Building energy-saving control method and system based on large model

Abstract

The invention provides a building energy-saving control method and system based on a large model, and relates to the technical field of data processing. The method comprises the following steps: acquiring environment data, and receiving a control instruction; determining similar instructions of the control instructions in the historical control instructions, and determining a fuzzy instruction based on the number of the similar instructions and the environmental similarity of the historical environmental data; for the real-time fuzzy instruction, determining a similar instruction with the highest matching degree with the time sequence and environment similarity as a matching instruction; for the non-real-time fuzzy instruction, determining a clear instruction of the same environment regulation subsystem, which is finally input in an adjacent time sequence, as a matching instruction; building state vectors are generated based on the matching instruction and the environment data at all moments in the execution period of the matching instruction, and user habit clusters are obtained through clustering; and based on the fluctuation similarity of each environment index in the user habit cluster, determining the environment index association degree of each environment adjustment subsystem, and adjusting the output power of the environment adjustment subsystem. According to the invention, the energy-saving control accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a building energy-saving control method and system based on a large model. BACKGROUND

[0002] As a key link for realizing low-carbon operation of buildings, the operation process of a building energy-saving control system is affected by various dynamic factors such as environmental changes, personnel activity conditions, and running states of equipment (such as air conditioners, lighting, etc.), and is a typical multivariable coupled system, so that precise energy-saving control thereof has high complexity.

[0003] The existing building energy-saving control technology analyzes historical fluctuations of various monitoring data of a single subsystem, predicts relevant environmental indicators, and adjusts the output of the corresponding equipment (such as increasing the air conditioner temperature, reducing the lighting brightness, etc.) when the predicted data exceeds the safety threshold of the corresponding indicator.

[0004] However, the existing building energy-saving control technology is prone to control conflicts in the multi-subsystem coordinated control scene (for example, when the lighting subsystem borrows natural light to maintain indoor appropriate brightness for energy saving, it will cause an increase in solar radiation of the building environment, and thus cause the environmental temperature to rise, thereby increasing the load of the temperature control system and reducing the overall energy-saving effect), and the behavior of personnel in the monitored environment is uncertain (for example, different users have different electricity usage habits, or sudden personnel gathering, the regulation and control demand of each subsystem will change suddenly), so it cannot accurately predict the environmental indicator fluctuations in the multi-system coordinated control, and it is also difficult to develop targeted strategies to deal with the above-mentioned sudden situations, resulting in insufficient accuracy of building energy-saving control. SUMMARY

[0005] In order to solve the problem that the related art cannot accurately predict the environmental indicator fluctuations in the multi-system coordinated control, and it is also difficult to develop targeted strategies to deal with the sudden situations, resulting in insufficient accuracy of building energy-saving control, the present application provides a building energy-saving control method based on a large model, and the technical solution adopted is as follows: Collecting environmental data of a target building and receiving a control instruction input by a user; Determining a similar instruction of the control instruction in historical control instructions of a historical time sequence, and determining a fuzzy instruction in the control instruction based on the number of the similar instructions and the environmental similarity between the corresponding historical environmental data of the similar instructions; When the fuzzy instruction is a real-time instruction, determining the similar instruction with the highest matching degree of the time sequence similarity and the environmental similarity as a matching instruction; When the fuzzy instruction is a non-real-time instruction, determining a clear instruction of the same environmental adjustment subsystem as the fuzzy instruction input by the user for the last time in adjacent time sequences as the matching instruction. generate a corresponding building state vector based on the matching instruction and the environment data at each time point in the execution period thereof, and perform clustering operation on the building state vector to obtain a user habit cluster; determine an environment index correlation degree of each of the environment regulation subsystems based on the fluctuation similarity of different environment indexes in the environment data corresponding to each of the building state vectors in the user habit cluster; adjust the output power of the environment regulation subsystems of the target building in the matching instruction based on the environment index correlation degree and execute.

[0006] Exemplarily, the similar instructions in the historical control instructions in the historical time sequence are determined, the fuzzy instructions in the control instructions are determined based on the number of the similar instructions and the environment similarity between the corresponding historical environment data, and the method comprises the following steps: vectors corresponding to the control instructions and the environment data are extracted, and are denoted as instruction vectors and environment vectors; cosine similarity of the instruction vectors and historical instruction vectors corresponding to the historical control instructions is calculated, the historical instruction vectors corresponding to the cosine similarity greater than a first preset threshold value are determined as similar instruction vectors, and the historical control instructions corresponding to the similar instruction vectors are the similar instructions; the number of the similar instructions corresponding to the control instructions in the historical time sequence is determined, and is denoted as a first fuzzy factor; cosine similarity between each two historical environment vectors corresponding to the control instructions is calculated and an average value is obtained, and is denoted as a second fuzzy factor; the fuzzy instructions in the control instructions are determined based on the first fuzzy factor and the second fuzzy factor.

[0007] Exemplarily, when the fuzzy instruction is a real-time instruction, the similar instruction with the highest matching degree of the time sequence similarity and the environment similarity is determined as a matching instruction, and the method comprises the following steps: when the fuzzy instruction is a real-time instruction, an environment vector of the environment data corresponding to the fuzzy instruction and historical environment vectors of the environment data corresponding to each of the similar instructions are extracted; cosine similarity of the environment vector corresponding to the fuzzy instruction and the historical environment vectors corresponding to each of the similar instructions is calculated, and normalization is performed to obtain the environment similarity corresponding to each of the similar instructions; a first time sequence of the fuzzy instruction in the monitoring period in which the fuzzy instruction is located and a second time sequence of each of the similar instructions in the monitoring period in which the similar instruction is located are obtained, a difference value of the first time sequence and each of the second time sequences is calculated, and normalization is performed to obtain the time sequence similarity corresponding to each of the similar instructions; the matching degree of each of the similar instructions and the fuzzy instruction is determined based on the environment similarity and the time sequence similarity corresponding to each of the similar instructions, and the similar instruction with the highest matching degree is determined as the matching instruction.

[0008] Exemplarily, when the fuzzy instruction is a non-real-time instruction, the clear instruction of the same environmental regulation subsystem as the fuzzy instruction input by the user for the last time in the adjacent time sequence is determined as the matching instruction, including: when the fuzzy instruction is a non-real-time instruction, determining the target environmental regulation subsystem corresponding to the fuzzy instruction; obtaining the time intervals of each adjacent historical control instruction input by the user to the target environmental regulation subsystem in the historical time sequence and calculating the average value, which is recorded as the user operation interval of the target environmental regulation subsystem; determining the user operation interval in the past up to the current moment as the adjacent time sequence, and obtaining the clear instruction input by the user to the target environmental regulation subsystem in the adjacent time sequence that is closest to the current moment, and determining it as the matching instruction.

[0009] Exemplarily, the method generates a corresponding building state vector based on the matching instruction and the environmental data at each moment during its execution period, and performs a clustering operation on the building state vector to obtain a user habit cluster, including: obtaining the time interval between the user inputting the matching instruction and the next adjacent historical control instruction, which is recorded as the execution period of the matching instruction; extracting the historical environment vector corresponding to the historical environment data collected at different moments during the execution period of the matching instruction, and constructing the building state vector at the corresponding moment based on the matching instruction and the historical environment vector; performing dimensionality reduction processing on the building state vector, and calculating the Euclidean distance between each pair of the building state vectors after the dimensionality reduction processing, and determining the clustering radius based on the Euclidean distance; performing a clustering operation on each building state vector after the dimensionality reduction processing with the clustering radius to obtain the user habit cluster; wherein the cluster size of the user habit cluster is not less than a second preset threshold.

