CN118210233A

CN118210233A - Intelligent control method and system for building refrigeration machine room

Info

Publication number: CN118210233A
Application number: CN202410437348.0A
Authority: CN
Inventors: 金惠志; 金础; 蔡志武
Original assignee: Shanghai Guorui Environmental Technology Co ltd
Current assignee: Shanghai Guorui Environmental Technology Co ltd
Priority date: 2024-04-11
Filing date: 2024-04-11
Publication date: 2024-06-18

Abstract

The application provides an intelligent control method and system for a building refrigeration machine room. In the process, part of the first state control factors are converted into the second state control factors, so that the fine description of the running state of the machine room is realized. Further, the control instruction matching is performed by using a machine room control knowledge base, and a control instruction matched with the refrigeration machine room of the target building is generated. Through the intelligent control method, the high-efficiency and accurate control of the building refrigeration machine room can be realized, the operation efficiency of the machine room is improved, the energy consumption is reduced, and the purposes of energy conservation and emission reduction are achieved. Meanwhile, the intelligent control system also has self-adaptability and expandability, can adapt to the control requirements of building refrigeration machine rooms of different scales and types, and provides a new solution for intelligent management of buildings.

Description

Intelligent control method and system for building refrigeration machine room

Technical Field

The application relates to the technical field of intelligent buildings, in particular to an intelligent control method and system for a building refrigeration machine room.

Background

Along with the continuous promotion of building intellectualization level, the building refrigeration computer lab is as the important component of building energy consumption, and its operating efficiency and control accuracy have important meaning to energy saving and emission reduction. The traditional control method of the building refrigeration machine room often depends on fixed control logic and manual operation, and is difficult to adapt to the changes under different environments and load conditions, so that the problems of low energy efficiency, energy consumption, and the like are caused.

In order to solve the above problems, in recent years, related art has begun to explore a method of controlling a building refrigeration room based on data driving and intelligent algorithms. The method learns the operation characteristics and the control rules of the machine room by collecting and analyzing the operation state data of the machine room so as to realize more accurate and efficient control. However, the existing control method still has some challenges when processing complex running state data, such as high data dimension, strong nonlinearity, dynamic change, and the like, which results in poor control effect.

Disclosure of Invention

Therefore, the present application aims to provide a method and a system for intelligent control of a building refrigeration room, which are used for generating an initial state trend conversion vector by extracting target running state chain data of a target building refrigeration room and performing multi-round state trend prediction conversion by combining a predefined control strategy. The predefined control strategy is selected from a plurality of basic control strategies through a cycle strategy learning and updating process, and can be adaptive to the operation characteristics and control requirements of the machine room, so that the high-efficiency and accurate control of the building refrigeration machine room can be realized, the operation efficiency of the machine room is improved, the energy consumption is reduced, a new solution is provided for intelligent management of the building, and the system also has expandability and adaptability, can be adaptive to the control requirements of the building refrigeration machine room with different scales and types, and has wide application prospect and market value.

According to a first aspect of the application, there is provided a smart control method for a building refrigeration room, the method comprising:

Extracting target running state chain data of a target building refrigeration machine room; wherein the target operating state chain data includes a plurality of first state control factors;

the method comprises the steps of obtaining a predefined control strategy, wherein the predefined control strategy is selected from a plurality of final control strategies which are finally generated after a cyclic strategy learning and updating process is executed according to a plurality of basic control strategies and strategy convergence requirements are met; in each strategy learning and updating process, according to a template running state chain data sequence, performing strategy learning and updating on a plurality of basic control strategies in the current strategy learning and updating process to generate a plurality of final control strategies, and if the strategy convergence requirement is not met, using the plurality of final control strategies as a plurality of basic control strategies in the next strategy learning and updating process;

according to the predefined control strategy, carrying out multi-round state trend prediction conversion on the target running state chain data to generate an initial state trend conversion vector; in the process of predicting and converting each state trend, converting part of the first state control factors in the plurality of first state control factors into one second state control factor in the initial state trend conversion vector;

And carrying out control instruction matching in a machine room control knowledge base according to the initial state trend conversion vector to generate a control instruction matching result corresponding to the target building refrigeration machine room.

In a possible implementation manner of the first aspect, the performing policy learning and updating on the plurality of basic control policies of the current policy learning and updating process according to the template running state chain data sequence, to generate a plurality of final control policies includes:

Aiming at the basic control strategies, according to one basic control strategy, converting each template operation state chain data in the template operation state chain data sequence into a corresponding template state trend conversion vector respectively, and determining a strategy utility value corresponding to the one basic control strategy according to each generated template state trend conversion vector;

determining policy attention weights corresponding to the plurality of basic control policies respectively according to policy utility values corresponding to the plurality of basic control policies respectively;

According to the strategy attention weights respectively corresponding to the plurality of basic control strategies, carrying out multi-round selection from the plurality of basic control strategies to generate a plurality of reference control strategies; wherein, each time a basic control strategy is selected from the plurality of basic control strategies as a reference control strategy;

and performing strategy learning on the multiple reference control strategies to generate multiple final control strategies.

In a possible implementation manner of the first aspect, the determining, according to the generated state trend conversion vectors of each template, a policy utility value corresponding to the one basic control policy includes:

Respectively carrying out standardized conversion on the state trend conversion vectors of each template to generate standardized state trend conversion vectors, taking one part of standardized state trend conversion vectors in each standardized state trend conversion vector as a reference vector sequence and the other part of standardized state trend conversion vectors as a tracking vector sequence;

For each reference vector, matching is carried out from the tracking vector sequence according to one reference vector, and a corresponding matching result is generated;

and determining evaluation parameter values of the generated matching results according to the control effect data corresponding to the reference vectors respectively, and taking the evaluation parameter values as strategy utility values of the basic control strategy.

In a possible implementation manner of the first aspect, the determining, according to the control effect data corresponding to each reference vector, an evaluation parameter value of each generated matching result includes:

for each reference vector, determining validity parameters of a matching result corresponding to one reference vector according to control effect data of the reference vector;

And generating the evaluation parameter value according to the validity parameters of the matching results respectively corresponding to the reference vectors.

