Building intelligent operation and maintenance system and method based on large model

Abstract

The invention discloses a building intelligent operation and maintenance system and method based on a large model, and relates to the field of intelligent detection, and the method comprises the steps: carrying out the one-dimensional convolution coding of a first equipment energy consumption time sequence, extracting local features, and capturing the dynamic change on a time axis through a time sequence propagation code; and then, based on the equipment attribute semantic association matrix, the first equipment energy consumption characteristic is mapped to a second equipment semantic space in a nonlinear manner, so that the second equipment energy consumption prediction is fused with the influence of the operation state of the upstream equipment. And identifying hidden energy consumption fluctuation of the second equipment caused by the first equipment by comparing the difference between the predicted data and the actual monitoring data. According to the method, the limitation of traditional isolated modeling is broken through, cross-device semantic space feature transmission is achieved, energy transmission anomalies between devices are accurately captured, the phenomena of false report and missing report are effectively reduced, the accuracy and efficiency of anomaly detection are improved, and meanwhile the requirement for manual checking is lowered.

Description

Technical Field

[0001] The present application relates to the field of intelligent detection, and more specifically, to a large-scale model-based intelligent building operation and maintenance system and method. Background Art

[0002] As modern building systems continue to grow in complexity, building operations and maintenance management has evolved from single-device monitoring to multi-system collaborative optimization. Currently, large public buildings are widely deploying intelligent platforms such as building automation systems (BAS) and energy management systems (EMS), which continuously collect operational parameters and energy consumption time series data for equipment such as HVAC, elevators, and lighting.

[0003] Traditional O&M methods primarily rely on threshold alarm mechanisms and isolated device analysis based on statistical models, making it difficult to effectively analyze the complex coupling relationships between devices. Especially in scenarios with high-density sensor deployments, the time series correlations of device operating states exhibit non-stationary and multi-scale characteristics. For example, an anomaly in the air conditioning system may indirectly trigger energy consumption fluctuations in ventilation equipment through air flow paths, but traditional rule-based correlation analysis methods are unable to capture such hidden cross-system fault propagation chains. Furthermore, existing time series prediction models often ignore the mapping relationship between the semantic space of device attributes and dynamic energy consumption characteristics when dealing with multi-device collaborative optimization, resulting in feature drift in cross-device anomaly reasoning. When faced with the complex energy transfer and operating condition coordination relationships between devices, traditional methods often generate a large number of false positives or missed negatives, forcing O&M personnel to repeatedly manually verify device status. This not only significantly increases O&M costs but also may lead to delayed responses to abnormal conditions in critical equipment.

[0004] Therefore, a building intelligent operation and maintenance solution based on large models is desired. Summary of the Invention

[0005] In order to solve the above technical problems, this application is proposed.

[0006] According to one aspect of the present application, a large-scale model-based intelligent building operation and maintenance method is provided, which includes:

[0007] Obtaining a time queue of energy consumption data of a first device in the building;

[0008] A time queue for obtaining energy consumption data of a second device in the building;

[0009] Performing recursive message passing encoding based on local timing on the time queue of the energy consumption data of the first device to obtain a message passing encoding vector of a first device energy consumption timing pattern;

[0010] Based on the energy consumption timing pattern message transmission coding vector of the first device, perform energy consumption timing encoding and decoding on the time queue of the energy consumption data of the second device to obtain a generation time queue of the energy consumption data of the second device;

[0011] Calculating a time series distribution difference value of energy consumption data between a time queue of energy consumption data of the second device and a generation time queue of energy consumption data of the second device;

[0012] Based on the comparison between the time series distribution difference value of the energy consumption data and a preset threshold, it is confirmed whether the energy consumption of the second device is abnormal.

[0013] According to another aspect of the present application, a large-scale model-based intelligent building operation and maintenance system is provided, comprising:

[0014] An energy consumption data acquisition module for the first device, configured to acquire a time queue of energy consumption data of the first device in the building;

[0015] The energy consumption data acquisition module of the second device is used to acquire the time queue of the energy consumption data of the second device in the building;

[0016] a first device energy consumption time series encoding module, configured to perform recursive message passing encoding based on local time series on the time queue of the energy consumption data of the first device to obtain a first device energy consumption time series pattern message passing encoding vector;

[0017] A second device energy consumption timing encoding module is configured to perform energy consumption timing encoding and decoding on the time queue of the energy consumption data of the second device based on the energy consumption timing mode message transmission encoding vector of the first device to obtain a generation time queue of the energy consumption data of the second device;

[0018] an energy consumption data difference calculation module, configured to calculate a time series distribution difference value of energy consumption data between a time queue of energy consumption data of the second device and a generation time queue of energy consumption data of the second device;

[0019] The abnormality judgment module is used to confirm whether there is an abnormality in the energy consumption of the second device based on the comparison between the time series distribution difference value of the energy consumption data and a preset threshold.

[0020] Compared with the existing technology, the present application provides a large-scale model-based intelligent building operation and maintenance system and method, which adopts a data processing and analysis algorithm based on large-scale model artificial intelligence. The technical concept uses a large-scale model to construct an energy consumption reasoning mechanism with dynamic associations between devices. First, the energy consumption time series of the first device is convolutionally encoded in one dimension to extract the local time domain correlation characteristics of its operating mode, and then the dynamic evolution law of the characteristics on the time axis is captured through time series propagation coding. Subsequently, based on the device attribute semantic association matrix, the energy consumption characteristics of the first device are nonlinearly mapped to the semantic space of the second device, so that the energy consumption prediction of the second device can integrate the operating status influence of the upstream device. Finally, by comparing the time series distribution difference between the prediction sequence and the actual monitoring data, the hidden energy consumption fluctuations of the second device, such as those indirectly caused by the abnormality of the first device, are identified. This solution breaks through the limitations of traditional isolated modeling. Through dynamic feature transmission across the semantic space of devices, it accurately captures the energy transfer anomaly caused by the failure of the first device to the second device, and integrates the originally scattered threshold alarm events into a complete hidden fault propagation chain traceability analysis. It significantly reduces false alarms and missed alarms caused by the unmodeled coupling of working conditions between devices, improves the accuracy and timeliness of anomaly detection, and reduces the workload of manual verification. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0022] Figure 1 This is a flowchart of a large-model-based intelligent building operation and maintenance method according to an embodiment of the present application.

[0023] Figure 2 This is a flowchart of step S130 in the large-model-based intelligent building operation and maintenance method according to an embodiment of the present application.

[0024] Figure 3 This is a flowchart of step S132 in the large-model-based building intelligent operation and maintenance method according to an embodiment of the present application.

[0025] Figure 4 This is a flowchart of step S140 in the large model-based building intelligent operation and maintenance method according to an embodiment of the present application.

[0026] Figure 5 This is a block diagram of a large-model-based building intelligent operation and maintenance system according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] The following describes embodiments of the present disclosure in more detail with reference to the accompanying drawings. While the drawings illustrate certain embodiments of the present disclosure, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.

[0028] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in a different order and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0029] To address the challenges presented in the aforementioned background technology, the present application employs a data processing and analysis algorithm based on large-scale artificial intelligence (AI). This approach utilizes a large-scale model to construct an energy consumption inference mechanism for dynamic inter-device correlations. First, one-dimensional convolutional coding is performed on the energy consumption time series of the first device to extract the local temporal correlation features of its operating mode. Time series propagation coding is then used to capture the dynamic evolution of these features along the time axis. Subsequently, based on the semantic correlation matrix of device attributes, the energy consumption features of the first device are nonlinearly mapped to the semantic space of the second device, enabling the energy consumption prediction of the second device to incorporate the influence of the operating status of upstream devices. Finally, by comparing the time series distribution differences between the predicted sequence and the actual monitored data, the implicit energy consumption fluctuations of the second device indirectly caused by the first device's anomaly are identified. This approach overcomes the limitations of traditional isolated modeling. By dynamically propagating features across the semantic space of devices, it accurately captures the energy transfer anomalies caused by a first device failure on the second device. This approach integrates previously discrete threshold alarm events into a complete implicit fault propagation chain traceability analysis, significantly reducing false positives and missed negatives caused by unmodeled inter-device operating condition coupling, improving the accuracy and timeliness of anomaly detection, and reducing the workload of manual verification.

