A Building Operation and Maintenance Management Method and System Based on Big Data

Abstract

The present invention discloses a building operation and maintenance management method and system based on big data, which relates to the field of building intelligent technologies. It includes collecting multi-source heterogeneous data, performing data standardization processing to generate a standard data stream; constructing a virtual building model based on digital twin technology, mapping the standard data stream to corresponding device nodes of the virtual building model to generate a device feature matrix; inputting the device feature matrix into a deep learning network, training to generate an energy consumption prediction model, and outputting an energy consumption prediction sequence; constructing a multi-layer perceptron health scoring model based on the energy consumption prediction sequence, and generating a fault warning priority list in combination with device weight coefficients; automatically generating maintenance work orders according to the fault warning priority list. Through maintenance effect evaluation and a secondary maintenance mechanism, the present invention forms a complete closed-loop management process, ensures the maintenance quality and continuously optimizes the maintenance strategy, thereby improving the overall operation and maintenance management level.

Description

Technical Field

[0001] The present invention relates to the technical field of building intelligence, and particularly to a building operation and maintenance management method and system based on big data. Background Art

[0002] With the rapid development of intelligent buildings, the intelligent and automated levels of building operation and maintenance management have been continuously improved. Currently, most intelligent buildings have installed various sensors and automated control systems, which can realize basic environmental monitoring, energy metering, and equipment control functions. However, these systems often operate independently and lack a unified data management and analysis platform, resulting in the inability to fully utilize data resources and low operation and maintenance management efficiency.

[0003] Traditional building operation and maintenance management mainly relies on manual inspections and regular maintenance, and has the following problems: it is difficult to grasp the real-time operation status of equipment, and the fault warning ability is insufficient; the maintenance plan lacks pertinence and cannot formulate a reasonable maintenance strategy according to the actual status of the equipment; the scheduling efficiency of maintenance resources is low, and the response time is long; the management of maintenance records is not standardized, and it is difficult to trace the whole-life cycle data of equipment. These problems seriously restrict the improvement of the building operation and maintenance management level. Summary of the Invention

[0004] In view of the problems existing in the existing building operation and maintenance management system, such as scattered data collection, insufficient prediction ability, and low maintenance efficiency, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is how to realize the unified collection and analysis of building equipment operation data through Internet of Things and big data technologies, establish an equipment health status evaluation and fault warning mechanism, improve the scheduling efficiency of maintenance resources, and thus improve the intelligent level of building operation and maintenance management.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a building operation and maintenance management method based on big data, which includes collecting multi-source heterogeneous data through an Internet of Things sensor network, performing data standardization processing to generate a standard data stream; constructing a virtual building model based on digital twin technology, mapping the standard data stream to corresponding device nodes of the virtual building model, and using a spatio-temporal attention mechanism to extract device operation features to generate a device feature matrix; inputting the device feature matrix into a deep learning network, training to generate an energy consumption prediction model, and outputting an energy consumption prediction sequence; constructing a multi-layer perceptron health scoring model based on the energy consumption prediction sequence, combining with device weight coefficients to generate a fault warning priority list; automatically generating maintenance work orders according to the fault warning priority list, performing work order allocation based on the equipment maintenance difficulty coefficient and the skill level of operation and maintenance personnel, pushing them to the operation and maintenance personnel, and feeding back the work order execution results to the virtual building model for status update.

[0008] As a preferred solution of the building operation and maintenance management method based on big data according to the present invention, wherein: constructing a virtual building model based on digital twin technology includes the following steps: obtaining building information model data of the building, constructing a three-dimensional geometric model of the building through a three-dimensional modeling engine, and generating a virtual building skeleton model; based on the virtual building skeleton model, calculating and constructing a spatial position relationship matrix; constructing a spatial attention network based on the spatial position relationship matrix, and generating a spatial association weight between equipment rooms;

[0009] Organize the standard data stream in time sequence, and extract the dynamic time sequence features of equipment operation through a time attention network; fuse the spatial association weight and the dynamic time sequence features to generate a spatio-temporal fusion feature vector; organize the spatio-temporal fusion feature vector into an equipment feature matrix.

[0010] As a preferred solution of the building operation and maintenance management method based on big data according to the present invention, wherein: training and generating an energy consumption prediction model includes the following steps: sliding and splitting the equipment feature matrix according to a preset time window to generate a sample set, and dividing the sample set into a training sample set and a validation sample set according to a preset ratio; constructing a bidirectional long short-term memory network as the backbone network of the energy consumption prediction model, inputting the training sample set into the bidirectional long short-term memory network, and extracting a time sequence feature vector; setting a multi-head attention calculation unit in the output layer of the bidirectional long short-term memory network to generate a fusion feature vector; inputting the fusion feature vector into a fully connected layer to obtain an energy consumption prediction value, and calculating the mean square error loss value between the energy consumption prediction value and the actual energy consumption value; according to the calculated mean square error loss value, using an adaptive learning rate optimization algorithm to update network parameters; inputting the validation sample set into the bidirectional long short-term memory neural network, calculating the validation loss value, and using an early stopping mechanism based on the convergence of the validation loss to select the optimal model parameters to obtain an energy consumption prediction model; inputting the real-time equipment feature matrix into the energy consumption prediction model to output an energy consumption prediction sequence.

[0011] As a preferred solution of the building operation and maintenance management method based on big data according to the present invention, wherein: generating a list of fault warning priorities includes the following steps: performing a time series difference operation on the energy consumption prediction sequence by using a difference calculation unit to obtain an energy consumption change trend array; calling a time series pattern recognition module to extract features from the energy consumption change trend array to obtain a fluctuation feature value, a trend slope value, and a maximum energy consumption difference; inputting the fluctuation feature value, the trend slope value, and the maximum energy consumption difference into a pre-trained multi-layer perceptron to output an equipment health risk score; reading the equipment weight coefficient in the equipment configuration database, multiplying the equipment weight coefficient by the equipment health risk score to obtain an equipment warning priority score; performing a descending sorting algorithm on the equipment warning priority score to generate a list of equipment fault warning priorities.

