Chiller control method and device, chiller and heating and ventilation system

Abstract

The application relates to a cold machine control method and device, a cold machine and a heating and ventilation system. The cold machine control method comprises the following steps: determining a plurality of candidate models; determining scores of the candidate models based on a current operation scene and a plurality of scoring dimensions; the scoring dimensions at least comprise an accuracy scoring dimension, a speed scoring dimension and a computing power scoring dimension; selecting a target model from the plurality of candidate models based on the scores, and controlling the operation of the cold machine based on the target model. In this way, in the intelligent control process of the cold machine, the candidate models are evaluated through the accuracy, speed and computing power dimensions, the model most suitable for the current operation scene can be accurately selected to intelligently control the operation of the cold machine, and the model selection flexibility in different scenes is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control, and particularly relates to a cold machine control method and device, a cold machine and a heating and ventilation system. BACKGROUND

[0002] In the related art, the cold machine is a key device in the heating and ventilation system, and an intelligent control based on artificial intelligence (AI) is introduced in the control layer. However, in actual application, different scenes (such as high-precision control scenes, low-computing-resource scenes, etc.) facing the cold machine operation have different requirements for the model, and the flexibility and adaptability of the model application are poor. SUMMARY

[0003] The present application provides a cold machine control method and device, a cold machine and a heating and ventilation system to solve the technical problem of poor flexibility of model application.

[0004] In a first aspect, the present application provides a cold machine control method, comprising: determining a plurality of candidate models; determining scores of the candidate models based on a current running scene and a plurality of score dimensions; the score dimensions at least include: an accuracy score dimension, a speed score dimension and a computing power score dimension; selecting a target model from the plurality of candidate models based on the scores, and controlling the running of the cold machine based on the target model.

[0005] In one possible implementation, the determining of the scores of the candidate models based on the current running scene and the plurality of score dimensions comprises: determining a sub-score of each score dimension of the candidate model based on the current running scene and at least one score index corresponding to each score dimension; performing normalization processing on the sub-scores; and determining the scores of the candidate models based on the normalized sub-scores and weights corresponding to the score dimensions.

[0006] In one possible implementation, the determining of the sub-score of each score dimension of the candidate model based on the current running scene and at least one score index corresponding to each score dimension comprises: performing a predetermined operation through the candidate model; and determining the sub-score of each score dimension of the candidate model based on the current running scene, an operation result of the predetermined operation and at least one score index corresponding to each score dimension.

[0007] In a possible implementation, the determining the sub-score of each scoring dimension of the candidate model based on the current running scenario, the operation result of the predetermined operation, and at least one scoring indicator corresponding to each scoring dimension comprises: determining a first sub-score of the accuracy scoring dimension of the candidate model based on the current running scenario, a data volume corresponding to the predetermined operation, an actual value and a model predicted value corresponding to the operation result of the predetermined operation, and a first scoring indicator corresponding to the accuracy scoring dimension; determining a second sub-score of the speed scoring dimension of the candidate model based on the current running scenario, a training start and end time and a prediction start and end time corresponding to the operation result of the predetermined operation, and a second scoring indicator corresponding to the speed scoring dimension; and determining a third sub-score of the computing power scoring dimension of the candidate model based on the current running scenario, a system performance parameter corresponding to the operation result of the predetermined operation, and a third scoring indicator corresponding to the computing power scoring dimension.

[0008] In a possible implementation, after the operation of the cold machine is controlled based on the target model, the method further comprises: determining whether the target model needs to be updated; in response to the target model needing to be updated, determining whether to maintain the current working state based on the current working state of the cold machine; if it is determined that the current working state is not maintained, controlling the operation of the cold machine based on a predetermined control algorithm, and suspending the control function of the target model; and updating the target model.

[0009] In a possible implementation, the updating the target model comprises: updating and deploying the target model based on an updated model; determining whether the updating and deployment is successful; if the updating and deployment fails, resuming the control function of the target model and outputting prompt information; and if the updating and deployment is successful, controlling the operation of the cold machine based on the updated target model.

[0010] In a possible implementation, the determining the plurality of candidate models comprises: processing predetermined training data based on a predetermined mechanism model; inputting a first result output by the predetermined mechanism model and the predetermined training data into a predetermined data-driven model; and training based on the first result and a second result output by the predetermined data-driven model to obtain the plurality of candidate models for cold machine control.

[0011] In a possible implementation, the processing the predetermined training data based on the predetermined mechanism model comprises: inputting the predetermined training data into the predetermined mechanism model to obtain a plurality of cold machine control processing data; and outputting a first result representing the performance of the cold machine based on the plurality of cold machine control processing data.

[0012] In a second aspect, the present application provides a cold machine control device, the device comprising: a determination unit configured to determine a plurality of candidate models; a scoring unit configured to determine scores of the candidate models based on a current running scenario and a plurality of scoring dimensions; the scoring dimensions at least comprising: an accuracy scoring dimension, a speed scoring dimension, and a computing power scoring dimension; and an execution unit configured to select a target model from the plurality of candidate models based on the scores, and control running of the cold machine based on the target model.

[0013] In a third aspect, the present application provides a cold machine device comprising the cold machine control device of any one of the preceding second aspect.

[0014] In a fourth aspect, the present application provides a heating and ventilation system comprising the cold machine device of any one of the preceding third aspect.

[0015] In a fifth aspect, the present application provides an electronic device comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory are in communication with each other via the communication bus; the memory is configured to store a computer program; and the processor is configured to execute the computer program to implement the method of any one of the preceding first aspect.

[0016] In a sixth aspect, the present application further provides a storage medium having a computer program stored thereon, the computer program being executable by a processor to implement the method of any one of the preceding first aspect.

