LLM-based scheduling management system

Abstract

The invention discloses an LLM-based scheduling management system, which belongs to the field of industrial and commercial enterprise scheduling management systems, adopts a data acquisition technology driven by bio-electricity signals, enriches data acquisition forms, can acquire various bio-electricity signals such as electroencephalogram, electrocardio, myoelectricity and the like, and improves the scheduling efficiency. And through a multi-channel parallel data transmission architecture and an advanced signal coding technology, the accuracy and stability of data transmission are ensured, key data in a human physiological state or a special environment are reflected, and some application scenes with relatively high requirements on data comprehensiveness and accuracy are met. The LLM-based scheduling module generates an optimal scheduling scheme by means of the powerful analysis and reasoning capability of a large language model in combination with a genetic algorithm and a simulated annealing algorithm, quantitative evaluation is carried out through a multi-objective optimization algorithm, various complex constraint conditions can be fully considered, and efficient scheduling is realized.

Description

Technical Field

[0001] The present invention relates to the field of scheduling management systems for industrial and commercial enterprises, and more specifically, to a scheduling management system based on LLM. Background Art

[0002] In the current era of rapid digital and intelligent development, scheduling management plays a crucial role in many fields, such as manufacturing, logistics and transportation, healthcare, etc. It is directly related to the rational utilization of resources, the efficient completion of tasks, and the effective control of costs, thereby affecting the operating efficiency and economic benefits of the entire system.

[0003] Currently, there are certain limitations in traditional scheduling management systems on the market. For example, in terms of data collection, most systems use conventional data collection technologies that rely on physical contact sensors or simple environmental monitoring devices, making it difficult to obtain key data such as bioelectrical signals that reflect the physiological state of the human body or in special environments, and it is difficult to meet some application scenarios with high requirements for data comprehensiveness and accuracy, such as personalized scheduling of medical rehabilitation equipment and task allocation of intelligent devices in special environments. Secondly, in terms of scheduling, as the core part of the system, the existing technologies for the scheduling module mostly generate scheduling plans based on fixed rules or simple mathematical models, and it is not very convenient to handle complex and changeable task and resource constraint conditions. Therefore, based on the above, we propose a scheduling management system based on LLM to solve the above existing problems. Summary of the Invention

[0004] 1. Technical Problems to be Solved

[0005] Aiming at the problems existing in the prior art, the purpose of the present invention is to provide a scheduling management system based on LLM. This system adopts a data collection technology driven by bioelectrical signals, enriches the form of data collection, can collect various bioelectrical signals such as electroencephalogram, electrocardiogram, and electromyogram, and through a multi-channel parallel data transmission architecture and advanced signal encoding technology, ensures the accuracy and stability of data transmission, achieves key data reflecting the physiological state of the human body or in special environments, and meets some application scenarios with high requirements for data comprehensiveness and accuracy;

[0006] The data preprocessing module uses a composite noise reduction method that combines deep learning algorithms with wavelet transform and adaptive filtering techniques, as well as analysis methods that comprehensively consider the time domain, frequency domain, and time-frequency domain. It can comprehensively extract data features and select appropriate data normalization methods according to requirements, greatly improving the quality of data processing. The scheduling module based on LLM utilizes the powerful analysis and reasoning capabilities of the large language model, combines genetic algorithms and simulated annealing algorithms to generate an optimal scheduling plan, and conducts quantitative evaluation through multi-objective optimization algorithms. It can fully consider various complex constraint conditions and achieve efficient scheduling.

[0007] 2. Technical Solution

[0008] To solve the above problems, the present invention adopts the following technical solutions.

[0009] A scheduling management system based on LLM includes a data acquisition module, a data preprocessing module, a communication module, a scheduling module based on LLM, and a storage module. Each module works collaboratively through an efficient and stable data transmission mechanism:

[0010] The data acquisition module operates based on data acquisition technology driven by bioelectric signals. The data acquisition module collects bioelectric signals, converts the collected bioelectric signals into digital signals, and realizes data output through a multi-channel parallel data transmission architecture. At the hardware level of the multi-channel parallel data transmission architecture, multiple high-speed data lines are arranged in parallel, and each data line corresponds to an independent data transmission channel. At the signal encoding level of the multi-channel parallel data transmission architecture, Manchester encoding or differential Manchester encoding technology is used to convert digital signals into signal forms suitable for transmission, enhancing the signal anti-interference ability, facilitating accurate restoration and extraction of the original data at the receiving end, and further ensuring the accuracy and stability of data transmission;