[0012] As a preferred solution of the building operation and maintenance management method based on big data according to the present invention, wherein: automatically generating a maintenance work order according to the fault warning priority list includes: calling a maintenance knowledge base, matching a corresponding standard maintenance process based on the equipment number and the warning level, and generating an initial maintenance work order; constructing an equipment maintenance difficulty evaluation model, and calculating an equipment maintenance difficulty coefficient according to the equipment type, the fault complexity, and the maintenance man-hours; establishing an operation and maintenance personnel skill evaluation system, and quantifying the skill level based on the work experience, professional qualifications, and historical maintenance success rate of the operation and maintenance personnel; querying the operation and maintenance personnel resource pool database, matching the equipment maintenance difficulty coefficient with the skill level, and selecting operation and maintenance personnel meeting the qualification requirements by using a task allocation algorithm to establish a work order allocation form; pushing the initial maintenance work order to the mobile terminal of the operation and maintenance personnel through the operation and maintenance management platform, triggering the work order execution process, and recording the work order receiving time; receiving the work order execution data transmitted back by the operation and maintenance personnel through the mobile terminal, and updating the status information of the corresponding equipment node in the virtual building model based on the work order execution data; monitoring the operation status of the equipment after maintenance, collecting the performance data of the first operation cycle after maintenance, comparing and analyzing the performance data with the performance data before maintenance, and generating a maintenance effect evaluation report; judging whether it is necessary to start a secondary maintenance process according to the result of the maintenance effect evaluation report.

[0013] As a preferred solution of the building operation and maintenance management method based on big data according to the present invention, wherein: obtaining the energy consumption change trend array includes the following steps: constructing a time window matrix, and segmenting the energy consumption prediction sequence according to a preset window length N; performing a first-order difference operation on the time window matrix, calculating the difference between adjacent predicted energy consumption values within each time window, and generating an initial difference array; performing median filtering on the initial difference array to generate a smoothed difference array; multiplying the smoothed difference array by a time step correction coefficient to obtain an energy consumption change rate array per unit time; performing wavelet transform on the energy consumption change rate array to extract the main trend component of the energy consumption change, and obtaining the energy consumption change trend array.

[0014] As a preferred solution of the building operation and maintenance management method based on big data according to the present invention, wherein: the construction process of the virtual building skeleton model includes: importing building information model data, and the building information model data includes floor plan drawings, equipment layout drawings, and pipeline distribution drawings; calling a geometric conversion engine to parse the building information model data and output a standard geometric description language file; parsing the standard geometric description language file, extracting building component features, and establishing a component feature table; performing a mesh dissection operation based on the standard geometric description language file to generate a three-dimensional mesh data structure; loading the three-dimensional mesh data structure and the building component feature table, and calling a hierarchical reconstruction algorithm to assemble spatial components to generate a virtual building skeleton model.

[0015] In a second aspect, an embodiment of the present invention provides a building operation and maintenance management system based on big data, which includes a data collection module for collecting multi-source heterogeneous data through an Internet of Things sensor network, performing data standardization processing, and generating a standard data stream; a mapping module for constructing a virtual building model based on digital twin technology, mapping the standard data stream to corresponding device nodes of the virtual building model, and using a spatio-temporal attention mechanism to extract device operation characteristics and generate a device feature matrix; a prediction and analysis module for inputting the device feature matrix into a deep learning network, training to generate an energy consumption prediction model, and outputting an energy consumption prediction sequence; an evaluation and early warning module for constructing a multi-layer perceptron health scoring model based on the energy consumption prediction sequence, and generating a fault early warning priority list in combination with device weight coefficients; a work order management module for automatically generating repair work orders according to the fault early warning priority list, performing work order allocation based on the device repair difficulty coefficient and the skills level of operation and maintenance personnel, pushing them to operation and maintenance personnel, and feeding back the work order execution results to the virtual building model for status update.

[0016] The beneficial effects of the present invention are as follows: Through the combination of the Internet of Things sensor network and digital twin technology, the present invention realizes the real-time mapping between the physical building and the virtual model, provides an intuitive three-dimensional visualization interface, and greatly improves the efficiency of device management and fault location. The deep learning architecture that combines bidirectional long short-term memory network and multi-head attention mechanism can accurately capture the temporal characteristics of device operation data, significantly improving the accuracy of energy consumption prediction and the prediction time span. The fault early warning mechanism based on device health scoring and weight coefficients, combined with the intelligent work order allocation system, realizes the timely discovery and rapid response to faults, improving the initiative and pertinence of operation and maintenance work. Through the maintenance effect evaluation and secondary maintenance mechanism, a complete closed-loop management process is formed, ensuring the maintenance quality and continuously optimizing the maintenance strategy, and improving the overall operation and maintenance management level. Brief Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0018] Figure 1 It is a framework flowchart of a building operation and maintenance management method based on big data.

[0019] Figure 2 It is a flowchart for obtaining a device feature matrix of a building operation and maintenance management method based on big data.

[0020] Figure 3 It is a flowchart for obtaining a fault early warning priority list of a building operation and maintenance management method based on big data. Detailed implementation manners

[0021] To make the above objects, features and advantages of the present invention more obvious and understandable, the following detailed description of the specific implementation manners of the present invention will be given with reference to the accompanying drawings of the specification.

[0022] In the following description, many specific details are set forth to facilitate a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0023] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure or characteristic that may be included in at least one implementation manner of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or selectively exclusive embodiments from other embodiments.