[0017] The above technical solution provided by the embodiments of the present application has the following advantages compared with the prior art: the method provided by the embodiments of the present application determines a plurality of candidate models; determines scores of the candidate models based on a current running scenario and a plurality of scoring dimensions; the scoring dimensions at least comprising: an accuracy scoring dimension, a speed scoring dimension, and a computing power scoring dimension; selects a target model from the plurality of candidate models based on the scores, and controls running of the cold machine based on the target model. In this way, according to the current running scenario and the scoring mechanism of the accuracy, speed, and computing power dimensions, a control model most suitable for the current running scenario requirement can be selected, so that when the cold machine is controlled based on the control model, the running status of the cold machine can be flexibly adapted, and the accuracy and efficiency of the control can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which are incorporated herein and form part of the specification, illustrate embodiments consistent with the present application and, together with the description, further serve to explain the principles behind the present application.

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, for those skilled in the art, other drawings can also be obtained based on these drawings without any creative effort.

[0020] One or more embodiments are illustrated by way of example with reference to the drawings, which are schematic and not drawn to scale, wherein the same or similar elements are denoted by the same reference numerals, and wherein the various features of the figures are not necessarily drawn to scale, and wherein the figures constitute a part of the detailed description of the embodiments.

[0021] Figure 1 A flowchart of a cold machine control method provided by an embodiment of the present application is shown in FIG. 1.

[0022] Figure 2 A flowchart of a cold machine control method provided by an embodiment of the present application is shown in FIG. 1.

[0023] Figure 3 A flowchart of a cold machine control method provided by an embodiment of the present application is shown in FIG. 1.

[0024] Figure 4 A flowchart of a cold machine control method provided by an embodiment of the present application is shown in FIG. 1.

[0025] Figure 5 A flowchart of a cold machine model training provided by an embodiment of the present application is shown in FIG. 1.

[0026] Figure 6 A flowchart of a cold machine model score promotion provided by an embodiment of the present application is shown in FIG. 1.

[0027] Figure 7 A flowchart of a cold machine model update deployment provided by an embodiment of the present application is shown in FIG. 1.

[0028] Figure 8 A schematic diagram of a cold machine control device provided by an embodiment of the present application is shown in FIG. 1.

[0029] Figure 9 A schematic diagram of an electronic device provided by an embodiment of the present application is shown in FIG. 1. DETAILED DESCRIPTION

[0030] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the following will be combined with the drawings in the embodiments of the present application to make a clear and complete description of the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work belong to the scope of protection of the present application. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0031] The following disclosure provides many different embodiments or examples for implementing different structures of the present application. In order to simplify the disclosure of the present application, the components and settings of specific examples are described below. Of course, they are only examples and the purpose is not to limit the present application. In addition, reference numerals and / or letters can be repeated in different examples in the present application. Such repetition is for the purpose of simplification and clarity, and does not indicate the relationship between the various embodiments and / or settings discussed.

[0032] In order to solve the technical problem of poor flexibility of model application in the prior art, the present application provides a cold machine control method, which scores the candidate models according to the current running scene and multiple scoring dimensions, so as to select the model most suitable for the current scene to control the running of the cold machine, and improve the flexibility and accuracy of model application.

[0033] Figure 1 A flowchart of a cold machine control method provided by the present embodiment is shown in Figure 1 The method can include:

[0034] S10: determining multiple candidate models;

[0035] S20: determining the scores of the candidate models based on the current running scene and multiple scoring dimensions; the scoring dimensions at least include: accuracy scoring dimension, speed scoring dimension and computing power scoring dimension;

[0036] S30: selecting a target model from the multiple candidate models based on the scores, and controlling the running of the cold machine based on the target model.

[0037] The cold machine control method provided by the present embodiment can be applied to a cold machine, or can also be applied to a server or other terminal with control capability for the cold machine, such as a terminal device connected to the cold machine through Bluetooth, local area network or other means to control the running of the cold machine. Here, the cold machine can be a device for realizing cooling and other functions in a heating and ventilation system.

[0038] In the embodiment, the determining the plurality of candidate models can comprise: in response to the chiller being woken up or the chiller starting to run, determining the plurality of candidate models. Wherein, the woken up can refer to switching from a sleep state or a shutdown state to a working state or a startup state, etc.

[0039] In one embodiment, the candidate model and the target model can also be referred to as a chiller model or a chiller control model, etc. The candidate model can be a model for chiller control, for example, different parameters of different candidate models or different weight distributions of sub-models contained. Here, the sub-models of the candidate model can include a predetermined mechanism model and a predetermined data-driven model.

[0040] For example, the predetermined mechanism model can be used to provide constraints on the physical characteristics of the system, and the predetermined data-driven model can be used to supplement the processing of nonlinear characteristics in a high-dimensional complex environment. The weights of the two can be dynamically adjusted under different working conditions of the chiller. For example, when the device approaches the rated working condition, the weight of the predetermined mechanism model is higher; and in a complex fluctuating working condition, the weight of the predetermined data-driven model is higher.

[0041] In one embodiment, the determining the score of the candidate model based on the current running scenario and the plurality of scoring dimensions can comprise: determining a sub-score of the candidate model under each scoring dimension based on the current running scenario; determining the score of the candidate model based on the sub-score corresponding to each scoring dimension. Wherein, the current running scenario can include at least one of the current working mode, the current working state, and the current working condition of the chiller.

[0042] In one embodiment, the determining the sub-score of the candidate model under each scoring dimension based on the current running scenario can comprise: executing a predetermined operation through the candidate model, and determining the sub-score of the candidate model under each scoring dimension based on the current running scenario, the operation result of the predetermined operation, and at least one scoring indicator corresponding to each scoring dimension.