In a possible implementation manner of the first aspect, the determining, according to the policy utility values corresponding to the plurality of basic control policies respectively, the policy attention weights corresponding to the plurality of basic control policies respectively includes:

And aiming at the plurality of basic control strategies, taking the strategy utility value corresponding to one basic control strategy and the ratio of the sum of the strategy utility values corresponding to the plurality of basic control strategies as the strategy attention weight corresponding to the one basic control strategy.

In a possible implementation manner of the first aspect, the performing policy learning on the plurality of reference control policies, generating the plurality of final control policies includes:

dividing the plurality of reference control strategies into a plurality of strategy combinations, each strategy combination comprising two reference control strategies;

When cross operation is randomly determined to be performed on one strategy combination according to a preset cross weight coefficient aiming at the plurality of strategy combinations, exchanging part of control knowledge vectors in one reference control strategy with corresponding part of control knowledge vectors in the other reference control strategy aiming at two reference control strategies covered by the one strategy combination to generate two reference control strategies after the cross operation;

And generating a plurality of final control strategies according to the plurality of reference control strategies after the cross operation and the plurality of reference control strategies without the cross operation.

In a possible implementation manner of the first aspect, generating the plurality of final control strategies according to the plurality of reference control strategies after the interleaving operation and the plurality of reference control strategies without interleaving operation includes:

for a plurality of reference control strategies after the cross operation and a plurality of reference control strategies without the cross operation, for each evolution factor in one reference control strategy, when the transition operation on the control knowledge vector is randomly determined according to the preset transition probability, the control knowledge vector is converted from an initial state to a final state, and the reference control strategy after the transition operation is generated;

and taking the multiple reference control strategies after the transition operation and the multiple reference control strategies without the transition operation as the multiple final control strategies.

In a possible implementation manner of the first aspect, the method further includes:

extracting reference running state chain data of a reference building refrigeration machine room aiming at each reference building refrigeration machine room covered in the machine room control knowledge base;

converting the reference running state chain data into corresponding reference state trend conversion vectors according to the predefined control strategy;

and carrying out standardized conversion on the reference state trend conversion vector to generate a standardized state trend conversion vector corresponding to the reference building refrigeration machine room.

In a possible implementation manner of the first aspect, the performing control instruction matching in a machine room control knowledge base according to the initial state trend conversion vector, to generate a control instruction matching result corresponding to the target building refrigeration machine room, includes:

Carrying out standardized conversion on the initial state trend conversion vector to generate a standardized state trend conversion vector corresponding to the target building refrigeration machine room;

And according to the standardized state trend conversion vector corresponding to the target building refrigeration machine room, matching the standardized state trend conversion vector corresponding to each reference control instruction set contained in the machine room control knowledge base, and generating a control instruction matching result corresponding to the building refrigeration machine room, wherein the control instruction matching result comprises at least one matched reference control instruction set.

According to a second aspect of the present application, there is provided a building refrigeration room intelligent control system, the building refrigeration room intelligent control system comprising a machine-readable storage medium storing machine-executable instructions and a processor, the processor implementing the aforementioned building refrigeration room intelligent control method when executing the machine-executable instructions.

According to a third aspect of the present application, there is provided a computer-readable storage medium having stored therein computer-executable instructions that, when executed, implement the aforementioned intelligent control method for a building refrigeration room.

According to any one of the aspects, the application has the technical effects that:

According to the embodiment of the application, the running state data of the target building refrigerating machine room is extracted, and the multi-round state trend prediction conversion is carried out by combining with the predefined control strategy, so that the initial state trend conversion vector is generated. In the process, part of the first state control factors are converted into the second state control factors, so that the fine description of the running state of the machine room is realized. Further, the control instruction matching is performed by using a machine room control knowledge base, and a control instruction matched with the refrigeration machine room of the target building is generated. Through the intelligent control method, the high-efficiency and accurate control of the building refrigeration machine room can be realized, the operation efficiency of the machine room is improved, the energy consumption is reduced, and the purposes of energy conservation and emission reduction are achieved. Meanwhile, the intelligent control system also has self-adaptability and expandability, can adapt to the control requirements of building refrigeration machine rooms of different scales and types, and provides a new solution for intelligent management of buildings.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the embodiments will be briefly described below, it being understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and other related drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is a schematic flow chart of a method for intelligent control of a building refrigeration machine room according to an embodiment of the present application.

Fig. 2 is a schematic diagram of a component structure of a building refrigeration room intelligent control system for implementing the building refrigeration room intelligent control method according to an embodiment of the present application.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application, and it should be understood that the accompanying drawings in the present application are for the purpose of illustration and description only, and are not intended to limit the scope of the present application. In addition, it should be understood that the schematic drawings are not drawn to scale. A flowchart, as used in this disclosure, illustrates operations implemented in accordance with some embodiments of the present application. It should be understood that the operations of the flow diagrams may be implemented out of order and that steps without logical context may be performed in reverse order or concurrently. In addition, one skilled in the art, under the direction of the present disclosure, may add a plurality of other operations to the flowchart, or may destroy a plurality of operations from the flowchart.

Fig. 1 is a schematic flow chart of a building refrigeration room intelligent control method and system according to an embodiment of the present application, and it should be understood that in other embodiments, the order of part of the steps in the building refrigeration room intelligent control method according to the present application may be shared with each other according to actual needs, or part of the steps may be omitted or maintained. The intelligent control method for the building refrigeration machine room comprises the following detailed steps:

And step S110, extracting target running state chain data of a target building refrigeration machine room. Wherein the target operating state chain data includes a plurality of first state control factors.

In detail, the target building refrigeration room may refer to a refrigeration room of a specific building requiring operation state monitoring and control, and is an important component of the interior of the building, responsible for providing cooling capacity to maintain a comfortable environment of the building. For example, a refrigerating machine room is arranged in a basement of an office building, equipment such as a water chilling unit, a cooling water pump, a chilled water pump, a pipeline system and a control system are arranged in the machine room, and a cooling tower is arranged outdoors to jointly form a refrigerating system.