[0024] Embodiment 1, referring to Figures 1 to 3 , which is the first embodiment of the present invention. This embodiment provides a building operation and maintenance management method based on big data. The framework flowchart is as Figure 1 shown, including

[0025] S1: Collect multi-source heterogeneous data through the Internet of Things sensor network, perform data standardization processing, and generate a standard data stream.

[0026] Specifically, configure the Internet of Things sensor network. The Internet of Things sensor network includes environmental monitoring sensors, energy monitoring devices, and occupancy sensors. Among them, the environmental monitoring sensors include temperature sensors, humidity sensors, and air quality sensors. The energy monitoring devices include smart meters and smart water meters. The occupancy sensors include people flow density sensors, space usage status sensors, and equipment operation status sensors; collect multi-source heterogeneous data of the Internet of Things sensor network through edge devices. The multi-source heterogeneous data includes temperature and humidity data, air quality index data, energy consumption data, equipment status data, and space occupancy data, and add a unified time stamp to the collected multi-source heterogeneous data for data annotation; perform format conversion on the multi-source heterogeneous data with the unified time stamp added, such as converting BIM data to IFC format and converting energy data to B2MML format to generate format-standardized data; perform integration processing on the format-standardized data, reorganize it according to the data type and time sequence relationship, and generate a standard data stream.

[0027] Preferably, by adopting an Internet of Things sensor network with a hierarchical architecture, the data acquisition efficiency is improved, and the system stability is enhanced. The data preprocessing function of the edge device effectively reduces the data transmission load. The addition of a unified timestamp solves the problem of inconsistent time series of data from different sources.

[0028] S2: Construct a virtual building model based on digital twin technology, map the standard data stream to the corresponding device nodes of the virtual building model, and adopt a spatio-temporal attention mechanism to extract the dynamic characteristics of the device operation state to generate a device feature matrix.

[0029] Specifically, the flowchart for obtaining the device feature matrix is as Figure 2 shown and includes the following steps:

[0030] S2.1: Obtain the building information model data of the building, construct a three-dimensional geometric model of the building through a three-dimensional modeling engine, and generate a virtual building skeleton model.

[0031] Specifically, import the building information model data, call the geometric conversion engine to perform format parsing on the building information model data, and output a standard geometric description language file. The building information model data includes floor plans, equipment layout plans, and pipeline distribution plans; parse the standard geometric description language file, extract the geometric parameters and spatial information of wall components, door and window components, and staircase components, and establish a component feature table. The component feature table records the spatial coordinates, shape parameters, and material properties of each component; read the floor plan data in the standard geometric description language file, set the mesh division parameters, perform adaptive mesh division operations, and generate a three-dimensional mesh data structure; load the three-dimensional mesh data structure and the building component feature table, call the hierarchical reconstruction algorithm, and assemble the spatial components in the vertical stratification order to output the virtual building skeleton model.

[0032] Furthermore, parse the spatial data of the equipment layout plan, calculate the spatial coordinates of the equipment installation location, mark the device nodes in the virtual building skeleton model, and generate a device location information table; read the topological data of the pipeline distribution plan, calculate the pipeline spatial path, draw the pipeline orientation in the virtual building skeleton model, and establish a pipeline connection topology table; perform a spatial interference detection algorithm on the virtual building skeleton model, the device location information table, and the pipeline connection topology table to eliminate component collision points and output an optimized three-dimensional geometric model; extract the material attribute data in the building component feature table, perform a material mapping operation, and attach the material data to the surface of the three-dimensional geometric model to generate a visual model with material effects; measure the geographical coordinate reference point of the physical building, calculate the coordinate transformation matrix, and map the local coordinates of the visual model to the geographical coordinate system to output a calibrated virtual building skeleton model.

[0033] In specific implementation, the geometric transformation engine selects Open Cascade or VTK library to implement BIM format parsing and conversion. The adaptive mesh generation is based on the quadtree / octree algorithm, with the basic mesh size set to 0.5m, the local refinement threshold set to 0.1m, and the mesh quality factor not less than 0.85 to ensure the mesh quality. The hierarchical reconstruction algorithm uses a spatial octree structure and, in the order from bottom to top, sequentially completes the assembly of walls, doors and windows, stairs, equipment, and pipelines within each layer. The spatial interference detection uses a hierarchical bounding box (AABB tree) for fast detection and performs precise triangular mesh interference detection on the colliding components. The coordinate transformation uses a seven-parameter transformation model, and the coordinates of the reference points are obtained through GPS measurement to ensure centimeter-level accuracy.

[0034] It should be noted that in this step, the adaptive mesh generation algorithm is used, which can automatically adjust the mesh density according to the complexity of different regions, optimizing the use of computing resources while ensuring the model accuracy.

[0035] S2.2: Based on the equipment location information table and pipeline connection topology table in the virtual building skeleton model, calculate and construct a spatial position relationship matrix.

[0036] Among them, the construction process of the spatial position relationship matrix is as follows: Extract the equipment node coordinates and floor numbers from the equipment location information table, and calculate the weighted Euclidean distance d between equipment i and equipment j ij , where the vertical weighting coefficient is set for the distance between different floors; Extract the pipeline type information from the pipeline connection topology table, and determine the connection coefficient c between equipment i and equipment j according to the weight coefficients of different pipeline types ij ; Use the association strength calculation formula to construct an N×N spatial position relationship matrix M, where N is the total number of equipment, and the matrix element m ij represents the spatial association strength between equipment i and equipment j.

[0037] Specifically, the association strength calculation formula is as follows:

[0038]

[0039] Among them, and are the weight coefficients of the distance term and the connection term respectively (0.6 and 0.4 in this embodiment), d0 is the distance attenuation coefficient (10 meters in this embodiment, which can be adjusted according to the actual building scale), d ij is the weighted Euclidean distance between equipment i and equipment j, and c ij is the weighted connection coefficient between equipment i and equipment j.