[0043] In one embodiment, the determining the score of the candidate model based on the sub-score corresponding to each scoring dimension can comprise: determining the score of the candidate model based on the sub-score corresponding to each scoring dimension and the weight corresponding to each scoring dimension. For example, it can comprise: performing normalization processing on the sub-score; determining the score of the candidate model based on the normalized sub-score and the weight corresponding to the scoring dimension.

[0044] In one embodiment, the scoring dimensions at least include one of the following: an accuracy scoring dimension, a speed scoring dimension, and a computing power scoring dimension. Wherein, the accuracy scoring dimension can be used to evaluate the prediction accuracy of the candidate model, the speed scoring dimension can be used to evaluate the training speed and / or the prediction speed of the candidate model, and the computing power scoring dimension can be used to evaluate the occupation of the computing power resources by the candidate model during the running process.

[0045] In an embodiment, each scoring dimension can correspond to at least one scoring indicator. For example, the first scoring indicator corresponding to the precision scoring dimension can include at least one of the following: data variance understanding coefficient, maximum deviation, average deviation, and relative error, etc.; the second scoring indicator corresponding to the speed scoring dimension can include at least one of the following: training duration and prediction duration, etc.; the third scoring indicator corresponding to the computing power scoring dimension can include at least one of the following: floating point calculation amount, processor utilization rate, and memory occupancy rate, etc.

[0046] In an embodiment, selecting a target model from the plurality of candidate models based on the scores can include: selecting the candidate model with the highest score from the plurality of candidate models as the target model.

[0047] In an embodiment, selecting a target model from the plurality of candidate models based on the scores can include: selecting a target model from the plurality of candidate models based on the current running scenario and the scores. For example, when the number of candidate models with the highest scores in the plurality of candidate models is greater than 1, selecting the candidate model with the highest matching degree to the current running scenario from the candidate models with the highest scores as the target model.

[0048] Here, the highest matching degree to the current running scenario can refer to the highest matching degree of the included sub-model weight distribution to the current running scenario.

[0049] In an embodiment, controlling the operation of the chiller based on the target model can include: executing a control function based on the target model to control the operation of the chiller.

[0050] In an embodiment, controlling the operation of the chiller based on the target model can include: determining whether there is a current model for controlling the operation of the chiller; if not, controlling the operation of the chiller based on the target model; if so, updating the current model based on the target model.

[0051] In an embodiment, updating the current model based on the target model can include: determining whether to maintain the current working state of the chiller based on the current working state; if it is determined not to maintain the current working state, controlling the operation of the chiller based on a predetermined control algorithm, suspending the control function of the current model, and updating the current model based on the target model.

[0052] In an embodiment, the above step of updating the current model based on the target model can include: updating and deploying the current model based on the target model; determining whether the update and deployment is successful; if the update and deployment fails, resuming the control function of the current model and outputting a prompt message; if the update and deployment is successful, controlling the operation of the chiller based on the target model.

[0053] In an embodiment, determining whether to maintain the current working state can comprise determining whether the chiller allows maintaining the current working state. Wherein, maintaining the current working state can refer to maintaining the current running condition, all running parameters remain unchanged, and no adjustment is made.

[0054] In an embodiment, the method can further comprise: if it is determined to maintain the current working state, maintaining the current running parameters unchanged, and updating the current model based on the target model.

[0055] In an embodiment, determining whether to maintain the current working state based on the current working state of the chiller can comprise: determining whether the current running parameters need to be changed based on the current working state of the chiller; if the current running parameters need to be changed, determining not to maintain the current working state; if the current running parameters do not need to be changed, determining to maintain the current working state.

[0056] Here, determining whether the current running parameters need to be changed can comprise: determining whether the current running parameters need to be changed within a predetermined time length. The predetermined time length can be related to the time length required for updating the current model based on the target model.

[0057] In an embodiment, controlling the running of the chiller based on the predetermined control algorithm can comprise: switching to a predetermined group control strategy and taking over the control of the running of the chiller by the predetermined control algorithm. The predetermined control algorithm here can be a conventional control algorithm.

[0058] In this way, according to the current running scenario and the scoring mechanism of multiple dimensions of precision, speed, and computing power, the control model that best matches the requirements of the current running scenario can be selected, so that when the control model is used for chiller running control, the running status of the chiller can be flexibly adapted, and the accuracy and efficiency of the control can be improved.

[0059] In order to integrate these scores for comprehensive evaluation, so that the scores of all dimensions are compared within the same dimension and the importance of different dimensions is balanced, therefore, Figure 2 A flowchart of a chiller control method provided by the embodiment is shown in Figure 2 As shown in the above step S20, determining the score of the candidate model based on the current running scenario and multiple scoring dimensions can comprise:

[0060] S21: determining a sub-score of the candidate model in each scoring dimension based on the current running scenario and at least one scoring indicator corresponding to each scoring dimension;

[0061] S22: performing normalization processing on the sub-score;

[0062] S23: determining the score of the candidate model based on the normalized sub-score and the weight corresponding to the scoring dimension.

[0063] In an embodiment, the sub-scores can be one-to-one corresponding to the score indicators, i.e., each score indicator corresponds to a sub-score.

[0064] In an embodiment, the normalization of the sub-scores can include: based on the minimum value and the maximum value in all sub-scores, normalizing each sub-score.

[0065] For example, the normalization formula is:

[0066] Wherein: X is the sub-score, X nor is the normalized sub-score, min(X) is the minimum value in all sub-scores, and max(X) is the maximum value in all sub-scores.