[0030] Figure 1 Flowchart of the intelligent building operation and maintenance method based on a large model according to an embodiment of the present application. Figure 1As shown, the large-model-based intelligent building operation and maintenance method according to the embodiment of the present application includes: S110, obtaining the time queue of energy consumption data of the first device in the building; S120, obtaining the time queue of energy consumption data of the second device in the building; S130, performing recursive message passing encoding based on local timing on the time queue of the energy consumption data of the first device to obtain the first device energy consumption timing pattern message passing encoding vector; S140, based on the first device energy consumption timing pattern message passing encoding vector, performing energy consumption timing encoding and decoding on the time queue of the energy consumption data of the second device to obtain the generation time queue of the energy consumption data of the second device; S150, calculating the energy consumption data timing distribution difference value between the time queue of the energy consumption data of the second device and the generation time queue of the energy consumption data of the second device; S160, based on the comparison between the energy consumption data timing distribution difference value and the preset threshold, confirming whether there is any abnormality in the energy consumption of the second device.

[0031] In step S110, a time series of energy consumption data for a first device in the building is obtained. It should be understood that a time series of energy consumption data for a first device in the building refers to a sequence of energy consumption data for the first device at different points in time, recorded in chronological order. For example, the energy consumption values of a first device (e.g., HVAC equipment) are collected at regular intervals (e.g., every minute, every hour, etc.), and these values are arranged in chronological order to form a time series of energy consumption data for that device. This data reflects the changes in energy consumption of the device over a period of time. In particular, the energy consumption of equipment in a building is an important basis for assessing the operating status, performance, and overall energy efficiency of the building. As one of the many devices in a building, the energy consumption data of the first device is crucial for understanding its operating status and its relationship to other devices. By obtaining a time series of energy consumption data, the device's energy consumption patterns and trends can be analyzed to identify possible anomalies or potential problems with the device. For example, if a device's energy consumption suddenly and significantly increases over a period of time, this may indicate a malfunction or abnormal operation, requiring further inspection and maintenance.

[0032] In step S120, a time queue of energy consumption data for a second device within the building is obtained. Accordingly, a time queue of energy consumption data for a second device within the building is a collection of energy consumption data for the second device at different times, arranged in chronological order. For example, for an elevator device in a building, power consumption is collected at fixed intervals (e.g., every 5 minutes). This time-varying power consumption data is arranged sequentially to form a time queue of energy consumption data for the elevator, the second device. This queue visually presents the dynamic changes in energy consumption for the device within a certain timeframe. As part of a building's equipment system, the energy consumption of the second device is an important factor in assessing the building's overall energy consumption and equipment operating status. Different devices have different energy consumption characteristics and patterns of change. Obtaining a time queue of energy consumption data for a second device facilitates analyzing the energy consumption pattern of each device individually, determining whether it exhibits abnormal energy consumption, such as excessive energy consumption or abnormal fluctuations, thereby promptly identifying potential equipment failures or inefficient operating conditions. In particular, in intelligent building operation and maintenance, devices do not operate in isolation but rather interact with each other in complex ways. Obtaining the time queue of the energy consumption data of the second device can be compared and correlated with the energy consumption data of the first device and other related devices, exploring the relationship between energy transfer, working condition coordination, etc. between devices, and providing data support for revealing the hidden fault propagation chain between devices.

[0033] In step S130 , recursive message passing encoding based on local timing is performed on the time queue of the energy consumption data of the first device to obtain a message passing encoding vector of the energy consumption timing pattern of the first device.

[0034] Figure 2 Flowchart of step S130 in the large model-based intelligent building operation and maintenance method according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 2 As shown, the step S130, performing local time series recursive message passing encoding on the time queue of the energy consumption data of the first device to obtain the first device energy consumption time series pattern message passing encoding vector, includes: S131, performing energy consumption time series pattern feature extraction based on one-dimensional convolutional coding on the time queue of the energy consumption data of the first device to obtain the sequence distribution of the local time domain correlation feature vector of the energy consumption of the first device; S132, performing time series propagation recursive structured encoding on the sequence distribution of the local time domain correlation feature vector of the energy consumption of the first device to obtain the first device energy consumption time series pattern message passing encoding vector.

[0035] Specifically, step S131 performs one-dimensional convolutional coding-based energy consumption time series pattern feature extraction on the time queue of the energy consumption data of the first device to obtain a sequence distribution of the local time-domain correlation feature vector of the first device's energy consumption. It should be understood that, given the complex non-stationary and multi-scale time series characteristics of device energy consumption data, traditional threshold alarm mechanisms and statistical analysis methods often use mean statistics or extreme value detection within a fixed time window, making it difficult to effectively capture spatially localized micro-time series patterns such as transient current spikes at the start and stop of air conditioner compressors and the step-shaped energy consumption curves unique to elevator acceleration. These features often carry potential signs of early equipment failure, such as abnormal fluctuations in motor starting current caused by bearing wear. However, due to the lack of refined extraction capabilities for local time-domain correlation features, traditional methods easily overwhelm these features within the overall energy consumption trend, resulting in insufficient sensitivity in anomaly detection. Therefore, the present application performs one-dimensional convolutional coding-based energy consumption time series pattern feature extraction on the time queue of the energy consumption data of the first device to obtain a sequence distribution of the local time-domain correlation feature vector of the first device's energy consumption. That is, deep feature mining is performed on the time queue of the energy consumption data of the first device through one-dimensional convolutional coding technology. Its essence is to use learnable convolution kernels to perform local sliding calculations along the time axis. This processing method can adaptively capture energy consumption fluctuation patterns at different time scales: larger convolution kernels identify long-period energy consumption transitions during the start-up and shutdown phases of the equipment, while smaller convolution kernels focus on high-frequency oscillation details during operation. By stacking multiple layers of convolution, a local time-domain correlation feature vector sequence with hierarchical characterization capabilities is eventually formed. For example, in HVAC equipment, the periodic energy consumption peak corresponding to the refrigerant cycle and the non-periodic energy consumption glitches caused by abnormal leakage can be accurately separated.

[0036] Specifically, step S132 performs time-series propagation recursive structured encoding on the sequence distribution of the local time-domain correlation feature vector of the first device's energy consumption to obtain the first device's energy consumption time-series pattern message transmission encoding vector. Furthermore, considering that the time-series characteristics of device energy consumption often exhibit complex dynamic evolution patterns, traditional methods employing statistical modeling with fixed time windows or simple recurrent neural network processing struggle to effectively capture implicit cross-time-step correlations, such as the gradual shift in ventilation equipment energy consumption within hours of an air conditioning system anomaly. While existing technologies can extract local time-domain features, they lack the ability to model the dynamic propagation mechanism of these features along the time axis, making it impossible to distinguish between fluctuations in normal equipment operating conditions and time-series pattern distortions caused by early anomalies. For example, when HVAC compressor bearings wear, their energy consumption curve exhibits a characteristic of gradually increasing harmonic components in specific frequency bands. These subtle time-series pattern changes, if analyzed only within isolated time windows, can be easily misidentified as random noise, ignoring their underlying systematic evolutionary trends. Based on this, the present application performs time-series propagation recursive structured coding on the sequence distribution of the local time-domain correlation feature vector of the energy consumption of the first device to obtain the first device energy consumption time-series pattern message transmission coding vector.