[0040] For example, the weight coefficients of pipeline types are set based on the importance of device energy flow and information flow. For example, the weight of the water supply pipeline is 1.0, the weight of the drainage pipeline is 1.0, the weight of the HVAC pipeline is 1.5, the weight of the power pipeline is 1.2, and the weight of the network pipeline is 0.8. In addition, when there is no connection between device i and device j, c ij is 0. When there is a connection between device i and device j, the weighted connection coefficient c ij is calculated according to the following formula:

[0041]

[0042] where K ij is the set of all pipeline types connecting device i and device j, is the weight coefficient corresponding to pipeline type k. When there are multiple types of pipeline connections between two devices, the maximum value of the weight coefficients of each pipeline type is taken as the weighted connection coefficient. This calculation method reflects the maximum influence degree of different types of pipelines on the device correlation.

[0043] In specific implementation, the distance between devices is calculated using the three-dimensional Euclidean distance formula, and the vertical weighting coefficient is set to 1.5 to enhance the influence between floors. To optimize the storage space, the device connection coefficients are stored using an adjacency matrix and represented using a sparse matrix, and the correlation strength matrix only stores the upper triangular part using its symmetry. All matrix elements are normalized, and the value range is limited to the interval [0,1].

[0044] S2.3: Construct a spatial attention network based on the spatial position relationship matrix to generate the spatial correlation weight between devices.

[0045] Specifically, the spatial position relationship matrix M is input into K attention heads, and each attention head contains a query matrix, a key matrix, and a value matrix; calculate the attention scores for each attention head; splice the outputs of the K attention heads, and obtain the spatial attention features through a linear transformation; normalize the spatial attention features to obtain the spatial correlation weight matrix between devices.

[0046] S2.4: Organize the standard data stream in time sequence and extract the dynamic time sequence features of device operation through a time attention network.

[0047] Specifically, group the standard data stream according to device ID, segment the time series data of each device using a sliding window, and the window size is T; construct a bidirectional long short-term memory network for the segmented sequence to extract local time series patterns; use the self-attention mechanism to calculate the importance weights at different time points, and weighted aggregation to obtain the dynamic time series feature vector.

[0048] In specific implementation, the sliding window size is set to 24 to capture the daily cycle pattern, and the step size is 1 to ensure fine-grained feature extraction. The BiLSTM network adopts a two-layer structure, the hidden layer dimension is 128, and the dropout rate is set to 0.3 to prevent overfitting. The temporal attention mechanism introduces 32-dimensional relative position encoding to enhance the temporal modeling ability. The final feature vector dimension is 256, including short-term and long-term temporal information. The batch normalization layer is used to process the features to accelerate the training convergence.

[0049] S2.5: Fuse the spatial correlation weight and the dynamic temporal features to generate a spatio-temporal fusion feature vector.

[0050] Specifically, perform matrix multiplication on the spatial correlation weight matrix W and the dynamic temporal feature vector of each device to obtain the temporal features considering spatial correlation; adaptively fuse the original temporal features and the spatially weighted features through a gating mechanism, and output the spatio-temporal fusion feature vector.

[0051] S2.6: Organize the spatio-temporal fusion feature vectors into a device feature matrix.

[0052] Furthermore, arrange the spatio-temporal fusion feature vectors of each device in the order of device ID to construct an N×d-dimensional device feature matrix, where N is the number of devices and d is the feature dimension. Each row represents the spatio-temporal fusion features of a device.

[0053] It should be noted that through the multi-head attention mechanism and the gating fusion mechanism, the present invention realizes the adaptive fusion of the spatial relationship and the temporal features between devices, and improves the feature expression ability.

[0054] Preferably, by introducing the spatio-temporal attention mechanism and the adaptive feature fusion strategy, the present invention realizes the dynamic feature extraction of the device operating state, and overcomes the limitation of the traditional method that ignores the spatial correlation between devices.

[0055] S3: Input the device feature matrix into the deep learning network, train to generate an energy consumption prediction model, and output an energy consumption prediction sequence.

[0056] Specifically, it includes the following steps:

[0057] S3.1: Slide and split the device feature matrix according to a preset time window to generate a sample set containing historical operation data segments, and divide the sample set into a training sample set and a validation sample set according to a preset ratio.

[0058] Among them, each sample contains a device operation parameter sequence and the corresponding actual energy consumption value, and the ratio of the training sample set to the validation sample set is set to 8:2.

[0059] S3.2: Construct a bidirectional long short-term memory network as the backbone network of the energy consumption prediction model, input the device operation parameter sequence of the training sample set into the bidirectional long short-term memory network, and extract the time series feature vector.

[0060] S3.3: Set a multi-head attention calculation unit in the output layer of the bidirectional long short-term memory network, calculate the weights of the time series feature vector, and generate a fused feature vector.

[0061] S3.4: Input the fused feature vector into the fully connected layer to obtain the energy consumption prediction value, and calculate the mean square error loss value between the energy consumption prediction value and the actual energy consumption value.

[0062] S3.5: According to the calculated mean square error loss value, use the adaptive learning rate optimization algorithm to update the network parameters.

[0063] Specifically, use the chain rule to calculate the gradient value of the network parameters with respect to the loss function, and use the backpropagation algorithm to update the gradient matrix layer by layer; initialize the Adam optimizer parameters, set the first moment estimation coefficient, the second moment estimation coefficient, and the numerical stability coefficient; calculate the exponentially weighted moving average of the gradient matrix, and update the first moment estimation vector; calculate the exponentially weighted moving average of the square of the gradient matrix, and update the second moment estimation vector; perform bias correction on the first moment estimation vector and the second moment estimation vector to obtain the corrected first moment estimation vector and the second moment estimation vector; calculate the adaptive learning rate according to the corrected moment estimation vector; update the network parameters based on the calculated adaptive learning rate and the corrected first moment estimation vector.