[0011] The data preprocessing module receives data from the data acquisition module, encapsulates the preprocessed data in a specific data format, and uses memory mapping file technology to achieve fast transmission at the memory level. The specific data format includes a data frame containing feature identification, data content, and checksum information, which is convenient for later transmission to the scheduling module based on LLM;

[0012] The LLM-based scheduling module receives the data transmitted by the data preprocessing module. The generated scheduling scheme data is serialized according to the requirements of the communication protocol and converted into a format suitable for network transmission. For example, when using Bluetooth transmission, the data is segmented into appropriate-sized data packets through the Bluetooth protocol stack for transmission. When using WiFi transmission, the TCP / IP protocol is utilized to establish a reliable connection and then the data is sent. When using ZigBee network transmission, the data is encapsulated into frames according to the ZigBee protocol and transmitted through the self-organizing network routing path, and the data is transmitted to the communication module.

[0013] The communication module receives the data transmitted by the LLM-based scheduling module and performs adaptive data transmission according to the communication protocol supported by the external device. For example, when communicating with a Bluetooth device, it follows the pairing, connection, and data transmission processes of the Bluetooth device. When communicating with a WiFi device, it performs data reception and transmission according to the access rules and data interaction methods of the WiFi network. When communicating with a ZigBee device, it utilizes the self-organizing network and routing functions of the ZigBee network to ensure that the data is accurately delivered to the target device.

[0014] The storage module uses a high-speed SATA interface or an NVMe protocol interface to interact with other modules. The temporary data collected by the data acquisition module, the intermediate data processed by the data preprocessing module, and part of the scheme data generated by the LLM-based scheduling module are quickly written into the storage module for storage.

[0015] Furthermore, the data acquisition module includes a bioelectric signal sensor, a signal amplifier, a filter, and a signal conditioning circuit.

[0016] The bioelectric signal sensor selects a high-sensitivity dry electrode sensor to stably collect bioelectric signals including but not limited to electroencephalogram (EEG), electrocardiogram (ECG), and electromyogram (EMG). The signal amplifier has a programmable gain function, and the amplification factor can be flexibly set through system software. A low-noise operational amplifier is used to ensure high-fidelity amplification of the signal. The filter uses a digital band-pass filter, and the passband frequency range of the digital band-pass filter can be dynamically adjusted according to different bioelectric signal types through software algorithms. For example, for EEG signals, the passband frequency range is set to 0.5 - 100 Hz; for ECG signals, it is set to 0.05 - 150 Hz; and for EMG signals, it is set to 20 - 500 Hz. Through precise filtering operations, power frequency interference, baseline drift, and other noises are effectively removed, and at the same time, the analog bioelectric signals are converted into digital signals for subsequent processing. The signal conditioning circuit is a passive circuit network composed of operational amplifiers, precision resistors, and capacitors.

[0017] Further, the data preprocessing module includes a noise reduction unit, a feature extraction unit, and a data normalization unit;

[0018] The noise reduction unit adopts a composite noise reduction method that combines deep learning algorithms with wavelet transform and adaptive filtering techniques. Specifically, the composite noise reduction method first uses a deep learning model to perform feature learning on bioelectric signals in a complex noise environment, identifies the noise feature patterns, then combines the time-frequency analysis characteristics of wavelet transform to perform multi-scale decomposition on the signal to further separate the noise and the effective signal. Finally, through the adaptive filtering algorithm, the filtering parameters are dynamically adjusted according to the real-time changes of the signal, so as to effectively remove the noise while retaining the key features of the bioelectric signal to the greatest extent.

[0019] Further, the feature extraction unit comprehensively uses time-domain, frequency-domain, and time-frequency-domain analysis methods. Specifically, the analysis methods are to extract feature parameters such as mean, variance, peak factor, and zero-crossing rate in the time domain, calculate power spectral density, center frequency, etc. in the frequency domain through Fourier transform, and extract time-frequency distribution features in the time-frequency domain using tools such as short-time Fourier transform and wavelet transform to provide comprehensive and rich data features for subsequent analysis;

[0020] The data normalization unit selects the minimum-maximum normalization or Z-score normalization method according to actual needs. The minimum-maximum normalization method specifically uses the formula

[0021]

[0022] to accurately map the extracted feature parameter X to the interval [0, 1], where X min and Xmax are respectively the minimum and maximum values of this feature parameter in the training dataset;

[0023] The Z-score normalization method specifically uses the formula

[0024]

[0025] to normalize the feature parameter to a standard distribution with a mean of 0 and a standard deviation of 1, where μ and σ are respectively the mean and standard deviation of this feature parameter in the training dataset, so as to eliminate the dimensional difference between different feature parameters and make the data more suitable for subsequent model analysis.

