre-set maintenance personnel's phone number, generates a DTMF signal to trigger the landline outbound dialing mechanism, and calls the modified RJ11 interface to realize the active dialing function. The interactive decision-making module includes a speech parsing submodule, an AI multimodal interactive decision-making submodule, and an autonomous learning optimization submodule. It converts the speech of maintenance personnel into text using an end-to-end ASR model, identifies keywords, and then uses AI to perform multimodal interaction and decision-making based on these keywords. The decision-making process incorporates voiceprint detection and current physical state verification to determine whether an anomaly has been resolved. Voiceprint detection checks for the presence of a 12kHz peak voiceprint in the data center, while current physical state verification calculates the current during the recovery phase after handling the anomaly. Compared with the reference stage current correlation coefficient To determine whether the treatment is effective. The calculation formula is as follows: ， A value >0.98 indicates that the handling was effective; and through multimodal interaction and decision-making, an immutable evidence chain of voiceprint + current + operation record multimodal decision log is generated; through the self-learning optimization submodule, the voiceprint of this fault is added to the edge sample library to enhance the security identification of operation and maintenance personnel; the Bayesian generative adversarial network (B-GAN) generator synthesizes similar spectra and updates the discriminator weights.

8. The computer room AI duty officer decision-making system according to claim 7, characterized in that: The anti-interference submodule employs RC low-pass filtering and TVS diode electrostatic protection. The RC low-pass filter includes a 100pF capacitor and a 10kΩ resistor to suppress electromagnetic interference in the computer room. The modified RJ11 interface uses a shielded twisted-pair current sensor to reduce signal attenuation and calculates the signal false trigger rate. To achieve anti-interference design and reduce false trigger rate The calculation formula is: in For noise current, The actual signal current, when the false trigger rate A value less than 0.1% is not a false trigger.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the computer room AI duty officer decision-making method according to any one of claims 1-6.

10. A computer device, comprising a processor (101), a memory (102), a display module (103), and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the computer room AI duty officer decision-making method according to any one of claims 1-6.

Images