AI large model-based computing power resource optimization method

Abstract

The invention relates to the technical field of data processing, in particular to a computing power resource optimization method based on an AI large model, and the method comprises the steps: S1, obtaining real-time computing power state data and working state data; s2, pushing the computing power alarm information and adjusting the proportion of the computing power alarm information; s3, performing decision adjustment on the computing power resource optimization decision, and performing decision updating on the decision adjustment process; the method comprises the steps of S1, obtaining a computing power optimization distribution sequence, carrying out decision making on a computing power intelligent optimization method, and carrying out easiness adjustment and easiness updating, and S5, carrying out computing power resource optimization according to the computing power intelligent optimization method, the problems of high-precision calculation and computing power resource waste are balanced, and the computing efficiency of the intelligent robot is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a computing power resource optimization method based on an AI large model. BACKGROUND

[0002] The positioning technology of AGV (Automatic Guided Vehicle) is intelligent from "environment-dependent" to "environment-independent", and the computing power demand continues to increase. From magnetic navigation to SLAM, each leap in AGV positioning accuracy comes at the cost of a surge in computing power demand. Computing power optimization is the key to resolving the contradiction between high precision and low cost, and will further become a watershed in the competitiveness of the AGV industry in the future.

[0003] Chinese Patent Publication No. CN118626224B discloses an AI server cluster computing power scheduling method and system. The method includes selecting computing power analysis indicators of the AI server cluster, using the computing power analysis indicators to measure the computing power resources in the AI server cluster, calculating the transmission delay and transmission energy consumption of the user's pending tasks from the user to the computing power resources, calculating the processing delay and processing energy consumption of the computing power resources processing the pending tasks, calculating the first computing power scheduling cost of the pending task scheduling the computing power resources, calculating the second computing power scheduling cost of the computing power resources scheduling the cooperative computing power resources, constructing the action value function and state value function of the pending task under the computing power scheduling state and computing power scheduling action, and optimizing the action value function and state value function to obtain the optimized action value function and optimized state value function. However, this invention cannot be applied to autonomous navigation, obstacle avoidance and multi-cooperative intelligent robots represented by AGV (Automatic Guided Vehicle), and cannot be adapted to optimize the computing power resource allocation and consumption of the AI large model in the intelligent robot. SUMMARY

[0004] Therefore, the present application provides a computing power resource optimization method based on an AI large model to overcome the problem of waste of computing power resources and low computing efficiency of intelligent robots caused by the inability to balance the optimization of computing power resource allocation and consumption of AI large models in intelligent robots and high-precision computing requirements in the prior art.

[0005] To achieve the above-mentioned purpose, the present application provides a computing power resource optimization method based on an AI large model, comprising: Step S1, collecting real-time computing power state data and working state data; Step S2, obtaining the current actual computing power proportion according to the real-time computing power state data, pushing the computing power alarm information according to the current actual computing power proportion, and adjusting the proportion of the current actual computing power proportion according to the predicted computing power proportion; Step S3, according to the current actual computing power ratio, the computing power resource optimization decision is obtained, and the computing power resource optimization decision is adjusted according to the pre-computing power influence degree, and the process of decision adjustment is updated according to the computing power data important coefficient; Step S4, according to the computing power data important coefficient, the computing power optimization distribution sequence is obtained, and the computing power intelligent optimization method is output according to the computing power optimization distribution sequence, the computing power optimization distribution sequence is adjusted according to the optimization easiness, and the easiness of the process of easiness adjustment is updated according to the regional accuracy; Step S5, according to the computing power intelligent optimization method, the computing power resource allocation of the warehouse AGV automatic guided vehicle robot is optimized.

[0006] Further, in the step S1, the real-time computing power state data and the working state data are collected, and the real-time computing power state data is stored in the computing power monitoring database; In the step S2, when the current actual computing power ratio is obtained according to the real-time computing power state data, and the computing power alarm information is pushed according to the current actual computing power ratio, the current actual computing power ratio sjp is calculated according to the current actual computing power sj and the total computing power si, and sjp=sj / si is set, the current actual computing power ratio sjp is obtained, the current actual computing power ratio sjp is compared with the first preset current actual computing power ratio sjp1 and the second preset current actual computing power ratio sjp2, the state of the current actual computing power ratio is judged according to the comparison result, and the computing power alarm information is pushed according to the judgment result, wherein: When sjp≤sjp1, it is determined that the state of the current actual computing power ratio is low ratio, and the low ratio state is pushed as the computing power alarm information; When sjp1<sjp≤sjp2, it is determined that the state of the current actual computing power ratio is medium ratio, and the medium ratio state is pushed as the computing power alarm information; When sjp>sjp2, it is determined that the state of the current actual computing power ratio is high ratio, and the high ratio state is pushed as the computing power alarm information.

[0007] Further, in the step S2, when the current actual computing power ratio is adjusted according to the predicted computing power ratio, the predicted computing power ratio model is constructed according to the historical computing power state data, and the predicted computing power ratio model is obtained, the current actual computing power ratio, the computing power utilization rate, the computing power response time and the computing power density are input into the predicted computing power ratio model, and the predicted computing power ratio output by the predicted computing power ratio model is obtained; In the step S2, when the predicted computing power proportion ycis compared with the preset predicted computing power proportion yc0, the compliance of the predicted computing power proportion is judged according to the comparison result, and the current actual computing power sj is adjusted according to the judgment result, wherein: When yc≤yc0, it is determined that the compliance of the predicted computing power proportion is up to the standard, and the current actual computing power sj is not adjusted; When yc>yc0, it is determined that the compliance of the predicted computing power proportion is not up to the standard, and the current actual computing power sj is adjusted. The current actual computing power sj is adjusted according to the adjustment coefficient tz, and tz=1.36-0.22×e is set, where e is the base of natural logarithm, and the adjusted current actual computing power sj` is obtained. Set sj`=sj×tz, replace the current actual computing power sj with the adjusted current actual computing power sj`, and recalculate the current actual computing power proportion sjp according to the current actual computing power sj and the total computing power si. -(yc-yc0)

[0008] Further, in the step S3, when the computing power resource optimization decision is obtained according to the current actual computing power proportion by the computing power resource optimization decision obtaining method, the computing power resource optimization decision obtaining method comprises: Step S41, the adjustable proportion space kt is calculated according to the current actual computing power proportion sjp and the total computing power si, and kt=(si-sjp)×100% is set, and the adjustable proportion space kt is obtained; Step S42, the adjustable proportion space kt is compared with the preset adjustable proportion space kt0, the state of the adjustable proportion space is judged according to the comparison result, and the computing power resource optimization decision is output according to the judgment result, wherein: When kt≥kt0, it is determined that the state of the adjustable proportion space is sufficient, and the computing power resource optimization is not performed as the computing power resource optimization decision is output; When kt<kt0, it is determined that the state of the adjustable proportion space is not sufficient, and the computing power resource optimization is performed as the computing power resource optimization decision is output.

[0009] Further, in the step S3, when the computing power resource optimization decision is adjusted according to the pre-computing power influence degree, the pre-computing power influence degree is obtained according to the real-time computing power state data by the pre-computing power influence degree obtaining method, and the pre-computing power influence degree obtaining method comprises: ​Step S51, input the computing power utilization rate influence value, the computing power response time influence value and the computing power density influence value as the first target influence value into the computing power influence prediction model to obtain the predicted computing power utilization rate influence value ap1, the predicted computing power response time influence value ap2 and the predicted computing power density influence value ap3 output by the computing power influence prediction model; Step S52, input the predicted computing power utilization rate influence value ap1, the predicted computing power response time influence value ap2 and the predicted computing power density influence value ap3 as the second target influence value; Step S53, calculate the pre-computing power influence degree ysl according to the second target influence value, the predicted computing power utilization rate influence value weight A1, the predicted computing power response time influence value weight A2 and the predicted computing power density influence value weight A3, set ysl=ap1×A1+ap2×A2+ap3×A3, and obtain the pre-computing power influence degree ysl.