[0026] Further, the scheduling module based on LLM includes a data parsing sub-module, a model calling sub-module, a solution generation sub-module, and a solution evaluation sub-module;

[0027] The data parsing sub-module uses lexical analysis, syntactic analysis, and semantic understanding algorithms in natural language processing technology, combined with the rule matching algorithm of the scheme evaluation sub-module, to deeply parse the preprocessed data. The data parsing sub-module accurately identifies key information such as task type, priority, time requirements, and resource requirements, and converts unstructured data into structured instructions, providing clear input for subsequent model analysis.

[0028] Furthermore, the model calling sub-module calls a large language model pre-trained with massive data, and inputs the parsed key information into the model for analysis and reasoning through a secure and efficient API interface. The large language model has the ability to be updated in real time, continuously optimizes the analysis results by continuously learning the latest data and knowledge, and improves the accuracy and rationality of the scheduling scheme.

[0029] The scheme generation sub-module combines genetic algorithms and simulated annealing algorithms to generate a scheduling scheme according to the output results of the large language model. The scheduling scheme comprehensively considers factors such as the order of tasks, resource allocation, time limits, and cost constraints, and generates an optimal scheduling scheme that meets multiple constraint conditions. The genetic algorithm searches for the optimal solution in the solution space by simulating selection, crossover, and mutation operations in the process of biological evolution, while the simulated annealing algorithm avoids the algorithm falling into a local optimal solution by introducing random perturbations.

[0030] Furthermore, the scheme evaluation sub-module uses the multi-objective optimization algorithm of non-dominated sorting genetic algorithm-II (NSGA-II) to quantitatively analyze the generated scheduling scheme from multiple dimensions such as task completion time, resource utilization rate, cost, and risk. Specifically, in the dimension of task completion time, the total completion time of all tasks in the scheme is accurately calculated and compared with the preset time threshold. In terms of resource utilization rate, the actual usage ratio of various resources such as manpower, material resources, and equipment to the total available amount is detailedly counted. The cost dimension comprehensively considers direct costs required to execute tasks, such as raw material procurement and equipment rental costs, and indirect costs, such as management costs and energy consumption costs. The risk dimension evaluates various risk factors that the scheme may face during execution, such as equipment failures and personnel changes. By quantitatively analyzing these dimensions, a detailed evaluation index vector is generated, and the Euclidean distance and cosine similarity methods are used to calculate the distance between the evaluation index vectors of different schemes and the ideal target vector, providing a scientific basis for the optimization of the scheme.

[0031] 3. Beneficial effects

[0032] Compared with the prior art, the advantages of the present invention are as follows:

[0033] (1) This solution adopts bioelectric signal-driven data acquisition technology, enriches the form of data acquisition, and can collect multiple bioelectric signals such as EEG, ECG, and EMG. It also uses a multi-channel parallel data transmission architecture and advanced signal coding technology to ensure the accuracy and stability of data transmission, and achieve key data that reflects the physiological state of the human body or special environments, meeting some application scenarios that require high data comprehensiveness and accuracy.

[0034] (2) In this solution, the data preprocessing module uses a composite noise reduction method that combines deep learning algorithms with wavelet transform and adaptive filtering technology, as well as comprehensive analysis methods in the time domain, frequency domain and time-frequency domain to comprehensively extract data features and select appropriate data normalization methods according to needs, which greatly improves the quality of data processing. The scheduling module based on LLM uses the powerful analytical reasoning capabilities of the large language model, combined with genetic algorithms and simulated annealing algorithms to generate the optimal scheduling plan, and performs quantitative evaluation through a multi-objective optimization algorithm, which can fully consider various complex constraints and achieve efficient scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of the main system architecture and data flow of the present invention;

[0036] Figure 2 A mind map of the overall system architecture of the present invention;

[0037] Figure 3 A mind map of the data acquisition module of the present invention;

[0038] Figure 4 A mind map of the data preprocessing module of the present invention;

[0039] Figure 5 A mind map of the scheduling module of the present invention;

[0040] Figure 6 This is a mind map of the communication module and storage module of the present invention.