[0037] Specifically, the recurrent neural network is first used to capture the preliminary temporal dependencies of the local characteristics of the energy consumption of the first device, and a sequence of initial coding vectors for sequence transfer is generated. Subsequently, the dynamic importance of the feature vectors at different time steps is quantified through the time domain credibility adjustment coefficient (such as the higher diagnostic value of identifying the transient characteristics of the compressor start-stop stage compared to the steady-state operation stage), and the spatial domain credibility adjustment coefficient is introduced to reflect the structural influence of the device in the network topology (such as the key node attributes of the air-conditioning host compared to the terminal fan). The message transmission spatiotemporal collaborative adjustment coefficient formed by the fusion of the two can dynamically adjust the information flow intensity of the feature on the time propagation path, such as enhancing the continuous propagation weight of the abnormal harmonic component on the time axis, and suppressing the transmission attenuation of the normal working condition fluctuation signal. In this way, the message transmission coding vector of the energy consumption time series pattern of the first device obtained is a high-level feature representation of the original energy consumption data after multi-layer processing. It not only integrates local and global energy consumption feature information, but also takes into account the dynamic propagation and interaction of features in time series. It can more accurately characterize the energy consumption behavior pattern of the first device, and provide more representative and discriminative feature vectors for subsequent tasks such as equipment status assessment, fault diagnosis, and energy consumption prediction, which helps to improve the accuracy and reliability of these tasks.

[0038] Figure 3 Flowchart of step S132 in the large model-based intelligent building operation and maintenance method according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 3As shown, step S132, performs time propagation recursive structured coding on the sequence distribution of the local time domain correlation feature vector of the energy consumption of the first device to obtain the first device energy consumption time pattern message transmission coding vector, including: S132-1, performs sequence coding based on a recurrent neural network on the sequence distribution of the local time domain correlation feature vector of the energy consumption of the first device to obtain the sequence distribution of the sequence-transmitted first device energy consumption local time domain correlation initial feature vector; S132-2, performs message transmission based on time-space credibility adjustment on the sequence distribution of the sequence-transmitted first device energy consumption local time domain correlation initial feature vector to obtain the first device energy consumption time pattern message transmission coding vector.

[0039] Specifically, step S132-1 performs sequence encoding based on a recurrent neural network on the sequence distribution of the local time-domain correlation feature vector of the first device energy consumption to obtain a sequence distribution of the initial feature vector of the local time-domain correlation of the first device energy consumption. This process can be expressed by the following formula:

[0040] I={v1,v2,...,v i ,...,v t}

[0041] RNN(I)={h1,h2,...,h i ,...,h t}

[0042] Where I is the sequence distribution of the local time domain correlation feature vector of the first device energy consumption, v1, v2, v i and v t are the first, second, i-th and t-th local time-domain correlation feature vectors of the energy consumption of the first device in the sequence distribution of the local time-domain correlation feature vector of the energy consumption of the first device, RNN(I) is the sequence encoding of I based on the RNN structure, h1, h2, h i and h t The first device energy consumption local time domain correlation initial feature vectors are transmitted for the first, second, i-th and t-th sequences in the sequence distribution of the first device energy consumption local time domain correlation initial feature vectors respectively.

[0043] It should be understood that the temporal characteristics of equipment energy consumption often have long-range dependencies and dynamic evolution characteristics. Although feature extraction using one-dimensional convolution can separate micro-fluctuations in equipment operation, the features of each time window are isolated, making it difficult to characterize the systematic evolution of energy consumption patterns during equipment aging. For example, when an HVAC evaporator is frosted, its energy consumption curve will show a long-term trend of periodic peak amplitude gradually decaying. If this feature correlation spanning dozens of sampling points relies solely on local convolutional coding, it is easily segmented into discrete abnormal fragments and loses key information about the evolution path. Therefore, by performing sequence encoding based on a recurrent neural network on the sequence distribution of the local time-domain correlation feature vectors of the first device energy consumption to perform deep time-series modeling on the sequence of the local time-domain correlation feature vectors of the first device energy consumption, a sequence distribution of the initial feature vectors of the local time-domain correlation of the first device energy consumption is obtained. The recurrent connection mechanism of the RNN gives it an inherent temporal memory capability, which can dynamically couple the convolutional features of the current moment with the hidden state of the previous time step. For example, when an air conditioner is in the variable frequency speed regulation phase, its energy consumption curve exhibits periodic oscillation characteristics. By continuously updating the hidden state, the RNN can couple the compressor frequency variation characteristics captured in the previous time window with the power fluctuation characteristics at the current moment for analysis. This temporal memory property enables the system to identify the characteristic propagation patterns generated by the air conditioning system during specific operating phases (such as cooling mode switching), and these patterns often have a strong correlation with the abnormal energy consumption of subsequent ventilation equipment with a time delay.

[0044] Specifically, the step S132-2 performs message transmission based on time-space credibility adjustment on the sequence distribution of the sequence-transmitted first device energy consumption local time domain associated initial feature vector to obtain the first device energy consumption timing pattern message transmission coding vector, including: calculating the first device energy consumption local time domain associated feature time domain credibility adjustment coefficient of each sequence-transmitted first device energy consumption local time domain associated initial feature vector in the sequence distribution of the sequence-transmitted first device energy consumption local time domain associated initial feature vector; calculating the first device energy consumption local time domain associated feature space domain credibility adjustment coefficient of each sequence-transmitted first device energy consumption local time domain associated initial feature vector in the sequence distribution of the sequence-transmitted first device energy consumption local time domain associated initial feature vector; based on The time domain credibility adjustment coefficient of the first device energy consumption local time domain association characteristic and the space domain credibility adjustment coefficient of the first device energy consumption local time domain association characteristic of each sequence transmission of the first device energy consumption local time domain association initial feature vector are used to construct the time-space coordination adjustment coefficient of the first device energy consumption local time domain association characteristic message transmission of the first device energy consumption local time domain association initial feature vector transmitted by each sequence; based on the time-space coordination adjustment coefficient of the first device energy consumption local time domain association characteristic message transmission of the first device energy consumption local time domain association initial feature vector transmitted by each sequence, the message transmission aggregation based on time-space regulation is performed on the first device energy consumption local time domain association initial feature vector transmitted by each sequence to obtain the first device energy consumption timing pattern message transmission coding vector.

[0045] Specifically, in the embodiment of the present application, the time domain credibility adjustment coefficient of the first device energy consumption local time domain correlation feature of each sequence-transmitted first device energy consumption local time domain correlation initial feature vector in the sequence distribution of the sequence-transmitted first device energy consumption local time domain correlation initial feature vector is calculated. This process can be expressed by the following formula:

[0046]

[0047] Among them, v i is the i-th first device energy consumption local time domain correlation feature vector in the sequence distribution of the first device energy consumption local time domain correlation feature vector, h i The first local time domain associated initial feature vector of energy consumption of the first device in the sequence transmission is the i-th sequence transmission of the first local time domain associated initial feature vector of energy consumption of the first device in the sequence distribution, W 1i It is h i The corresponding learnable weight matrix, W 2i It is v i The corresponding learnable weight matrix, α and β are trainable weighted hyperparameters, tanh is the hyperbolic tangent activation function, is matrix multiplication, v t is the scoring weight vector, It is hi The corresponding first device energy consumption local time domain correlation feature energy score, softmax is a normalization function, It is h i The corresponding time-domain credibility adjustment coefficient of the local time-domain correlation characteristic of the energy consumption of the first device.

