Search

CN-121997302-A - Digital employee semantic interaction enhancement method and medium based on large language model

CN121997302ACN 121997302 ACN121997302 ACN 121997302ACN-121997302-A

Abstract

The invention relates to the technical field of semantic processing, in particular to a digital employee semantic interaction enhancement method and medium based on a large language model. The method comprises the steps of obtaining flow measurement compensation data after delay compensation at each moment according to valve opening data and flow measurement data distribution at different moments in a neighborhood range at each moment, judging whether an actuator-sensor is communicated or not, obtaining operator adjustment feature labels at each moment according to the valve opening data at different moments in the neighborhood range at each moment and the flow measurement compensation data distribution after delay compensation if the actuator-sensor is communicated, setting the operator adjustment feature labels at each moment to be in a fault mode if the actuator-sensor is not communicated, constructing constraint prompt words based on the operator adjustment feature labels and the communication state of the actuator-sensor, and inputting the constraint prompt words into a large language model to generate interactive feedback information. The invention improves the physical consistency of semantic interaction by accurately converting physical facts into semantic constraints.

Inventors

  • WANG JINGNI

Assignees

  • 陕西易普科技有限责任公司

Dates

Publication Date
20260508
Application Date
20260128

Claims (10)

  1. 1. A digital employee semantic interaction enhancement method based on a large language model, the method comprising: acquiring valve opening data and flow measurement data of a fluid control scene at each moment; According to valve opening data and flow measurement data distribution at different moments in a neighborhood range at each moment, flow measurement compensation data after delay compensation at each moment is obtained, and whether an actuator-sensor is communicated or not is judged; if the flow is communicated, obtaining an adjustment-response isotropy product of each moment according to valve opening data of different moments in a local range of each moment and flow measurement compensation data distribution after delay compensation; Acquiring a homeotropic intensity integral value and a homeotropic inverse count of each moment according to the homeotropic intensity integral distribution of different moments in the neighborhood range of each moment, and acquiring a homeotropic intensity threshold value and an inverse inversion threshold value; If the two types of the input language models are not communicated, the operator adjusting feature labels at each moment are set to be in a fault mode, constraint prompt words are constructed based on the operator adjusting feature labels and the communication state of the actuator and the sensor, and the constraint prompt words are input into the large language models to generate interactive feedback information.
  2. 2. The method for enhancing semantic interaction of digital staff based on large language model as claimed in claim 1, wherein the method for acquiring flow measurement compensation data comprises: Judging whether the valve is in a silence steady state or not according to valve opening data at different moments in a neighborhood range at each moment; If the time is not in the silence steady state, acquiring accumulated values of products between valve opening data at different times in a neighborhood range of each time and flow measurement data corresponding to any preset delay time; And acquiring the flow measurement data after each moment and passing through the transmission delay moment in a sequence formed by all flow measurement data corresponding to the neighborhood range of each moment, and taking the flow measurement data after each moment and the delay moment as flow measurement compensation data after each moment and the delay compensation.
  3. 3. The method for enhancing semantic interaction of digital employee based on big language model according to claim 2, wherein the method for obtaining the judgment whether the digital employee is in silence steady state comprises: obtaining fluctuation degrees of valve opening data at all times in a neighborhood range of each time, and judging that the valve opening data is not in a silence steady state if a fluctuation degree result is larger than or equal to a preset silence threshold value; And if the fluctuation degree result is smaller than the preset silence threshold, judging that the silence is in a silence steady state.
  4. 4. A method for enhancing semantic interaction of digital employee based on large language model according to claim 3, wherein the method for obtaining the connectivity of actuator-sensor comprises: If the device is in a silence steady state, judging the communication between the actuator and the sensor; If the flow measurement compensation data is not in the silence steady state, obtaining the correlation coefficient of a sequence formed by valve opening data at all times in the neighborhood range of each time and flow measurement compensation data after corresponding delay compensation, and if the correlation coefficient is larger than or equal to a preset connectivity threshold value, judging that the actuator and the sensor are communicated, otherwise, judging that the actuator and the sensor are not communicated.
  5. 5. A method for enhancing semantic interaction of digital employees based on a large language model according to claim 1, wherein the method for obtaining the adjustment-response homography product comprises: taking valve opening data in a local range at each moment or flow measurement compensation data after delay compensation as analysis data, and performing first-order differential fitting on the analysis data corresponding to different moments to obtain a data change rate; The product of the rate of change of data between the valve opening data and the flow measurement compensation data is obtained as a tuning-response co-directional product.
  6. 6. A digital employee semantic interaction enhancement method based on a large language model according to claim 5, wherein the method for acquiring the homeopathic adjusted intensity integral value and the retrospective adjusted retrospective count comprises: If the adjustment-response coherence product of the existing time is larger than a preset coherence product threshold, taking the corresponding time as the homeotropic time, obtaining the adjustment-response coherence product of each homeotropic time, multiplying the time difference between each homeotropic time and the previous adjacent time as a first product, and calculating the accumulated value of the first products corresponding to all homeotropic times in the neighborhood range of each time as a homeotropic adjustment intensity integral value; If the adjustment-response isotropy product is smaller than a preset isotropy product threshold value at a certain moment and the data change rate of valve opening data between the corresponding moment and the previous moment is unequal, setting the condition value at the corresponding moment as a positive integer 1, otherwise setting the condition value as 0, and accumulating the condition values at all moments in the neighborhood range at each moment to be used as counter potential adjustment counter counting.
  7. 7. The method for enhancing semantic interaction of digital staff based on large language model as claimed in claim 1, wherein the method for obtaining the homeopathic strength threshold and the inverse homeotropic inversion threshold comprises: Acquiring an intensity mean value of the homeopathic regulation intensity integral value at all moments in a neighborhood range of each moment, acquiring a product of a preset first confidence coefficient and the intensity standard deviation, and calculating the sum of a product result and the intensity mean value to serve as a homeopathic intensity threshold value; And obtaining the inversion mean value of the inversion count regulated by the inversion potential at all time points in the neighborhood range at each moment, inverting the standard deviation, obtaining the product of a preset second confidence coefficient and the inversion standard deviation, and calculating the sum of the product result and the inversion mean value to be used as an inversion threshold value of the inversion potential.
  8. 8. A method for enhancing semantic interaction of digital employees based on large language models according to claim 1, wherein the method for obtaining the operator-adjusted feature labels comprises: If the homeopathic adjusted intensity integral value at each moment is larger than the homeopathic intensity threshold value and the inversion count of the inversion adjustment is smaller than or equal to the inversion threshold value, setting the operator adjusted characteristic label at the corresponding moment as an expert aggressive mode; if the counter-potential regulating counter-potential counting at each moment is larger than the counter-potential threshold, setting the operator regulating characteristic label at the corresponding moment into a novice panic mode; And if the homeopathic adjusted intensity integral value at each moment is smaller than or equal to the homeopathic intensity threshold value and the inversion count of the inversion adjustment is smaller than or equal to the inversion threshold value, setting the operator adjusted characteristic label at the corresponding moment as a standard mode.
  9. 9. The large language model-based digital employee semantic interaction enhancement method of claim 4, wherein the correlation coefficient obtaining method is pearson correlation coefficient.
  10. 10. A digital employee semantic interaction enhancement medium based on a large language model, the medium comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of a digital employee semantic interaction enhancement method based on a large language model as claimed in any one of claims 1 to 9 when the computer program is executed by the processor.