[0067] In an embodiment, each score dimension corresponds to a weight, for example, the precision score dimension corresponds to a first weight The speed score dimension corresponds to a second weight The computing power score dimension corresponds to a third weight

[0068] In an embodiment, the first score indicator corresponding to the precision score dimension includes: data variance understanding coefficient RS, maximum deviation MD, average deviation AD, and relative error AR; the second score indicator corresponding to the speed score dimension can include: training time TT and prediction time PT; the third score indicator corresponding to the computing power score dimension can include: floating point calculation amount FL, processor utilization rate GC, and memory occupancy rate MU.

[0069] Correspondingly, the sub-scores corresponding to the first score indicators under the precision score dimension after normalization are respectively: RS nor , MD nor , AD nor , and AR nor ; the sub-scores corresponding to the second score indicators under the speed score dimension after normalization are respectively: TT nor , and PT nor ; the sub-scores corresponding to the third score indicators under the computing power score dimension after normalization are respectively: FL nor , GC nor , and MU nor .

[0070] In an embodiment, the step S23 can include: determining a first product of a sum of all first score indicators corresponding sub-scores in the precision score dimension and a first weight; determining a second product of a sum of all second score indicators corresponding sub-scores in the speed score dimension and a second weight; determining a third product of a sum of all third score indicators corresponding sub-scores in the computing power score dimension and a third weight; and calculating a sum of the first product, the second product and the third product as the score of the candidate model.

[0071] That is, the score of the candidate model

[0072] In this way, the sub-scores of the plurality of score indicators in different dimensions are normalized, and then weighted calculation is performed based on the weights of different dimensions, so that the obtained score can be more consistent with the performance of the model in different dimensions, and thus the score can more accurately describe the characteristics of the model.

[0073] In some embodiments, the step S21 can include: performing a predetermined operation by the candidate model; and determining sub-scores of the candidate model in each score dimension based on a current running scenario, an operation result of the predetermined operation, and at least one score indicator corresponding to each score dimension.

[0074] In an embodiment, performing a predetermined operation by the candidate model can refer to processing predetermined data by the candidate model, such as performing a training operation and / or a prediction operation based on the predetermined data by the candidate model.

[0075] In an embodiment, the operation result of the predetermined operation can refer to at least one of an actual value corresponding to the operation result of the predetermined operation and a model predicted value, a training start and end time and a prediction start and end time corresponding to the operation result of the predetermined operation, and a system performance parameter corresponding to the operation result of the predetermined operation.

[0076] Here, the actual value is an actual value corresponding to the predetermined data, and the model predicted value is a predicted value output by the model when performing a prediction operation based on the predetermined data.

[0077] In this way, the actual running performance of the model is reflected by the processing result of the model, and then combined with a plurality of indicators in different dimensions, the score can more comprehensively and stereoscopically represent the characteristics of the model itself.

[0078] In an embodiment, the determination of the sub-scores of the candidate model in each score dimension based on the current running scenario, the operation result of the predetermined operation, and the at least one score indicator corresponding to each score dimension includes:

[0079] Step a: Based on the current operating scenario, the amount of data corresponding to the predetermined operation, the actual value and model prediction value corresponding to the operation result of the predetermined operation, and the first scoring index corresponding to the accuracy scoring dimension, determine the first sub-score of the candidate model under the accuracy scoring dimension;

[0080] Step b: Based on the current running scenario, the training start and end times and prediction start and end times corresponding to the operation results of the predetermined operation, and the second scoring index corresponding to the speed scoring dimension, determine the second sub-score of the candidate model under the speed scoring dimension;

[0081] Step c: Based on the current operating scenario, the system performance parameters corresponding to the operation result of the predetermined operation, and the third scoring index corresponding to the computing power scoring dimension, determine the third sub-score of the candidate model under the computing power scoring dimension.

[0082] In one embodiment, each first scoring indicator corresponds to a first sub-scoring. The first scoring indicators corresponding to the accuracy scoring dimension include: data variance understanding coefficient RS, maximum deviation MD, average deviation AD, and relative error AR.

[0083] Accordingly, step a may include: based on the current operating scenario, the data volume n corresponding to the predetermined operation, and the actual value y corresponding to the operation result of the predetermined operation. i Compared with model predictions and the average of actual values The first sub-score corresponding to the coefficient of variation RS is determined as follows:

[0084]

[0085] Here, the data variance understanding coefficient represents the ability of the candidate model to understand the data variance. The closer the coefficient is to 1, the better the understanding ability.

[0086] Accordingly, step a may also include: based on the current operating scenario, the actual value y corresponding to the operation result of the predetermined operation. i Compared with model predictions The maximum difference between the two values ​​determines the first sub-score corresponding to the maximum deviation MD: Here, the maximum deviation characterizes the error of the alternative model under the worst-case scenario.

[0087] Accordingly, step a may also include: based on the current operating scenario, the data volume n corresponding to the predetermined operation, and the actual value y corresponding to the operation result of the predetermined operation. i Compared with model predictions The first sub-score corresponding to the mean deviation AD is determined as follows:

[0088]

[0089] Correspondingly, step a can further include: based on the current running scenario, the data quantity n corresponding to the predetermined operation, and the actual value y corresponding to the operation result of the predetermined operation i and the model prediction value The first sub-score corresponding to the relative error AR is determined as:

[0090]

[0091] In an embodiment, each second score indicator corresponds to a second sub-score, and the second score indicator corresponding to the speed score dimension can include: a training duration TT and a prediction duration PT. The training start and end time can include a training start time and a training end time, and the prediction start and end time can include a prediction start time and a prediction end time.