[0010] Further, in the step S3, when the pre-computing power influence degree ysl is compared with the preset computing power influence degree ysl0, the compliance of the pre-computing power influence degree is judged according to the comparison result, and the preset adjustable allocation proportion space kt0 is adjusted according to the judgment result, wherein: When ysl≤ysl0, it is determined that the compliance of the pre-computing power influence degree is up to standard, and the preset adjustable allocation proportion space kt0 is not adjusted; When ysl>ysl0, it is determined that the compliance of the pre-computing power influence degree is not up to standard, and the preset adjustable allocation proportion space kt0 is adjusted, the preset adjustable allocation proportion space kt0 is adjusted according to the decision adjustment coefficient jc, set jc=1.42-0.23×e -(ysl-ysl0) , obtain the adjusted preset adjustable allocation proportion space kt0`, set kt0`=kt0×jc, replace the preset adjustable allocation proportion space kt0 with the adjusted preset adjustable allocation proportion space kt0`, and compare the adjustable allocation proportion space kt with the preset adjustable allocation proportion space kt0 again.

[0011] Further, in the step S3, when the process of decision adjustment is updated according to the computing power data importance coefficient, the computing power data importance coefficient E is calculated according to the calling frequency d1, the accuracy requirement d2, the average error d3, the error frequency d4, the calling frequency weight β1, the accuracy requirement weight β2, the average error weight β3 and the error frequency weight β4, set E=β1×d1+β2×d2+β3×d3+β4×d4, and obtain the computing power data importance coefficient E. In the step S3, when updating the decision of the process of decision adjustment according to the computing power data importance coefficient, the computing power data importance coefficient E is compared with a preset computing power data importance coefficient E0, the attribute of the computing power data importance coefficient is judged according to the comparison result, and the preset computing power influence degree ysl0 is updated according to the judgment result, wherein: When E≤E0, the step S3 judges that the attribute of the computing power data importance coefficient is unimportant, and the preset computing power influence degree ysl0 is not updated; When E>E0, the step S3 judges that the attribute of the computing power data importance coefficient is important, and the preset computing power influence degree ysl0 is updated, the preset computing power influence degree ysl0 is updated according to the decision update coefficient gx, gx=1.46+0.23×e -(E-E0) , the updated preset computing power influence degree ysl0` is obtained, ysl0`=ysl0×gx, the preset computing power influence degree ysl0 is replaced by the updated preset computing power influence degree ysl0`, and the preset computing power influence degree ysl0 is compared with the preset computing power influence degree ysl0 again.

[0012] Further, in the step S4, when obtaining the computing power optimization distribution order according to the computing power data importance coefficient, the computing power important coefficient set is obtained according to the computing power data importance coefficient, the computing power important coefficient set is sorted according to the importance, and the computing power optimization distribution order is obtained; In the step S4, when deciding the computing power intelligent optimization method according to the computing power optimization distribution order, the computing power optimization distribution order is input into the computing power intelligent optimization method decision tree, and the computing power intelligent optimization method is obtained; In the step S4, when adjusting the ease degree of the computing power optimization distribution order according to the optimization ease degree, the optimization ease degree is obtained according to the working state data by an optimization ease degree obtaining method, and the optimization ease degree obtaining method comprises: Step S81, input the point cloud region into the point cloud feature conversion model to obtain the point cloud feature value, and input the shared positioning state into the shared feature conversion model to obtain the shared positioning state feature value; Step S82, normalize the point cloud feature value and the shared positioning state feature value to obtain the normalized point cloud feature value dc and the normalized shared positioning state feature value gz; Step S83, calculate the optimization ease degree yd according to the normalized point cloud feature value dc, the normalized shared positioning state feature value gz, the normalized point cloud feature value weight α and the normalized shared positioning state feature value weight β, set yd=α×(1-dc)+β×gz, and obtain the optimization ease degree yd.

[0013] Further, in the step S4, when the optimization easiness yd is compared with the preset optimization easiness yd0, the state of the optimization easiness is judged according to the comparison result, and the importance coefficient E of the computing power data is adjusted according to the judgment result, wherein: When yd≤yd0, it is determined that the state of the optimization easiness is not easy to optimize, and the importance coefficient E of the computing power data is not adjusted; When yd>yd0, it is determined that the state of the optimization easiness is easy to optimize, and the importance coefficient E of the computing power data is adjusted, the importance coefficient E of the computing power data is adjusted according to the easiness adjustment coefficient, yt=1.33-0.22×e -(yd-yd0) is set, the importance coefficient E of the computing power data is replaced by the adjusted importance coefficient E` of the computing power data, the importance coefficient set of the computing power is reacquired according to the importance coefficient E of the computing power data, and the importance coefficient set of the computing power is re-sequenced according to the importance, and the computing power optimization distribution sequence is obtained.

[0014] Further, in the step S4, when the optimization easiness yd is compared with the preset optimization easiness yd0, the state of the optimization easiness is judged according to the comparison result, and the importance coefficient E of the computing power data is adjusted according to the judgment result, wherein: When gj≤gj0, it is determined that the state of the region accuracy is low, and the preset optimization easiness yd0 is not updated; When gj>gj0, it is determined that the state of the region accuracy is high, and the preset optimization easiness yd0 is updated, the preset optimization easiness yd0 is updated according to the easiness update coefficient yx, yx=1.35+0.28×e -(gj-gj0) is set, the importance coefficient E of the computing power data is replaced by the adjusted importance coefficient E` of the computing power data, the importance coefficient set of the computing power is reacquired according to the importance coefficient E of the computing power data, and the importance coefficient set of the computing power is re-sequenced according to the importance, and the computing power optimization distribution sequence is obtained. In the step S5, the computing power resource of the warehouse AGV automatic guided vehicle robot is optimized according to the computing power intelligent optimization method.

[0015] Compared with the prior art, the method has the beneficial effects that, by step S1, the real-time computing power state data and the working state data are collected, so as to subsequently perform computing power intelligent optimization according to the real-time computing power state data and the working state data, by step S2, the computing power alarm information is pushed and the proportion is adjusted, so as to monitor the use of the current actual computing power in real time and timely push the computing power alarm information, thereby improving the computing power resource optimization efficiency, by step S3, the computing power resource optimization decision is adjusted and the process of the decision adjustment is updated, so as to reasonably reduce the influence of the pre-computing power and the computing power data important coefficient on the computing power resource optimization decision, thereby improving the accuracy of optimizing the computing power resource distribution, by step S4, the computing power optimization distribution sequence is obtained and the computing power intelligent optimization method is decided, and the easiness is adjusted and updated, so as to output the computing power intelligent optimization method according to the computing power importance degree of the unit task and reduce the influence of the optimization easiness on the computing power optimization distribution sequence, thereby improving the efficiency of optimizing the computing power resource, and by step S5, the computing power resource is optimized according to the computing power intelligent optimization method, so as to implement and execute the decided computing power intelligent optimization method, thereby improving the efficiency of the robot computing power resource optimization. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 FIG. 1 is a flowchart of the AI big model-based computing power resource optimization method of the present embodiment. DETAILED DESCRIPTION

[0017] In order to make the objectives and advantages of the present application clearer, the present application will be further described below with examples; it should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0018] The preferred embodiments of the present application will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present application and do not limit the protection scope of the present application.