[0010] Exemplarily, the method of determining the correlation between the environmental indicators of each environmental regulation subsystem based on the fluctuation similarity of different environmental indicators in the environmental data corresponding to each building state vector in the user habit cluster includes: sorting the building state vectors in the user habit cluster based on the order of monitoring moments to obtain a building state vector sequence; for the current environmental regulation subsystem, extracting the environmental data of each first environmental indicator corresponding to it from the building state vector sequence to form a first environmental data sequence; extracting the environmental data of each second environmental indicator corresponding to other environmental regulation subsystems from the building state vector sequence to form a second environmental data sequence; for each first environmental data sequence, determining the fluctuation similarity between it and each second environmental data sequence, and determining the correlation between the environmental indicators based on the average value of each fluctuation similarity and the absolute value range thereof.

[0011] Exemplarily, for each first environmental data sequence, determining the fluctuation similarity between it and each second environmental data sequence includes: calculating the first difference of the environmental data at the left and right endpoints of the first environmental data sequence, and the second difference of the environmental data at the left and right endpoints of the second environmental data sequence, and recording the difference between the first difference and the second difference as a first fluctuation factor; based on the first environmental data sequence, taking time as the independent variable and the first environmental indicator as the dependent variable, constructing a corresponding time-first environmental indicator curve; based on the second environmental data sequence, taking time as the independent variable and the second environmental indicator as the dependent variable, constructing a corresponding time-second environmental indicator curve; determining the difference in the number of maximum points between the time-first environmental indicator curve and the time-second environmental indicator curve, recording it as a second fluctuation factor; and determining the fluctuation similarity between the first environmental data sequence and the corresponding second environmental data sequence based on the first fluctuation factor and the second fluctuation factor.

[0012] Exemplarily, the adjusting and executing of the output power of the environmental conditioning subsystem of the target building in the matching instruction based on the correlation of the environmental indicators includes: for any monitoring moment, if the user habit cluster to which the corresponding building state vector belongs is different from that of the previous monitoring moment, recording it as the target moment; determining an abnormality indicator based on the number of target moments up to the current moment, the average time interval between each target moment, and the correlation of the environmental indicators of each target moment and its most recent non-target moment, and determining the moment type of the current moment based on the abnormality indicator; wherein the moment type includes system control abnormality moment, user control state sudden change moment and normal fluctuation moment; adjusting and executing the output power of the environmental conditioning subsystem of the target building in the matching instruction based on the moment type.

[0013] Exemplarily, the adjusting and executing of the output power of the environmental control subsystem of the target building in the matching instruction based on the moment type includes: when the current moment is the moment of sudden change of the user control state or the moment of normal fluctuation, calculating the differential mean of the environmental data corresponding to each environmental indicator in the current environmental control subsystem, adjusting and executing the output power of each environmental control subsystem based on the differential mean, the abnormal indicator and the correlation degree of the environmental indicator; and sending an alarm message when the current moment is the moment of abnormal system control.

[0014] Correspondingly, the present application also provides a building energy-saving control system based on a large model, comprising: The data acquisition module is used to collect environmental data of the target building and receive control instructions input by the user; a data processing module, configured to determine similar instructions to the control instruction in the historical control instructions of the historical time sequence, and determine fuzzy instructions in the control instruction based on the number of similar instructions and the environmental similarity between the corresponding historical environmental data; The data processing module is further configured to, when the fuzzy instruction is a real-time instruction, determine the similar instruction with the highest matching degree with the fuzzy instruction in terms of timing similarity and environmental similarity as a matching instruction; The data processing module is further configured to, when the fuzzy instruction is a non-real-time instruction, determine the clear instruction of the same environment adjustment subsystem as the fuzzy instruction, which is the last input by the user in an adjacent time sequence, as the matching instruction; The data processing module is further configured to generate a corresponding building state vector based on the matching instruction and the environmental data at each moment during its execution, and perform a clustering operation on the building state vector to obtain a user habit cluster; The data processing module is further configured to determine the correlation between the environmental indicators of each of the environmental adjustment subsystems based on the fluctuation similarity of different environmental indicators in the environmental data corresponding to each of the building state vectors in the user habit cluster; An energy-saving control module is used to adjust the output power of the environmental conditioning subsystem of the target building in the matching instruction based on the correlation degree of the environmental indicators and execute it.

[0015] This application may have some or all of the following beneficial effects: In the large model-based building energy-saving control method provided in the present application, by identifying fuzzy instructions in the control instructions, and selecting the similar instructions with the highest matching degree for real-time fuzzy instructions in historical control instructions, and matching the last clear instructions of the same subsystem in adjacent time sequences for non-real-time fuzzy instructions, fuzzy instructions can be accurately processed, effectively solving the control deviation problem caused by the semantic ambiguity of user input instructions, and improving the system response accuracy; based on the matching instructions and their execution period environmental data, the building state vector is generated and clustered to obtain user habit clusters, so as to accurately capture the user's control preferences in different environments, which helps to improve user experience; by analyzing the similarity of environmental indicator fluctuations of different environmental adjustment subsystems in the user habit cluster, the environmental indicator correlation between subsystems is determined, and the correlation rules between each environmental adjustment subsystem are clarified, so that multi-system collaborative optimization can be achieved; based on the environmental indicator correlation, the output power of each environmental adjustment subsystem is dynamically adjusted, which improves the system's adaptability to complex scenarios, reduces ineffective energy consumption while meeting user needs, and improves the overall accuracy and energy efficiency of building energy-saving control. It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0017] Figure 1 A flow chart of a building energy-saving control method based on a large model according to an exemplary embodiment of the present application is shown; Figure 2 A schematic block diagram of a building energy-saving control system based on a large model according to an exemplary embodiment of the present application is shown. DETAILED DESCRIPTION

[0018] To further illustrate the technical means and effectiveness of this application to achieve the intended invention objectives, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation, structure, features, and effectiveness of the large-scale model-based building energy-saving control method and system proposed in this application. In the following description, different references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0019] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0020] The specific scheme of the building energy-saving control method and system based on the large model provided by this application is described in detail below with reference to the accompanying drawings.

[0021] See also Figure 1 , which shows a method flow chart of a building energy-saving control method based on a large model provided by an embodiment of the present application, such as Figure 1 As shown, the building energy-saving control method based on the large model specifically includes the following steps: S110: Collecting environmental data of the target building and receiving control instructions input by the user; S120: determining similar instructions of the control instructions in the historical control instructions of the historical time sequence, and determining fuzzy instructions in the control instructions based on the number of similar instructions and the environmental similarity between the corresponding historical environmental data; S130: When the fuzzy instruction is a real-time instruction, a similar instruction with the highest matching degree in terms of timing similarity and environment similarity is determined as a matching instruction; S140: When the fuzzy instruction is a non-real-time instruction, the clear instruction of the same environment adjustment subsystem as the fuzzy instruction, which is input by the user last time in the adjacent time sequence, is determined as a matching instruction; S150: Generate a corresponding building state vector based on the matching instruction and the environmental data at each moment during its execution, and perform a clustering operation on the building state vector to obtain a user habit cluster; S160: Determine the correlation between the environmental indicators of each environmental regulation subsystem based on the fluctuation similarity of different environmental indicators in the environmental data corresponding to each building state vector in the user habit cluster; S170: Adjust the output power of the environmental regulation subsystem of the target building in the matching instruction based on the environmental indicator correlation and execute it.

[0022] The following describes in detail the various steps of the large model-based building energy-saving control method: In step S110 , environmental data of the target building is collected, and a control instruction input by a user is received.

[0023] In the embodiment of the present application, the target building refers to a building that requires energy-saving control. The large-model-based building energy-saving control method provided in the embodiment of the present application can collaboratively control multiple environmental regulation subsystems of the target building, wherein the environmental regulation subsystem refers to a functional module that performs real-time monitoring, regulation, and control of a specific type of environmental indicator within the target building. Each environmental regulation subsystem is independently responsible for regulating the corresponding environmental parameters and may be interrelated or coupled. For example, the environmental regulation subsystem may include a lighting subsystem, a temperature subsystem, a humidity regulation subsystem, etc.