[0064] S3.6: Input the validation sample set into the bidirectional long short-term memory neural network, calculate the validation loss value, and use the early stopping mechanism based on the convergence of the validation loss to select the optimal model parameters to obtain the energy consumption prediction model.

[0065] Among them, using the early stopping mechanism based on the convergence of the validation loss to select the optimal model parameters includes: setting a validation loss monitoring window, recording the validation loss value after each round of training, calculating the moving average of the validation loss within the monitoring window, and constructing a smoothed loss curve; calculating the change rate of the validation loss between adjacent rounds, judging the rising or falling trend of the loss according to the positive or negative of the change rate, and counting the number of consecutive rising rounds; judging whether the number of consecutive rising rounds exceeds the preset threshold, if it exceeds the threshold, trigger the early stopping condition, and record the training round number when triggered; trace back the historical validation loss records, locate the training round corresponding to the minimum validation loss value, and extract the network parameters of this round as the optimal model parameters.

[0066] S3.6: Input the real-time device feature matrix into the energy consumption prediction model, and output the energy consumption prediction sequence for the next M time points.

[0067] Preferably, by using the bidirectional long short-term memory network structure and the multi-head attention mechanism in combination, the accuracy of energy consumption prediction is improved, and the prediction lead time is extended. The adaptive learning rate optimization algorithm is adopted to accelerate the model convergence speed.

[0068] S4: Construct a multi-layer perceptron health scoring model based on the energy consumption prediction sequence, and generate a fault warning priority list in combination with the device weight coefficient.

[0069] Specifically, the flowchart for obtaining the fault warning priority list is as Figure 3 shown, including the following steps:

[0070] S4.1: Use the differential calculation unit to perform time series difference operation on the energy consumption prediction sequence to obtain the energy consumption change trend array.

[0071] Specifically, construct a time window matrix, segment the energy consumption prediction sequence according to the preset window length N, where each time window contains the predicted energy consumption values of adjacent N time points; perform a first-order difference operation on the time window matrix, calculate the difference between adjacent predicted energy consumption values within each time window to generate an initial difference array; perform median filtering on the initial difference array to eliminate the sudden interference generated during the difference process and generate a smoothed difference array; multiply the smoothed difference array by the time step correction coefficient to obtain the energy consumption change rate array per unit time; perform wavelet transform on the energy consumption change rate array to extract the main trend component of the energy consumption change, filter out the high-frequency noise, and obtain the energy consumption change trend array.

[0072] S4.2: Call the time series pattern recognition module to extract features from the energy consumption change trend array to obtain the fluctuation feature value, trend slope value, and maximum energy consumption difference.

[0073] Specifically, resample the energy consumption change trend array at a fixed sampling frequency to generate an equally spaced time series sample to ensure the time consistency of data analysis; use a sliding window variance calculator to calculate the local variance value of the equally spaced time series sample, and select the variance peak value as the fluctuation feature value; call the least squares fitting module to perform linear regression operation on the equally spaced time series sample, and extract the slope of the fitting line as the trend slope value; search for the maximum difference between adjacent data points in the equally spaced time series sample based on the dynamic programming algorithm, and record this difference as the maximum energy consumption difference; combine the fluctuation feature value, trend slope value, and maximum energy consumption difference into a feature vector for subsequent health risk score calculation.

[0074] S4.3: Input the fluctuation feature value, trend slope value, and maximum energy consumption difference into the pre-trained multi-layer perceptron to output the device health risk score.

[0075] Specifically, normalization processing is performed on the fluctuation eigenvalue, trend slope value, and maximum energy consumption difference to generate a feature normalization vector. The normalization uses the maximum-minimum normalization method. A three-layer perceptron network structure is constructed. The first layer contains 8 neurons, the second layer contains 4 neurons, and the output layer contains 1 neuron. The ReLU activation function is used to connect between layers. The feature normalization vector is input into the neurons of the first layer, and the output vector of the first layer is calculated through the weight matrix W1 and the bias vector b1. The output vector of the first layer is input into the neurons of the second layer after being processed by the ReLU activation function, and the output vector of the second layer is calculated through the weight matrix W2 and the bias vector b2. The output vector of the second layer is input into the neurons of the output layer after being processed by the ReLU activation function, and the health risk score is calculated through the weight matrix W3 and the bias vector b3.

[0076] S4.4: Read the device weight coefficient in the device configuration database, multiply the device weight coefficient by the device health risk score to obtain the device warning priority score.

[0077] S4.5: Execute the descending sorting algorithm to sort the device warning priority scores and generate a device failure warning priority list.

[0078] Among them, the device failure warning priority list includes the device number, warning score, and warning level.

[0079] S5: Automatically generate a maintenance work order according to the failure warning priority list, perform work order allocation based on the device maintenance difficulty coefficient and the skill level of the operation and maintenance personnel, push it to the operation and maintenance personnel, and feedback the work order execution result to the virtual building model for status update.

[0080] In a specific embodiment, the implementation process is as follows:

[0081] S5.1: Call the maintenance knowledge base, match the corresponding standard maintenance process based on the device number and warning level, and generate an initial maintenance work order.