[0019] Please refer to Figure 1 FIG. 1 is a flowchart of the AI big model-based computing power resource optimization method of the present embodiment, the method comprises: Step S1, collecting real-time computing power state data and working state data; Step S2, obtaining the current actual computing power proportion according to the real-time computing power state data, pushing the computing power alarm information according to the current actual computing power proportion, and adjusting the proportion of the current actual computing power proportion according to the predicted computing power proportion; Step S3, obtain the computing power resource optimization decision according to the current actual computing power proportion, and adjust the computing power resource optimization decision according to the pre-computing power influence degree, and update the decision adjustment process according to the computing power data importance coefficient; Step S4, obtain the computing power optimization distribution sequence according to the computing power data importance coefficient, output the computing power intelligent optimization method according to the computing power optimization distribution sequence, adjust the computing power optimization distribution sequence according to the optimization easiness, and update the easiness adjustment process according to the regional accuracy; Step S5, optimize the computing power resource according to the computing power intelligent optimization method.

[0020] Specifically, the method is applied to the intelligent control cloud of autonomous navigation, obstacle avoidance and multi-cooperation of intelligent robots represented by AGV, and the computing power resource allocation of AI large model in the intelligent robot is optimized to balance the problem of high-precision calculation and computing power resource waste, thereby improving the computing efficiency of the intelligent robot. The method collects real-time computing power state data and working state data through step S1, so as to facilitate subsequent computing power intelligent optimization according to real-time computing power state data and working state data. The method also pushes the computing power alarm information and adjusts the proportion through step S2, so as to monitor the use of the current actual computing power in real time and timely push the computing power alarm information, thereby improving the computing power resource optimization efficiency. The method also adjusts the computing power resource optimization decision and updates the decision adjustment process through step S3, so as to reasonably reduce the influence of pre-computing power influence degree and computing power data importance coefficient on computing power resource optimization decision, thereby improving the accuracy of optimizing computing power resource allocation. The method also obtains the computing power optimization distribution sequence and makes decisions on the computing power intelligent optimization method through step S4, and adjusts and updates the easiness, so as to output the computing power intelligent optimization method according to the importance of the unit task, and reduce the influence of optimization easiness on the computing power optimization distribution sequence, thereby improving the efficiency of optimizing computing power resources. The method also optimizes the computing power resource according to the computing power intelligent optimization method through step S5, so as to implement the decision of the computing power intelligent optimization method, thereby improving the efficiency of robot computing power resource optimization.

[0021] Specifically, in the step S1, the real-time computing power state data and working state data are collected, and the real-time computing power state data is stored in the computing power monitoring database.

[0022] In particular, the real-time computing power state data includes current actual computing power, total computing power, computing power utilization rate influence value, computing power response time influence value, and computing power density influence value, the working state data includes point cloud area and shared positioning state, the current actual computing power refers to the computing power required by the warehouse AGV robot to execute the task request at the current moment, the total computing power refers to all the computing power that can be called when the robot executes the task request, and the specific acquisition method of the current actual computing power and the total computing power is not limited in the embodiment, and the related personnel in the field can freely choose according to the actual demand, such as directly measuring the computing power by the central processing unit, the computing power utilization rate influence value refers to the value for measuring the influence degree of the actual use efficiency of the computing power resource on the system performance, the computing power response time influence value refers to the value for measuring the influence degree of the time interval from the initiation of the task request to the actual processing of the task request by the computing power resource on the system performance, the computing power density influence value refers to the value for measuring the influence degree of the computing power in a unit physical space on the system performance, the unit physical space refers to the geometric measurement of a specific device unit, such as the cabinet and chip packaging area per square meter, and the specific acquisition method of the computing power utilization rate influence value, the computing power response time influence value and the computing power density influence value is not limited in the embodiment, and the related personnel in the field can freely choose according to the actual demand, such as obtaining by expert evaluation, the expert evaluation refers to the way that the expert who has the setting ability of the computing power utilization rate influence value, the computing power response time influence value and the computing power density influence value sets the computing power utilization rate influence value, the computing power response time influence value and the computing power density influence value, and the specific acquisition method of the expert evaluation is not limited in the embodiment, such as obtaining the computing power utilization rate influence value, the computing power response time influence value and the computing power density influence value input by the expert in the terminal interaction window through the cloud, the point cloud area refers to the area composed of discrete points in the visual space obtained by the warehouse AGV robot, and the specific acquisition method of the point cloud area is not limited in the embodiment, and the related personnel in the field can freely choose according to the actual demand, such as obtaining the point cloud area by laser radar, the shared positioning state refers to the state that the warehouse AGV robot shares the respective positioning information within a preset range when working cooperatively, and the shared positioning state includes self pose, navigation state, environment data and task state, the self pose refers to the driving pose of the warehouse AGV robot at the current moment, such as the current position coordinates, the heading angle and the driving speed, the navigation state refers to the navigation mode of the warehouse AGV robot at the current moment, such as currently using the two-dimensional code guiding mode, the environment data refers to the change of obstacles in the visual space obtained by the warehouse AGV robot at the current moment, such as temporary shelf position change, and the task state refers to the task condition of the warehouse AGV robot at the current moment, such as currently being in the charging state.The preset range refers to a circular range with the position of the warehouse AGV robot as the center, such as a circular range with d as the radius, and d=10m is set. The computing power monitoring database refers to a database for storing real-time computing power state data. The present embodiment does not limit the computing power monitoring database, and a person skilled in the art can freely select according to actual needs, such as constructing a data storage library in the cloud and taking the data storage library as the computing power monitoring database.

[0023] Specifically, in the step S1, the real-time computing power state data and the working state data are collected, so that the computing power is intelligently optimized according to the real-time computing power state data and the working state data, thereby improving the efficiency of computing power resource optimization.

[0024] Specifically, in the step S2, the current actual computing power ratio is obtained according to the real-time computing power state data, and when the computing power alarm information is pushed according to the current actual computing power ratio, the current actual computing power ratio sjp is calculated according to the current actual computing power sj and the total computing power si, and sjp=sj / si is set to obtain the current actual computing power ratio sjp. The current actual computing power ratio sjp is compared with the first preset current actual computing power ratio sjp1 and the second preset current actual computing power ratio sjp2, the state of the current actual computing power ratio is judged according to the comparison result, and the computing power alarm information is pushed according to the judgment result, wherein: When sjp≤sjp1, it is determined that the state of the current actual computing power ratio is low ratio, and the low ratio state is pushed as the computing power alarm information; When sjp1<sjp≤sjp2, it is determined that the state of the current actual computing power ratio is medium ratio, and the medium ratio state is pushed as the computing power alarm information; When sjp>sjp2, it is determined that the state of the current actual computing power ratio is high ratio, and the high ratio state is pushed as the computing power alarm information.

[0025] Specifically, the first preset current actual computing power proportion refers to a preset lower limit value for judging the state of the current actual computing power proportion, and the second preset current actual computing power proportion refers to a preset upper limit value for judging the state of the current actual computing power proportion. The first preset current actual computing power proportion and the second preset current actual computing power proportion are not limited in the embodiment, and a person skilled in the art can freely select them according to actual needs. For example, the first preset current actual computing power proportion sjp1 = 0.60 and the second preset current actual computing power proportion sjp2 = 0.75 are set in the embodiment. The state of the current actual computing power proportion refers to the degree of the current actual computing power occupying the total computing power. The state of the current actual computing power proportion includes low proportion, medium proportion, and high proportion. The specific method of pushing the computing power warning information is not limited in the embodiment, and a person skilled in the art can freely select it according to actual needs. For example, the computing power warning information is pushed to a mobile terminal device of a person skilled in the art.