[0024] In the embodiments of the present application, the aforementioned environmental data refers to data collected by monitoring equipment related to the target building's environment and equipment operation. For example, the aforementioned environmental data may include environmental indicator data and equipment operation indicator data; the aforementioned environmental indicator data may include data such as indoor and outdoor temperature, humidity, CO2 concentration, and light intensity, while the aforementioned equipment operation indicator data may include data such as the total output power of lighting equipment and the total output power of temperature control equipment (such as air conditioners).

[0025] Exemplarily, the above-mentioned collection of environmental data of the target building can be achieved as follows: monitoring points are evenly set up in the target building, and each monitoring point is equipped with monitoring equipment such as temperature sensors, humidity sensors, CO2 concentration sensors, and light sensors. The monitoring points need to avoid easily interfered areas such as air-conditioning outlets and direct light source areas; monitoring points are set up outside the target building with the same density as indoors (the nearest monitoring points indoors and outdoors are matched, such as being set on the inside and outside of the same facade), and temperature and light sensors are installed to avoid areas near heat sources or building obstructions; the real-time power of lighting equipment and temperature control equipment (such as the total output power of lamps and the total output power of air conditioners) is synchronously monitored and recorded.

[0026] In a specific implementation of the embodiment of the present application, 24 hours can be set as a monitoring period, the above-mentioned environmental data and equipment operation data can be collected in real time, and all data can be timestamp aligned and preliminarily filtered (such as eliminating obvious outliers, etc.).

[0027] In an embodiment of the present application, the above-mentioned control instructions refer to instructions for adjusting the operating status of each environmental adjustment subsystem; illustratively, the above-mentioned control instructions can be divided into clear instructions and fuzzy instructions, wherein clear instructions refer to control instructions input by the user with clear semantics and specific adjustment parameters, for example, "raise the temperature to 28 degrees"; fuzzy instructions refer to control instructions input by the user with vague semantics and no clear specific adjustment parameters, for example, "raise the temperature".

[0028] Exemplarily, the above-mentioned receiving of the control instruction input by the user can be implemented as follows: receiving the control instruction input by the user through the subsystem control panel, remote control or intelligent voice, etc.

[0029] In this embodiment of the present application, after acquiring environmental data and control instructions, the collected environmental data and received control instructions are transmitted and stored to an edge node via a wireless transmission module. This edge node is a key node in the building energy-saving control system responsible for data storage, processing, and instruction generation. For example, historical environmental data and control instructions from the past month can be stored on the edge node to serve as basic data for subsequent analysis of user habits and generation of control strategies.

[0030] In step S120 , similar instructions of the control instructions are determined in the historical control instructions of the historical time sequence, and fuzzy instructions in the control instructions are determined based on the number of similar instructions and the environmental similarity between the corresponding historical environmental data.

[0031] In the embodiment of the present application, the above-mentioned historical time series refers to the time series of historical control instructions and corresponding environmental data stored in the above-mentioned edge node in chronological order, reflecting the user's past control behavior and environmental change rules.

[0032] In the embodiment of the present application, the above-mentioned historical control instructions refer to the control instructions input by the user in the past recorded in the historical time series, which are basic data for analyzing the similarity of the current control instructions.

[0033] In the embodiment of the present application, the above-mentioned similar instructions refer to instructions in historical control instructions that are semantically similar to the current control instructions.

[0034] Exemplarily, the above-mentioned determination of similar instructions of control instructions in historical control instructions of historical time series can be implemented as follows: extract the vectors corresponding to the control instructions and environmental data, recorded as instruction vectors and environmental vectors; calculate the cosine similarity between the instruction vector and the historical instruction vector corresponding to the historical control instruction, and determine the historical instruction vector corresponding to the cosine similarity greater than the first preset threshold as a similar instruction vector, and the corresponding historical control instruction is a similar instruction.

[0035] In a specific implementation of an embodiment of the present application, the process of extracting the instruction vector and the environment vector can be implemented by a text extraction model, which is specifically implemented as follows: using a bidirectional encoder representation from transformers (BERT) model to extract the vectors of the current control instruction and the historical control instruction, which are respectively recorded as the current instruction vector and the historical instruction vector; obtaining the environment data at the same time as the current control instruction or the historical control instruction and arranging them in a preset fixed dimensional order to obtain the corresponding environment vectors, which are respectively recorded as the current environment vector and the historical environment vector; using the current instruction vector For example, calculate the current instruction vector The cosine similarity between each historical instruction vector is calculated. The historical environment vector with a cosine similarity greater than 0.7 is determined as a similar instruction vector, and its corresponding historical control instruction is recorded as a similar instruction to the current control instruction.

[0036] In the embodiment of the present application, the environmental similarity between the above-mentioned historical environmental data refers to the degree of similarity between the historical environmental data corresponding to similar instructions, which is used to reflect the consistency of the environment in which the similar instructions are located.

[0037] Exemplarily, the above-mentioned determination of fuzzy instructions in control instructions based on the number of similar instructions and the environmental similarity between their corresponding historical environmental data can be implemented as follows: determine the number of similar instructions corresponding to the control instructions in the historical time series, recorded as the first fuzzy factor; calculate the cosine similarity between the historical environmental vectors corresponding to each control instruction and calculate the average value, recorded as the second fuzzy factor; determine the fuzzy instructions in the control instructions based on the first fuzzy factor and the second fuzzy factor.

[0038] Specifically, with the current instruction vector For example, the current instruction vector The fuzzy instruction corresponding to the current control instruction can be determined by the following process: obtain all similar instructions of the current control instruction through the above method and count their number, which is recorded as the first fuzzy factor ; Extract the historical environment vectors corresponding to all similar instructions, calculate the cosine similarity between these historical environment vectors and calculate the average value, which is recorded as the second fuzzy factor mentioned above ; Calculate the fuzziness of the current control instruction using the following formula: Because of its clear semantics, clear instructions are more distinguishable from other instructions, and usually there are fewer similar instructions. However, because of its unclear semantics, fuzzy instructions can find multiple similar instructions in historical control instructions. That is, in the above formula, the first fuzzy factor The larger the value of is, the more similar instructions there are to the current control instruction, and the higher the fuzziness of the current control instruction is. It is a quantitative index used to measure the consistency of the environment of similar instructions in the current instruction vector. Clear instructions have clear semantics, and their similar instructions usually correspond to a more consistent environment. The corresponding second fuzzy factor The value of is large. Fuzzy instructions are semantically vague, and similar instructions are often distributed in a variety of environments with large differences. The corresponding second fuzzy factor The value of is small; in summary, the first fuzzy factor of the current control instruction The larger the value of The smaller the value of , the greater the possibility that the current control instruction is a fuzzy instruction.

[0039] In an embodiment of the present application, after determining the ambiguity level of the current control instruction using the above method, the ambiguity level is normalized, and based on the result of the normalization process, a determination is made as to whether the current control instruction is an ambiguous instruction. For example, if the normalized value of the ambiguity level of the current control instruction is greater than 0.5, the instruction is determined to be an ambiguous instruction; otherwise, the instruction is determined to be a clear instruction. It should be noted that the preset threshold for determining whether the current control instruction is an ambiguous instruction can also be set to a value other than 0.5 in the interval (0, 1) based on the needs of the actual scenario.

[0040] In step S130 , when the fuzzy instruction is a real-time instruction, a similar instruction with the highest matching degree in terms of timing similarity and environment similarity is determined as a matching instruction.

[0041] In an embodiment of the present application, the above-mentioned real-time instructions refer to fuzzy instructions that are input by the user in real time and processed immediately by the system synchronously, and the corresponding environmental data are the current environmental data collected in real time (such as the current indoor temperature, light, etc.).

[0042] In the embodiments of the present application, the timing similarity is a parameter used to measure the degree of temporal proximity between the current real-time fuzzy instruction and its similar instructions. For example, the timing similarity can be determined by the timing difference between the current real-time fuzzy instruction and its similar instructions during their respective monitoring cycles. The smaller the timing difference, the higher the timing similarity between the current real-time fuzzy instruction and its similar instructions.