The target operation state chain data may be a series of continuous data sets generated in the operation process of the refrigeration machine room and describing the operation state of the refrigeration machine room, and the data sets are arranged in time sequence to form a chain, and reflect the operation state change of the machine room from one moment to the other moment. For example, the target operating state chain data may include measurements of chilled and chilled water temperatures, pressures, flows, etc. that are collected in real time by sensors installed in the machine room and transmitted to a database of the intelligent control system of the building refrigeration machine room.

In the target operation state chain data, the first state control factors refer to parameters or variables which can directly influence the operation state of the refrigeration machine room, and are usually key factors to be monitored and controlled. For example, the first state control factor may include an inlet temperature of the cooling water, a fan speed of the cooling tower, a compressor operating frequency of the chiller, and the like. These changes in the first state control factor directly affect refrigeration efficiency, energy consumption, and equipment life.

Thus, in this embodiment, the intelligent control system of the building refrigeration machine room is used as a server, and is first connected to the monitoring system of the target building refrigeration machine room, where the monitoring system records the running state data of the machine room in real time, for example, a plurality of parameters including the temperature of the chilled water and the cooling water, the flow rate of the cooling water, the fan rotation speed of the cooling tower, the running state of the compressor, and the like, and these parameters form a running state chain of the machine room. The server extracts the state chain data from the monitoring system at regular time or in real time through a preset data interface, and stores the state chain data in a memory or a database of the server for subsequent processing.

After extracting the original state chain data, the server may perform a series of data preprocessing operations to ensure accuracy and availability of the data. These operations may include data cleansing (e.g., outlier removal, missing value filling), data transformation (e.g., unity, data normalization), and data aggregation (e.g., calculating statistics of average, maximum, minimum, etc.). The preprocessed data is more suitable for subsequent strategy learning and state trend prediction.

Step S120, a predefined control strategy is obtained, wherein the predefined control strategy is to execute a cyclic strategy learning and updating process according to a plurality of basic control strategies, and is selected from a plurality of final control strategies finally generated after meeting the strategy convergence requirement. In each strategy learning and updating process, according to a template running state chain data sequence, strategy learning and updating are carried out on a plurality of basic control strategies in the current strategy learning and updating process, a plurality of final control strategies are generated, and if the strategy convergence requirement is not met, the plurality of final control strategies are used as a plurality of basic control strategies in the next strategy learning and updating process.

In detail, the predefined control strategy refers to a set of control logic or method defined in advance according to a specific algorithm or rule in the control process of the refrigeration machine room, and can be used for guiding how to adjust the control parameters according to the actual operation state of the machine room to achieve the expected operation target. For example, the predefined control strategy may be a temperature control strategy based on a PID algorithm, or may be a complex control strategy based on an intelligent algorithm such as fuzzy logic or a neural network. These control strategies typically dynamically adjust the operating parameters of the apparatus based on real-time collected operating state data, such as adjusting the rotational speed of the cooling water pump to control the circulating flow of cooling water. The basic control strategy refers to simple or basic control logic or methods as initial inputs or references during the formation of the predefined control strategy, which may not be directly applied to the actual control process, but provide a basis or starting point for a more complex control strategy. For example, the base control strategy may be a manual control strategy based on empirical rules, such as "when the indoor temperature is above 26 ℃, the chiller is turned on; and when the indoor temperature is lower than 24 ℃, the water chilling unit is turned off. These simple rules, while perhaps not adaptable to all complex and varying operating conditions, provide an initial framework or source of inspiration for higher level automated control strategies.

The loop strategy learning and updating process refers to an iterative optimization method that gradually improves the control effect by continuously repeating the steps of learning, evaluating and updating the control strategy. In this process, the current control strategy is evaluated and adjusted and optimized accordingly. For example, the server may evaluate the performance of the current control strategy based on the real-time collected target operating state chain data and the historical control effect record, and automatically adjust the set values of the control parameters or change the control logic to improve the future control effect based on the evaluation result. This process of evaluating, adjusting, and re-evaluating is repeated until a satisfactory control effect is achieved or certain termination conditions are met. During this cycle, the control strategy of each stage will be the basis for learning and updating of the next stage.

The policy convergence requirement refers to a set of conditions or criteria set during the loop policy learning and updating process, which are used to determine whether the current control policy has reached a sufficiently good performance level, and may stop further iterative optimization. These requirements typically include indicators of control accuracy, stability, energy consumption, and the like. For example, the policy convergence requirement may include a specific indicator such as "the decrease in control error is less than 1% after three consecutive iterations" or "the total energy consumption decreases below a certain preset value". When these requirements are met, the server will stop further iterative optimization and output the current control strategy as the final selected predefined control strategy to the actuator for execution. If the convergence requirement is not met, iterative optimization needs to be continued until the condition is met.

Thus, in this embodiment, the server may incorporate a policy learning module that is based on reinforcement learning or other machine learning algorithms. In an initial stage, a plurality of basic control strategies are defined in the strategy learning module, and the basic control strategies can be formulated based on expert experience, historical data or simple control logic. The server uses these basic control strategies in combination with the template run state chain data sequence (i.e., historical or simulated run state data) to perform multiple rounds of strategy learning and updating processes.

During each round of strategy learning and updating, the server evaluates the performance of each basic control strategy according to the current control effect (such as energy consumption, temperature stability and other indexes), and adjusts or optimizes the strategy according to the evaluation result. This process may involve adjustment of parameters, modification of rules, generation of new policies, etc.

When the policy convergence requirement is met (e.g., the difference between the learning results of consecutive rounds is less than a certain threshold, or a predetermined number of learning rounds is reached), the server selects a policy that performs best from the plurality of control policies that are finally generated as the predefined control policy. If the convergence requirement is not met, the plurality of control strategies generated by the current round are used as the input of the next round of strategy learning and updating process.

Step S130, performing multiple rounds of state trend prediction conversion on the target running state chain data according to the predefined control strategy, and generating an initial state trend conversion vector. And in the process of predicting and converting each state trend, converting part of the first state control factors in the plurality of first state control factors into one second state control factor in the initial state trend conversion vector.