[0026] Specifically, in the step S2, the current actual computing power proportion is calculated, and the state of the current actual computing power proportion is judged, so as to monitor the use of the current actual computing power in real time and timely push the computing power warning information, thereby improving the computing power resource optimization efficiency.

[0027] Specifically, in the step S2, when the current actual computing power proportion is adjusted according to the predicted computing power proportion, the predicted computing power proportion model is constructed according to the historical computing power state data, the current actual computing power proportion, the computing power utilization rate, the computing power response time, and the computing power density are input into the predicted computing power proportion model, and the predicted computing power proportion output by the predicted computing power proportion model is obtained.

[0028] Specifically, the historical computing power state data refers to historical occurred computing power state data and corresponding predicted computing power proportion obtained from the computing power monitoring database, the historical occurred computing power state data includes actual computing power proportion, computing power utilization rate, computing power response time and computing power density, the predicted computing power proportion model refers to a convolutional neural network model taking current actual computing power proportion, computing power utilization rate, computing power response time and computing power density as input data and taking predicted computing power proportion as output data, the computing power utilization rate refers to the proportion of the theoretical maximum computing capacity effectively used at the current time, the computing power response time refers to the time delay between task request submission and start of execution, the computing power density refers to the number of completed task requests per unit time, and the predicted computing power proportion refers to the current actual computing power reached in the future time step obtained according to the predicted computing power proportion model. The present embodiment does not limit the future time step, for example, the future time step is set to 3 seconds in the present embodiment. The present embodiment does not limit the specific construction method of the predicted computing power proportion model, and the related person skilled in the art can freely set it according to the actual needs, for example, the historical computing power state data is taken as a training set to train the predicted computing power proportion model to obtain the predicted computing power proportion model.

[0029] Specifically, in the step S2, the current actual computing power at the future time is predicted by constructing the predicted computing power proportion model, so as to timely optimize the computing power and save computing power resources.

[0030] Specifically, in the step S2, when adjusting the current actual computing power proportion according to the predicted computing power proportion, the predicted computing power proportion yc is compared with the preset predicted computing power proportion yc0, the compliance of the predicted computing power proportion is judged according to the comparison result, and the current actual computing power sj is adjusted according to the judgment result, wherein: When yc≤yc0, it is determined that the compliance of the predicted computing power proportion is up to standard, and the current actual computing power sj is not adjusted; When yc>yc0, it is determined that the compliance of the predicted computing power proportion is not up to standard, and the current actual computing power sj is adjusted, the current actual computing power sj is adjusted according to the adjustment coefficient tz, and tz=1.36-0.22×e is set, where e is the base of natural logarithm, the adjusted current actual computing power sj` is obtained, sj` is set as sj×tz, the current actual computing power sj is replaced by the adjusted current actual computing power sj`, and the current actual computing power proportion sjp is recalculated according to the current actual computing power sj and the total computing power si. -(yc-yc0)

[0031] ​Specifically, the preset prediction computing power ratio refers to a preset value for judging the standard reaching condition of the prediction computing power ratio. The specific value of the preset prediction computing power ratio is not limited in the embodiment, and can be freely selected by a person skilled in the art according to actual needs. For example, the preset prediction computing power ratio yc0 is set to 65% in the embodiment. The standard reaching condition of the prediction computing power ratio refers to the standard reaching degree of the prediction computing power ratio judged according to the prediction computing power ratio and the preset prediction computing power ratio. The standard reaching condition of the prediction computing power ratio includes reaching the standard and not reaching the standard.

[0032] Specifically, in the step S2, when the prediction computing power ratio does not reach the standard, the current actual computing power is quickly increased according to the adjustment coefficient to timely adjust the current actual computing power ratio. When the prediction computing power ratio reaches the late stage of not reaching the standard, the increasing trend of the current actual computing power tends to be stable. The adjustment coefficient is set to match this change trend. The constant term 1.36 in the adjustment coefficient refers to the maximum value that the adjustment coefficient can reach. The constant coefficient 0.22 refers to the change rate of the adjustment coefficient. The change trend of the adjustment coefficient increases from 1.14 to 1.36, so as to reasonably avoid the influence of the prediction computing power ratio on the current actual computing power ratio, thereby improving the computing power resource optimization efficiency of the AI large model in the intelligent warehouse AGV robot.

[0033] Specifically, in the step S3, when the computing power resource optimization decision is obtained by the computing power resource optimization decision obtaining method according to the current actual computing power ratio, the computing power resource optimization decision obtaining method includes: In step S41, the adjustable proportion space kt is calculated according to the current actual computing power ratio sjp and the total computing power si, and kt=(si-sjp)×100% is set to obtain the adjustable proportion space kt. In step S42, the adjustable proportion space kt is compared with the preset adjustable proportion space kt0. The state of the adjustable proportion space is judged according to the comparison result, and the computing power resource optimization decision is output according to the judgment result, wherein: When kt≥kt0, it is determined that the state of the adjustable proportion space is sufficient, and the computing power resource optimization is not performed as the computing power resource optimization decision. When kt<kt0, it is determined that the state of the adjustable proportion space is not sufficient, and the computing power resource optimization is performed as the computing power resource optimization decision.

[0034] Specifically, the preset adjustable proportion space refers to a preset value for judging the state of the adjustable proportion space. The present embodiment does not limit the preset adjustable proportion space, and a person skilled in the art can freely select it according to actual needs. For example, the present embodiment sets the preset adjustable proportion space kt0=25%. The state of the adjustable proportion space refers to the sufficiency of the adjustable proportion space. The state of the adjustable proportion space includes insufficient and sufficient.

[0035] Specifically, in the step S3, the state of the adjustable proportion space is judged to determine whether to optimize the computing resource according to the adjustable proportion space in different states, so as to save the computing resource and improve the sustainability and optimization efficiency of the computing resource allocation.

[0036] Specifically, in the step S3, when the pre-computing power influence degree is used to adjust the computing resource optimization decision, the pre-computing power influence degree is obtained from real-time computing power state data by a pre-computing power influence degree obtaining method. The pre-computing power influence degree obtaining method includes: Step S51, the computing power utilization rate influence value, the computing power response time influence value and the computing power density influence value are input into the computing power influence prediction model as the first target influence value, and the predicted computing power utilization rate influence value ap1, the predicted computing power response time influence value ap2 and the predicted computing power density influence value ap3 output by the computing power influence prediction model are obtained; Step S52, the predicted computing power utilization rate influence value ap1, the predicted computing power response time influence value ap2 and the predicted computing power density influence value ap3 are input as the second target influence value; Step S53, the pre-computing power influence degree ysl is calculated according to the second target influence value, the predicted computing power utilization rate influence value weight A1, the predicted computing power response time influence value weight A2 and the predicted computing power density influence value weight A3, and ysl=ap1xA1+ap2xA2+ap3xA3 is set to obtain the pre-computing power influence degree ysl.