[0043] In the present embodiment, the aforementioned environmental similarity is a parameter used to measure the degree of similarity between the environments of the current real-time fuzzy instruction and similar instructions. For example, the environmental similarity can be determined by the cosine similarity between the real-time environmental vector corresponding to the current real-time fuzzy instruction and the historical environmental vector corresponding to the similar instruction. A larger cosine similarity value indicates a greater degree of similarity between the environments of the current real-time fuzzy instruction and similar instructions.

[0044] In an embodiment of the present application, the above-mentioned matching instruction is a historical control instruction that is screened out from all similar instructions of the current real-time fuzzy instruction and has the highest comprehensive matching degree with the current real-time fuzzy instruction in terms of timing similarity and environmental similarity, and is used to correct the semantic ambiguity problem of the current real-time fuzzy instruction.

[0045] Exemplarily, when the fuzzy instruction is a real-time instruction, the similar instruction with the highest matching degree in timing similarity and environmental similarity is determined as the matching instruction, which can be implemented as follows: when the fuzzy instruction is a real-time instruction, the environmental vector corresponding to the fuzzy instruction and the historical environmental vector corresponding to each similar instruction are extracted; the cosine similarity between the environmental vector corresponding to the fuzzy instruction and the historical environmental vector corresponding to each similar instruction is calculated, and normalized to obtain the environmental similarity corresponding to each similar instruction; the first timing of the fuzzy instruction in its monitoring period and the second timing of each similar instruction in its monitoring period are obtained, the difference between the first timing and each second timing is calculated, and normalized to obtain the timing similarity corresponding to each similar instruction; the matching degree of each similar instruction and the fuzzy instruction is determined based on the environmental similarity and timing similarity corresponding to each similar instruction, and the similar instruction with the highest matching degree is determined as the matching instruction.

[0046] Specifically, assuming that the current instruction vector If the corresponding control instruction is the current fuzzy instruction, the implementation of the above-mentioned determination of the matching instruction can be as follows: determine whether the input time of the current fuzzy instruction is synchronized with the system processing time. If synchronized, determine that it is the current real-time fuzzy instruction; obtain the real-time environment vector and its timestamp corresponding to the current real-time fuzzy instruction at the same moment (that is, the above-mentioned first time sequence), as well as the historical environment vector and its timestamp corresponding to each similar instruction determined by the current real-time fuzzy instruction at the same moment (that is, the above-mentioned second time sequence); for each similar instruction, calculate the timestamp difference between its corresponding historical environment vector at the same moment and the real-time environment vector at the same moment corresponding to the current real-time fuzzy instruction, and normalize the timestamp difference, which is recorded as the timing similarity between the corresponding similar instruction and the current real-time fuzzy instruction. For example, assuming that the system takes 24 hours as a monitoring cycle, assuming that the timestamp of a similar instruction is 8:40 in the 24-hour monitoring cycle on Wednesday, and the timestamp of the current real-time fuzzy instruction is 8:00 in the 24-hour monitoring cycle on Thursday, then the timestamp difference between the similar instruction and the current real-time fuzzy instruction is 40 minutes, and the time series similarity is obtained by normalizing them; for each similar instruction, the cosine similarity of its historical environment vector and the real-time environment vector at the same time is calculated and normalized, and recorded as the environment similarity between the corresponding similar instruction and the current real-time fuzzy instruction; with a similar instruction vector For example, the matching degree between its corresponding similar instruction and the current real-time fuzzy instruction can be calculated by the following formula: in, Similar instruction vector The matching degree between the corresponding similar instructions and the current real-time fuzzy instructions; For the similar instruction vector With the current instruction vector The cosine similarity between the two, the larger the value, the greater the similarity between the similar instruction and the environment of the current real-time fuzzy instruction; is the timing difference between the similar instruction and the current real-time fuzzy instruction in its monitoring cycle. The smaller the value, the greater the timing similarity between the similar instruction and the current real-time fuzzy instruction (in each monitoring cycle, the user's adjustment habits under similar timing are more likely to be the same, for example, the user is accustomed to lowering the room brightness during lunch break, etc.), and the 0.01 in the denominator is used to avoid the denominator being 0; in summary, the embodiment of the present application uses the above formula to comprehensively consider the timing similarity and environmental similarity of each similar instruction with the current real-time fuzzy instruction to determine its matching degree with the current real-time fuzzy instruction.

[0047] In the embodiment of the present application, after determining the matching degree between each similar instruction and the current real-time fuzzy instruction, the similar instruction with the highest matching degree is determined as the matching instruction of the current real-time fuzzy instruction.

[0048] In step S140, when the ambiguous instruction is a non-real-time instruction, a clear instruction inputted by the user last time in the adjacent time sequence and corresponding to the ambiguous instruction is determined as the matching instruction.

[0049] In the embodiment of the present application, the non-real-time instruction refers to an ambiguous instruction whose input time is not synchronized with the system processing time. For example, when the user inputs an ambiguous instruction, the system is in a fault state, and the system processes the ambiguous instruction after the system fault is recovered.

[0050] In the embodiment of the present application, the adjacent time sequence refers to a time period covered by the user operation interval, which is taken as a reference point. The user operation interval is used to reflect the regular operation interval of the user to the current environment adjustment subsystem and is calculated based on historical data. For example, if the user inputs an instruction to the temperature subsystem every two hours on average, the adjacent time sequence is two hours before the current time.

[0051] In the embodiment of the present application, the same environment adjustment subsystem refers to the matching instruction and the current non-real-time ambiguous instruction regulating the same type of environment adjustment subsystem. For example, the current non-real-time ambiguous instruction regulates the temperature subsystem, and the clear instruction inputted by the user last time to the temperature subsystem in the adjacent time sequence is required.

[0052] In the embodiment of the present application, the matching instruction refers to the clear instruction inputted by the user last time and corresponding to the current non-real-time ambiguous instruction, which is selected from the adjacent time sequence and used to correct the semantic ambiguity of the current non-real-time ambiguous instruction.

[0053] Exemplarily, when the ambiguous instruction is a non-real-time instruction, the clear instruction inputted by the user last time and corresponding to the ambiguous instruction in the adjacent time sequence is determined as the matching instruction, which can be implemented as follows: when the ambiguous instruction is a non-real-time instruction, the target environment adjustment subsystem corresponding to the ambiguous instruction is determined; the time interval of each adjacent historical control instruction inputted by the user to the target environment adjustment subsystem in the historical time sequence is obtained and averaged, which is recorded as the user operation interval of the target environment adjustment subsystem; the past user operation interval until the current time is determined as the adjacent time sequence, and the clear instruction inputted by the user to the target environment adjustment subsystem closest to the current time in the adjacent time sequence is obtained, which is determined as the matching instruction.

[0054] In a specific implementation of an embodiment of the present application, the process of determining the matching instruction can be implemented as follows: checking the difference between the input timestamp of the current fuzzy instruction and the system processing timestamp. If the difference exceeds a preset threshold, it is determined to be a non-real-time fuzzy instruction; parsing the control object of the non-real-time fuzzy instruction, and determining the environmental adjustment subsystem to which it belongs based on this, which is recorded as the target environmental adjustment subsystem. For example, if the non-real-time fuzzy instruction is "raising the temperature", then the corresponding target environmental adjustment subsystem is the temperature subsystem; if the non-real-time fuzzy instruction is "dimming the lights", then the corresponding target environmental adjustment subsystem is the lighting subsystem; extracting the timestamps of all historical control instructions input by the user to the target environmental adjustment subsystem from the historical data stored in the edge node, calculating the time intervals of all adjacent historical control instructions and averaging them to obtain the user operation interval, which is the duration of the adjacent time series; filtering out the clear instructions belonging to the target subsystem from the historical control instructions of the adjacent time series, and sorting them by timestamp; selecting the clear instruction with the latest timestamp in the above filtering results, and determining it as the matching instruction of the current non-real-time fuzzy instruction.

[0055] In addition, if the adjustment result of the target environment adjustment subsystem based on the matching instruction of the above-mentioned current non-real-time fuzzy instruction does not meet the actual requirements of the user, the user can input the control instruction again through the subsystem control panel, remote control or intelligent voice.