[0092] Correspondingly, step b can include: based on the current running scenario, the training start time TT corresponding to the operation result of the predetermined operation s and the training end time TT e , determining the second sub-score corresponding to the training duration TT as: TT = TT s -TT e .

[0093] Correspondingly, step b can further include: based on the current running scenario, the prediction start time PT corresponding to the operation result of the predetermined operation s and the prediction end time PT e , determining the second sub-score corresponding to the prediction duration PT as: PT = PT s -PT e .

[0094] In an embodiment, each third score indicator corresponds to a third sub-score, and the third score indicator corresponding to the computing power score dimension can include: a floating point calculation amount FL, a processor utilization rate GC, and a memory occupancy rate MU.

[0095] Correspondingly, the system performance parameter can include the number of floating point operations. Step c can include: based on the current running scenario, the number of floating point operations corresponding to the operation result of the predetermined operation, determining the third sub-score corresponding to the floating point calculation amount FL. For example, the number of floating point operations can be the number of floating point operations of all operations in the execution process of the candidate model, and the FL of the candidate model can be calculated and analyzed across frameworks through the ONNX tool.

[0096] Correspondingly, the system performance parameter can further include: a processor execution operation duration, and a total duration of the candidate model running. Step c can further include: based on the current running scenario, the processor execution operation duration Tactive and the total running time T of the alternative model total The third sub-score corresponding to the processor utilization GC is determined as follows:

[0097]

[0098] The processor execution time is the time actually used by the processor to perform the operation, which can be obtained through a system performance monitoring tool (such as 'nvidia-smi'). The total running time of the alternative model is the total time of the entire alternative model running, including waiting, data loading, etc., which can be obtained through log recording or system tools.

[0099] Correspondingly, the system performance parameters can also include: the actual memory usage M used and the total system memory M total The third sub-score corresponding to the memory occupancy rate MU is determined as follows:

[0100]

[0101] In this way, the scores calculated based on different parameters obtained from the model processing process and results in different dimensions can more accurately represent the performance characteristics of the model in different dimensions, thereby obtaining more accurate and scene-conforming scores.

[0102] Since there is a need to update or iteratively upgrade the model after deployment, the related art often requires manual updating and upgrading after shutdown to avoid device inoperability during the updating process, which is more troublesome. Therefore, Figure 3 The flowchart of a cold machine control method provided by the present embodiment is shown in FIG. 4. After the step S30 of controlling the operation of the cold machine based on the target model, the method can further include: Figure 3

[0103] S40: determining whether the target model needs to be updated;

[0104] S50: in response to the need to update the target model, determining whether to maintain the current working state based on the current working state of the cold machine;

[0105] S60: if it is determined not to maintain the current working state, controlling the operation of the cold machine based on a predetermined control algorithm and suspending the control function of the target model;

[0106] S70: updating the target model.

[0107] ​In one embodiment, determining whether to maintain the current working state can comprise determining whether the chiller allows the current working state to be maintained. Here, maintaining the current working state can refer to maintaining the current operating condition, i.e. all operating parameters remain unchanged and no adjustment is made.

[0108] In one embodiment, the method can further comprise, if it is determined to maintain the current working state, maintaining the current operating parameters unchanged and updating the target model.

[0109] In one embodiment, determining whether to maintain the current working state based on the current working state of the chiller can comprise determining whether the current operating parameters need to be changed based on the current working state of the chiller; if the current operating parameters need to be changed, determining not to maintain the current working state; and if the current operating parameters do not need to be changed, determining to maintain the current working state.

[0110] Here, determining whether the current operating parameters need to be changed can comprise determining whether the current operating parameters need to be changed within a predetermined time period. The predetermined time period can be related to the time period required for updating the target model.

[0111] In one embodiment, controlling the operation of the chiller based on the predetermined control algorithm in step S60 can comprise switching to a predetermined group control strategy and taking over the control of the operation of the chiller by the predetermined control algorithm. The predetermined control algorithm here can be a conventional control algorithm.

[0112] In one embodiment, suspending the control function of the target model can comprise freezing the control logic of the target model and blocking the adjustment instructions of the target model. Thus, the system will no longer respond to the adjustment signals brought by the model control, avoiding fluctuations caused by model switching.

[0113] In one embodiment, suspending the control function of the target model can comprise suspending the control function of the target model and monitoring the operating state of the chiller during the suspension of the control function of the target model. For example, the operating state of the equipment is continuously monitored to ensure that no abnormal operation or equipment failure occurs during the model update.

[0114] In this way, it is ensured that the chiller can smoothly implement online update and deployment of the control model during normal operation, avoiding the impact on the overall operation performance of the equipment due to shutdown or interruption of the control logic. Sufficient flexibility is provided so that the model update can be smoothly carried out in different application scenarios, which can greatly reduce the risk of system interruption.

[0115] In one embodiment, updating the target model can comprise updating and deploying the target model based on the updated model; determining whether the updating and deployment is successful; if the updating and deployment fails, resuming the control function of the target model and outputting a prompt information; and if the updating and deployment is successful, controlling the operation of the chiller based on the updated target model.

[0116] In one embodiment, determining whether the update deployment is successful can comprise determining whether the model update deployment is successful according to a version number of the target model.

[0117] In one embodiment, if the update deployment fails can mean that the update deployment fails and the number of failures reaches a predetermined number of times. For example, the predetermined number of times can be 2, 3, or 5, etc.

[0118] In one embodiment, resuming the control function of the target model and outputting prompt information can mean aborting the update deployment and rolling back to the old model, while giving a clear error prompt to notify the user to manually intervene.