[0037] Specifically, the computing power influence prediction model refers to a convolutional neural network model taking the first target influence value as input data and the second target influence value as output data. The specific construction method of the computing power influence prediction model is not limited in the embodiment, and a person skilled in the art can freely select according to actual needs. For example, the historical first target influence value and the corresponding second target influence value are taken as a training set to train the convolutional neural network model to obtain the computing power influence prediction model. The predicted computing power utilization influence value refers to a value obtained according to the computing power influence prediction model, which measures the influence degree of the actual use efficiency of the current computing power resource on the system performance in the future time step. The predicted computing power response time influence value refers to a value obtained according to the computing power influence prediction model, which measures the influence degree of the time interval from the initiation of the task request to the actual processing of the task request by the computing power resource on the system performance in the future time step. The predicted computing power density influence value refers to a value obtained according to the computing power influence prediction model, which measures the influence degree of the current computing power in a unit physical space on the system performance in the future time step. The predicted computing power utilization influence value weight refers to a value that measures the importance of the predicted computing power utilization influence value in the pre-computing power influence degree. The predicted computing power response time influence value weight refers to a value that measures the importance of the predicted computing power response time influence value in the pre-computing power influence degree. The predicted computing power density influence value weight refers to a value that measures the importance of the predicted computing power density influence value in the pre-computing power influence degree. The embodiment does not limit the predicted computing power utilization influence value weight A1, the predicted computing power response time influence value weight A2, and the predicted computing power density influence value weight A3. A person skilled in the art can freely select according to actual needs, as long as the requirement of A1+A2+A3=1 is met. For example, the embodiment sets A1=0.4, A2=0.3, and A3=0.3.

[0038] Specifically, in the step S3, the pre-computing power influence degree is obtained to adjust the computing power resource optimization decision according to the pre-computing power influence degree, predict the computing power influence in real time, and improve the efficiency of computing power resource deployment.

[0039] Specifically, in the step S3, when the pre-computing power influence degree is adjusted according to the pre-computing power influence degree, the pre-computing power influence degree ysl is compared with the preset computing power influence degree ysl0. The compliance of the pre-computing power influence degree is judged according to the comparison result, and the preset adjustable deployment proportion space kt0 is adjusted according to the judgment result, wherein: When ysl≤ysl0, it is determined that the pre-computing power influence degree meets the standard, and the preset adjustable deployment proportion space kt0 is not adjusted; When ysl> ysl0, it is determined that the pre-computed force influence degree does not meet the standard, the preset adjustable proportion space kt0 is adjusted, the preset adjustable proportion space kt0 is adjusted according to the decision adjustment coefficient jc, jc is set as 1.42-0.23x e -(ysl-ysl0) , the adjusted preset adjustable proportion space kt0` is obtained, kt0` is set as kt0x jc, the preset adjustable proportion space kt0 is replaced by the adjusted preset adjustable proportion space kt0`, and the adjustable proportion space kt is compared with the preset adjustable proportion space kt0 again.

[0040] Specifically, the preset computing power influence degree refers to a preset value for judging the pre-computed force influence degree, and the specific value of the preset computing power influence degree is not limited in the embodiment, and can be freely selected by a person skilled in the art according to actual needs. For example, the preset computing power influence degree ysl0=0.75 is set in the embodiment. The pre-computed force influence degree meeting the standard refers to the pre-computed force influence degree meeting the standard according to the pre-computed force influence degree and the preset computing power influence degree. The pre-computed force influence degree meeting the standard includes meeting the standard and not meeting the standard.

[0041] Specifically, in the step S3, when the pre-computed force influence degree does not meet the standard, the preset adjustable proportion space is increased quickly by the decision adjustment coefficient, so as to timely adjust the decision according to the pre-computed force influence degree. When the pre-computed force influence degree remains in the state of not meeting the standard in the later period, the trend of increasing the preset adjustable proportion space tends to be stable. The decision adjustment coefficient is set to match this change trend. The constant term 1.42 in the decision adjustment coefficient represents the maximum value that the decision adjustment coefficient can reach, and the constant coefficient 0.23 represents the change rate of the decision adjustment coefficient, so that the change trend of the decision adjustment coefficient is from 1.19 to 1.42, so as to reasonably reduce the influence of the pre-computed force influence degree on the optimization decision of the computing power resource, thereby improving the efficiency of optimizing the allocation of the computing power resource.

[0042] Specifically, in the step S3, when the decision adjustment process is updated according to the computing power data importance coefficient, the computing power data importance coefficient E is calculated according to the calling frequency d1, the accuracy requirement d2, the average error d3, the error frequency d4, the calling frequency weight β1, the accuracy requirement weight β2, the average error weight β3 and the error frequency weight β4. E is set as β1xd1+β2xd2+β3xd3+β4xd4, and the computing power data importance coefficient E is obtained.

[0043] Specifically, the number of calls refers to the number of calls of a single unit to the working state data in the warehouse AGV robot operation task request. The embodiment does not limit the specific acquisition method of the number of calls, and a person skilled in the art can freely choose according to actual needs, such as acquiring through local logs. The precision requirement refers to the data precision requirement of a single unit to the working state data in the warehouse AGV robot operation task request. The embodiment does not limit the specific acquisition method of the precision requirement, and a person skilled in the art can freely choose according to actual needs, such as acquiring the precision requirement according to the industrial vehicle safety standard, such as positioning error ≤ 30 mm. The average error refers to a value that measures the degree of deviation between the calculation result and the true value in the warehouse AGV robot operation task request. For example, if the true value is ag, the calculation result is pg, and the calculation times are cs=56, then the average error = |(ag-pg) / cs|. The error times refer to the cumulative number of task operation failures, result errors, and process interruptions caused by abnormal working state data in the warehouse AGV robot operation task request. The embodiment does not limit the specific acquisition method of the error times, and a person skilled in the art can freely choose according to actual needs, such as acquiring the error times through the native monitoring of the warehouse AGV robot operating system. The number of call weights refers to a value that measures the importance of the number of calls in the important coefficient of the computing power data. The precision requirement weight refers to a value that measures the importance of the precision requirement in the important coefficient of the computing power data. The average error weight refers to a value that measures the importance of the average error in the important coefficient of the computing power data. The error times weight refers to a value that measures the importance of the error times in the important coefficient of the computing power data. The embodiment does not limit the number of call weights β1, the precision requirement weight β2, the average error weight β3, and the error times weight β4, and a person skilled in the art can freely choose according to actual needs, as long as β1+β2+β3+β4=1 is met. For example, if β1=0.23, β2=0.16, β3=0.33, and β4=0.28 are set.

[0044] Specifically, in the step S3, the important degree of the working state data is quantified by acquiring the computing power data important coefficient, which facilitates subsequent allocation of computing power resources based on the computing power data important coefficient, thereby improving the efficiency of computing power resource optimization.

[0045] Specifically, in the step S3, when updating the decision-making of the process according to the computing power data important coefficient, the computing power data important coefficient E is compared with the preset computing power data important coefficient E0. The attribute of the computing power data important coefficient is judged according to the comparison result, and the preset computing power influence degree ysl0 is updated according to the judgment result, wherein: When E≤E0, it is determined that the attribute of the computing power data important coefficient is unimportant, and the preset computing power influence degree ysl0 is not updated; When E>E0, it is determined that the attribute of the computing power data important coefficient is important, the preset computing power influence degree ysl0 is updated, the preset computing power influence degree ysl0 is updated according to the decision update coefficient gx, and gx=1.46+0.23×e -(E-E0) , the updated preset computing power influence degree ysl0` is obtained, ysl0`=ysl0×gx is set, the preset computing power influence degree ysl0 is replaced by the updated preset computing power influence degree ysl0`, and the preset computing power influence degree ysl0 is compared with the preset computing power influence degree ysl0 again.

[0046] Specifically, the preset computing power data important coefficient refers to a preset value for judging the attribute of the computing power data important coefficient, and the preset computing power data important coefficient is not limited in the embodiment, and can be freely selected by a person skilled in the art according to actual needs. For example, the preset computing power data important coefficient E0=0.82 is set in the embodiment. The attribute of the computing power data important coefficient refers to the importance of the computing power data important coefficient in the warehouse AGV robot operation task request. The attribute of the computing power data important coefficient includes unimportant and important.