[0041] Description of the numbers in the figure:

[0042] 1. Data acquisition module; 101. Bioelectric signal sensor; 102. Signal amplifier; 103. Filter; 104. Signal conditioning circuit;

[0043] 2. Data preprocessing module; 201. Noise reduction unit; 202. Feature extraction unit; 203. Data normalization unit;

[0044] 3. Communication module;

[0045] 4. Scheduling Module; 401. Data Parsing Sub-module; 402. Model Invocation Sub-module; 403. Solution Generation Sub-module; 404. Solution Evaluation Sub-module;

[0046] 5. Storage Module. Detailed Implementation Manner

[0047] Next, in conjunction with the accompanying drawings of the present invention specification, the technical solutions in the embodiments of the present invention will be clearly and completely described; obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0048] Embodiment:

[0049] In conjunction with the accompanying drawings of the specification Figures 1 - 6 , in a large comprehensive hospital, the principle of using this scheduling management system based on LLM for scheduling management is as follows:

[0050] First, in each examination department, such as the Department of Cardiology, Physical Examination Center, etc., the bioelectric signal sensor 101 of the data acquisition module 1 is placed at a specific part of the patient's body. Taking the acquisition of electrocardiogram signals as an example, the highly sensitive dry electrode sensor can accurately and stably obtain the electrocardiogram activity signals of the patient. These signals are extremely weak, so the signal amplifier 102 plays an important role. The technical personnel in the hospital will flexibly set the amplification factor through the system software according to the conditions of different patients. For example, for elderly patients with weaker signals, the amplification factor is appropriately increased; for younger patients with relatively stronger signals, an appropriate factor is set to ensure that the signals can be clearly amplified without distortion. During the amplification process, the low-noise operational amplifier ensures the high fidelity of the signals, minimizing the noise mixed in the signals during the amplification process. The filter 103 is also indispensable in the entire data acquisition process. Due to various interference sources in the hospital environment, such as power frequency interference generated by electrical equipment and baseline drift caused by the movement of the patient's body, etc., the filter 103 strictly sets the passband frequency range to 0.05 - 150 Hz according to the characteristics of the electrocardiogram signals. Within this frequency range, most of the noise can be effectively removed while the key characteristics of the electrocardiogram signals can be completely retained. At the same time, the filter 103 is also responsible for converting the analog bioelectric signals into digital signals for subsequent digital processing.

[0051] The signal conditioning circuit 104 consists of an operational amplifier, precision resistors, and capacitors to form a passive circuit network. It further conditions the signals after amplification and filtering, stabilizing the amplitude and phase of the signals to ensure that the signal quality meets the requirements of subsequent transmission and processing.

[0052] The data after the above processing is transmitted through a multi-channel parallel data transmission architecture. At the hardware level, multiple high-speed data lines are arranged in parallel, and each data line corresponds to an independent data transmission channel, achieving parallel data transmission and greatly improving the transmission efficiency; at the signal coding level, Manchester coding technology is adopted to convert digital signals into signal forms suitable for transmission. The characteristic of Manchester coding is that there is a level transition in the middle of each symbol, and this transition serves as both a clock signal and a data signal, enhancing the anti-interference ability of the signal. Even in the complex electromagnetic environment of a hospital, it can ensure that the data is accurately transmitted to the data preprocessing module 2.

[0053] Then, after the data preprocessing module 2 receives the data transmitted by the data acquisition module 1, the noise reduction unit 201 starts to work. The deep learning algorithm will first learn a large amount of historical electrocardiogram signal data, which contains electrocardiogram signals in various complex noise environments. Through learning, the model can identify the characteristic patterns of different noises. When actually processing newly acquired electrocardiogram signals, combined with the time-frequency analysis characteristics of wavelet transform, the signal is decomposed at multiple scales. Wavelet transform can analyze the signal at different time and frequency resolutions, decomposing the signal into sub-signals in different frequency bands, so as to more accurately separate noise and effective signals. Finally, the adaptive filtering algorithm dynamically adjusts the filtering parameters according to the real-time changes of the signal. For example, when the patient suddenly coughs or the body moves, causing abnormal fluctuations in the signal, the adaptive filtering algorithm can adjust in time, while effectively removing noise, maximizing the retention of key features of the electrocardiogram signal, such as P waves, QRS complexes, T waves, etc.