[0119] In one embodiment, controlling the operation of the cold machine based on the updated target model can mean releasing the control logic of the frozen target model and resuming the adjustment instructions of the target model. Therefore, the new model is put into actual control tasks.

[0120] In this way, the model update failure can recover the original model control in time to avoid long-term impact on system control. The model update deployment can be successfully completed without affecting the current operation. This not only reduces the complexity of system maintenance, but also improves the flexibility and reliability of the cold machine AI control, ensuring that the system can continue to operate efficiently and effectively.

[0121] In some embodiments, determining a plurality of candidate models in the step S10 can comprise:

[0122] processing predetermined training data based on a predetermined mechanism model;

[0123] inputting a first result output by the predetermined mechanism model and the predetermined training data into a predetermined data-driven model;

[0124] training based on the first result and a second result output by the predetermined data-driven model to obtain a plurality of candidate models for cold machine control.

[0125] In one embodiment, inputting the first result output by the predetermined mechanism model and the predetermined training data into the predetermined data-driven model can comprise inputting the first result output by the predetermined mechanism model into the predetermined data-driven model as a constraint condition of the data-driven model; and inputting the predetermined training data into the predetermined data-driven model for training. In this way, the data-driven model not only relies on historical data, but also follows the physical laws of the device, improving the interpretability and robustness of the model.

[0126] In an embodiment, the predetermined data-driven model can be driven by a predetermined algorithm, for example, the predetermined algorithm can be a random forest, a support vector machine (SVM), a K-nearest neighbor, a neural network, an XGBoost, etc.

[0127] In an embodiment, before processing the predetermined training data based on the predetermined mechanism model, the method can further include: importing the predetermined training data; and pre-processing the predetermined training data. Wherein, importing the predetermined training data can be to obtain historical running data corresponding to a predetermined time range as the predetermined training data, or directly upload the predetermined training data.

[0128] In an embodiment, pre-processing the predetermined training data can include removing outliers (such as frequent start-stop, abnormal on-off events, etc.), data denoising and normalization processing of the predetermined training data. Ensuring the quality and consistency of the input data, meeting the needs of subsequent model training.

[0129] In this way, through the hybrid training mechanism of the two models, by taking the mechanism model as one of the features input by the data-driven model, the accuracy of model training can be better improved, the device running mechanism and the characteristics of the training data are consistent, thereby improving the model performance.

[0130] In an embodiment, processing the predetermined training data based on the predetermined mechanism model can include:

[0131] inputting the predetermined training data into the predetermined mechanism model to obtain a plurality of cold machine control processing data;

[0132] outputting a first result representing the performance of the cold machine based on the plurality of cold machine control processing data.

[0133] In an embodiment, the cold machine control processing data can include at least one of: a heat exchange efficiency coefficient CAPFT, a heat balance coefficient EIRFT, and a refrigeration load coefficient EIRFPLR.

[0134] In an embodiment, obtaining the plurality of cold machine control processing data can include: determining a plurality of cold machine running data based on the predetermined training data; and determining the plurality of cold machine control processing data based on the cold machine running data.

[0135] Here, the cold machine running data can include at least one of: a refrigeration total pipe water supply temperature T eo , a refrigeration total pipe return water temperature T ei , a cooling total pipe return water temperature T ci , a cooling total pipe return water temperature T co , a cold machine refrigeration capacity Q e , a cold machine rated refrigeration capacity Q e,ref .

[0136] In an embodiment, determining the plurality of cold machine control processing data based on the cold machine running data can comprise: determining a first calculation coefficient corresponding to each of the cold machine running data based on model fitting;

[0137] determining the plurality of cold machine control processing data based on the cold machine running data and the first calculation coefficient.

[0138] For example, CAPFT= a0+ a1T eo + a2T eo 2 + a3T ci + a4T ci 2 + a5T eo T ci ;

[0139] EIRFT= b0+ b1T ei + b2T ei 2 + b3T co + b4T co 2 + b5T eo T co ;

[0140] EIRFPLR= g0+ g1Q e + g2Q e 2 + g3T co + g4T co 2 + g5Q e T co .

[0141] wherein a0, a1, a2, a3, a4, a5, b0, b1, b2, b3, b4, b5, g0, g1, g2, g3, g4, g5 are the first calculation coefficients.

[0142] In an embodiment, outputting the first result representing the performance of the cold machine based on the plurality of cold machine control processing data can comprise: determining a second calculation coefficient based on model fitting; outputting the first result representing the performance of the cold machine based on the plurality of cold machine control processing data and the second calculation coefficient.

[0143] the first result.

[0144] For example, the outputted first result can be a performance coefficient (COP) of the cold machine, expressed as:

[0145]

[0146] wherein d COPTo calculate the second coefficient.

[0147] In this way, a new performance coefficient expression is introduced in the training process, so that the output of the first

[0148] One result can more accurately and evenly represent the cold performance of different aspects, improving the reliability of the model.

[0149] As a possible implementation, as shown in Figure 4 , a cold machine control method is provided.

[0150] Specifically, the method can include:

[0151] 1. Model training module. The model training module includes data processing, model customization selection, parameter optimization and model training, as shown in Figure 5 To achieve higher prediction accuracy and control effect under different operating conditions, a cold machine model is trained based on a mechanism model and a data-driven model. The specific implementation method is as follows:

[0152] Mechanism model as a constraint condition for data-driven model: when training the data-driven model, the output of the mechanism model can be used as an additional input feature to participate in the training process of the data-driven model. This can make the data-driven model not only rely on historical data, but also follow the physical laws of the device, improving the interpretability and robustness of the model.