[0048] Accordingly, the time series data of equipment energy consumption often has significant temporal heterogeneity characteristics. For example, the operating status of the air-conditioning system during the morning and evening peak hours will fluctuate violently due to sudden changes in ambient temperature, while the energy consumption curve during regular time periods is stable. If the time series processing method based on uniform weighting is used, it is difficult to distinguish the difference in data quality between different time periods. When occasional noise from the sensor or short-term overload of the equipment causes a sudden drop in the credibility of the feature vector of the local time period, if all time step features are treated equally without discrimination, it is very easy to introduce a noise amplification effect in the subsequent cross-device reasoning link, resulting in misjudgment of abnormal detection. In the present application, the time domain credibility adjustment coefficient of the first device energy consumption local time domain correlation feature of each sequence-transmitted first device energy consumption local time domain correlation initial feature vector in the sequence distribution of the sequence-transmitted first device energy consumption local time domain correlation initial feature vector is calculated. That is, the time domain credibility adjustment coefficient is introduced to dynamically adjust the weight of the sequence-transmitted first device energy consumption local time domain correlation initial feature vector. This mechanism, based on an attention mechanism, constructs an adaptive evaluation model capable of capturing fluctuations in data quality across different air conditioning system operation phases. During stable operation, feature vectors receive high confidence scores due to data stability. However, during periods of compressor startup and shutdown, or when sensors experience occasional packet loss, the system automatically reduces the weight coefficient for the corresponding time step. For example, when dust accumulation on the air conditioning filter causes a gradual increase in fan power, the time-domain confidence adjustment coefficient increases the weight of mid-term operating data, suppressing interference from the initial stable phase and the later abnormal saturation phase, and precisely focusing on the critical time window for fault feature propagation.

[0049] Specifically, in the embodiment of the present application, the spatial credibility adjustment coefficient of the first device energy consumption local time domain correlation feature of each sequence-transmitted first device energy consumption local time domain correlation initial feature vector in the sequence distribution of the sequence-transmitted first device energy consumption local time domain correlation initial feature vector is calculated. This process can be expressed by the following formula:

[0050]

[0051] Among them, h i The first local time domain associated initial feature vector of energy consumption of the first device in the sequence transmission is the i-th sequence transmission of the first local time domain associated initial feature vector of energy consumption of the first device, h i T It is h i The transposed vector of To calculate the square of the F norm, s(hi ) is h i The corresponding first device energy consumption local time domain correlation feature space similarity score value, exp is the exponential function value with the natural constant e as the base, It is h i The corresponding spatial domain credibility adjustment coefficient of the local time domain correlation feature of the first device energy consumption.

[0052] It should be understood that when analyzing the energy consumption characteristics of the first device, not only the time series information is important, but also the spatial structural relationship of the device in the entire system will have an impact on its energy consumption characteristics. Calculating the spatial credibility adjustment coefficient can take into account the spatial relationship and structural importance of the node (device) in the structured network, and comprehensively analyze the energy consumption characteristics. Therefore, by calculating the spatial credibility adjustment coefficient of the first device energy consumption local time domain correlation characteristic of each sequence-transmitted first device energy consumption local time domain correlation initial feature vector in the sequence distribution of the sequence-transmitted first device energy consumption local time domain correlation initial feature vector. That is, a spatial credibility adjustment coefficient is assigned to each vector in the sequence distribution of the sequence-transmitted first device energy consumption local time domain correlation initial feature vector, similar to the spatial confidence constraint factor that assigns a quantitative value to the node spatial confidence, so as to characterize the importance of the device (node) in the energy consumption characteristic spatial structure and the information dissemination potential.

[0053] Specifically, in an embodiment of the present application, based on the temporal credibility adjustment coefficient of the first device energy consumption local time domain correlation feature and the spatial credibility adjustment coefficient of the first device energy consumption local time domain correlation feature, each sequence transmits the first device energy consumption local time domain correlation initial feature vector, and constructs the spatiotemporal coordination adjustment coefficient of the first device energy consumption local time domain correlation feature message transmission. This process can be expressed by the following formula:

[0054]

[0055] in, It is h i The corresponding time domain credibility adjustment coefficient of the local time domain correlation characteristic of the first device energy consumption, It is h i The corresponding spatial domain credibility adjustment coefficients of the local time domain correlation characteristics of the first device energy consumption, ω1 and ω2 are and The contribution adjustment weighting coefficient, sigmoid is the sigmoid function, It is h i The corresponding first device energy consumption local time domain correlation characteristic message transmission time and space coordination adjustment coefficient.

[0056] Accordingly, the propagation of equipment abnormality characteristics is subject to the dual constraints of temporal dynamics and spatial structure. The bearing wear fault of the air-conditioning system presents a characteristic of gradual energy consumption increase in the time dimension, and the spatial path of the terminal equipment affected by the air duct network topology has primary and secondary differences: the ventilation equipment closer to the fault source will show abnormalities first, while the remote equipment may show delayed response due to the damping effect of the pipe network. The traditional single-dimensional confidence assessment method cannot capture this spatiotemporal coupling characteristic, and it is easy to treat the early abnormal characteristics of key nodes and the random fluctuations of edge nodes in the same way, resulting in the dilution of the core propagation path information during cross-device reasoning. To this end, based on the time domain credibility adjustment coefficient of the first device energy consumption local time domain correlation feature and the spatial domain credibility adjustment coefficient of the first device energy consumption local time domain correlation feature, the first device energy consumption local time domain correlation feature message transmission spatiotemporal collaborative adjustment coefficient of the sequence transmission of the first device energy consumption local time domain correlation initial feature vector is constructed. This step utilizes a gated fusion mechanism to dynamically balance the contribution weights of the temporal credibility adjustment coefficient (such as the diagnostic value of transient characteristics during compressor startup and shutdown) and the spatial credibility adjustment coefficient (such as the structural importance of the host node in the duct network). When processing HVAC systems, the model automatically increases the collaborative weights of specific time windows (periods of significant leak characteristics) and key nodes (main air valve controllers) during the initial stages of a refrigerant leak. This allows the cyclical pressure fluctuations caused by host bearing wear to be tracked continuously in terms of their amplitude growth trend in the temporal dimension while also strengthening their propagation intensity at core nodes in the spatial dimension, forming an anomaly propagation map with joint spatiotemporal and temporal sensitivity.