[0054] The feature extraction unit 202 comprehensively uses time-domain, frequency-domain, and time-frequency-domain analysis methods to fully extract useful information from electrocardiogram signals. In the time domain, feature parameters such as mean, variance, peak factor, and zero-crossing rate are extracted. The mean reflects the average level of the electrocardiogram signal over a period of time, the variance reflects the degree of signal fluctuation, the peak factor is of great significance for detecting abnormal peaks in the electrocardiogram signal, and the zero-crossing rate can reflect the change frequency of the signal. In the frequency domain, the time-domain signal is converted into a frequency-domain signal through Fourier transform, and features such as power spectral density and center frequency are calculated. The power spectral density can show the energy distribution of different frequency components, and the center frequency reflects the frequency range where the signal energy is concentrated. In the time-frequency domain, tools such as short-time Fourier transform and wavelet transform are used to extract time-frequency distribution features. The short-time Fourier transform can observe the frequency changes of the signal in a short time, and the wavelet transform can more carefully capture the local features of the signal at different times and frequencies. These rich feature parameters provide comprehensive data support for subsequent analysis.

[0055] The data normalization unit 203 selects the Z-score normalization method according to actual requirements. In the case data of the hospital, a large number of electrocardiogram signal characteristic parameters of different patients are collected. By calculating the mean (μ) and standard deviation (σ) of these parameters in the training dataset, using the formula Normalize the extracted characteristic parameters to a standard distribution with a mean of 0 and a standard deviation of 1. This eliminates the dimensional differences between different characteristic parameters. For example, parameters with different dimensions such as amplitude and frequency can be compared and analyzed on the same scale after normalization, making the data more suitable for subsequent model analysis.

[0056] The preprocessed data is encapsulated in a data frame format containing feature identification, data content, and checksum information. The feature identification is used to clarify the type of data, such as whether it is an electrocardiogram signal or other types of medical data; the data content is the actual electrocardiogram signal characteristic data after processing; the checksum information is used to detect whether data errors occur during data transmission to ensure the integrity and accuracy of the data. The encapsulated data uses the memory mapping file technology to achieve fast transmission at the memory level and efficiently transmits the data to the scheduling module 4 based on the LLM.

[0057] After receiving the preprocessed data, the data parsing sub-module 401 uses lexical analysis, syntactic analysis, and semantic understanding algorithms in natural language processing technology, combined with the rule matching algorithm solution evaluation sub-module 404, to deeply parse the data. For example, when the data contains information such as "Patient Zhang San, scheduled for an electrocardiogram examination at 10 am, is an emergency patient", lexical analysis breaks the sentence into individual words or phrases, syntactic analysis determines the grammatical relationships between these words, and the semantic understanding algorithm understands the actual meaning of the sentence. Through these operations, combined with the rule matching algorithm, accurately identify that the task type is an electrocardiogram examination, the priority is for emergency patients first, the time requirement is 10 am, and the resource requirement is an electrocardiogram examination device. In this way, the unstructured data is converted into structured instructions, providing clear input for subsequent model analysis.

[0058] The model invocation sub-module 402 invokes the large language model pre-trained with massive medical data through a secure and efficient API interface. These medical data include the usage records and scheduling plans of medical devices in different hospitals, different time periods, and various patient situations. After receiving the key information input by the data parsing sub-module 401, the large language model performs analysis and reasoning. At the same time, the large language model has the ability to be updated in real time. The hospital continuously uploads new medical data to the model training platform, and the model continuously optimizes the analysis results by continuously learning these latest data and knowledge, improving the accuracy and rationality of the scheduling plan.