[0082] Specifically, read the device number, obtain the device type code, device model parameters, and device installation location information from the device basic database, and construct a device feature identification set; use the feature vector matching algorithm to retrieve the maintenance knowledge base based on the device feature identification set, and extract the corresponding device maintenance procedure data, including standard maintenance steps, safety precautions, and maintenance tool requirements; input the warning level into the preset maintenance classification rule table, and extract the maintenance response requirements corresponding to the level, including maintenance response time limit, maintenance personnel qualification requirements, and maintenance acceptance standards; extract the real-time operation parameters, historical maintenance records, and historical exception records of the corresponding device from the virtual building model, and combine them with the maintenance procedure data to generate a fault analysis report, including abnormal parameter description, possible fault causes, and recommended inspection items; match the spare parts management database based on the fault analysis report to generate a recommended spare parts list, including spare parts number, spare parts name, and recommended quantity; organize and integrate the maintenance procedure data, maintenance response requirements, fault analysis report, and recommended spare parts list, and typeset them according to the work order template format standard to generate an initial maintenance work order.

[0083] S5.2: Construct a device maintenance difficulty evaluation model, and calculate the device maintenance difficulty coefficient according to the device type, fault complexity, and maintenance man-hours.

[0084] First, establish a device complexity benchmark score table based on the device type, and the benchmark score range for different types of devices is 0-100; second, construct a fault complexity evaluation matrix, and the matrix dimensions include fault type, fault range, and fault level, and each dimension is set with a weight coefficient; then, train a random forest model based on historical maintenance data to predict the maintenance man-hours; finally, perform a weighted sum of the benchmark score, fault complexity score, and standardized maintenance man-hours to obtain the device maintenance difficulty coefficient, and the coefficient range is 0-1.

[0085] S5.3: Establish an operation and maintenance personnel skill evaluation system, and quantify the skill level based on the work experience, professional qualifications, and historical maintenance success rate of the operation and maintenance personnel.

[0086] First, construct a skill evaluation index system, including basic indexes such as working years, professional qualification level, and certification status, as well as performance indexes such as historical maintenance completion rate and maintenance quality score; second, use the analytic hierarchy process to determine the weight of each index; then, calculate the comprehensive skill score based on the fuzzy comprehensive evaluation method; finally, use the Min-Max normalization method to map the comprehensive score to the 0-1 interval to obtain the standardized skill level score.

[0087] S5.4: Query the operation and maintenance personnel resource pool database, match the device maintenance difficulty coefficient with the skill level, and use the task assignment algorithm to select operation and maintenance personnel who meet the qualification requirements to establish a work order assignment table.

[0088] First, the system obtains information such as the on-duty status and workload of the operation and maintenance personnel resource pool in real time; uses the cosine similarity algorithm to calculate the matching degree between the equipment repair difficulty coefficient and the personnel skill level; constructs a task assignment priority queue based on the matching degree; combines the personnel workload and response time limit requirements, and uses the Hungarian algorithm for optimal task assignment to output the work order assignment form.

[0089] S5.5: Push the initial repair work order to the mobile terminal of the operation and maintenance personnel through the operation and maintenance management platform, trigger the work order execution process, and record the work order reception time.

[0090] Among them, the mobile terminal of the operation and maintenance personnel displays the work order details, equipment location navigation, and repair operation guidance interface.

[0091] S5.6: Receive the work order execution data sent back by the operation and maintenance personnel through the mobile terminal, and update the status information of the corresponding equipment node in the virtual building model based on the work order execution data.

[0092] Among them, the work order execution data includes fault confirmation information, repair operation records, and replacement parts lists, and the status information includes equipment health status, repair timestamp, and repair measure codes.

[0093] S5.7: Monitor the operation status of the equipment after maintenance, collect the performance data of the first operation cycle after maintenance, compare and analyze it with the performance data before maintenance, and generate a maintenance effect evaluation report.

[0094] Among them, the performance data includes equipment energy consumption indicators, operation parameters, and status indicators.

[0095] S5.8: According to the results of the maintenance effect evaluation report, judge whether it is necessary to start the secondary maintenance process. If secondary maintenance is required, return to S5.1 to regenerate the work order; otherwise, complete the current maintenance process.

[0096] Specifically, the system extracts the performance improvement amplitude and fault improvement indicators in the evaluation report, compares the extracted indicators with the preset thresholds; if the indicators are lower than the thresholds, mark that secondary maintenance is required and return to S5.1 to regenerate the work order; if the indicators meet the standards, update the equipment status to normal and end the current maintenance process.

[0097] Furthermore, this embodiment also provides a building operation and maintenance management system based on big data, including a data acquisition module for collecting multi-source heterogeneous data through an Internet of Things sensor network, performing data standardization processing, and generating a standard data stream; a mapping module for constructing a virtual building model based on digital twin technology, mapping the standard data stream to corresponding device nodes of the virtual building model, and using a spatio-temporal attention mechanism to extract device operation characteristics and generate a device feature matrix; a prediction and analysis module for inputting the device feature matrix into a deep learning network, training to generate an energy consumption prediction model, and outputting an energy consumption prediction sequence; an evaluation and early warning module for constructing a multi-layer perceptron health scoring model based on the energy consumption prediction sequence, and generating a fault early warning priority list in combination with device weight coefficients; a work order management module for automatically generating maintenance work orders according to the fault early warning priority list, allocating work orders based on the device maintenance difficulty coefficient and the skill level of operation and maintenance personnel, pushing them to operation and maintenance personnel, and feeding back the work order execution results to the virtual building model for status update.

[0098] In summary, through the combination of the Internet of Things sensor network and digital twin technology, the present invention realizes the real-time mapping between the physical building and the virtual model, provides an intuitive three-dimensional visualization interface, and greatly improves the efficiency of device management and fault location. The deep learning architecture that combines bidirectional long short-term memory network and multi-head attention mechanism can accurately capture the temporal characteristics of device operation data, significantly improving the accuracy of energy consumption prediction and the prediction time span. The fault early warning mechanism based on device health scoring and weight coefficients, combined with the intelligent work order allocation system, realizes the timely discovery and rapid response to faults, improving the initiative and pertinence of operation and maintenance work. Through the evaluation of maintenance effects and the secondary maintenance mechanism, a complete closed-loop management process is formed, ensuring the maintenance quality and continuously optimizing the maintenance strategy, and improving the overall operation and maintenance management level.