Description

Digital employee semantic interaction enhancement method and medium based on large language model Technical Field The invention relates to the technical field of semantic processing, in particular to a digital employee semantic interaction enhancement method and medium based on a large language model. Background In industrial fluid control scenarios, operators need to adjust actuators such as valves, pumps, etc. in real time according to process requirements. With the development of large language model LLM technology, industrial sites began to introduce LLM-based digital staff for assisting operators in understanding operating conditions and providing decision advice. In the prior art, the general large language model performs reasoning based on text probability when processing industrial time sequence data, but lacks priori knowledge on dynamic characteristics of a physical system, particularly response lag caused by transmission delay of a fluid pipeline and fluid inertia, so that an auxiliary system outputs alarm or shutdown suggestion which is inconsistent with field facts, normal production scheduling is interfered, and the effectiveness of semantic interaction is poorer. Disclosure of Invention In order to solve the technical problem of lack of priori knowledge of dynamic delay characteristics of a physical system, the worse the effectiveness of semantic interaction, the invention aims to provide a digital employee semantic interaction enhancement method and medium based on a large language model, and the adopted technical scheme is as follows: the invention provides a digital employee semantic interaction enhancement method based on a large language model, which comprises the following steps: acquiring valve opening data and flow measurement data of a fluid control scene at each moment; According to valve opening data and flow measurement data distribution at different moments in a neighborhood range at each moment, flow measurement compensation data after delay compensation at each moment is obtained, and whether an actuator-sensor is communicated or not is judged; if the flow is communicated, obtaining an adjustment-response isotropy product of each moment according to valve opening data of different moments in a local range of each moment and flow measurement compensation data distribution after delay compensation; Acquiring a homeotropic intensity integral value and a homeotropic inverse count of each moment according to the homeotropic intensity integral distribution of different moments in the neighborhood range of each moment, and acquiring a homeotropic intensity threshold value and an inverse inversion threshold value; If the two types of the input language models are not communicated, the operator adjusting feature labels at each moment are set to be in a fault mode, constraint prompt words are constructed based on the operator adjusting feature labels and the communication state of the actuator and the sensor, and the constraint prompt words are input into the large language models to generate interactive feedback information. Further, the method for acquiring the flow measurement compensation data comprises the following steps: Judging whether the valve is in a silence steady state or not according to valve opening data at different moments in a neighborhood range at each moment; If the time is not in the silence steady state, acquiring accumulated values of products between valve opening data at different times in a neighborhood range of each time and flow measurement data corresponding to any preset delay time; and acquiring the flow measurement data after the transmission delay time at each moment in a sequence formed by the flow measurement data at different moments in the neighborhood range of each moment, and taking the flow measurement data after the delay compensation at each moment as flow measurement compensation data. Further, the acquiring method for judging whether the silence steady state exists comprises the following steps: obtaining fluctuation degrees of valve opening data at all times in a neighborhood range of each time, and judging that the valve opening data is not in a silence steady state if a fluctuation degree result is larger than or equal to a preset silence threshold value; And if the fluctuation degree result is smaller than the preset silence threshold, judging that the silence is in a silence steady state. Further, the acquisition method for judging whether the actuator-sensor is communicated comprises the following steps: If the device is in a silence steady state, judging the communication between the actuator and the sensor; If the flow measurement compensation data is not in the silence steady state, obtaining the correlation coefficient of a sequence formed by valve opening data at all times in the neighborhood range of each time and flow measurement compensation data after corresponding delay compensation, and if the correlation coefficient is larger