[0153] Hybrid model design: by training the mechanism model and the data-driven model in parallel, a hybrid model framework is formed. The mechanism model provides constraints on the physical properties of the system, and the data-driven model supplements the processing of nonlinear features in high-dimensional complex environments. In different operating intervals of the cold machine, the weights of the two can be dynamically adjusted. For example, when the device is close to the rated operating condition, it relies more on the mechanism model; while in complex fluctuating conditions, it relies more on the prediction ability of the data-driven model.

[0154] Parameter optimization and model training: after the model is selected, the hyperparameters of the model are automatically or semi-automatically (user can customize parameter range) adjusted, the model performance is optimized, and the training is performed.

[0155] 2. Model optimization module. The model optimization module mainly scores the model from the three dimensions of accuracy priority, speed priority and computing power priority, as shown in Figure 6 According to the comprehensive score of each model, the model with the highest comprehensive score is automatically selected as the optimal model and output.

[0156] 3. Model updating and deployment module. Ensure that the cold machine system can smoothly realize online updating and deployment of the control model during normal operation, avoiding affecting the overall operation performance of the device due to shutdown or interruption of the control logic. This module includes four main steps, as shown in Figure 7 .

[0157] Through the model deployment module, the cold machine system can successfully complete model update and deployment without affecting the current operation. This not only reduces the complexity of system maintenance, but also improves the flexibility and reliability of cold machine AI control, ensuring that the system can continue to operate efficiently.

[0158] As shown in Figure 8 The cold machine control device provided by the embodiments of the present application can include:

[0159] The determining unit 100 is configured to determine a plurality of candidate models.

[0160] The scoring unit 200 is configured to determine scores of the candidate models based on a current operation scenario and a plurality of scoring dimensions; the scoring dimensions at least include: an accuracy scoring dimension, a speed scoring dimension, and a computing power scoring dimension.

[0161] The execution unit 300 is configured to select a target model from the plurality of candidate models based on the scores, and control the operation of the cold machine based on the target model.

[0162] In some embodiments, the scoring unit 200 is specifically configured to: determine sub-scores of each scoring dimension of the candidate models based on the current operation scenario and at least one scoring indicator corresponding to each scoring dimension; perform normalization processing on the sub-scores; and determine the scores of the candidate models based on the normalized sub-scores and weights corresponding to the scoring dimensions.

[0163] In one possible implementation, the scoring unit 200 is specifically configured to: perform a predetermined operation through the candidate models; and determine sub-scores of each scoring dimension of the candidate models based on the current operation scenario, an operation result of the predetermined operation, and at least one scoring indicator corresponding to each scoring dimension.

[0164] In one possible implementation, the scoring unit 200 is specifically configured to: determine a first sub-score of the accuracy scoring dimension of the candidate models based on the current operation scenario, a data volume corresponding to the predetermined operation, actual values and model predicted values corresponding to an operation result of the predetermined operation, and a first scoring indicator corresponding to the accuracy scoring dimension; determine a second sub-score of the speed scoring dimension of the candidate models based on the current operation scenario, training start and end time and prediction start and end time corresponding to the operation result of the predetermined operation, and a second scoring indicator corresponding to the speed scoring dimension; and determine a third sub-score of the computing power scoring dimension of the candidate models based on the current operation scenario, system performance parameters corresponding to the operation result of the predetermined operation, and a third scoring indicator corresponding to the computing power scoring dimension.

[0165] In a possible implementation, the execution unit 300 is further configured to: determine whether the target model needs to be updated; in response to the target model needing to be updated, determine whether to maintain the current working state of the chiller based on the current working state of the chiller; if it is determined that the current working state is not maintained, control the operation of the chiller based on a predetermined control algorithm, and suspend the control function of the target model; and update the target model.

[0166] In a possible implementation, the execution unit 300 is specifically configured to: update and deploy the target model based on an updated model; determine whether the update and deployment are successful; if the update and deployment fail, restore the control function of the target model and output prompt information; and if the update and deployment are successful, control the operation of the chiller based on the updated target model.

[0167] In a possible implementation, the determination unit 100 is specifically configured to: process predetermined training data based on a predetermined mechanism model; input a first result output by the predetermined mechanism model and the predetermined training data into a predetermined data-driven model; and train based on the first result and a second result output by the predetermined data-driven model to obtain a plurality of candidate models for chiller control.

[0168] In a possible implementation, the determination unit 100 is specifically configured to: input predetermined training data into a predetermined mechanism model to obtain a plurality of chiller control processing data; and output a first result representing chiller performance based on the plurality of chiller control processing data.

[0169] The embodiment of the present application further provides a chiller device, which comprises the chiller control apparatus according to any one or more of the preceding embodiments.

[0170] The embodiment of the present application further provides a heating and ventilation system, which comprises the chiller device according to any one or more of the preceding embodiments.

[0171] As shown in FIG. 1, Figure 9 The embodiment of the present application provides an electronic device, which comprises a processor 111, a communication interface 112, a memory 113 and a communication bus 114, wherein the processor 111, the communication interface 112 and the memory 113 complete communication with each other through the communication bus 114,

[0172] The memory 113 is used to store a computer program.

[0173] In an embodiment of the present application, the processor 111, when executing the program stored in the memory 113, implements the cold machine control method provided by any one or more of the method embodiments described above, at least including: determining a plurality of candidate models; determining scores of the candidate models based on a current running scenario and a plurality of scoring dimensions; the scoring dimensions at least including: an accuracy scoring dimension, a speed scoring dimension, and a computing power scoring dimension; selecting a target model from the plurality of candidate models based on the scores, and controlling the running of the cold machine based on the target model.

[0174] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the steps of the cold machine control method provided by any one or more of the method embodiments described above.

[0175] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments.