[0099] Example 2, referring to Figures 1 to 3 , is the second embodiment of the present invention. This embodiment provides a building operation and maintenance management method based on big data. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0100] To verify the effectiveness of this building operation and maintenance management method, the research team selected a Grade A office building in Nanshan District, Shenzhen as the experimental object, with a building area of approximately 45,000 square meters and a total of 25 floors. The experimental period was from June 2023 to December 2023, covering two typical climate cycles of summer and winter.

[0101] In the initial stage of the experiment, the research team deployed a comprehensive Internet of Things sensor network, including 350 temperature sensors, 280 humidity sensors, 180 air quality sensors, 25 smart electricity meters, 25 smart water meters, and 160 people flow density sensors. The sensor sampling frequency was set at 5 minutes per time. The edge computing device used Huawei Atlas 200 DK for data preprocessing and local caching. The data standardization process adopted a unified UTC+8 time zone timestamp to ensure data synchronization.

[0102] In the process of constructing the virtual building model, the IFC format building information model exported by Autodesk Revit 2023 was used as the basic data. The three-dimensional reconstruction accuracy reached the centimeter level. The model contained more than 12,000 geometric entities, covering all key equipment nodes. Tetrahedral meshes were used for mesh generation, with the minimum mesh size of 0.1 meter and the total number of meshes of about 2.8 million. The coordinate system adopted the WGS84 standard, and the positioning accuracy was better than 0.5 meter.

[0103] The training dataset of the energy consumption prediction model contained 180 days of historical operation data with a sampling interval of 5 minutes, totaling about 51,840 data samples. The bidirectional long short-term memory network adopted a two-layer structure, with each layer containing 128 hidden units. The multi-head attention mechanism was set with 8 attention heads. The training used the Adam optimizer, with the initial learning rate set at 0.001, the batch size of 64, and the number of training epochs of 200. The average prediction error on the validation set was 3.2%, and the prediction lead time could reach 48 hours.

[0104] As shown in Technical Comparison Table 1, the present invention was comprehensively compared with the existing building management system:

[0105] Table 1 Performance Comparison between the Present Invention and the Existing Building Management System

[0106]

[0107] It can be seen from the comparison data that the method of the present invention is significantly superior to the existing technical solutions in all key performance indicators. Especially in terms of the prediction lead time and the accuracy of fault warning, they are increased by 100% and 15.5% respectively, providing sufficient preparation time for preventive maintenance.

[0108] During the six - month experimental operation period, the system processed a total of 2,860 device anomaly warnings and generated 1,245 maintenance work orders. Through the hierarchical processing of the health scoring model, 85% of the potential faults were timely processed before their impacts expanded. The average repair time of equipment faults was reduced from the original 120 minutes to 45 minutes, and the one - time repair completion rate was increased to 92%. The system also accurately tracked the energy consumption changes of various types of equipment. The monthly energy consumption of the central air - conditioning system decreased by 18.5%, the energy consumption of the lighting system decreased by 12.3%, and the energy consumption of the water supply and drainage system decreased by 8.7%.

[0109] The experimental results show that this method not only significantly improves the building operation and maintenance efficiency but also achieves remarkable energy - saving effects. Through the combination of digital twin technology and deep learning, the system can accurately predict equipment anomalies and ensure the optimal allocation of maintenance resources through an intelligent work order allocation mechanism. Notably, the adaptive learning ability of the system enables the accuracy of the prediction model to continuously improve with the increase in operation time. By the end of the experiment, the average error of energy consumption prediction has been reduced to less than 3.2%.

[0110] The tracking and evaluation of the maintenance effect show that the average degree of performance recovery after equipment maintenance reaches over 95% of the factory standard, far exceeding the 80% level of traditional maintenance methods. At the same time, the implementation of preventive maintenance has extended the average fault - free operation time of equipment by 25%, and it is expected to save the owner more than 30% of the annual maintenance cost. These data fully prove the significant advantages and good economic benefits of this method in practical applications.

[0111] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not restrictive. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A building operation and maintenance management method based on big data, characterized in that: including, collecting multi-source heterogeneous data through an Internet of Things sensor network, performing data standardization processing, and generating a standard data stream; constructing a virtual building model based on digital twin technology, mapping the standard data stream to corresponding device nodes of the virtual building model, and using a spatio-temporal attention mechanism to extract device operation characteristics to generate a device feature matrix; inputting the device feature matrix into a deep learning network, training to generate an energy consumption prediction model, and outputting an energy consumption prediction sequence; constructing a multi-layer perceptron health scoring model based on the energy consumption prediction sequence, and generating a fault warning priority list in combination with device weight coefficients; automatically generating maintenance work orders according to the fault warning priority list, performing work order allocation based on the device maintenance difficulty coefficient and the skill level of operation and maintenance personnel, pushing them to operation and maintenance personnel, and feeding back the work order execution results to the virtual building model for status update; The constructing of the virtual building model based on digital twin technology includes the following steps: obtaining building information model data of the building, constructing a three-dimensional geometric model of the building through a three-dimensional modeling engine, and generating a virtual building skeleton model; calculating and constructing a spatial position relationship matrix based on the virtual building skeleton model; constructing a spatial attention network based on the spatial position relationship matrix to generate spatial association weights between devices; organizing the standard data stream in time series, and extracting dynamic time series characteristics of device operation through a time attention network; fusing the spatial association weights and the dynamic time series characteristics to generate a spatio-temporal fusion feature vector; organizing the spatio-temporal fusion feature vector into a device feature matrix; The training to generate the energy consumption prediction model includes: constructing a bidirectional long short-term memory network as the backbone network of the energy consumption prediction model, inputting a training sample set into the bidirectional long short-term memory network, and extracting a time series feature vector; setting a multi-head attention calculation unit in the output layer of the bidirectional long short-term memory network to generate a fused feature vector; inputting the fused feature vector into a fully connected layer to obtain an energy consumption prediction value, and calculating the mean square error loss value between the energy consumption prediction value and the actual energy consumption value; updating network parameters using an adaptive learning rate optimization algorithm according to the calculated mean square error loss value; inputting a validation sample set into the bidirectional long short-term memory network, calculating a validation loss value, and selecting optimal model parameters using an early stopping mechanism based on the convergence of the validation loss to obtain an energy consumption prediction model; The generating of the fault warning priority list includes the following steps: performing a time series difference operation on the energy consumption prediction sequence using a difference calculation unit to obtain an energy consumption change trend array; invoking a time series pattern recognition module to extract features from the energy consumption change trend array to obtain energy consumption change feature data; inputting the energy consumption change feature data into a pre-trained multi-layer perceptron to output a device health risk score; reading device weight coefficients from a device configuration database, multiplying the device weight coefficients by the device health risk score to obtain a device warning priority score; performing a descending sorting algorithm on the device warning priority scores to generate a device fault warning priority list.