In detail, the state trend prediction conversion refers to processing the target operation state chain data by utilizing a predefined control strategy to predict the possible state change trend of the refrigerating machine room in the future. Such conversion typically involves analysis of the data, application of models, and calculation of algorithms intended to convert current state data into predictions of future states. For example, assuming that the temperature in the current building room is rising, the server predicts that the temperature will continue to rise and possibly exceed the comfort range if no action is taken, based on predefined control strategies and historical data. This predictive process is a state trend predictive conversion that converts current temperature data into predictions of future temperature changes.

The second state control factor is generated in the state trend prediction conversion process and is used for describing a predicted parameter or variable of the future state trend of the refrigeration machine room. These factors are typically calculated based on the current first state control factor and a predefined control strategy. For example, in the above example, the predicted future temperature may be considered a second state control factor. It is calculated based on the current first state control factor (such as current temperature, outdoor temperature, etc.) and a predefined control strategy (such as temperature control algorithm) for describing future state trends of the refrigeration machine room.

Thus, in this embodiment, once the predefined control strategy is obtained, the server will use the predefined control strategy to make multiple rounds of state trend predictive transitions on the extracted target operating state chain data. In each cycle of switching, the server will convert a portion of the first state control factors (e.g., current temperature, humidity, etc.) in the state chain data into second state control factors (e.g., predicted temperature, predicted humidity, etc.) that predict the indoor operating state of the building for a period of time in the future, according to rules and logic in the predefined control strategy.

Through the multiple rounds of switching, the server is able to generate an initial state trend switching vector containing a plurality of second state control factors, the initial state trend switching vector describing an expected operating state of the machine room over a period of time in the future.

In an alternative embodiment, the present embodiment may set the number of rounds (e.g., N rounds) of state trend predictive conversion, and prepare a first list of state control factors required per round of conversion.

For each of the rounds of conversion,

A. Selecting a first state control factor: a partial factor is selected from the plurality of first state control factors as an input to the present round of conversion. The selection of these factors may be based on rules or algorithms in a predefined control strategy.

B. Applying a predictive model: the selected first state control factor is converted to a second state control factor using a predefined predictive model (e.g., a time series analysis model, a machine learning model, etc.). This conversion process may involve mathematical operations, statistical analysis, or model reasoning of the data, etc.

C. Updating the state trend transition vector: and adding the second state control factor obtained by conversion into the initial state trend conversion vector. As each conversion takes place, this vector will be refined and contain more information about the trend of the machine room status.

Judging the end condition: when the set number of conversion rounds is reached, the loop execution is ended. At this point, the initial state trend transition vector already contains the results of the multi-round predictive transitions.

And finally, integrating and formatting the results obtained by the multi-round state trend prediction conversion to generate a complete initial state trend conversion vector. This initial state trend transition vector will serve as the basis for subsequent control instruction matching and machine room operation optimization. The initial state trend translation vector is then stored in a database of the server and transmitted to other related systems or modules for further processing and analysis as needed.

And step 140, performing control instruction matching in a machine room control knowledge base according to the initial state trend conversion vector, and generating a control instruction matching result corresponding to the target building refrigeration machine room.

In detail, the machine room control knowledge base is a database or knowledge management system that stores a large number of control instructions, rules, policies, and related knowledge. These knowledge are accumulated based on expert experience, historical data, best practices, etc., for guiding the control operations of the refrigeration room. For example, the machine room control knowledge base may include various control instructions, such as "when the indoor temperature exceeds 28 ℃, the second chiller is turned on" or "when the outdoor humidity is lower than 30%, the humidification device is turned off". These control instructions are based on understanding of the operation rules of the refrigeration machine room and summary of long-term practical experience, and aim to realize efficient, stable and safe operation of the machine room.

The control instruction matching refers to a process of searching and selecting a control instruction applicable to the current refrigeration machine room state in a machine room control knowledge base. This process typically involves the analysis of the initial state trend transition vectors, the comparison with control instructions in the knowledge base, and the use of matching algorithms. For example, after the server generates the initial state trend translation vector, it searches the machine room control knowledge base for a control command matching the initial state trend translation vector. For example, if the initial state trend transition vector indicates that the temperature is rising and may exceed the comfort range, the server may find and execute a control command in the knowledge base to "turn on the chiller to reduce the indoor temperature". This matching and execution process is an implementation of control instruction matching.

That is, in this embodiment, the server has a machine room control knowledge base, and the machine room control knowledge base stores a large amount of control instructions and information such as applicable scenes and effect evaluations corresponding to the control instructions. These control commands may be specific operations of turning on or off a certain device, adjusting device parameters, switching operation modes, etc. After the initial state trend conversion vector is generated, the server can match control instructions in a machine room control knowledge base. The matching process may include steps of calculating a similarity between the initial state trend transition vector and a scene in which each control instruction in the knowledge base is applicable, evaluating an improvement effect of each control instruction on an expected operation state, and the like. Finally, the server generates one or more control instruction matching results according to the matching results, and the control instruction matching results are used as the basis for the actual control of the subsequent machine room. The control instruction can be sent to the control system of the machine room through the output interface of the server, so that the real-time adjustment and optimization of the running state of the machine room are realized.

Based on the steps, the embodiment of the application extracts the running state data of the target building refrigeration machine room and performs multi-round state trend prediction conversion by combining with a predefined control strategy so as to generate an initial state trend conversion vector. In the process, part of the first state control factors are converted into the second state control factors, so that the fine description of the running state of the machine room is realized. Further, the control instruction matching is performed by using a machine room control knowledge base, and a control instruction matched with the refrigeration machine room of the target building is generated. Through the intelligent control method, the high-efficiency and accurate control of the building refrigeration machine room can be realized, the operation efficiency of the machine room is improved, the energy consumption is reduced, and the purposes of energy conservation and emission reduction are achieved. Meanwhile, the intelligent control system also has self-adaptability and expandability, can adapt to the control requirements of building refrigeration machine rooms of different scales and types, and provides a new solution for intelligent management of buildings.