[0056] In step S150, a corresponding building state vector is generated based on the matching instruction and the environmental data at each moment during its execution, and a clustering operation is performed on the building state vector to obtain a user habit cluster.

[0057] In the embodiment of the present application, the above-mentioned matching instruction is a control instruction determined by the above-mentioned method for correcting a real-time fuzzy instruction or a non-real-time fuzzy instruction, which has clear semantics and contains specific control parameters.

[0058] In this embodiment of the present application, the execution period of the matching instruction refers to the period from the moment the matching instruction is issued to the moment the user enters the next control instruction. During this execution period, the control logic of the matching instruction continues to be effective. For example, the period from the time the user enters the matching instruction "Raise the temperature to 26°C" until the user enters the next control instruction "Lower the temperature" is recorded as the execution period of the matching instruction "Raise the temperature to 26°C".

[0059] In this embodiment of the present application, the building state vector is a vector formed by combining the instruction vector corresponding to the matching instruction and the environment vector at each time during the execution period. For example, the building state vector for a matching instruction may be in the form of: [instruction vector, environment vector at time 1, environment vector at time 2, ..., environment vector at time n].

[0060] Exemplarily, the above-mentioned generation of the corresponding building state vector based on the matching instruction and the environmental data at each moment during its execution period can be implemented as follows: obtain the time interval between the user input of the matching instruction and its next adjacent historical control instruction, which is recorded as the execution period of the matching instruction; extract the historical environment vector corresponding to the historical environment data collected at different moments during the execution period of the matching instruction, and construct the building state vector at the corresponding moment based on the matching instruction and the historical environment vector.

[0061] In a specific implementation of the embodiment of the present application, taking the lighting-temperature coupled intelligent control system as an example, the above-mentioned construction of the building state vector can be implemented as follows: extract the timestamp of the matching instruction and the timestamp of the next control instruction after it , calculate the time interval , is recorded as the execution period of the matching instruction; during the execution period T, environmental data (including indoor and outdoor temperature, humidity, light intensity, equipment power, etc.) are collected according to the preset sampling frequency; the environmental data at each moment in the execution period T are arranged according to the preset fixed dimension order (such as "indoor temperature → outdoor temperature → indoor and outdoor temperature difference → ... → lighting power → air conditioning power") to generate the environmental vector at the corresponding moment [ ], where n is the number of samples during the execution period; extract the instruction vector corresponding to the matching instruction, recorded as , combine the instruction vector with the environment vector at each moment during the execution period to obtain the building state vector corresponding to the matching instruction [ ].

[0062] In the embodiment of the present application, the above-mentioned user habit cluster refers to a sample cluster obtained by clustering the building state vector, which is used to reflect the user's typical control habits under similar environmental conditions. The building state vectors within the same user habit cluster have similar instruction characteristics and environmental characteristics (such as temperature adjustment habits in the summer afternoon and lighting adjustment habits in the winter night).

[0063] Exemplarily, the above-mentioned clustering operation on the building state vectors to obtain the user habit clusters can be implemented as follows: perform dimensionality reduction processing on the building state vectors, calculate the Euclidean distance between each pair of the building state vectors after the dimensionality reduction processing, and determine the clustering radius based on the Euclidean distance; perform clustering operation on each building state vector after the dimensionality reduction processing with the clustering radius to obtain the user habit clusters; wherein, the cluster size of the user habit cluster is not less than a second preset threshold.

[0064] In a specific implementation of an embodiment of the present application, the above-mentioned clustering operation can be implemented as follows: constructing a target matrix based on the building state vector corresponding to the matching instruction; using the principal component analysis (PCA) algorithm to reduce the dimension of the target matrix to retain the main features in the target matrix (such as retaining the principal components with a cumulative contribution rate ≥ 90%); calculating the Euclidean distance between all building state vectors after dimensionality reduction; using the 30% quantile of the Euclidean distance as the clustering radius and 1% of the total sample volume (that is, the above-mentioned second preset threshold) as the minimum cluster sample number, and using the density-based spatial clustering of applications with noise (DBSCAN) algorithm to cluster the building state vectors after dimensionality reduction, each sample cluster obtained by clustering is a user habit cluster, and each cluster represents the user's stable control habits in a similar environment.

[0065] Because the environmental indicators corresponding to each environmental regulation subsystem may be correlated (for example, the "light intensity" environmental indicator of the lighting subsystem affects the "indoor temperature" environmental indicator of the temperature subsystem), embodiments of the present application can also assign weights to data points in a user habit cluster based on the degree to which they conform to this characteristic. Exemplarily, this process can be implemented as follows: for any user habit cluster at a certain monitoring point, the distance from each data point in the user habit cluster to its cluster center is calculated. The smaller the distance, the more consistent the data point is with the overall characteristics of the user habit cluster. Based on the distance between each data point and the cluster center, a weight is assigned to the corresponding building state vector. Taking building state vector j as an example, the weight corresponding to building state vector j can be determined based on the following formula: in, is the weight of building state vector j; is the distance between the data point corresponding to the building state vector j and the cluster center point. The 0.01 in the denominator is used to avoid the denominator being 0. In the above formula, the weight and distance Negative correlation, that is, the closer the data point corresponding to the building state vector j is to the cluster center, the larger the weight corresponding to the building state vector j should be, which means that it can better reflect user habits.

[0066] Furthermore, the original building state vector j is adjusted based on the weight of the above building state vector j by the following formula: in, is the vector after the building state vector j is adjusted; through this adjustment, the influence of the core data points that conform to user habits can be strengthened, and the interference of data points that deviate from the cluster characteristics can be weakened.

[0067] In step S160 , the environmental indicator correlation of each environmental adjustment subsystem is determined based on the fluctuation similarity of different environmental indicators in the environmental data corresponding to each building state vector in the user habit cluster.

[0068] In the embodiments of this application, the aforementioned environmental indicators refer to specific quantifiable parameters of the environmental conditioning subsystem and are the core monitoring targets of the environmental conditioning subsystem. For example, the environmental indicators of the lighting subsystem may include indicators such as indoor brightness, outdoor brightness, and the difference between indoor and outdoor brightness; the environmental indicators of the temperature subsystem may include indicators such as indoor temperature, outdoor temperature, and the difference between indoor and outdoor temperature.

[0069] In the embodiments of the present application, the fluctuation similarity is a parameter used to measure the degree of similarity between the temporal variation trends of two environmental indicators. A larger value indicates a more consistent variation trend between the corresponding environmental indicators. For example, if the temperature increases as the lighting brightness increases, the fluctuation similarity between the corresponding brightness and temperature indicators is high.

[0070] In an embodiment of the present application, the above-mentioned environmental indicator correlation is a quantitative indicator obtained by integrating multiple fluctuation similarities, reflecting the degree of correlation between environmental indicators of different environmental adjustment subsystems (such as the degree of correlation between the indoor brightness of the lighting subsystem and the indoor temperature of the temperature subsystem). The larger the value, the more significant the environmental correlation between the corresponding environmental adjustment subsystems.

[0071] Exemplarily, the above-mentioned determination of the correlation between environmental indicators of each environmental regulation subsystem based on the fluctuation similarity of different environmental indicators in the environmental data corresponding to each building state vector in the user habit cluster can be implemented as follows: sorting the building state vectors in the user habit cluster based on the order of monitoring time to obtain a building state vector sequence; for the current environmental regulation subsystem, extracting the environmental data of each first environmental indicator corresponding to it from the building state vector sequence to form a first environmental data sequence; extracting the environmental data of each second environmental indicator corresponding to other environmental regulation subsystems from the building state vector sequence to form a second environmental data sequence; for each first environmental data sequence, determining its fluctuation similarity with each second environmental data sequence, and determining the correlation between environmental indicators based on the average value of each fluctuation similarity and the absolute value range thereof.