[0047] Specifically, in the step S3, when the attribute of the computing power data important coefficient is important, the preset computing power influence degree is quickly reduced by the decision update coefficient, and the decision updating process is updated in time. When the attribute of the computing power data important coefficient reaches the late stage of the important state, the trend of reducing the preset computing power influence degree tends to be stable, and the decision update coefficient is used to match this change trend. The constant term 1.46 in the decision update coefficient is the maximum value that the decision update coefficient can reach, and the constant coefficient 0.23 is the change rate of the decision update coefficient. The change trend of the decision update coefficient increases from 1.23 to 1.46, so as to reasonably reduce the influence of the computing power data important coefficient on the decision adjustment process, thereby improving the accuracy of allocating computing power resources.

[0048] Specifically, in the step S4, when the computing power data important coefficient is used to obtain the computing power optimization distribution order, the computing power important coefficient set is obtained according to the computing power data important coefficient, the computing power important coefficient set is obtained, the important degree of the computing power important coefficient set is sorted, and the computing power optimization distribution order is obtained.

[0049] Specifically, the set of computing power importance coefficients refers to a data set composed of computing power data importance coefficients of all units running when the warehouse AGV robot runs a request task. When a task request is executed, all units running are obtained through a central processor, computing power data importance coefficients of all units are calculated according to step S3, and the computing power data importance coefficients of all units are taken as the set of computing power importance coefficients. The importance degree sorting refers to a process of arranging computing power data importance coefficients in the set of computing power importance coefficients in descending order according to the size of the numerical value to obtain a computing power optimization distribution order. The embodiment does not limit the specific implementation manner of the importance degree sorting, and a person skilled in the art can freely select according to actual needs, such as sorting the importance degree by Python.

[0050] Specifically, in step S4, the computing power optimization distribution order is obtained to output the computing power intelligent optimization method according to the computing power importance degree of the unit task, thereby improving the efficiency of computing power resource optimization.

[0051] Specifically, in step S4, when the computing power intelligent optimization method is decided according to the computing power optimization distribution order, the computing power optimization distribution order is input into the computing power intelligent optimization method decision tree to obtain the computing power intelligent optimization method.

[0052] Specifically, the computing power intelligent optimization method decision tree refers to a decision tree model taking the computing power optimization distribution order as input and the computing power intelligent optimization method as output. The embodiment does not limit the specific construction manner of the computing power intelligent optimization method decision tree, and a person skilled in the art can freely select according to actual needs, such as constructing the intelligent optimization method decision tree by taking the predicted computing power proportion, the pre-computing power influence degree and the computing power intelligent optimization method as the execution path of the intelligent optimization method decision tree.

[0053] Specifically, in step S4, the computing power intelligent optimization method is decided by constructing the computing power intelligent optimization method decision tree to preferentially optimize the computing power of important task units, thereby improving the efficiency of computing power optimization.

[0054] Specifically, in step S4, when the ease of optimization is adjusted according to the optimization ease of the computing power optimization distribution order, the optimization ease is obtained according to the working state data by the optimization ease obtaining method. The optimization ease obtaining method includes: Step S81, input the point cloud region into the point cloud feature conversion model to obtain the point cloud feature value, and input the shared positioning state into the shared feature conversion model to obtain the shared positioning state feature value; Step S82, normalize the point cloud feature value and the shared positioning state feature value to obtain the normalized point cloud feature value dc and the normalized shared positioning state feature value gz. In step S83, the optimization easiness yd is calculated according to the normalized point cloud feature value dc, the normalized shared positioning state feature value gz, the normalized point cloud feature value weight a, and the normalized shared positioning state feature value weight β, and yd = a x (1 - dc) + β x gz is set to obtain the optimization easiness yd.

[0055] Specifically, the point cloud feature conversion model refers to a convolutional neural network model that takes a point cloud region as input and outputs a point cloud feature value. The specific construction method of the point cloud feature conversion model is not limited in the embodiment, and a person skilled in the art can freely select according to actual needs. For example, a historical point cloud region and its corresponding point cloud feature value are taken as a training set to train the convolutional neural network model to obtain the convolutional neural network model. The point cloud feature value refers to a vector value reflecting the characteristics of the point cloud region output by the point cloud feature conversion model. The shared feature conversion model refers to a convolutional neural network model that takes a shared positioning state as input data and outputs a shared positioning state feature value. The specific construction method of the shared feature conversion model is not limited in the embodiment, and a person skilled in the art can freely select according to actual needs. For example, a historical shared positioning state and its corresponding shared positioning state feature value are taken as a training set to train the convolutional neural network model to obtain the shared feature conversion model. The shared positioning state feature value refers to a feature vector value reflecting the characteristics of the shared positioning state obtained by the shared feature conversion model. The normalization process refers to an operation process of mapping the point cloud feature value and the shared positioning state feature value to the [0, 1] interval to obtain a normalized point cloud feature value and a normalized shared positioning state feature value. The specific method of normalization is not limited in the embodiment, and a person skilled in the art can freely select according to actual needs. It is only required to meet the requirement that the closer the normalized point cloud feature value dc is to 1, the higher the sparsity of the point cloud, and the closer the normalized shared positioning state feature value gz is to 1, the better the shared state. For example, normalization is performed by Python to obtain the normalized point cloud feature value and the normalized shared positioning state feature value. The normalized point cloud feature value weight refers to a coefficient for measuring the importance of the normalized point cloud feature value in the optimization easiness. The normalized shared positioning state feature value weight refers to a coefficient for measuring the importance of the normalized shared positioning state feature value in the optimization easiness. The normalized point cloud feature value weight a and the normalized shared positioning state feature value weight β are not limited in the embodiment, and a person skilled in the art can freely select according to actual needs. It is only required to meet the requirement that a + β = 1. For example, a = 0.4 and β = 0.6 are set in the embodiment.

[0056] Specifically, in step S4, the optimization ease is obtained. When the optimization ease is higher, the difficulty of optimizing the computing power of the running task is lower. When the optimization ease is lower, the difficulty of optimizing the computing power of the running task is higher. This makes it easier to intuitively quantify the difficulty of computing power optimization, thereby improving the efficiency of optimizing computing power resources.

[0057] Specifically, in step S4, when the order of computing power optimization distribution is adjusted according to the optimization ease, the optimization ease yd is compared with the preset optimization ease yd0, and the optimization ease is judged according to the comparison result. The importance coefficient E of the computing power data is adjusted according to the judgment result, wherein: When yd≤yd0, the optimization difficulty is determined to be difficult to optimize, and the difficulty of the important coefficient E of the computing power data is not adjusted; When yd>yd0, the optimization difficulty is determined to be easy to optimize, and the difficulty of the computing power data important coefficient E is adjusted according to the difficulty adjustment coefficient, and yt=1.33-0.22×e is set. -(yd-yd0) , obtain the adjusted computing power data importance coefficient E', set E'=E×yt, replace the computing power data importance coefficient E with the adjusted computing power data importance coefficient E', and re-acquire the computing power important coefficient set according to the computing power data importance coefficient E, and re-sort the computing power important coefficient set by importance to obtain the computing power optimization distribution order.

[0058] Specifically, the preset optimization ease refers to a preset value for judging the optimization ease situation. This embodiment does not limit the preset optimization ease. Relevant technical personnel in this field can freely choose according to actual needs. For example, this embodiment sets the preset optimization ease yd0=0.45. The optimization ease situation refers to the degree of ease of computing power optimization reflected by the optimization ease judged based on the optimization ease and the preset optimization ease. The optimization ease situation includes difficult to optimize and easy to optimize.