In one possible implementation, step S120 may include:

step S121, for the plurality of basic control policies, converting each template running state chain data in the sequence of template running state chain data into a corresponding template state trend conversion vector according to one basic control policy, and determining a policy utility value corresponding to the one basic control policy according to each generated template state trend conversion vector.

In this embodiment, the server first obtains a sequence of template running state chain data of the refrigeration machine room, where the data are records of the running state of the machine room at different time points. Next, the server selects one of a plurality of basic control strategies, such as a "time period based temperature control strategy". According to this strategy, the server converts each piece of data in the sequence of template run state chain data into a template state trend conversion vector describing a future state trend. For example, a piece of data may show that the temperature continues to rise for a certain period of time, and according to the underlying control strategy, the server predicts that the temperature will continue to rise if no action is taken, thus generating a template state trend transition vector representing the trend of the temperature rise.

Then, the server evaluates the utility of the basic control strategy in coping with different state trends according to the generated state trend conversion vectors of the templates. The utility value may be a value that indicates the effectiveness of the strategy in achieving the control objective (e.g., maintaining the temperature within a comfort range). For example, if a strategy is effective in controlling temperature within a target range over multiple simulations, its utility value will be high.

Step S122, determining the policy attention weights corresponding to the plurality of basic control policies according to the policy utility values corresponding to the plurality of basic control policies.

In this embodiment, the server assigns attention weights to the plurality of basic control policies calculated in the previous step according to their policy utility values. The attention weight reflects the importance of the server to each basic control strategy in the subsequent strategy selection process. For example, if a basic control strategy would in most cases exhibit a higher utility value, it would be given a higher attention weight. Conversely, if a policy performs poorly in most cases, its attention is weighted lower.

Step S123, performing multiple rounds of selection from the multiple basic control strategies according to the strategy attention weights respectively corresponding to the multiple basic control strategies, so as to generate multiple reference control strategies. Wherein one basic control strategy is selected as a reference control strategy at a time from the plurality of basic control strategies.

In this embodiment, the server performs a multi-round selection process according to the attention weights of the respective basic control policies. In each round of selection, the server randomly or according to a certain rule selects a basic control strategy as a reference control strategy according to the weight. This process may be repeated multiple times to generate multiple reference control strategies. These reference control strategies are the subject of the server's focus in the subsequent strategy learning process.

And step S124, performing strategy learning on the plurality of reference control strategies to generate a plurality of final control strategies.

In this embodiment, the server further learns the multiple reference control policies generated in the previous step. In this process, the server may utilize a machine learning algorithm, an optimization algorithm, or other intelligent techniques to adjust and optimize the reference control strategy to generate a final control strategy that is more adapted to the current refrigeration room operating state. For example, the server may test the effects of different control strategies through simulation experiments, and adjust the strategies according to the experimental results until a set of final control strategies capable of effectively maintaining stable operation of the machine room under various conditions is found.

In one possible implementation, step S121 may include:

Step S1211, performing normalized conversion on the state trend conversion vectors of the templates, generating normalized state trend conversion vectors, and taking a part of normalized state trend conversion vectors in the normalized state trend conversion vectors as a reference vector sequence and another part of normalized state trend conversion vectors as a tracking vector sequence.

In this embodiment, the server first processes the state trend conversion vectors of each template and performs standardized conversion on the state trend conversion vectors. The purpose of the normalized conversion is to eliminate the dimension and scale differences between the different data so that the various state trend conversion vectors are compared and analyzed on the same scale. For example, the state trend transition vector may contain data in multiple dimensions of temperature, humidity, pressure, etc., and the numerical range and units for each dimension may be different. Through standardized conversion, the server may convert these data into a uniform range of values, such as between 0 and 1 for all values.

After the normalization conversion, the server generates each normalized state trend conversion vector. These normalized vectors have the same scale and can be used directly for subsequent comparison and matching operations.

Next, the server classifies the normalized state trend transition vectors into two groups: a reference vector sequence and a tracking vector sequence. The sequence of reference vectors is used as a reference criterion for evaluating the effect of the underlying control strategy. The tracking vector sequence is used for simulating the state change in the actual running process.

For example, the server may select a portion of the normalized vectors as a sequence of reference vectors that represent the operational trends of the room under ideal conditions. The server then takes the remaining normalized vectors as a sequence of tracking vectors that simulate the various state changes that the machine room may encounter during actual control.

Step S1212, for each reference vector, performs matching from the tracking vector sequence according to one reference vector, and generates a corresponding matching result.

In this embodiment, the server performs a matching operation in the tracking vector sequence for each reference vector. The purpose of the matching is to find the tracking vector that is most similar to the reference vector to evaluate the effect of the control strategy in actual operation. The matching process may employ different algorithms and metrics such as calculating the distance between vectors, similarity, etc.

For example, the server may select a reference vector and then calculate its distance from each vector in the sequence of tracking vectors. The tracking vector with the smallest distance is considered to be the most similar to the reference vector and is therefore selected as the matching result. This process is repeated for each reference vector until a match is found for all reference vectors.

Step S1213, determining an evaluation parameter value of each generated matching result according to the control effect data corresponding to each reference vector, and taking the evaluation parameter value as a policy utility value of the one basic control policy.

Finally, the server determines the evaluation parameter value of each matching result according to the control effect data corresponding to each reference vector. The control effect data may be predefined evaluation indicators such as temperature deviation, energy consumption, etc. The evaluation parameter value reflects the extent to which the control strategy achieves the desired effect in actual operation.

For example, if a reference vector represents an ideal temperature control trend, and the corresponding matching result shows that the temperature deviation is large in actual operation, the evaluation parameter value is low. The server calculates the evaluation parameter values for each matching result and takes these values as the policy utility values for the current underlying control policy. This utility value will be used in subsequent policy learning and optimization processes.

In one possible implementation, step S1213 may include:

step S1213-1, for each reference vector, determining a validity parameter of a matching result corresponding to one reference vector according to control effect data of the one reference vector.