[0176] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by means of software plus a general hardware platform, and of course can also be implemented by hardware. Based on such understanding, the above technical solutions or the part that contributes to the related art can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in the various embodiments or some parts of the embodiments.

[0177] It should be understood that the terms used herein are for the purpose of describing particular example embodiments only and are not intended to be limiting. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. The terms "comprises", "comprising", "includes", "including" and "has" are to be construed as containing, comprising, including or having, as set forth herein, and therefore should be understood to be appropriate in the context with the statement being used to avoid the use of the phrase "consisting of" or "consisting essentially of". The method steps, processes, and operations described herein are not to be construed as necessarily requiring their performance in the particular order in which they are described, unless specifically identified as an order dependent step. It is also to be understood that additional or alternative steps can be employed.

[0178] In the case of no contradiction, each step in a certain embodiment or example can be implemented as an independent example, and the steps can be combined arbitrarily, for example, the scheme after removing part of the steps in a certain embodiment or example can also be implemented as an independent example, and the order of the steps in a certain embodiment or example can be exchanged arbitrarily, in addition, the optional mode or optional example in a certain embodiment or example can be combined arbitrarily; in addition, the embodiments or examples can be combined arbitrarily, for example, part or all steps of different embodiments or examples can be combined arbitrarily, a certain embodiment or example can be combined with the optional mode or optional example of other embodiments or examples.

[0179] The above description is merely a specific implementation of the present application, enabling those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features applied herein.

Claims

1. A cold machine control method, characterized by, The method comprises: determining a plurality of candidate models; determining scores of the candidate models based on a current running scenario and a plurality of scoring dimensions; the scoring dimensions at least comprise: an accuracy scoring dimension, a speed scoring dimension, and a computing power scoring dimension; selecting a target model from the plurality of candidate models based on the scores, and controlling running of a chiller based on the target model.

2. The method of claim 1, wherein, The determining of the scores of the candidate models based on the current running scenario and the plurality of scoring dimensions comprises: determining sub-scores of each scoring dimension of the candidate models based on the current running scenario and at least one scoring indicator corresponding to each scoring dimension; normalizing the sub-scores; determining the scores of the candidate models based on the normalized sub-scores and weights corresponding to the scoring dimensions.

3. The method of claim 2, wherein, The determining of the sub-scores of each scoring dimension of the candidate models based on the current running scenario and at least one scoring indicator corresponding to each scoring dimension comprises: executing a predetermined operation by the candidate models; determining the sub-scores of each scoring dimension of the candidate models based on the current running scenario, an operation result of the predetermined operation, and at least one scoring indicator corresponding to each scoring dimension.

4. The method of claim 3, wherein, The determining of the sub-scores of each scoring dimension of the candidate models based on the current running scenario, an operation result of the predetermined operation, and at least one scoring indicator corresponding to each scoring dimension comprises: determining a first sub-score of the accuracy scoring dimension of the candidate models based on the current running scenario, a data volume corresponding to the predetermined operation, actual values and model predicted values corresponding to the operation result of the predetermined operation, and a first scoring indicator corresponding to the accuracy scoring dimension; determining a second sub-score of the speed scoring dimension of the candidate models based on the current running scenario, training start and end time and prediction start and end time corresponding to the operation result of the predetermined operation, and a second scoring indicator corresponding to the speed scoring dimension; determining a third sub-score of the computing power scoring dimension of the candidate models based on the current running scenario, system performance parameters corresponding to the operation result of the predetermined operation, and a third scoring indicator corresponding to the computing power scoring dimension.

5. The method of claim 1, wherein, After the controlling of the running of the chiller based on the target model, the method further comprises: determining whether the target model needs to be updated; in response to the target model needing to be updated, determining whether to maintain a current working state of the chiller based on the current working state; if it is determined that the current working state is not maintained, controlling the running of the chiller based on a predetermined control algorithm, and suspending a control function of the target model; updating the target model.

6. The method of claim 5, wherein, The updating of the target model comprises: updating and deploying the target model based on an updated model; determining whether the updating and deployment is successful; if the updating and deployment fails, resuming the control function of the target model, and outputting prompt information; if the updating and deployment is successful, controlling the running of the chiller based on the updated target model.

7. The method of claim 1, wherein, The determining of the plurality of candidate models comprises: processing predetermined training data based on a predetermined mechanism model; inputting the first result output by the predetermined mechanism model and the predetermined training data into a predetermined data-driven model; training based on the first result and a second result output by the predetermined data-driven model to obtain a plurality of candidate models for cold machine control.

8. The method of claim 7, wherein, The processing of the predetermined training data based on the predetermined mechanism model comprises: inputting the predetermined training data into the predetermined mechanism model to obtain a plurality of cold machine control processing data; outputting a first result representing the performance of the cold machine based on the plurality of cold machine control processing data.

9. A cold machine control device, characterized by, The device comprises: a determination unit configured to determine a plurality of candidate models; a scoring unit configured to determine a score of the candidate models based on a current running scenario and a plurality of scoring dimensions; the scoring dimensions at least include: an accuracy scoring dimension, a speed scoring dimension, and a computing power scoring dimension; an execution unit configured to select a target model from the plurality of candidate models based on the score, and control the running of the cold machine based on the target model.

10. A cold appliance characterized in that The cold machine device comprises the cold machine control device of claim 9.

11. A heating and ventilation system, characterised in that The heating and ventilation system comprises the cold machine device of claim 10.

12. An electronic device, comprising: comprise: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory complete mutual communication through the communication bus; the memory is configured to store a computer program; and the processor is configured to execute the computer program to implement the method of any one of claims 1-8.

13. A storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the method of any one of claims 1-8.

Images