[0059] The scheduling plan generation sub-module 403 combines the genetic algorithm and the simulated annealing algorithm to generate a scheduling plan based on the output results of the large language model. When generating the plan, factors such as the sequence of tasks, resource allocation, time limit, and cost constraints are comprehensively considered. For example, for the electrocardiogram examination task of emergency patients, equipment and medical staff are arranged preferentially; for multiple examination tasks that require continuous use of the same equipment, the order is reasonably arranged to reduce the idle time of the equipment. The genetic algorithm searches for the optimal solution in the solution space by simulating the selection, crossover, and mutation operations in the biological evolution process. The selection operation selects excellent individuals according to the fitness value of the individuals (which can be understood as the quality of the plan in scheduling). The crossover operation exchanges part of the genes of two excellent individuals to generate new individuals. The mutation operation randomly changes the genes of individuals with a certain probability to introduce new solutions. The simulated annealing algorithm avoids the algorithm falling into a local optimal solution by introducing random perturbations. In the initial stage, the algorithm searches at a higher temperature (corresponding to a larger random perturbation). As the search progresses, the temperature gradually decreases, and the random perturbation gradually decreases, finally converging to the global optimal solution or an approximate global optimal solution. By combining these two algorithms, an optimal scheduling plan that meets multiple constraint conditions is generated.

[0060] The plan evaluation sub-module 404 uses the multi-objective optimization algorithm of the non-dominated sorting genetic algorithm II (NSGA-II) to quantitatively analyze the generated scheduling plan from multiple dimensions such as task completion time, resource utilization rate, cost, and risk. In the dimension of task completion time, the total completion time of all tasks in the plan is accurately calculated and compared with the preset time threshold. For example, for the electrocardiogram examination of emergency patients, the preset time threshold may be 15 minutes. If the actual calculated completion time exceeds this threshold, it indicates that there may be problems with the time arrangement in the plan. In terms of resource utilization rate, the ratio of the actual usage amount of various resources to the total available amount is detailedly counted. For example, the ratio of the actual usage duration of the electrocardiogram examination equipment to the total available duration of the equipment, and the ratio of the actual working time of medical staff to the total scheduled working time, etc. The cost dimension comprehensively considers the direct costs required to execute the tasks, such as the maintenance costs of the equipment and the procurement costs of consumables such as electrode patches, and indirect costs, such as the management costs of the hospital and the energy consumption costs. The risk dimension evaluates various risk factors that the plan may face during execution, such as the risk of equipment failure and the risk of personnel changes caused by sudden illness of medical staff. Through quantitative analysis of these dimensions, a detailed evaluation index vector is generated. Then, methods such as Euclidean distance and cosine similarity are used to calculate the distance between the evaluation index vectors of different plans and the ideal target vector. The ideal target vector is set according to the management objectives and historical experience of the hospital. The closer the distance, the closer the plan is to the ideal state, providing a scientific basis for the optimization of the plan.

[0061] The hospital uses a WiFi network for data transmission. According to the TCP / IP protocol, the communication module 3 first establishes a reliable connection with the internal network devices of the hospital. During the connection establishment process, mechanisms such as the three-way handshake are used to ensure the stability and reliability of the connection. After the scheduling module 4 based on the LLM generates the scheduling scheme data, the communication module 3 encapsulates and transmits the data according to the requirements of the TCP / IP protocol. The data is segmented into data packets of appropriate sizes, and each data packet contains information such as the destination address, source address, and data content. These data packets are transmitted through the WiFi network to various medical device terminals and the mobile devices of medical staff. For example, the tablet computer at the nurse station will receive the latest patient examination arrangement information, and the doctor's mobile phone can also receive real-time reminders of the examination tasks of the patients they are responsible for, ensuring that the device usage arrangements are promptly communicated to relevant personnel and guaranteeing the efficient progress of medical examinations.

[0062] The storage module 5 uses a high-speed SATA interface and has the capabilities of large-capacity storage and high-speed reading and writing. The temporary electrocardiogram data collected by the data acquisition module 1 will be quickly written into the storage module 5 for storage. These temporary data may be used for reexamination or further research and analysis in the future. The intermediate data processed by the data preprocessing module 2, such as the data after noise reduction, feature extraction, and normalization, will also be stored in the storage module 5. In addition, some of the scheme data generated by the scheduling module 4 based on the LLM will also be stored here.

[0063] The storage module 5 provides the functions of data backup and historical data query for the entire system, facilitating the hospital to review and statistically analyze the usage of medical devices, so as to continuously optimize the scheduling scheme and improve the service quality and management level of the hospital.

[0064] The above is only a preferred specific embodiment of the present invention; however, the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution of the present invention and its improved concept, makes equivalent substitutions or changes, and all should be covered by the protection scope of the present invention.