In this embodiment, the server now processes the reference vectors and determines validity parameters of the matching result for each reference vector from its control effect data. The control effect data is typically pre-collected for quantifying the effect of the control strategy in actual applications. These data may include key indicators of percent energy consumption reduction, temperature maintenance stability, etc.

Taking server room temperature control as an example, a reference vector may represent a desired room temperature drop profile. The server matches the actual temperature data in the sequence of tracking vectors according to the curve, and finds a closest tracking vector as a matching result. The server then evaluates the control effect of this matching result, such as whether the actual temperature drops smoothly according to the expected curve, whether out-of-range fluctuations occur, etc.

Based on these control effect data, the server calculates a validity parameter for the matching result of each reference vector. The validity parameter may be a numerical score reflecting how close the matching result is to the desired control effect. For example, if the actual temperature profile is very consistent with the desired profile, without large deviations or fluctuations, the validity parameter will be high; conversely, if the actual temperature profile is far from the desired profile, and frequent out-of-range conditions occur, the validity parameter will be low.

Step S1213-2, generating the evaluation parameter value according to the validity parameters of the matching results corresponding to the reference vectors.

In this embodiment, after collecting validity parameters of the matching results of all reference vectors, the server further processes these data to generate an evaluation parameter value. This evaluation parameter value is a comprehensive quantitative evaluation of the effect of the underlying control strategy.

In particular, the server may integrate these validity parameters into one evaluation parameter value in different ways. For example, the server may calculate an average value, a weighted average value, a minimum value, a maximum value, or the like of all the validity parameters as a whole evaluation parameter value. This value will reflect the overall behavior of the control strategy in handling the various room conditions.

The server may also consider the importance or weight of different reference vectors during the integration process. Some reference vectors may represent more critical or common machine room states and should therefore be given higher weight when calculating the evaluation parameter values. In this way, the server can more accurately evaluate the comprehensive effect of the control strategy in practical application.

Finally, the evaluation parameter value obtained by the server is used as a basis for measuring the quality of the basic control strategy, and an important reference is provided for subsequent strategy learning, updating and optimizing.

In one possible implementation, step S122 may include: and aiming at the plurality of basic control strategies, taking the strategy utility value corresponding to one basic control strategy and the ratio of the sum of the strategy utility values corresponding to the plurality of basic control strategies as the strategy attention weight corresponding to the one basic control strategy.

In this embodiment, the server has calculated the policy utility values corresponding to the plurality of basic control policies, respectively. These utility values reflect the effect of different control strategies in dealing with a particular problem. Now, the server needs to determine the policy focus weight for each underlying control policy based on these utility values.

Assume that the server has three basic control policies A, B and C, which correspond to policy utility values U_ A, U _B and U_C, respectively. The server first calculates the sum of the three utility values, i.e., u_total=u_a+u_b+u_c.

Then, the server regards, for each basic control policy, the ratio of the corresponding policy utility value to the total utility value as the policy attention weight of the policy. Specifically:

For the basic control strategy a, its strategy focuses on the weight w_a=u_a/u_total.

For the base control strategy B, its strategy focuses on the weights w_b=u_b/u_total.

For the base control strategy C, its strategy focuses on the weights w_c=u_c/u_total.

Thus, the server gets the policy focus weight for each underlying control policy. These weights reflect the importance or priority of the different control strategies in solving the problem. Higher weighted strategies will receive more attention and importance in subsequent strategy learning and updating processes.

In this way, the server can dynamically adjust the degree of interest in the policies based on their actual performance, thereby more effectively performing policy learning and optimization. This is of great importance for dealing with complex, variable practical problems.

In one possible implementation, step S124 may include:

In step S1241, the plurality of reference control strategies are divided into a plurality of strategy combinations, each strategy combination containing two reference control strategies.

In this embodiment, for example, the server has six reference control strategies, labeled S1, S2, S3, S4, S5, and S6. The server divides these policies into a plurality of policy combinations, each combination containing two reference control policies. Possible partitioning modes are: (S1, S2), (S3, S4), (S5, S6), and the like. In this way, the server obtains a plurality of combinations of policies, each containing two policies that can be learned and crossed with each other.

Step S1242, for the plurality of policy combinations, when determining that a cross operation is performed on one policy combination according to a preset cross weight coefficient, exchanging part of control knowledge vectors in one reference control policy with corresponding part of control knowledge vectors in another reference control policy for two reference control policies covered by the one policy combination, so as to generate two reference control policies after the cross operation.

In this embodiment, before policy learning, the server presets a cross weight coefficient, which is used to determine the probability of performing a cross operation. The interleaving operation refers to interchanging part of control knowledge vectors in two reference control strategies to generate a new strategy.

Assuming that the cross weight coefficient is set to 0.5, this means that at each iteration the server has a 50% probability of selecting a policy combination to cross. The server may randomly select a policy combination, such as the combination (S1, S2) selected.

Once the server has determined to cross-operate on a certain policy combination, it will begin performing specific cross-over steps. The interleaving operation typically involves the interchange of control knowledge vectors.

Taking (S1, S2) as an example, it is assumed that S1 and S2 each contain a series of control parameters and rules. The server randomly selects a part of the control knowledge vectors (e.g. parameters related to temperature control) in S1 and then exchanges with the corresponding part of the control knowledge vectors (also parameters related to temperature control) in S2. Thus, S1 and S2 obtain a part of control knowledge of each other, respectively, and new strategies S1 'and S2' after the crossover operation are generated.

Step S1243, generating the final control policies according to the reference control policies after the interleaving operation and the reference control policies without interleaving operation.

After multiple iterations and interleaving operations, the server obtains a series of interleaving operation reference control strategies and original reference control strategies without interleaving operation. Together, these policies constitute the final policy repository.

The server may further evaluate and optimize these policies and select the ones that perform best as the final control policies. These final control strategies will be applied in the actual control system to achieve efficient management and control of the target system.

In one possible embodiment, step S1243 may include:

Step S1243-1, for a plurality of reference control strategies after the cross operation and a plurality of reference control strategies without the cross operation, for each evolution factor in one reference control strategy, when the transition operation on the control knowledge vector is randomly determined according to the preset transition probability, the control knowledge vector is converted from an initial state to a final state, and the reference control strategy after the transition operation is generated.