[0072] Among them, the above-mentioned determination of the fluctuation similarity between each first environmental data sequence and each second environmental data sequence can be achieved as follows: calculate the first difference of the environmental data at the left and right endpoints of the first environmental data sequence, and the second difference of the environmental data at the left and right endpoints of the second environmental data sequence, and record the difference between the first difference and the second difference as the first fluctuation factor; based on the first environmental data sequence, with time as the independent variable and the first environmental indicator as the dependent variable, construct the corresponding time-first environmental indicator curve; based on the second environmental data sequence, with time as the independent variable and the second environmental indicator as the dependent variable, construct the corresponding time-second environmental indicator curve; determine the difference in the number of maximum points between the time-first environmental indicator curve and the time-second environmental indicator curve, and record it as the second fluctuation factor; determine the fluctuation similarity between the first environmental data sequence and the corresponding second environmental data sequence based on the first fluctuation factor and the second fluctuation factor.

[0073] In a specific implementation of the embodiment of the present application, taking the above-mentioned lighting-temperature coupled intelligent control system as an example, the process of determining the correlation degree of environmental indicators between the lighting subsystem and the temperature subsystem in the control system can be implemented as follows: based on the current analysis target (such as the lighting-temperature adjustment correlation of a certain monitoring point), a target user habit cluster is selected from multiple user habit clusters obtained by clustering; for the target user habit cluster, all building state vectors in the cluster are sorted by time based on the order of monitoring moments to form a building state vector sequence; for the lighting subsystem, the environmental data of all corresponding first environmental indicators (such as indoor brightness, outdoor brightness) are extracted from the building state vector sequence, and arranged in time order to form a corresponding first environmental data sequence L=[ ], where t is the time step number, which can be the number of environmental data collected at a preset sampling frequency in the first environmental data sequence; for the temperature subsystem, the environmental data of all corresponding second environmental indicators (such as indoor temperature and outdoor temperature) are extracted from the building state vector sequence and arranged in chronological order to form the corresponding second environmental data sequence I=[ ]; For each set of the first environmental data sequence (such as the indoor brightness sequence L) and the second environmental data sequence (such as the indoor temperature sequence I), calculate the difference between the left and right endpoints of the first environmental data sequence = - , and the difference between the left and right endpoints of the second environmental indicator = - ,Will and The difference is recorded as the first fluctuation factor; the time-first environmental index curve is constructed with time as the horizontal axis and the first environmental index as the vertical axis, and the time-second environmental index curve is constructed with time as the horizontal axis and the second environmental index as the vertical axis, and the number of maximum points of the time-first environmental index curve is counted respectively. and the number of maximum points of the time-second environmental index curve ,Will and The difference is recorded as the second volatility factor; the volatility similarity of this group of indicators is calculated based on the following formula: in, For the first environment data sequence L=[ ] and the second environment data sequence I=[ ] in the period T (the period corresponding to the time step t); The first fluctuation factor reflects the difference in the overall change range between the first environmental data series and the second environmental data series within time period T. The smaller its absolute value, the closer the change range of the two indicator series. The second fluctuation factor reflects the closeness of the peak occurrence frequencies of the first environmental data series and the second environmental data series in time period T. The smaller its absolute value is, the closer the peak occurrence frequencies of the two environmental indicator series in time period T are, and the higher the synchronization of the fluctuation trends is. The role of 0.01 in the denominator is to avoid the denominator being 0.

[0074] In an embodiment of the present application, the fluctuation similarity between other groups of first environmental data sequences and second environmental data sequences can be obtained in the same way as above, and further, the process of determining the correlation between the environmental indicators of the lighting subsystem and the temperature subsystem based on the fluctuation similarity between each group of data sequences can be implemented as follows: for all first environmental indicators of the lighting subsystem (such as indoor brightness, outdoor brightness, etc.), calculate the fluctuation similarity with all second environmental indicators of the temperature subsystem (such as indoor temperature, outdoor temperature, etc.) and calculate the average value to obtain the average fluctuation similarity within time period T. ; Calculate the absolute value range of all fluctuation similarities within period T (That is, the difference between the maximum and minimum values ​​of all fluctuation similarities); determine the environmental indicator correlation between the lighting subsystem and the temperature subsystem based on the following formula: in, is the environmental index correlation between the lighting subsystem and the temperature subsystem within the time period T; is the above average fluctuation similarity, reflecting the overall fluctuation trend of the environmental indicators of the lighting subsystem and the temperature subsystem within the time period T. The larger its value, the closer the overall fluctuation trend of the environmental indicators of the two environmental control subsystems within the time period T is. It is the absolute value range of all fluctuation similarities within the above period T, reflecting the degree of dispersion of the fluctuation similarities of different pairs of environmental indicators within the period T. The smaller the value, the closer the similarities of each fluctuation are and the more stable the consistency of the overall fluctuation trend.

[0075] In step S170 , the output power of the environment conditioning subsystem of the target building in the matching instruction is adjusted based on the environmental indicator correlation and executed.

[0076] In the embodiment of the present application, the output power is the power of the corresponding environment conditioning subsystem when it is in operation, for example, the cooling / heating power of an air conditioner, the luminous power of a lighting device, etc.

[0077] Exemplarily, the above-mentioned adjustment of the output power of the environmental regulation subsystem of the target building in the matching instruction based on the correlation of environmental indicators and execution can be implemented as follows: for any monitoring moment, if the user habit cluster to which the corresponding building state vector belongs is different from that of the previous monitoring moment, it is recorded as the target moment; based on the number of target moments up to the current moment, the average time interval between each target moment, and the correlation of the environmental indicators of each target moment and its most recent non-target moment, the abnormality indicator is determined, and the moment type of the current moment is determined based on the abnormality indicator; wherein the moment type includes the system control abnormal moment, the user control state sudden change moment and the normal fluctuation moment; based on the moment type, the output power of the environmental regulation subsystem of the target building in the matching instruction is adjusted and executed.

[0078] In the embodiments of the present application, the monitoring time refers to the specific time point at which the building energy-saving control system performs real-time or scheduled monitoring of the building environment status (such as temperature and brightness) of the target building and user control behavior. For example, the monitoring time point is once every 10 minutes.

[0079] In the embodiment of the present application, the target time refers to the monitoring time when the user habit cluster to which the building state vector belongs is different from the previous monitoring time, that is, the time when the user habit cluster is switched.

[0080] In an embodiment of the present application, the above-mentioned abnormality indicator is an indicator of the deviation of the monitoring time from the normal range based on a comprehensive judgment of the number of target moments, the mean time interval and the correlation of environmental indicators, and is used to identify abnormal conditions (such as too frequent switching of habit clusters and sudden changes in correlation).

[0081] In an embodiment of the present application, the above-mentioned system control abnormality moment refers to the monitoring moment when the user habit cluster switching is abnormal or the environmental indicator correlation is abnormal due to the system's own control logic failure (such as device power abnormality); the above-mentioned user control state sudden change moment refers to the monitoring moment when the user habit cluster is switched due to a sudden change in the user's control habits (such as switching from energy-saving mode to comfort mode); the above-mentioned normal fluctuation moment is the monitoring moment when the user habit cluster switching frequency, the time interval between adjacent target moments and the environmental indicator correlation are all within the normal range, which belongs to natural environmental changes or routine user adjustments.