[0059] Specifically, in step S4, the optimization ease is judged. When the optimization ease is easy to optimize, the important coefficients of the computing power data are promptly increased according to the ease adjustment coefficient, so as to give priority to optimizing the computing power of the task units with high optimization ease. When the optimization ease maintains the easy optimization state to a certain extent, the increase in the important coefficients of the computing power data tends to be gentle, and the ease adjustment coefficient is set to match this change trend. The constant term 1.33 in the ease adjustment coefficient refers to the maximum value that the ease adjustment coefficient can reach, and the constant coefficient 0.22 is the coefficient for controlling the change rate of the ease adjustment coefficient, so that the change trend of the ease adjustment coefficient increases from 1.11 to 1.33, so as to reduce the impact of the optimization ease on the order of computing power optimization distribution, thereby improving the efficiency of optimizing computing power resources.

[0060] Specifically, in the step S4, when updating the ease degree according to the ease degree adjustment process of the area accuracy, the area accuracy gj is compared with the preset area accuracy gj0, the state of the area accuracy is judged according to the comparison result, and the preset optimization ease degree yd0 is updated according to the judgment result, wherein: When gj≤ gj0, it is determined that the state of the area accuracy is low accuracy, and the preset optimization ease degree yd0 is not updated; When gj> gj0, it is determined that the state of the area accuracy is high accuracy, and the preset optimization ease degree yd0 is updated, the preset optimization ease degree yd0 is updated according to the ease degree updating coefficient yx, yx=1.35+0.28×e -(gj-gj0) , the updated preset optimization ease degree yd0` is obtained, yd0`=yd0×yx, the preset optimization ease degree yd0 is replaced by the updated preset optimization ease degree yd0`, and the optimization ease degree yd is compared with the preset optimization ease degree yd0 again.

[0061] Specifically, the area accuracy refers to an area that needs to be identified with high accuracy, for example, when a warehouse AGV robot is transporting goods to a destination, the destination is set as a high-accuracy area, the preset area accuracy refers to a preset value for judging the state of the area accuracy, and the preset area accuracy is not limited in the embodiment, and can be freely selected by a person skilled in the art according to actual needs, for example, the preset area accuracy gj0=2.32 is set in the embodiment, the state of the area accuracy refers to the degree of the area accuracy reflected by the area accuracy, and the state of the area accuracy includes low accuracy and high accuracy.

[0062] Specifically, in the step S4, by judging the state of the area accuracy, when the state of the area accuracy is high accuracy, the preset optimization ease degree is quickly reduced by the ease degree updating coefficient for timely updating the ease degree, and when the state of the area accuracy reaches the later stage of high accuracy, the trend of reducing the preset optimization ease degree tends to be flat, and the ease degree updating coefficient is set to match this change trend, the constant term 1.35 in the ease degree updating coefficient refers to the maximum value that the ease degree updating coefficient can reach, and the constant coefficient 0.28 represents the change rate of the ease degree updating coefficient, so that the change trend of the ease degree updating coefficient is from 1.07 to 1.35, so as to reasonably reduce the influence of the high-accuracy area on the optimization ease degree, thereby improving the accuracy and efficiency of the computing power resource optimization.

[0063] Specifically, in the step S5, the computing power resource allocation of the warehouse AGV robot is optimized according to the computing power intelligent optimization method.

[0064] Specifically, the computing resource allocation of the warehouse AGV robot refers to a manner of deploying computing resources by the warehouse AGV robot in the process of executing a task request.

[0065] Specifically, in the step S5, the computing resource optimization is performed according to the computing power intelligent optimization method, so as to implement the decision-making computing power intelligent optimization method, thereby improving the efficiency of computing resource optimization.

[0066] So far, the technical solutions of the present application have been described in conjunction with the preferred embodiments shown in the drawings, but those skilled in the art can easily understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application, and the technical solutions after the changes or replacements will fall within the protection scope of the present application.

Claims

1. A computing resource optimization method based on AI big model, characterized by: include: Step S1, collecting real-time computing power status data and working status data; Step S2: Obtain the current actual computing power share based on the real-time computing power status data, push computing power alarm information based on the current actual computing power share, and adjust the current actual computing power share based on the predicted computing power share; Step S3: Obtain a computing power resource optimization decision based on the current actual computing power ratio, adjust the computing power resource optimization decision based on the estimated computing power impact, and update the decision adjustment process based on the computing power data importance coefficient; Step S4: Obtain the computing power optimization distribution order based on the computing power data importance coefficient, output the computing power intelligent optimization method based on the computing power optimization distribution order, adjust the computing power optimization distribution order based on the optimization ease, and update the ease of the ease adjustment process based on the regional accuracy; Step S5: Optimize the computing power resource allocation of the warehouse AGV automatic guided vehicle robot according to the computing power intelligent optimization method.

2. The computing resource optimization method based on the AI ​​large model according to claim 1 is characterized in that: In step S1, real-time computing power status data and working status data are collected, and the real-time computing power status data is stored in a computing power monitoring database; In step S2, the current actual computing power ratio is obtained according to the real-time computing power status data, and when the computing power alarm information is pushed according to the current actual computing power ratio, the current actual computing power ratio sjp is calculated according to the current actual computing power sj and the total computing power si, and sjp=sj / si is set to obtain the current actual computing power ratio sjp. The current actual computing power ratio sjp is compared with the first preset current actual computing power ratio sjp1 and the second preset current actual computing power ratio sjp2. The status of the current actual computing power ratio is judged according to the comparison result, and the computing power alarm information is pushed according to the judgment result, wherein: When sjp≤sjp1, the current actual computing power ratio is determined to be low, and the low ratio status is pushed as a computing power alarm message; When sjp1<sjp≤sjp2, the current actual computing power share is determined to be medium, and the medium share status is pushed as computing power alarm information; When sjp>sjp2, the current actual computing power ratio is determined to be high, and the high ratio status is pushed as computing power alarm information.

3. The computing resource optimization method based on the AI ​​large model according to claim 2 is characterized in that: In step S2, when adjusting the current actual computing power share according to the predicted computing power share, a predicted computing power share model is constructed based on the historical computing power status data to obtain a predicted computing power share model, and the current actual computing power, computing power utilization rate, computing power response time, and computing power density share are input into the predicted computing power share model to obtain a predicted computing power share output by the predicted computing power share model; In step S2, when adjusting the current actual computing power ratio according to the predicted computing power ratio, the predicted computing power ratio yc is compared with the preset predicted computing power ratio yc0, and the compliance of the predicted computing power ratio with the standard is judged according to the comparison result, and the current actual computing power sj is adjusted according to the judgment result, wherein: When yc≤yc0, the predicted computing power ratio is considered to have met the standard, and the current actual computing power sj is not adjusted; When yc>yc0, the predicted computing power ratio is judged as not meeting the standard, and the current actual computing power sj is adjusted according to the adjustment coefficient tz, setting tz=1.36-0.22×e -(yc-yc0) , where e is the base of the natural logarithm. The adjusted current actual computing power sj` is obtained, and sj`=sj×tz is set. The current actual computing power sj is replaced with the adjusted current actual computing power sj`, and the current actual computing power proportion sjp is recalculated based on the current actual computing power sj and the total computing power si.

4. The computing resource optimization method based on the AI ​​large model according to claim 3 is characterized in that: In step S3, when the computing power resource optimization decision is obtained according to the current actual computing power ratio by the computing power resource optimization decision obtaining method, the computing power resource optimization decision obtaining method includes: Step S41: Calculate the available allocation space kt based on the current actual computing power share sjp and the total computing power si, and set kt = (si - sjp) × 100% to obtain the available allocation space kt; Step S42: compare the adjustable share kt with the preset adjustable share kt0, determine the status of the adjustable share based on the comparison result, and output a computing power resource optimization decision based on the determination result, wherein: When kt ≥ kt0, the state of the available allocation space is determined to be sufficient, and no computing power resource optimization will be performed as the computing power resource optimization decision output; When kt<kt0, it is determined that the state of the allocated space is insufficient, and computing power resource optimization is performed as an output as a computing power resource optimization decision.