[0057] More specifically, in an embodiment of the present application, based on the spatiotemporal coordination adjustment coefficient of the first device energy consumption local time domain correlation feature message transmission of each sequence-transmitted first device energy consumption local time domain correlation initial feature vector, the message transmission aggregation based on spatiotemporal regulation is performed on each sequence-transmitted first device energy consumption local time domain correlation initial feature vector to obtain the first device energy consumption timing pattern message transmission coding vector, including:

[0058] The spatiotemporal collaborative adjustment coefficient of the first device energy consumption local time domain correlation characteristic message transmission of each sequence transmitting the first device energy consumption local time domain correlation initial characteristic vector is subjected to curvature-driven spatiotemporal multimode fusion geometric regularization to obtain the spatiotemporal collaborative optimization adjustment coefficient of the first device energy consumption local time domain correlation characteristic message transmission of each sequence transmitting the first device energy consumption local time domain correlation initial characteristic vector. This process can be expressed by the following formula:

[0059]

[0060]

[0061] in, It is h iThe corresponding time domain credibility adjustment coefficient of the local time domain correlation characteristic of the first device energy consumption, It is h i The corresponding spatial domain credibility adjustment coefficient of the local time domain correlation feature of the first device energy consumption, It is h i The corresponding first device energy consumption local time domain correlation feature message transmission time and space coordination adjustment coefficient, G i It is h i The corresponding first device energy consumption local time domain correlation characteristic constant curvature space mapping value, sin is the sine function, π is the pi, R i It is h i The corresponding hypersphere asymptotic representation value of the local time-domain correlation characteristic of the first device energy consumption, It is h i The corresponding first device energy consumption local time domain correlation characteristic spatiotemporal coupling metric weight factor, yes The optimized spatiotemporal collaborative optimization adjustment coefficient of the first device energy consumption local time domain correlation feature message transmission;

[0062] Based on the spatiotemporal collaborative optimization adjustment coefficient of the first device energy consumption local time domain correlation feature message transmission of each sequence transmission first device energy consumption local time domain correlation initial feature vector, the first device energy consumption local time domain correlation initial feature vector transmitted by each sequence is weighted and calculated by position to obtain the first device energy consumption timing pattern message transmission encoding vector. This process can be expressed by the following formula:

[0063]

[0064] Among them, h i The i-th sequence transmission first device energy consumption local time domain correlation initial feature vector in the sequence distribution of the sequence transmission first device energy consumption local time domain correlation initial feature vector, yes The optimized first device energy consumption local time domain correlation feature message transmission spatiotemporal collaborative optimization adjustment coefficient, t is the number of vectors in the sequence distribution of the sequence transmission first device energy consumption local time domain correlation initial feature vector, z is the first device energy consumption timing pattern message transmission encoding vector.

[0065] In particular, the time domain credibility adjustment coefficient of the local time domain correlation characteristic of the first device energy consumption and the spatial domain credibility adjustment coefficient of the local time domain correlation characteristic of the first device energy consumption Characterize the local time domain correlation initial feature vector h of the first device energy consumption of each sequence transmission iUnder the condition of the weight distribution of information in the time and space dimensions, if the time and space dimensions are regarded as a global space-time single degree of freedom modeling paradigm, it is necessary to avoid the non-flat representation of the space-time composite manifold caused by the negative correlation effect in the unimodal attention mechanism to prevent the degradation of the representation efficiency of the multidimensional fusion space.

[0066] Based on this, firstly, the local time domain associated initial feature vector h of the first device energy consumption is transmitted according to each sequence i Corresponding time domain credibility adjustment coefficient of the local time domain correlation characteristic of the first device energy consumption Spatial domain credibility adjustment coefficient of the local temporal domain correlation feature of the first device energy consumption Construct the constant curvature space mapping value G of the local time domain correlation feature of the first device energy consumption i And the hypersphere asymptotic representation value R of the local time domain correlation characteristic of the first device energy consumption i :

[0067]

[0068] Then, the metric tensor benchmark of the collaborative space is used as the spatiotemporal joint constraint benchmark for message passing, and the initial feature vector h associated with the local time domain energy consumption of the first device is transmitted for each sequence. i The local time domain correlation characteristic message transmission time and space coordination adjustment coefficient of the first device energy consumption Perform iterative optimization:

[0069]

[0070] When the constraints When established, the unimodal representation in the space-time dimension and The approach approaches the cooperative attractor of a quasi-flat spacetime, thus reflecting the quasi-Euclidean properties of the message transfer adjustment coefficient on a composite manifold. This method achieves geometry-preserving optimization of the fusion space by regularizing the negative curvature of a single modality, significantly improving the multimodal fusion representation capability of the spacetime cooperative adjustment coefficient.

[0071] Finally, based on the spatiotemporal collaborative optimization adjustment coefficient of the first device energy consumption local time domain correlation feature message transmission of the first device energy consumption local time domain correlation initial feature vector transmitted by each sequence, a weighted sum calculation is performed on the first device energy consumption local time domain correlation initial feature vector transmitted by each sequence to obtain the first device energy consumption time series pattern message transmission encoding vector. In this way, through weighted sum calculation, the model can adaptively enhance the abnormal characteristics transmitted by the air-conditioning system through the critical path (main air duct pressure monitoring point) during a specific operating stage (such as the end of the refrigerant charging cycle), so that the ventilation equipment prediction sequence generated in the subsequent cross-device mapping stage can accurately reflect the impact of such spatiotemporal coupling type anomalies.

[0072] In step S140 , based on the energy consumption timing pattern message transmission coding vector of the first device, energy consumption timing encoding and decoding is performed on the time queue of the energy consumption data of the second device to obtain a generation time queue of the energy consumption data of the second device. Figure 4 Flowchart of step S140 in the large model-based intelligent building operation and maintenance method according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 4 As shown, based on the energy consumption timing pattern message transmission coding vector of the first device, the time queue of the energy consumption data of the second device is energy consumption timing encoded and decoded to obtain the generation time queue of the energy consumption data of the second device, including: S141, based on the energy consumption timing pattern message transmission coding vector of the first device, the time queue of the energy consumption data of the second device is energy consumption timing mapping inference encoded to obtain the second device energy consumption timing pattern inference coding vector; S142, energy consumption timing feature decoding and prediction of the energy consumption timing pattern inference coding vector of the second device to obtain the generation time queue of the energy consumption data of the second device.

[0073] Specifically, in an embodiment of the present application, S141, based on the energy consumption timing pattern message passing coding vector of the first device, the time queue of the energy consumption data of the second device is subjected to energy consumption timing mapping inference coding to obtain the energy consumption timing pattern inference coding vector of the second device, including: constructing an attribute semantic association matrix between the first device and the second device; using the attribute semantic association matrix to perform energy consumption timing feature cross-semantic space mapping on the energy consumption timing pattern message passing coding vector of the first device to obtain the energy consumption timing pattern inference coding vector of the second device.

[0074] It should be understood that implicit associations between devices often involve complex physical coupling and energy transfer mechanisms. For example, while the air supply temperature of an air conditioning system and the fan speed of a ventilation system are different device parameters, they form a strong correlation through duct static pressure balance. Meanwhile, the illumination adjustment of a lighting system and the air conditioning load may only have a weak energy coupling relationship. Rule-based or statistical association analysis methods, limited by the static and one-sided nature of manually defined association rules, struggle to adaptively capture these dynamic semantic associations across systems. This leads to frequent misjudgments, such as anomaly detection models incorrectly associating air conditioning compressor failures with lighting system energy consumption fluctuations. Based on this, the technical solution of this application constructs an attribute semantic association matrix between the first and second devices. In other words, the attribute semantic association matrix is a quantitative representation of the relationships between devices, providing a foundational data structure for various subsequent analysis tasks. Whether diagnosing equipment faults, predicting energy consumption, or optimizing equipment operation strategies, understanding the relationships between devices is essential. The attribute semantic association matrix clearly presents these relationships, providing a basis for further calculations and reasoning.

[0075] Specifically, the attribute semantic association matrix between the first device and the second device is constructed, and the specific implementation process is as follows:

[0076] The first step is data collection and organization. With the help of intelligent platforms such as the building automation system (BAS) and energy management system (EMS) deployed in the building, the operating parameters and energy consumption time series data of the first device and the second device are continuously collected. These data cover the basic properties of the equipment (such as equipment type, specifications, etc.), operating status parameters (such as temperature, pressure, speed, etc.) and energy consumption data (energy consumption values at different time points). At the same time, the spatial location information of the equipment is collected to clarify their layout in the building, because the spatial location may affect the energy transfer and interaction between devices. For example, air-conditioning equipment and ventilation equipment on the same floor and close to each other may be more closely related.