[0082] Specifically, the process of determining the type of monitoring moment can be implemented as follows: for each monitoring moment, extract the corresponding building state vector and determine the user habit cluster to which the building state vector belongs; compare the user habit clusters of the current monitoring moment with those of the previous monitoring moment, and if they are inconsistent, mark the current monitoring moment as the target moment; count the number of target moments up to the current monitoring moment ; Calculate the time interval between adjacent target moments and find the average value ; Determine the environmental index correlation of each target moment and its nearest non-target moment using the above method to determine the environmental index correlation, calculate the difference and find the average value ; Calculate the range of correlation between environmental indicators at each target moment ; Calculate the abnormal index at the current monitoring moment using the following formula: in, is the abnormal indicator at the current monitoring moment; is the number of target moments, reflecting the frequency of user habit cluster switching. The larger the value, the more frequent the user habit cluster switching in a short period of time, and the greater the possibility of anomalies (such as system failures or sudden changes in user needs); is the mean time interval between adjacent target moments, reflecting the temporal distribution characteristics of user habit cluster switching. The smaller its value, the shorter the time interval between target moments, the more frequent the user habit cluster switching, and the higher the possibility of abnormality. The denominator is added with 0.01 to avoid division by zero error. is the average value of the difference between the environmental index correlation of each target moment and its nearest non-target moment, reflecting the degree of environmental index correlation mutation. The larger its absolute value, the more obvious the correlation mutation of the environmental regulation subsystem is when the user habit cluster switches, and the greater the possibility of anomaly. It is the extreme difference of the correlation between the environmental indicators at the target moments, reflecting the stability of the correlation at the target moments. The larger the value, the more drastic the fluctuation of the correlation between the environmental indicators at different target moments, the more unstable the correlation characteristics, and the higher the possibility of abnormality.

[0083] After determining the abnormal index of the current monitoring moment by the above formula, the abnormal index is normalized, and the type of the current monitoring moment is determined according to the normalization result. In a specific implementation, the determination rule can be as follows: If the normalization result is , then the current monitoring moment is determined to be the moment of system control abnormality, , then the current monitoring moment is determined to be the moment of sudden change of the user control state, and the rest of the moments are normal fluctuation moments.

[0084] In an embodiment of the present application, if the current monitoring moment is a normal fluctuation moment, the output power of each environmental regulation subsystem increases until it reaches the energy-saving load limit power (subject to the actual situation); if the current monitoring moment is a sudden change in the user regulation state, the energy-saving load limit is ignored, and the output power of each system increases until it reaches the maximum load of each subsystem (the single regulation amplitude does not exceed 30% of the current output power); if the current moment is a system regulation abnormal moment, a warning message is directly output without adjusting the output power. Exemplarily, the above-mentioned process of adjusting the output power at normal fluctuation moments and sudden changes in user regulation state can be implemented as follows: calculate the differential mean of the environmental data corresponding to each environmental indicator in the current environmental regulation subsystem, and adjust the output power of each environmental regulation subsystem based on the differential mean, abnormal indicators, and the correlation between environmental indicators.

[0085] Specifically, taking the above-mentioned lighting-temperature coupled intelligent control system as an example, under normal fluctuation moments and moments when the user control state suddenly changes, the adjustment of the output power of the lighting subsystem and the temperature subsystem can be achieved as follows: calculate the differential mean d between the current environmental data of each dimension of the lighting subsystem and the previous moment; if the differential mean d is positive, it is determined that the output power needs to be increased. The specific adjustment logic is as follows: if The correlation between the environmental indicators of the lighting subsystem and the temperature subsystem is negative, indicating that there is a conflict between the control of the lighting subsystem and the temperature subsystem. In this case, the output power of the lighting subsystem is adjusted to: in, Adjusted output power for the lighting subsystem; is the current output power of the lighting subsystem; is the abnormal indicator at the current monitoring moment; is the result obtained by normalizing the above difference mean d; Adjust the temperature subsystem output power to: in, The output power adjusted for the temperature subsystem; is the current output power of the temperature subsystem; is the abnormal indicator at the current monitoring moment; is the result obtained by normalizing the above difference mean d; like A positive correlation between the environmental indicators of the lighting and temperature subsystems indicates that the lighting and temperature subsystems have similar fluctuation trends and can be coordinated. For example, the output power of the lighting and temperature subsystems can be coordinated and increased based on the temperature subsystem's control logic (see the temperature subsystem's power adjustment formula). If the above-mentioned difference mean d is positive, it is determined that the output power needs to be reduced. The specific control logic is as follows: the output power of the lighting subsystem is adjusted to: in, Adjusted output power for the lighting subsystem; is the current output power of the lighting subsystem; is the abnormal indicator at the current monitoring moment; is the result obtained by normalizing the above difference mean d; Adjust the temperature subsystem output power to: in, The output power adjusted for the temperature subsystem; is the current output power of the temperature subsystem; is the abnormal indicator at the current monitoring moment; is the result obtained by normalizing the above difference mean d; In the above process, the embodiment of the present application dynamically adjusts the output power amplitude of the two systems according to the increase and decrease trend of the lighting value, combined with the subsystem correlation (whether there is conflict) and the moment type. While meeting user needs, it balances the coordination and conflict between subsystems and realizes energy-saving regulation.

[0086] The above mainly introduces the solution provided by the embodiment of the present application from the perspective of method. In order to realize the above functions, it includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should easily appreciate that, in combination with the units and algorithm steps of each example described in the embodiments disclosed herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in a hardware or computer software driven hardware manner depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0087] Correspondingly, the embodiment of the present application also provides a building energy-saving control system based on a large model, referring toFigure 2 As shown, the building energy-saving control system 200 based on the large model may include a data acquisition module 210, a data processing module 220 and an energy-saving control module 230, wherein: The data acquisition module is used to collect environmental data of the target building and receive control instructions input by the user; a data processing module, configured to determine similar instructions of the control instructions in the historical control instructions of the historical time sequence, and determine fuzzy instructions in the control instructions based on the number of similar instructions and the environmental similarity between the corresponding historical environmental data; The data processing module is further configured to, when the fuzzy instruction is a real-time instruction, determine a similar instruction with the highest matching degree of timing similarity and environment similarity as a matching instruction; The data processing module is further configured to, when the fuzzy instruction is a non-real-time instruction, determine the clear instruction of the same environment adjustment subsystem as the fuzzy instruction, which is the last input by the user in an adjacent time sequence, as a matching instruction; The data processing module is further used to generate a corresponding building state vector based on the matching instruction and the environmental data at each moment during its execution, and perform clustering operation on the building state vector to obtain user habit clusters; The data processing module is further used to determine the correlation between the environmental indicators of each environmental regulation subsystem based on the fluctuation similarity of different environmental indicators in the environmental data corresponding to each building state vector in the user habit cluster; The energy-saving control module is used to adjust the output power of the environmental regulation subsystem of the target building in the matching instruction based on the correlation of environmental indicators and execute it.

[0088] The specific implementation details of the above-mentioned large-model-based building energy-saving control system have been described in detail in the corresponding position of the large-model-based building energy-saving control method, so they will not be repeated here.

[0089] It should be noted that the order of the embodiments of the present application is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0090] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. A building energy-saving control method based on a large model, characterized in that: The method comprises: Collect environmental data of the target building and receive control instructions input by the user; Determining similar instructions to the control instruction in the historical control instructions of the historical time sequence, and determining fuzzy instructions in the control instruction based on the number of the similar instructions and the environmental similarity between the corresponding historical environmental data; When the fuzzy instruction is a real-time instruction, the similar instruction with the highest matching degree of timing similarity and environment similarity is determined as a matching instruction; When the fuzzy instruction is a non-real-time instruction, the clear instruction of the same environment adjustment subsystem as the fuzzy instruction input by the user for the last time in the adjacent time sequence is determined as the matching instruction; Generate a corresponding building state vector based on the matching instruction and the environmental data at each moment during its execution, and perform a clustering operation on the building state vector to obtain a user habit cluster; Determining the correlation between the environmental indicators of each of the environmental adjustment subsystems based on the fluctuation similarity of different environmental indicators in the environmental data corresponding to each of the building state vectors in the user habit cluster; The output power of the environmental conditioning subsystem of the target building in the matching instruction is adjusted based on the environmental indicator correlation and executed.