Claims

1. A scheduling and dispatching management system based on LLM, characterized in that: It includes a data acquisition module (1), a data preprocessing module (2), a communication module (3), an LLM-based scheduling module (4) and a storage module (5), and each module works together through an efficient and stable data transmission mechanism: The data acquisition module (1) operates based on a data acquisition technology driven by bioelectric signals, and the data acquisition module (1) acquires bioelectric signals and converts the acquired bioelectric signals into digital signals, and realizes data output with the help of a multi-channel parallel data transmission architecture. The hardware level of the multi-channel parallel data transmission architecture adopts a plurality of high-speed data lines arranged in parallel, and each data line corresponds to an independent data transmission channel. The signal encoding level of the multi-channel parallel data transmission architecture adopts Manchester encoding or differential Manchester encoding technology to convert digital signals into a signal form suitable for transmission, thereby enhancing the signal's anti-interference ability, facilitating accurate restoration and extraction of original data at the receiving end, and further ensuring the accuracy and stability of data transmission; The data preprocessing module (2) receives data from the data acquisition module (1), encapsulates the preprocessed data according to a specific data format, and uses memory mapping file technology to achieve fast transmission at the memory level. The specific data format includes a data frame of feature identification, data content and check code information, which is convenient for subsequent transmission to the scheduling module (4) based on LLM; The scheduling module (4) based on LLM receives the data transmitted by the data preprocessing module (2), and performs serialization processing on the generated scheduling plan data according to the requirements of the communication protocol, and converts it into a format suitable for network transmission. For example, if Bluetooth transmission is adopted, the data is divided into data packets of appropriate size for transmission through the Bluetooth protocol stack; if WiFi transmission is adopted, the TCP / IP protocol is used to establish a reliable connection before data transmission; if ZigBee network transmission is adopted, the data is encapsulated into frames according to the ZigBee protocol, and transmitted through the self-organizing network routing path, and the data is transmitted to the communication module (3); The communication module (3) receives data transmitted by the scheduling module (4) based on the LLM, and performs adaptive data transmission according to the communication protocol supported by the external device. For example, when communicating with a Bluetooth device, the pairing, connection and data transmission process of the Bluetooth device is followed. When communicating with a WiFi device, data is sent and received according to the access rules and data interaction method of the WiFi network. When communicating with a ZigBee device, the self-organizing network and routing functions of the ZigBee network are used to ensure that the data is accurately delivered to the target device. The storage module (5) uses a high-speed SATA interface or an NVMe protocol interface to exchange data with other modules. The temporary data collected by the data collection module (1), the intermediate data processed by the data preprocessing module (2), and the partial scheme data generated by the scheduling module (4) based on LLM are quickly written into the storage module (5) for storage.

2. The LLM-based scheduling management system according to claim 1, characterized in that: The data acquisition module (1) comprises a bioelectric signal sensor (101), a signal amplifier (102), a filter (103) and a signal conditioning circuit (104); The bioelectric signal sensor (101) uses a high-sensitivity dry electrode sensor to stably collect bioelectric signals including but not limited to EEG, ECG, and EMG. The signal amplifier (102) has a programmable gain function, and the amplification factor is flexibly set through system software. A low-noise operational amplifier is used to ensure high-fidelity amplification of the signal. The filter (103) uses a digital bandpass filter, and the passband frequency range of the digital bandpass filter can be dynamically adjusted according to different bioelectric signal types through a software algorithm. For example, for EEG signals, the passband frequency range is set to 0.5-100 Hz, for ECG signals, it is set to 0.05-150 Hz, and for EMG signals, it is set to 20-500 Hz. Through precise filtering operations, power frequency interference, baseline drift and other noises are effectively removed, and the analog bioelectric signals are converted into digital signals for subsequent processing. The signal conditioning circuit (104) is a passive circuit network composed of an operational amplifier, precision resistors and capacitors.

3. The LLM-based scheduling and dispatching management system according to claim 1, characterized in that: The data preprocessing module (2) comprises a noise reduction unit (201), a feature extraction unit (202) and a data normalization unit (203); The noise reduction unit (201) adopts a composite noise reduction method that combines a deep learning algorithm with wavelet transform and adaptive filtering technology. Specifically, the composite noise reduction method first uses a deep learning model to perform feature learning on bioelectric signals in a complex noise environment to identify noise feature patterns, then combines the time-frequency analysis characteristics of wavelet transform to perform multi-scale decomposition on the signal to further separate noise and effective signals, and finally uses an adaptive filtering algorithm to dynamically adjust the filtering parameters according to the real-time changes of the signal, thereby effectively removing noise while retaining the key features of the bioelectric signal to the greatest extent.