[0077] Next, feature engineering is performed. The collected equipment operating parameters and energy consumption data are cleaned to remove outliers and noise to ensure data quality. Key features are then extracted, such as trends in equipment energy consumption, periodicity, and correlations with other equipment. For different types of equipment, appropriate features are selected based on their operating principles and characteristics. For example, for air conditioning systems, cooling capacity, heating power, and compressor operating frequency are important features; for lighting systems, illumination and operating duration are more critical. These features are quantified and standardized to make the features of different devices comparable.

[0078] The similarity between the devices is then calculated. Based on the extracted and processed features, an appropriate similarity metric is used to calculate the degree of similarity between the first and second devices. Common methods include Euclidean distance and cosine similarity. Taking Euclidean distance as an example, the distance between two devices in feature space is calculated. The closer the distance, the higher the similarity between the devices and the greater the potential for association. For example, if the Euclidean distance between the first air conditioning device and the second ventilation device is small in terms of multiple features such as energy consumption trends and operating time, it indicates that they are similar in these aspects and may be associated.

[0079] Finally, an attribute semantic association matrix is constructed. Based on the calculated similarity results between devices, a two-dimensional matrix is constructed. The rows and columns of the matrix correspond to the first device and the second device respectively, and the values of the matrix elements represent the degree of association between the two devices. For device pairs with high similarity, the values of the corresponding elements in the matrix are larger; for device pairs with low similarity, the values of the corresponding elements are smaller. For example, if the first device and the second device have high similarity in multiple features, the values of the elements at the corresponding positions in the matrix may be close to 1; if the two devices have large differences in most features, the values of the corresponding elements may be close to 0. The attribute semantic association matrix constructed through the above implementation process can effectively quantify the complex relationship between the first device and the second device, and provide strong support for the subsequent use of the matrix to map the energy consumption timing pattern message encoding vector of the first device across the semantic space, thereby improving the accuracy and reliability of equipment energy consumption anomaly detection in building intelligent operation and maintenance.

[0080] Accordingly, it is considered that the propagation of anomalies between devices often involves complex nonlinear associations and mapping difficulties in heterogeneous semantic spaces. If a simple linear weighting or statistical correlation analysis is used, it is difficult to model the non-uniform impact of air pressure fluctuations in an air-conditioning system on ventilation equipment through a specific topological path. For example, when the static pressure of the air-conditioning host increases due to filter clogging, the degree of its impact on the ventilation equipment on each floor depends not only on the physical distance, but also on aerodynamic properties such as the duct branch angle and the local resistance coefficient. Due to the lack of a dynamic mapping mechanism across semantic spaces, the existing technology often simplifies the association between devices into a fixed weight transfer, resulting in the inability to accurately reflect the attenuation law and superposition effect of air-conditioning anomaly characteristics in the pipeline network when predicting the energy consumption of ventilation equipment. Based on this, the present application uses the attribute semantic association matrix to map the energy consumption timing characteristics of the first device energy consumption timing pattern message transmission encoding vector across the semantic space to obtain the second device energy consumption timing pattern inference encoding vector. In this way, after obtaining the second device energy consumption timing pattern inference encoding vector, the energy consumption of the second device can be predicted based on this. Because this encoding vector incorporates the energy consumption characteristics of the first device and the semantic associations between the attributes of the devices, the prediction results can more comprehensively reflect the actual energy consumption of the second device, achieving higher accuracy than traditional methods that rely solely on the second device's own historical data. For example, when predicting the energy consumption of ventilation equipment, considering the energy consumption characteristics of air conditioning equipment and the associations between them can more accurately predict the energy consumption of ventilation equipment under different operating conditions.

[0081] Specifically, in this embodiment of the present application, step S142, decoding and predicting the energy consumption time series characteristics of the inferred energy consumption time series pattern encoding vector of the second device to obtain the generation time queue of the energy consumption data of the second device, includes: using an RNN-based decoder to decode and predict the energy consumption time series characteristics of the inferred energy consumption time series pattern encoding vector of the second device to obtain the generation time queue of the energy consumption data of the second device. It should be understood that the inferred energy consumption time series pattern encoding vector of the second device inherently contains energy consumption characteristic information related to the time series. RNNs have a natural advantage in processing time series data. Their internal loop structure allows information to be transferred between different time steps, effectively capturing temporal dependencies in the data. In energy consumption data, the current energy consumption status of a device is often related to its past energy consumption status, such as during the device's warm-up process or gradual changes in load. RNNs can effectively model such dynamic changes and are therefore suitable for decoding and predicting the encoding vector. In particular, the energy consumption patterns of building equipment are often complex and may exhibit periodic and seasonal variations, as well as sudden abnormal fluctuations. RNNs are capable of learning these complex patterns, memorizing past information by continuously updating hidden states, and making predictions based on current input and memorized information. The RNN-based decoder can decode the complex energy consumption features contained in the energy consumption timing pattern inference encoding vector of the second device, thereby more accurately restoring the time series of energy consumption data.

[0082] Specifically, the process of using the RNN-based decoder to decode and predict the energy consumption timing feature of the second device energy consumption timing pattern inference coding vector to obtain the generation time queue of the energy consumption data of the second device is as follows:

[0083] Before decoding prediction, the accuracy and validity of the inferred encoding vector for the second device's energy consumption timing pattern must be ensured. This encoding vector is obtained by constructing an attribute semantic association matrix between the first and second devices and mapping the encoding vector of the first device's energy consumption timing pattern message transmission across a semantic space. It integrates the energy consumption characteristics of the first device and the semantic association information between the devices.

[0084] Once the encoding vector is prepared, it can be input into the RNN-based decoder. RNN has a unique advantage in processing time series data. Its internal loop structure allows information to be transferred between different time steps, effectively capturing the temporal dependencies in the data. This is crucial for analyzing the energy consumption data of building equipment that varies over time.

[0085] Inside the RNN decoder, the hidden state is first initialized. The hidden state acts like a "memory cell" in the RNN, storing information from previous time steps for use in subsequent calculations. The initial hidden state is typically set to an all-zero vector or an empirically determined initial value.

[0086] As the encoding vector is input sequentially over time, the RNN gradually updates its hidden state. At each time step, the RNN combines the current input encoding vector with the hidden state from the previous time step to perform calculations. Specifically, using a specific weight matrix and activation function, the input and hidden state are linearly combined and transformed to produce a new hidden state. For example, using activation functions such as tanh or ReLU to perform nonlinear transformations on the result of the linear combination allows the RNN to learn more complex time series patterns.

[0087] While updating the hidden state, the RNN decoder also generates a prediction output based on the current hidden state. The prediction output is an estimate of the energy consumption of the second device at the current time step. This prediction output is based on the energy consumption time series characteristics learned by the RNN and information from previous time steps.

[0088] After calculating over multiple time steps, the RNN decoder gradually generates a series of predicted outputs. These outputs, arranged in chronological order, form the time queue for the generation of the second device's energy consumption data. To improve the accuracy and stability of the predictions, a large amount of historical energy consumption data is used as training samples when training the RNN decoder. By continuously adjusting the weight parameters in the RNN, the decoder's predicted outputs are kept as close as possible to the actual energy consumption data. Common training methods include the backpropagation algorithm, which calculates the prediction error and updates the weights based on the gradient of the error, allowing the RNN to gradually learn the inherent patterns of the energy consumption data.

[0089] In practical applications, RNN-based decoders can be optimized and expanded based on specific needs. For example, the number of RNN layers can be increased to form a deep RNN to learn more complex features, or improved RNN structures such as LSTM (Long Short-Term Memory) and GRU (Gated Recurrent Unit) can be adopted. These structures can better handle long-term dependencies and avoid gradient vanishing or gradient exploding phenomena, thereby further improving the accuracy of energy consumption prediction.