In this embodiment, the server now has multiple reference control policies after the interleaving operation and multiple reference control policies without interleaving operation. To further enhance the diversity and adaptability of these strategies, the server will make transition operations on them.

The transition operation refers to randomly selecting an evolution factor (such as a control parameter, a rule or a part of an algorithm) in a certain reference control strategy according to a preset transition probability, and converting a corresponding control knowledge vector from an initial state to a final state. Such a transition may be a fine-tuning in value, a change in logic rules, an adjustment in the structure of an algorithm, or the like.

Taking a specific reference control strategy S1 as an example, it contains an evolution factor F1 regarding temperature control. The server first checks the pre-set transition probability, e.g. set to 0.3. This means that there is a 30% probability that F1 will be selected for change when the transition operation is performed. If the server randomly determines that a transition operation is to be performed on F1, it adjusts the control knowledge vector corresponding to F1 according to some rule or algorithm (e.g., random perturbation, mutation operation in genetic algorithm, etc.). For example, the control parameter of F1 is 25 ℃, and after the transition operation, the parameter may be adjusted to 26 ℃ or 24 ℃.

Through such transition operations, the server can introduce new variations and possibilities for each reference control strategy, thereby increasing the diversity and innovativeness of the strategy library.

Step S1243-2, a plurality of reference control strategies after the transition operation and a plurality of reference control strategies without the transition operation are used as the plurality of final control strategies.

After the transition operation, the server obtains a series of reference control strategies after transition and an original reference control strategy without transition operation. Together, these policies constitute the final policy repository.

The server may further evaluate and optimize these policies and select the ones that perform best as the final control policies. The evaluation basis can be indexes such as performance, stability, energy consumption and the like of the strategy in an actual control system. By comparing and analyzing these metrics, the server can pick the final control strategy that is most appropriate for the current control task.

These final control strategies will be applied in the actual control system to achieve efficient management and control of the target system. For example, in a temperature control task of a server room, the server may select a final control strategy that can stably maintain the room temperature within a set range. Thus, the equipment in the machine room can be ensured to run under good environmental conditions, and the reliability and performance of the system are improved.

In one possible embodiment, the method further comprises:

and step A110, extracting reference operation state chain data of one reference building refrigeration machine room aiming at each reference building refrigeration machine room covered in the machine room control knowledge base.

In this embodiment, the server first accesses a room control knowledge base, and the access room control knowledge base stores operation state data of a plurality of reference building refrigeration rooms. Each reference building refrigeration room has a series of operating state chain data that record the operating state of the room at various points in time, such as temperature, flow, equipment load, etc.

The server selects one of the reference building refrigeration rooms and extracts its reference operating state chain data. These data may exist in time series, one state record for each point in time. For example, the server may extract hourly operational status data of a certain machine room over the past 24 hours, including parameters such as indoor temperature, outdoor temperature, cooling water temperature, equipment current, etc.

Step a120, converting the reference running state chain data into a corresponding reference state trend conversion vector according to the predefined control strategy.

The server then converts the extracted reference operating state chain data into a reference state trend conversion vector according to a predefined control strategy. This conversion process aims to convert the raw operating state data into a form that is easier to analyze and process.

The predefined control strategy may define a series of data processing rules and algorithms for extracting key information from the raw data and generating state trend transition vectors. For example, the server may calculate the amount of change in the state parameter between adjacent points in time and combine these amounts of change into a vector. This vector represents the trend of the machine room status.

Taking the indoor temperature as an example, the server may calculate the amount of change in the temperature in each hour (the current hour temperature minus the previous hour temperature) in the indoor space and combine these amounts of change into one temperature trend conversion vector. Similarly, the server may also generate corresponding trend transition vectors for other state parameters.

And step A130, carrying out standardized conversion on the reference state trend conversion vector to generate a standardized state trend conversion vector corresponding to the reference building refrigeration machine room.

In order to facilitate comparison and subsequent processing between different machine rooms, the server needs to perform standardized conversion on the reference state trend conversion vector. The purpose of the standardized transformation is to transform data of different ranges and units onto a uniform scale.

The server may employ some common normalization methods, such as min-max normalization or Z-score normalization. In min-max normalization, the server scales the amount of change in each state parameter to a fixed range (e.g., between 0 and 1). In the Z-score normalization, the server then calculates a standard score for each variation, i.e., a standard deviation multiple of the variation from the average.

After standardized conversion, the server obtains a standardized state trend conversion vector corresponding to the reference building refrigeration machine room. This vector can be used in subsequent strategy learning, optimization and control operations.

In one possible implementation, step S140 may include:

And step S141, carrying out standardized conversion on the initial state trend conversion vector to generate a standardized state trend conversion vector corresponding to the target building refrigeration machine room.

In this embodiment, the server has obtained an initial state trend conversion vector of the target building refrigeration machine room, and this vector reflects the trend of the current state of the machine room. However, direct comparison and matching may lead to inaccurate results, as there may be differences in the status parameter ranges and units of the different machine rooms. Therefore, the server needs to perform normalized conversion on the initial state trend conversion vector.

The purpose of the normalized conversion is to convert the raw data to a uniform scale for comparison and analysis. The server may employ a min-max normalization approach to scale the amount of change in each state parameter to a fixed range (e.g., between 0 and 1). Thus, the state trend conversion vectors of different machine rooms can be compared on the same scale.

By executing the standardized conversion algorithm, the server converts each element in the initial state trend conversion vector into a standardized value, thereby generating a standardized state trend conversion vector corresponding to the target building refrigeration machine room.

Step S142, according to the standardized state trend conversion vector corresponding to the target building refrigeration machine room, matching the standardized state trend conversion vector corresponding to each reference control instruction set contained in the machine room control knowledge base, and generating a control instruction matching result corresponding to the building refrigeration machine room, where the control instruction matching result includes at least one matched reference control instruction set.