2. The building operation and maintenance management method based on big data according to claim 1, characterized in that: The training to generate the energy consumption prediction model further includes: The device feature matrix is sliced by sliding according to a preset time window to generate a sample set, and the sample set is divided into a training sample set and a validation sample set according to a preset ratio; The real-time device feature matrix is input into the energy consumption prediction model, and an energy consumption prediction sequence is output.

3. The building operation and maintenance management method based on big data according to claim 1, characterized in that: The energy consumption change feature data includes a fluctuation feature value, a trend slope value, and a maximum energy consumption difference.

4. The building operation and maintenance management method based on big data according to claim 1, characterized in that: Automatically generating a maintenance work order according to the fault warning priority list includes: Invoking the maintenance knowledge base, matching the corresponding standard maintenance process based on the device number and the warning level, and generating an initial maintenance work order; Constructing a device maintenance difficulty evaluation model, and calculating a device maintenance difficulty coefficient according to the device type, the fault complexity, and the maintenance man-hours; Establishing an operation and maintenance personnel skill evaluation system, and quantifying the skill level based on the work experience, professional qualifications, and historical maintenance success rate of the operation and maintenance personnel; Querying the operation and maintenance personnel resource pool database, matching the device maintenance difficulty coefficient with the skill level, selecting operation and maintenance personnel meeting the qualification requirements by using a task assignment algorithm, and establishing a work order assignment form; Pushing the initial maintenance work order to the operation and maintenance personnel's mobile terminal through the operation and maintenance management platform, triggering the work order start execution process, and recording the work order reception time; Receiving the work order execution data transmitted back by the operation and maintenance personnel through the mobile terminal, and updating the status information of the corresponding device node in the virtual building model based on the work order execution data; Monitoring the operation status of the device after maintenance, collecting the performance data of the first operation cycle after maintenance, comparing and analyzing it with the performance data before maintenance, and generating a maintenance effect evaluation report; Judging whether it is necessary to start a secondary maintenance process according to the result of the maintenance effect evaluation report.

5. The building operation and maintenance management method based on big data according to claim 1, wherein: The steps for obtaining the energy consumption change trend array include: Constructing a time window matrix, and segmenting the energy consumption prediction sequence according to a preset window length N; Performing a first-order difference operation on the time window matrix, calculating the difference between adjacent predicted energy consumption values within each time window, and generating an initial difference array; Performing median filtering on the initial difference array to generate a smoothed difference array; Multiplying the smoothed difference array by a time step correction coefficient to obtain an energy consumption change rate array per unit time; Performing wavelet transform on the energy consumption change rate array, extracting the main trend component of the energy consumption change, and obtaining the energy consumption change trend array.

6. The building operation and maintenance management method based on big data according to claim 1, characterized in that: The construction process of the virtual building skeleton model includes: Importing building information model data, which includes floor plan, equipment layout plan, and pipeline distribution plan; Invoking a geometric conversion engine to parse the building information model data and output a standard geometric description language file; Parsing the standard geometric description language file, extracting building component features, and establishing a component feature table; Based on the standard geometric description language file, performing a mesh dissection operation to generate a three-dimensional mesh data structure; Loading the three-dimensional mesh data structure and the component feature table, and invoking a hierarchical reconstruction algorithm to assemble spatial components to generate a virtual building skeleton model.

7. A building operation and maintenance management system based on big data, based on the big data-based building operation and maintenance management method according to any one of claims 1 to 6, characterized in that: It also includes, A data acquisition module, which is used to collect multi-source heterogeneous data through an Internet of Things sensor network, perform data standardization processing, and generate a standard data stream; A mapping module, which is used to build a virtual building model based on digital twin technology, map the standard data stream to the corresponding device nodes of the virtual building model, extract device operation features by using a spatio-temporal attention mechanism, and generate a device feature matrix; A prediction and analysis module, which is used to input the device feature matrix into a deep learning network, train and generate an energy consumption prediction model, and output an energy consumption prediction sequence; An evaluation and early warning module, which is used to build a multi-layer perceptron health scoring model based on the energy consumption prediction sequence, and generate a list of fault warning priorities in combination with device weight coefficients; A work order management module, which is used to automatically generate maintenance work orders according to the list of fault warning priorities, allocate work orders based on the device maintenance difficulty coefficient and the skill level of operation and maintenance personnel, push them to the operation and maintenance personnel, and feedback the work order execution results to the virtual building model for status update.

Images