[0090] In step S150, the difference in the time series distribution of energy consumption data between the time series of the energy consumption data of the second device and the time series of the energy consumption data generated by the second device is calculated. Accordingly, since the time series of the energy consumption data generated by the second device is predicted by the model, while the actual time series of the energy consumption data is observed, calculating the difference between the two can directly reflect the degree of deviation between the prediction results generated by the RNN-based decoder and the actual situation, thereby evaluating the accuracy of the energy consumption prediction model.

[0091] Specifically, first, we need to choose a suitable calculation method to measure the difference between the two. Common methods include mean square error (MSE), mean absolute error (MAE), and dynamic time warping (DTW) distance.

[0092] Taking the mean squared error (MSE) as an example, the calculation process is relatively straightforward. The energy consumption data for corresponding time points in the two time queues are subtracted to obtain the error value for each time point. These error values are then squared to amplify the effect of the error and eliminate the possibility of positive and negative errors canceling each other out. Next, the squared errors for all time points are added together and divided by the total number of time points to obtain the mean squared error value. A larger value indicates a greater difference between the two time queues; conversely, a smaller value indicates a smaller difference.

[0093] The mean absolute error (MAE) is calculated by directly taking the absolute difference between the energy consumption data at corresponding time points in two time queues, adding up all the absolute errors, and dividing by the total number of time points. This method provides a more intuitive reflection of the average error between the predicted and actual values.

[0094] The calculation of the Dynamic Time Warping distance is relatively complex. It is suitable for situations where there is a local time offset between two time series. This method finds the best matching path between the two time series and calculates the cumulative distance along this path as the difference value.

[0095] In practice, the appropriate calculation method is selected based on specific needs and data characteristics. Once the difference value is calculated, it serves as an important basis for determining whether the second device's energy consumption is abnormal. It quantifies the degree of deviation between the predicted data and the actual data, providing objective data support for subsequent determination of abnormal device energy consumption based on preset thresholds, helping to promptly identify potential device failures or abnormal operating conditions.

[0096] In step S160, based on the comparison between the time series distribution difference value of the energy consumption data and the preset threshold, it is confirmed whether there is an abnormality in the energy consumption of the second device. It should be understood that the preset threshold is a reference value set based on factors such as the historical energy consumption data, operating characteristics, and relevant standards of the device. The time series distribution difference value of the energy consumption data reflects the degree of deviation between the predicted energy consumption and the actual energy consumption. By comparing it with the preset threshold, the judgment of energy consumption anomalies can be quantified, avoiding the uncertainty of subjective judgment and providing an objective and measurable evaluation standard. Specifically, the energy consumption of the equipment usually has a certain regularity and stability. When the time series distribution difference value of the energy consumption data exceeds the preset threshold, it means that there is a large deviation between the actual energy consumption and the expected energy consumption. This is likely to be caused by factors such as equipment failure, changes in the operating environment, and improper human operation. Timely comparison can detect these potential problems as early as possible so that corresponding measures can be taken to deal with them and prevent the problem from further deteriorating. By confirming whether there is any abnormality in the energy consumption of the second device, possible failures or other problems of the equipment can be discovered in a timely manner, so that operation and maintenance personnel can conduct timely inspections and maintenance to ensure the normal operation of the equipment, avoid adverse consequences such as production interruptions and service quality degradation caused by equipment failure, and ensure the stability and reliability of the entire system.

[0097] In summary, the large-scale model-based intelligent building operation and maintenance method according to the embodiment of the present application is explained, which adopts a data processing and analysis algorithm based on large-scale model artificial intelligence. The technical concept constructs an energy consumption reasoning mechanism with dynamic associations between devices through a large model. First, the energy consumption time series of the first device is convolutionally encoded in one dimension to extract the local time domain correlation features of its operating mode, and then the dynamic evolution law of the features on the time axis is captured through time series propagation coding. Subsequently, based on the semantic association matrix of the device attributes, the energy consumption features of the first device are nonlinearly mapped to the semantic space of the second device, so that the energy consumption prediction of the second device can integrate the operating status influence of the upstream device. Finally, by comparing the time series distribution difference between the prediction sequence and the actual monitoring data, the implicit energy consumption fluctuation of the second device indirectly caused by the abnormality of the first device is identified. This solution breaks through the limitations of traditional isolated modeling. Through dynamic feature transmission across the semantic space of devices, it accurately captures the energy transfer anomaly caused by the failure of the first device to the second device, and integrates the originally scattered threshold alarm events into a complete hidden fault propagation chain traceability analysis. It significantly reduces false alarms and missed alarms caused by the unmodeled coupling of working conditions between devices, improves the accuracy and timeliness of anomaly detection, and reduces the workload of manual verification.

[0098] Figure 5 FIG is a block diagram of a building intelligent operation and maintenance system based on a large model according to an embodiment of the present application. Figure 5As shown, according to an embodiment of the present application, the large-scale model-based intelligent building operation and maintenance system 100 includes: a first device energy consumption data acquisition module 110, which is used to obtain a time queue of energy consumption data of the first device in the building; a second device energy consumption data acquisition module 120, which is used to obtain a time queue of energy consumption data of the second device in the building; a first device energy consumption timing encoding module 130, which is used to perform recursive message passing encoding based on local timing on the time queue of the energy consumption data of the first device to obtain a first device energy consumption timing pattern message passing encoding vector; a second device energy consumption timing encoding module 140, which is used to perform energy consumption timing encoding and decoding on the time queue of the energy consumption data of the second device based on the first device energy consumption timing pattern message passing encoding vector to obtain a generation time queue of the energy consumption data of the second device; an energy consumption data difference calculation module 150, which is used to calculate the energy consumption data timing distribution difference value between the time queue of the energy consumption data of the second device and the generation time queue of the energy consumption data of the second device; and an abnormality judgment module 160, which is used to confirm whether there is an abnormality in the energy consumption of the second device based on a comparison between the energy consumption data timing distribution difference value and a preset threshold.

[0099] Here, those skilled in the art will appreciate that the specific operations of each step in the above-mentioned large-scale model-based building intelligent operation and maintenance system have been described in detail in the referenced examples. Figures 1 to 4 The description of the large model-based building intelligent operation and maintenance method has been introduced in detail, and therefore, its repeated description will be omitted.

[0100] As described above, the large-model-based intelligent building operation and maintenance system 100 according to the embodiment of the present disclosure can be implemented in various wireless terminals, such as a server equipped with a large-model-based intelligent building operation and maintenance algorithm. In one possible implementation, the large-model-based intelligent building operation and maintenance system 100 according to the embodiment of the present disclosure can be integrated into the wireless terminal as a software module and / or hardware module. For example, the large-model-based intelligent building operation and maintenance system 100 can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the large-model-based intelligent building operation and maintenance system 100 can also be one of the many hardware modules of the wireless terminal.

[0101] Alternatively, in another example, the large model-based building intelligent operation and maintenance system 100 and the wireless terminal may also be separate devices, and the large model-based building intelligent operation and maintenance system 100 may be connected to the wireless terminal through a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0102] In summary, it is intended that the above detailed description be regarded as illustrative rather than restrictive, and it should be understood that the above embodiments should be construed as merely for illustrating the present invention rather than for limiting the scope of protection of the present invention.