After the standardized conversion is completed, the server starts to match control instructions in a machine room control knowledge base according to the standardized state trend conversion vector corresponding to the target building refrigeration machine room. A plurality of reference control instruction sets are stored in the machine room control knowledge base, and each set corresponds to a standardized state trend conversion vector.

In detail, the server adopts a certain matching algorithm (such as cosine similarity matching or euclidean distance matching, etc.) to compare the standardized state trend conversion vector of the target machine room with the standardized state trend conversion vector corresponding to each reference control instruction set in the knowledge base. By calculating indexes such as similarity or distance, the server can evaluate which reference control instruction set is most matched with the current state of the target machine room.

Through the matching process, the server generates a control instruction matching result corresponding to the building refrigeration machine room. This result may include one or more matching sets of reference control instructions, each set containing a series of control instructions for the equipment in the machine room. These control instructions can be used to adjust the operating conditions of the machine room to achieve a more efficient and stable cooling effect.

The intelligent control system 100 for a building refrigeration room shown in fig. 2 includes: a processor 1001 and a memory 1003. The processor 1001 is coupled to the memory 1003, such as via a bus 1002. Optionally, the intelligent control system 100 of the building refrigeration room may further include a transceiver 1004, where the transceiver 1004 may be used for data interaction between the server and other servers, such as data transmission and/or data reception, etc. It should be noted that, the transceiver 1004 is not limited to one embodiment in actual dispatching, and the structure of the intelligent control system 100 for a building refrigeration room is not limited to the embodiment of the present application.

The Processor 1001 may be a CPU (Central Processing Unit ), general purpose Processor, DSP (DIGITAL SIGNAL Processor, data signal Processor), ASIC (Application SpecificIntegrated Circuit ), FPGA (Field Programmable GATE ARRAY, field programmable gate array) or other programmable logic device, transistor logic device, hardware component, or any combination thereof. Which may implement or perform the various exemplary logic blocks, modules and circuits described in connection with this disclosure. The processor 1001 may also be a combination that implements computing functionality, such as a combination comprising one or more microprocessors, a combination of a DSP and a microprocessor, or the like.

Bus 1002 may include a path to transfer information between the components. Bus 1002 may be a PCI (PERIPHERAL COMPONENT INTERCONNECT, peripheral component interconnect standard) bus, or an EISA (ExtendedIndustry Standard Architecture ) bus, or the like. The bus 1002 may be divided into an address bus, a data bus, a control bus, and the like. For ease of illustration, only one thick line is shown in fig. 2, but not only one bus or one type of bus.

The Memory 1003 may be, but is not limited to, ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, RAM (Random Access Memory ) or other type of dynamic storage device that can store information and instructions, EEPROM (ELECTRICALLY ERASABLEPROGRAMMABLE READ ONLY MEMORY ), CD-ROM (Compact DiscRead Only Memory, compact disc Read Only Memory) or other optical disk storage, optical disk storage (including compact discs, laser discs, optical discs, digital versatile discs, blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium that can be used to carry or store program code and that can be Read by a computer.

The memory 1003 is used for storing program codes for executing the embodiments of the present application and is controlled to be executed by the processor 1001. The processor 1001 is configured to execute the program code stored in the memory 1003 to implement the steps shown in the foregoing method embodiment.

Embodiments of the present application provide a computer readable storage medium having program code stored thereon, which when executed by a processor, implements the steps of the foregoing method embodiments and corresponding content.

It should be understood that, although various operation steps are indicated by arrows in the flowcharts of the embodiments of the present application, the order in which these steps are implemented is not limited to the order indicated by the arrows. In some implementations of embodiments of the application, the implementation steps in the flowcharts may be performed in other orders based on demand, unless explicitly stated herein. Furthermore, some or all of the steps in the flowcharts may include a plurality of sub-steps or a plurality of stages, depending on the actual implementation scenario. Some or all of these sub-steps or phases may be performed at the same time, or each of these sub-steps or phases may be performed at different times, respectively. In the case of different execution timings, the execution order of the sub-steps or stages may be flexibly configured based on requirements, which is not limited by the embodiment of the present application.

The foregoing is merely an optional implementation manner of some of the implementation scenarios of the present application, and it should be noted that, for those skilled in the art, other similar implementation manners according to the technical idea of the present application may be adopted without departing from the technical idea of the solution of the present application, which is also within the protection scope of the embodiments of the present application.

Claims

1. An intelligent control method for a building refrigeration machine room is characterized by comprising the following steps:

2. The intelligent control method of a building refrigeration machine room according to claim 1, wherein the performing policy learning and updating on a plurality of basic control policies in the current policy learning and updating process according to the template operation state chain data sequence to generate a plurality of final control policies comprises:

3. The intelligent control method of a building refrigeration machine room according to claim 2, wherein determining the policy utility value corresponding to the one basic control policy according to the generated state trend conversion vectors of the templates comprises:

4. The intelligent control method of the building refrigeration machine room according to claim 3, wherein determining the evaluation parameter value of each generated matching result according to the control effect data corresponding to each reference vector respectively comprises:

5. The intelligent control method of a building refrigeration machine room according to claim 2, wherein determining the policy attention weights respectively corresponding to the plurality of basic control policies according to the policy utility values respectively corresponding to the plurality of basic control policies comprises:

6. The intelligent control method of a building refrigeration room according to any one of claims 2 to 5, wherein the performing policy learning on the plurality of reference control policies to generate the plurality of final control policies includes:

7. The intelligent control method of a building refrigeration room according to claim 6, wherein generating the plurality of final control strategies according to the plurality of reference control strategies after the cross operation and the plurality of reference control strategies without the cross operation comprises:

8. The intelligent control method of a building refrigeration room according to any one of claims 1 to 5, further comprising:

9. The intelligent control method of the building refrigeration machine room according to claim 8, wherein the performing control instruction matching in a machine room control knowledge base according to the initial state trend conversion vector to generate a control instruction matching result corresponding to the target building refrigeration machine room comprises:

10. A building refrigeration room intelligent control system comprising a processor and a computer readable storage medium storing machine executable instructions that when executed by the processor implement the building refrigeration room intelligent control method of any one of claims 1-9.