2. The building energy-saving control method based on a large model according to claim 1 is characterized in that: Determining similar instructions to the control instruction in the historical control instructions of the historical time sequence, and determining the fuzzy instructions in the control instruction based on the number of the similar instructions and the environmental similarity between the corresponding historical environmental data, includes: Extracting vectors corresponding to the control instruction and the environmental data, and recording them as instruction vector and environmental vector; Calculating the cosine similarity between the instruction vector and the historical instruction vector corresponding to the historical control instruction, and determining the historical instruction vector corresponding to a cosine similarity greater than a first preset threshold as a similar instruction vector, and the corresponding historical control instruction is the similar instruction; Determine the number of similar instructions corresponding to the control instruction in the historical time sequence, and record it as a first fuzzy factor; Calculate the cosine similarity between each pair of historical environment vectors corresponding to the control instructions and calculate the average value, which is recorded as the second fuzzy factor; The fuzzy instruction in the control instruction is determined based on the first fuzzy factor and the second fuzzy factor.

3. The building energy-saving control method based on a large model according to claim 1 is characterized in that: When the fuzzy instruction is a real-time instruction, determining the similar instruction with the highest matching degree of timing similarity and environment similarity as the matching instruction includes: When the fuzzy instruction is a real-time instruction, extracting an environment vector of the environment data corresponding to the fuzzy instruction and a historical environment vector of the environment data corresponding to each of the similar instructions; Calculating the cosine similarity between the environment vector corresponding to the fuzzy instruction and the historical environment vector corresponding to each of the similar instructions, and normalizing them to obtain the environment similarity corresponding to each of the similar instructions; Obtaining a first time sequence of the fuzzy instruction in the monitoring period in which it is located, and a second time sequence of each of the similar instructions in the monitoring period in which it is located, calculating a difference between the first time sequence and each of the second time sequences, and normalizing them to obtain the time sequence similarity corresponding to each of the similar instructions; The matching degree between each of the similar instructions and the fuzzy instruction is determined based on the environmental similarity and the time sequence similarity corresponding to each of the similar instructions, and the similar instruction with the highest matching degree is determined as the matching instruction.

4. The building energy-saving control method based on a large model according to claim 1 is characterized in that: When the fuzzy instruction is a non-real-time instruction, determining the clear instruction of the environment adjustment subsystem input by the user for the last time in the adjacent time sequence as the matching instruction includes: When the fuzzy instruction is a non-real-time instruction, determining a target environment adjustment subsystem corresponding to the fuzzy instruction; Obtaining the time intervals between adjacent historical control instructions input by the user to the target environment adjustment subsystem in the historical time sequence and calculating the average value, which is recorded as the user operation interval of the target environment adjustment subsystem; The user operation interval in the past up to the current moment is determined as the adjacent time sequence, and the clear instruction input by the user to the target environment adjustment subsystem within the adjacent time sequence closest to the current moment is obtained and determined as the matching instruction.

5. The building energy-saving control method based on a large model according to claim 1 is characterized in that: The generating of a corresponding building state vector based on the matching instruction and the environmental data at each moment during its execution, and performing a clustering operation on the building state vector to obtain a user habit cluster, includes: Obtaining a time interval between the user inputting the matching instruction and the next adjacent historical control instruction, and recording the time interval as the execution period of the matching instruction; Extracting historical environment vectors corresponding to the historical environment data collected at different times during the execution of the matching instruction, and constructing the building state vector at the corresponding time based on the matching instruction and the historical environment vectors; Performing dimensionality reduction processing on the building state vectors, calculating the Euclidean distance between each of the building state vectors after the dimensionality reduction processing, and determining a clustering radius based on the Euclidean distance; A clustering operation is performed on each of the building state vectors after dimensionality reduction processing using the clustering radius to obtain the user habit cluster; wherein the cluster size of the user habit cluster is not less than a second preset threshold.

6. The building energy-saving control method based on a large model according to claim 1 is characterized in that: The determining of the correlation between the environmental indicators of each environmental adjustment subsystem based on the fluctuation similarity of different environmental indicators in the environmental data corresponding to each building state vector in the user habit cluster includes: Sort the building state vectors in the user habit cluster based on the order of monitoring moments to obtain a building state vector sequence; For the current environmental adjustment subsystem, extract the environmental data of each corresponding first environmental indicator from the building state vector sequence to form a first environmental data sequence; Extracting the environmental data of each second environmental indicator corresponding to other environmental adjustment subsystems from the building state vector sequence to form a second environmental data sequence; For each of the first environmental data sequences, the fluctuation similarity between it and each of the second environmental data sequences is determined, and the environmental indicator correlation is determined based on the average value of each of the fluctuation similarities and the absolute value range thereof.

7. The building energy-saving control method based on a large model according to claim 6 is characterized in that: The determining, for each of the first environment data sequences, the fluctuation similarity between the first environment data sequence and each of the second environment data sequences includes: Calculate a first difference between the environmental data at the left and right endpoints of the first environmental data sequence, and a second difference between the environmental data at the left and right endpoints of the second environmental data sequence, and record the difference between the first difference and the second difference as a first fluctuation factor; Based on the first environmental data sequence, with time as the independent variable and the first environmental indicator as the dependent variable, construct a corresponding time-first environmental indicator curve; Based on the second environmental data sequence, with time as the independent variable and the second environmental indicator as the dependent variable, construct a corresponding time-second environmental indicator curve; Determine the difference in the number of maximum value points between the time-first environmental index curve and the time-second environmental index curve, and record it as a second fluctuation factor; The fluctuation similarity between the first environment data sequence and the corresponding second environment data sequence is determined based on the first fluctuation factor and the second fluctuation factor.

8. The building energy-saving control method based on a large model according to claim 1 is characterized in that: The adjusting and executing the output power of the environmental conditioning subsystem of the target building in the matching instruction based on the environmental indicator correlation degree includes: For any monitoring moment, if the user habit cluster to which the corresponding building state vector belongs is different from that at the previous monitoring moment, it is recorded as the target moment; Determining an abnormality indicator based on the number of target moments up to the current moment, the average time interval between each target moment, and the correlation between the environmental indicators of each target moment and its nearest non-target moment, and determining the moment type of the current moment based on the abnormality indicator; wherein the moment type includes a system control abnormality moment, a user control state sudden change moment, and a normal fluctuation moment; The output power of the environmental conditioning subsystem of the target building in the matching instruction is adjusted based on the time type and executed.

9. The building energy-saving control method based on a large model according to claim 8 is characterized in that: The adjusting and executing the output power of the environment conditioning subsystem of the target building in the matching instruction based on the moment type includes: When the current moment is the moment of sudden change of the user control state or the moment of normal fluctuation, calculating the differential mean of the environmental data corresponding to each environmental indicator in the current environmental control subsystem, and adjusting the output power of each environmental control subsystem based on the differential mean, the abnormal indicator, and the correlation degree of the environmental indicator, and executing the adjustment; When the current moment is a moment of abnormal system control, an alarm message is sent.

10. A building energy-saving control system based on a large model, characterized in that: The system comprises: The data acquisition module is used to collect environmental data of the target building and receive control instructions input by the user; a data processing module, configured to determine similar instructions to the control instruction in the historical control instructions of the historical time sequence, and determine fuzzy instructions in the control instruction based on the number of similar instructions and the environmental similarity between the corresponding historical environmental data; The data processing module is further configured to, when the fuzzy instruction is a real-time instruction, determine the similar instruction with the highest matching degree with the fuzzy instruction in terms of timing similarity and environmental similarity as a matching instruction; The data processing module is further configured to, when the fuzzy instruction is a non-real-time instruction, determine the clear instruction of the same environment adjustment subsystem as the fuzzy instruction, which is the last input by the user in an adjacent time sequence, as the matching instruction; The data processing module is further configured to generate a corresponding building state vector based on the matching instruction and the environmental data at each moment during its execution, and perform a clustering operation on the building state vector to obtain a user habit cluster; The data processing module is further configured to determine the correlation between the environmental indicators of each of the environmental adjustment subsystems based on the fluctuation similarity of different environmental indicators in the environmental data corresponding to each of the building state vectors in the user habit cluster; An energy-saving control module is used to adjust the output power of the environmental conditioning subsystem of the target building in the matching instruction based on the correlation degree of the environmental indicators and execute it.

Images