4. The LLM-based scheduling and dispatching management system according to claim 3 is characterized in that: The feature extraction unit (202) comprehensively uses time domain, frequency domain and time-frequency domain analysis methods. The analysis method specifically extracts feature parameters such as mean, variance, peak factor, zero-crossing rate, etc. in the time domain, calculates features such as power spectrum density and center frequency by Fourier transform in the frequency domain, and extracts time-frequency distribution features by using tools such as short-time Fourier transform and wavelet transform in the time-frequency domain, so as to provide comprehensive and rich data features for subsequent analysis; The data normalization unit (203) selects the minimum-maximum normalization method or the Z-score normalization method according to actual needs. The minimum-maximum normalization method is specifically to use the formula The extracted feature parameter X is accurately mapped to the interval [0,1], where X min Xmax and Xmax are the minimum and maximum values ​​of the feature parameter in the training data set, respectively; The Z-score normalization method is specifically based on the formula The feature parameters are normalized to a standard distribution with a mean of 0 and a standard deviation of 1, where μ and σ are the mean and standard deviation of the feature parameters in the training data set, respectively. This eliminates the dimensional differences between different feature parameters and makes the data more suitable for subsequent model analysis.

5. The LLM-based scheduling management system according to claim 1, characterized in that: The scheduling module (4) based on LLM includes a data analysis submodule (401), a model calling submodule (402), a solution generation submodule (403) and a solution evaluation submodule (404); The data analysis submodule (401) uses lexical analysis, syntactic analysis and semantic understanding algorithms in natural language processing technology, combined with the rule matching algorithm solution evaluation submodule (404), to perform in-depth analysis on the pre-processed data. The data analysis submodule (401) accurately identifies key information such as task type, priority, time requirement and resource demand, converts unstructured data into structured instructions, and provides clear input for subsequent model analysis.

6. The LLM-based scheduling and dispatching management system according to claim 5 is characterized in that: The model calling submodule (402) calls a large language model pre-trained with massive data, and inputs the parsed key information into the model for analysis and reasoning through a safe and efficient API interface. The large language model has the ability to update in real time, and continuously optimizes the analysis results by continuously learning the latest data and knowledge, thereby improving the accuracy and rationality of the scheduling and dispatching scheme. The scheme generation submodule (403) combines a genetic algorithm and a simulated annealing algorithm to generate a scheduling scheme according to the output result of the large language model, and the scheduling scheme comprehensively considers factors such as the order of tasks, resource allocation, time constraints, cost constraints, etc., to generate an optimal scheduling scheme that meets multiple constraints. The genetic algorithm searches for the optimal solution in the solution space by simulating the selection, crossover and mutation operations in the biological evolution process, and the simulated annealing algorithm introduces random disturbances to avoid the algorithm from falling into a local optimal solution.

7. The LLM-based scheduling and dispatching management system according to claim 5 is characterized in that: The scheme evaluation submodule (404) uses a multi-objective optimization algorithm of a non-dominated sorting genetic algorithm-II (NSGA-II) to perform quantitative analysis on the generated scheduling scheme from multiple dimensions, including task completion time, resource utilization, cost, and risk. Specifically, in the dimension of task completion time, the total completion time of all tasks in the scheme is accurately calculated and compared with a preset time threshold. In terms of resource utilization, the ratio of the actual usage of various resources, such as manpower, material resources, and equipment, to the total available amount is statistically analyzed in detail. In the cost dimension, the direct cost required to perform the task, such as raw material procurement and equipment rental fees, and indirect costs, such as management fees and energy consumption costs, are comprehensively considered. In the risk dimension, various risk factors that may be faced by the scheme during execution, such as equipment failure and personnel changes, are evaluated. By performing quantitative analysis on these dimensions, a detailed evaluation index vector is generated. The distance between the evaluation index vectors of different schemes and the ideal target vector is calculated using the Euclidean distance and cosine similarity methods, thereby providing a scientific basis for the optimization of the scheme.

Images