5. The computing resource optimization method based on the AI ​​large model according to claim 4 is characterized in that: In step S3, when adjusting the computing power resource optimization decision based on the pre-computing power influence degree, the pre-computing power influence degree is obtained based on the real-time computing power status data using a pre-computing power influence degree obtaining method, and the pre-computing power influence degree obtaining method includes: Step S51: Input the computing power utilization impact value, computing power response time impact value, and computing power density impact value as first target impact values ​​into the computing power impact prediction model, and obtain the predicted computing power utilization impact value ap1, predicted computing power response time impact value ap2, and predicted computing power density impact value ap3 output by the computing power impact prediction model; Step S52: using the predicted computing power utilization impact value ap1, the predicted computing power response time impact value ap2, and the predicted computing power density impact value ap3 as second target impact values; Step S53, calculate the pre-computing power influence degree ysl according to the second target impact value, the predicted computing power utilization influence value weight A1, the predicted computing power response time influence value weight A2 and the predicted computing power density influence value weight A3, set ysl=ap1×A1+ap2×A2+ap3×A3, and obtain the pre-computing power influence degree ysl.

6. The computing resource optimization method based on the AI ​​large model according to claim 5 is characterized in that: In step S3, when adjusting the computing power resource optimization decision based on the pre-computing power impact, the pre-computing power impact ysl is compared with the preset computing power impact ysl0, and the compliance of the pre-computing power impact is judged based on the comparison result. Based on the judgment result, the preset adjustable space kt0 is adjusted, where: When ysl≤ysl0, the step S3 determines that the pre-calculation power influence degree meets the standard, and no decision adjustment is made to the preset adjustable space kt0; When ysl>ysl0, the step S3 determines that the pre-calculated power influence degree is not up to standard, and makes a decision adjustment on the preset adjustable space kt0. The preset adjustable space kt0 is adjusted according to the decision adjustment coefficient jc, and jc=1.42-0.23×e -(ysl-ysl0) , obtain the adjusted preset adjustable proportion space kt0`, set kt0`=kt0×jc, replace the preset adjustable proportion space kt0 with the adjusted preset adjustable proportion space kt0`, and re-compare the adjustable proportion space kt with the preset adjustable proportion space kt0.

7. The computing resource optimization method based on the AI ​​large model according to claim 6 is characterized in that: In step S3, when the decision adjustment process is updated according to the computing power data importance coefficient, the computing power data importance coefficient E is calculated according to the number of calls d1, the accuracy requirement d2, the average error d3, the number of errors d4, the number of calls weight β1, the accuracy requirement weight β2, the average error weight β3 and the number of errors weight β4, and E=β1×d1+β2×d2+β3×d3+β4×d4 is set to obtain the computing power data importance coefficient E; In step S3, when the decision adjustment process is updated according to the computing power data importance coefficient, the computing power data importance coefficient E is compared with the preset computing power data importance coefficient E0, and the attribute of the computing power data importance coefficient is judged according to the comparison result, and the preset computing power influence degree ysl0 is updated according to the judgment result, where: When E≤E0, step S3 determines that the attribute of the important coefficient of the computing power data is unimportant, and does not make a decision update on the preset computing power influence degree ysl0; When E>E0, the step S3 determines that the attribute of the important coefficient of the computing power data is important, and makes a decision update on the preset computing power influence degree ysl0. The decision update coefficient gx is used to update the preset computing power influence degree ysl0, and gx=1.46+0.23×e -(E-E0) , obtain the updated preset computing power influence degree ysl0`, set ysl0`=ysl0×gx, replace the preset computing power influence degree ysl0 with the updated preset computing power influence degree ysl0`, and re-compare the pre-computing power influence degree ysl with the preset computing power influence degree ysl0.

8. The computing resource optimization method based on AI large model according to claim 7 is characterized in that: In step S4, when obtaining the computing power optimization distribution order according to the computing power data importance coefficient, a computing power important coefficient set is obtained according to the computing power data importance coefficient to obtain the computing power important coefficient set, and the computing power important coefficient set is sorted by importance to obtain the computing power optimization distribution order; In step S4, when making a decision on the computing power intelligent optimization method based on the computing power optimization distribution order, the computing power optimization distribution order is input into the computing power intelligent optimization method decision tree to obtain the computing power intelligent optimization method; In step S4, when the order of computing power optimization distribution is adjusted according to the optimization ease, the optimization ease is obtained according to the working status data using an optimization ease obtaining method, and the optimization ease obtaining method includes: Step S81: input the point cloud region into the point cloud feature conversion model to obtain a point cloud feature value, and input the shared positioning state into the shared feature conversion model to obtain a shared positioning state feature value; Step S82, normalizing the point cloud eigenvalues ​​and the shared positioning state eigenvalues ​​to obtain a normalized point cloud eigenvalue dc and a normalized shared positioning state eigenvalue gz; In step S83, the optimization ease yd is calculated based on the normalized point cloud eigenvalue dc, the normalized shared positioning state eigenvalue gz, the normalized point cloud eigenvalue weight α, and the normalized shared positioning state eigenvalue weight β, and yd=α×(1-dc)+β×gz is set to obtain the optimization ease yd.

9. The computing resource optimization method based on the AI ​​large model according to claim 8 is characterized in that: In step S4, when the order of computing power optimization distribution is adjusted according to the optimization ease, the optimization ease yd is compared with the preset optimization ease yd0, and the optimization ease is judged according to the comparison result. The importance coefficient E of the computing power data is adjusted according to the judgment result, wherein: When yd≤yd0, the optimization difficulty is determined to be difficult to optimize, and the difficulty of the important coefficient E of the computing power data is not adjusted; When yd>yd0, the optimization difficulty is determined to be easy to optimize, and the difficulty of the computing power data important coefficient E is adjusted according to the difficulty adjustment coefficient, and yt=1.33-0.22×e is set. -(yd-yd0) , obtain the adjusted computing power data importance coefficient E', set E'=E×yt, replace the computing power data importance coefficient E with the adjusted computing power data importance coefficient E', and re-acquire the computing power important coefficient set according to the computing power data importance coefficient E, and re-sort the computing power important coefficient set by importance to obtain the computing power optimization distribution order.

10. The computing resource optimization method based on AI large model according to claim 9 is characterized in that: In step S4, when the difficulty is updated in the process of adjusting the difficulty according to the regional accuracy, the regional accuracy gj is compared with the preset regional accuracy gj0, and the state of the regional accuracy is judged according to the comparison result. The preset optimized difficulty yd0 is updated according to the judgment result, wherein: When gj≤gj0, the step S4 determines that the state of the regional accuracy is low accuracy, and does not update the preset optimization ease yd0; When gj>gj0, the step S4 determines that the state of the regional accuracy is high precision, updates the preset optimization ease yd0, and updates the preset optimization ease yd0 according to the ease update coefficient yx, setting yx=1.35+0.28×e -(gj-gj0) , obtain the updated preset optimization ease yd0`, set yd0`=yd0×yx, replace the preset optimization ease yd0 with the updated preset optimization ease yd0`, and re-compare the optimization ease yd with the preset optimization ease yd0; In step S5, the computing power resource allocation of the warehouse AGV automatic guided vehicle robot is optimized according to the computing power intelligent optimization method.

Images