Claims

1. A building intelligent operation and maintenance method based on a large model, characterized in that: include: Obtaining a time queue of energy consumption data of a first device in the building; A time queue for obtaining energy consumption data of a second device in the building; Performing recursive message passing encoding based on local timing on the time queue of the energy consumption data of the first device to obtain a message passing encoding vector of a first device energy consumption timing pattern; Based on the energy consumption timing pattern message transmission coding vector of the first device, perform energy consumption timing encoding and decoding on the time queue of the energy consumption data of the second device to obtain a generation time queue of the energy consumption data of the second device; Calculating a time series distribution difference value of energy consumption data between a time queue of energy consumption data of the second device and a generation time queue of energy consumption data of the second device; Based on the comparison between the time series distribution difference value of the energy consumption data and a preset threshold, it is confirmed whether the energy consumption of the second device is abnormal.

2. The large-model-based intelligent building operation and maintenance method according to claim 1, characterized in that: Performing recursive message passing encoding based on local time sequence on the time queue of the energy consumption data of the first device to obtain a message passing encoding vector of the energy consumption time sequence pattern of the first device includes: Performing energy consumption time series pattern feature extraction based on one-dimensional convolutional coding on the time queue of the energy consumption data of the first device to obtain a sequence distribution of a local time domain correlation feature vector of the energy consumption of the first device; Perform time propagation recursive structured coding on the sequence distribution of the local time domain correlation feature vector of the energy consumption of the first device to obtain the first device energy consumption time sequence pattern message transmission coding vector.

3. The large-model-based intelligent building operation and maintenance method according to claim 2, characterized in that: Performing time-series propagation recursive structured coding on the sequence distribution of the local time-domain correlation feature vector of the energy consumption of the first device to obtain the first device energy consumption time-series pattern message transmission coding vector, including: Performing sequence encoding based on a recurrent neural network on the sequence distribution of the local time-domain correlation feature vector of the energy consumption of the first device to obtain a sequence distribution of the local time-domain correlation initial feature vector of the energy consumption of the first device transmitted sequentially; Message transmission based on time-space credibility adjustment is performed on the sequence distribution of the sequence transmission first device energy consumption local time domain associated initial feature vector to obtain the first device energy consumption timing pattern message transmission coding vector.

4. The large-model-based intelligent building operation and maintenance method according to claim 3 is characterized in that: Performing message transmission based on spatiotemporal credibility adjustment on the sequence distribution of the sequence transmission first device energy consumption local time domain associated initial feature vector to obtain the first device energy consumption timing pattern message transmission coding vector, including: Calculating a time domain credibility adjustment coefficient of a first device energy consumption local time domain correlation feature of each sequence transmitting the first device energy consumption local time domain correlation initial feature vector in the sequence distribution of the sequence transmitting the first device energy consumption local time domain correlation initial feature vector; Calculating a spatial domain credibility adjustment coefficient of a first device energy consumption local time domain correlation feature of each sequence-transmitted first device energy consumption local time domain correlation initial feature vector in the sequence distribution of the sequence-transmitted first device energy consumption local time domain correlation initial feature vector; Based on the time domain credibility adjustment coefficient of the first device energy consumption local time domain correlation feature and the spatial domain credibility adjustment coefficient of the first device energy consumption local time domain correlation feature of each sequence transmitting the first device energy consumption local time domain correlation initial feature vector, constructing the time and space coordination adjustment coefficient of the first device energy consumption local time domain correlation feature message transmission of each sequence transmitting the first device energy consumption local time domain correlation initial feature vector; Based on the spatiotemporal collaborative adjustment coefficient of the first device energy consumption local time domain correlation characteristic message transmission of the local time domain correlation initial characteristic vectors of the first device energy consumption transmitted in each sequence, the spatiotemporal modulation-based message transmission aggregation is performed on the local time domain correlation initial characteristic vectors of the first device energy consumption transmitted in each sequence to obtain the first device energy consumption timing pattern message transmission coding vector.

5. The large-model-based intelligent building operation and maintenance method according to claim 4 is characterized in that: Based on the first device energy consumption local time domain correlation characteristic message transmission spatiotemporal coordination adjustment coefficient of each sequence transmission first device energy consumption local time domain correlation initial characteristic vector, performing spatiotemporal modulation-based message transmission aggregation on each sequence transmission first device energy consumption local time domain correlation initial characteristic vector to obtain the first device energy consumption timing pattern message transmission coding vector, including: Performing curvature-driven spatiotemporal multi-mode fusion geometric regularization on the spatiotemporal collaborative adjustment coefficient of the first device energy consumption local time domain correlation characteristic message transmission of each sequence transmitting the first device energy consumption local time domain correlation initial characteristic vector to obtain the spatiotemporal collaborative optimization adjustment coefficient of the first device energy consumption local time domain correlation characteristic message transmission of each sequence transmitting the first device energy consumption local time domain correlation initial characteristic vector; Based on the spatiotemporal collaborative optimization adjustment coefficient of the first device energy consumption local time domain associated initial feature vector transmitted in each sequence, the first device energy consumption local time domain associated initial feature vector transmitted in each sequence is weighted and calculated by position to obtain the first device energy consumption timing pattern message transmission coding vector.

6. The large-model-based intelligent building operation and maintenance method according to claim 5, characterized in that: Based on the energy consumption timing pattern message transmission coding vector of the first device, performing energy consumption timing encoding and decoding on the time queue of the energy consumption data of the second device to obtain a generation time queue of the energy consumption data of the second device, including: Based on the first device energy consumption timing pattern message transmission coding vector, performing energy consumption timing mapping reasoning coding on the time queue of the energy consumption data of the second device to obtain the second device energy consumption timing pattern reasoning coding vector; The energy consumption timing pattern inference coding vector of the second device is decoded and predicted to obtain a generation time queue of the energy consumption data of the second device.

7. The large-model-based intelligent building operation and maintenance method according to claim 6, characterized in that: Based on the energy consumption timing pattern message transmission coding vector of the first device, performing energy consumption timing mapping inference coding on the time queue of the energy consumption data of the second device to obtain the energy consumption timing pattern inference coding vector of the second device, including: Constructing an attribute semantic association matrix between the first device and the second device; The attribute semantic association matrix is used to perform energy consumption timing feature cross-semantic space mapping on the first device energy consumption timing pattern message delivery encoding vector to obtain the second device energy consumption timing pattern reasoning encoding vector.

8. The large-model-based intelligent building operation and maintenance method according to claim 7, characterized in that: The energy consumption timing pattern inference coding vector of the second device is decoded and predicted with energy consumption timing characteristics to obtain the generation time queue of the energy consumption data of the second device, including: using an RNN-based decoder to decode and predict the energy consumption timing pattern inference coding vector of the second device to obtain the generation time queue of the energy consumption data of the second device.

9. A building intelligent operation and maintenance system based on a large model, characterized by: include: An energy consumption data acquisition module for the first device, configured to acquire a time queue of energy consumption data of the first device in the building; The energy consumption data acquisition module of the second device is used to acquire the time queue of the energy consumption data of the second device in the building; a first device energy consumption time series encoding module, configured to perform recursive message passing encoding based on local time series on the time queue of the energy consumption data of the first device to obtain a first device energy consumption time series pattern message passing encoding vector; A second device energy consumption timing encoding module is configured to perform energy consumption timing encoding and decoding on the time queue of the energy consumption data of the second device based on the energy consumption timing mode message transmission encoding vector of the first device to obtain a generation time queue of the energy consumption data of the second device; an energy consumption data difference calculation module, configured to calculate a time series distribution difference value of energy consumption data between a time queue of energy consumption data of the second device and a generation time queue of energy consumption data of the second device; The abnormality judgment module is used to confirm whether there is an abnormality in the energy consumption of the second device based on the comparison between the time series distribution difference value of the energy consumption data and a preset threshold.

Images