A differential pressure power generation engineering construction energy consumption monitoring method and system based on an internet of things

Abstract

The application discloses a differential pressure power generation engineering construction energy consumption monitoring method and system based on Internet of Things, and belongs to the technical field of Internet of Things and energy monitoring. The method comprises the following steps: acquiring energy consumption time sequence data of the whole cycle of differential pressure power generation engineering construction, working state time sequence data of construction machinery and construction progress scene data; performing multidimensional interference mode mining on the energy consumption time sequence data and the working state time sequence data, generating an interference mode abstract and constructing a construction energy consumption node network; performing traversal decision on the interference mode through a fuzzy interval decision tree, determining a construction latent energy consumption node and generating a dynamic evolution sequence; performing compound discrimination on candidate latent energy consumption nodes according to the dynamic evolution sequence and an incremental threshold value, so as to obtain an energy consumption abnormal node; performing vector quantization on data corresponding to the energy consumption abnormal node and generating a multidimensional energy consumption abnormal feature vector; and performing spatial positioning and link matching on the construction energy consumption abnormality according to the feature vector.

Description

Technical Field

[0001] This invention relates to the fields of Internet of Things (IoT) and energy monitoring technology, specifically to an IoT-based method and system for monitoring energy consumption during the construction of differential pressure power generation projects. Background Technology

[0002] Differential pressure power generation projects involve complex construction sites and intensive collaborative operations of large machinery, resulting in significant fluctuations and coupling in energy consumption. Existing construction energy consumption monitoring methods often employ single-threshold alarms or simple statistical analysis, identifying abnormal energy consumption behavior by setting fixed upper limits. An alarm is triggered when the real-time energy consumption reading of a particular piece of equipment or process exceeds a preset threshold, alerting management personnel for intervention. This approach has significant drawbacks. Fixed thresholds cannot adapt to the dynamic changes in construction procedures. Energy demand varies significantly across different construction stages, and a uniform judgment standard can misclassify normal peak values ​​during high-load operations as abnormal, while overlooking inefficient energy consumption during transition periods. Furthermore, energy consumption data is analyzed in isolation from machinery operating conditions, focusing only on the rise and fall of energy consumption readings without incorporating operating parameters such as equipment vibration, speed, and angle into the same analytical framework. This fails to reveal the correlation mechanism between abnormal machinery conditions and abnormal energy consumption. Locating the source of anomalies relies on manual experience, resulting in a long path from alarm signals to confirmation of specific problematic equipment and processes, leading to a delayed response. For latent energy consumption nodes that have not yet reached obvious anomalies but are in a continuous deterioration trend, existing methods lack effective identification and tracking capabilities. The problem this invention aims to solve is how to establish a dynamic interference analysis mechanism between the working status of construction machinery and energy consumption data to detect latent energy consumption anomalies, and how to accurately identify and spatially match energy consumption anomaly nodes in the process of evolution. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for monitoring energy consumption during construction of differential pressure power generation projects based on the Internet of Things. By mining multi-dimensional interference modes, a dynamic correlation between energy consumption and mechanical status is constructed. Fuzzy interval decision trees and incremental threshold composite discrimination are used to achieve accurate identification and spatial positioning of abnormal nodes in construction potential energy consumption.

[0004] To achieve the above objectives, the present invention provides the following technical solution: This application provides an IoT-based method for monitoring energy consumption during the construction of differential pressure power generation projects, comprising: acquiring time-series energy consumption data throughout the entire construction cycle of the differential pressure power generation project; acquiring time-series data of the working status of construction machinery and construction progress scenario data during the construction process; performing multi-dimensional interference pattern mining on the energy consumption time-series data and the working status time-series data to generate an interference pattern summary; constructing a construction energy consumption node network based on the interference pattern summary; performing traversal decision-making on the interference patterns using a fuzzy interval decision tree to determine construction potential energy consumption nodes and generate a dynamic evolution sequence of the construction energy consumption node network; performing composite discrimination on candidate potential energy consumption nodes based on the dynamic evolution sequence and incremental thresholds to obtain energy consumption anomaly nodes; performing vector quantization on the energy consumption time-series data corresponding to the energy consumption anomaly nodes to generate a multi-dimensional energy consumption anomaly feature vector; and performing spatial localization and stage matching of construction energy consumption anomalies based on the multi-dimensional energy consumption anomaly feature vector. This method constructs a network of construction energy consumption nodes and performs dynamic evolution analysis to uncover the deep interference relationship between energy consumption and mechanical status throughout the construction cycle. It also utilizes fuzzy interval decision trees and incremental threshold composite discrimination to effectively improve the accuracy and timeliness of identifying abnormal energy consumption nodes.

[0005] Preferably, the energy consumption time-series data for the entire construction cycle of a differential pressure power generation project is obtained by: deploying three-phase power sensors and flow sensors at multiple key equipment locations during the construction of the differential pressure power generation project; collecting voltage, current, power factor, and medium flow data during the construction process using the three-phase power sensors and flow sensors at a fixed sampling frequency; and synchronizing and aligning the data according to timestamps to generate a multi-dimensional synchronous data stream; then, performing sliding window segmentation on the multi-dimensional synchronous data stream to obtain multiple energy consumption time-series data segments. This acquisition method ensures the time base uniformity of multi-source energy consumption parameters and the orderliness of data segmentation, providing a high-quality data foundation for subsequent multi-dimensional interferometric mode mining.

[0006] As one technical solution of this invention, acquiring the time-series data of the working status of construction machinery and the scene data of construction progress during the construction process involves: deploying vibration sensors, speed sensors, and angle sensors at the hydraulic system, power output shaft, and slewing mechanism of the construction machinery, respectively, to collect the time-series data of the machinery's operating parameters; simultaneously, using multiple fixed cameras deployed at the construction site to acquire video stream data of the construction area, extracting the start and end time markers of each construction stage from the video stream data, and generating construction progress scene data. This achieves a spatiotemporal correlation between the operating status of the construction machinery and the construction progress, enabling energy consumption analysis to accurately correspond to specific construction stages.

[0007] Preferably, multi-dimensional interference pattern mining is performed on energy consumption time-series data and operating state time-series data to generate an interference pattern summary. This includes: defining an energy consumption feature space and an operating state feature space, and symbolically representing the energy consumption time-series data and operating state time-series data respectively; calculating the correlation coefficient matrix between each symbol sequence in the energy consumption feature space and the operating state feature space; detecting local extrema of the correlation coefficient matrix through a sliding time window to extract the symbol sequence pairs exhibiting interference; and compressing and encoding all the symbol sequence pairs exhibiting interference to generate an interference pattern summary. This method can effectively capture the subtle interference relationship between energy consumption changes and mechanical operating states, significantly compress redundant information, and highlight key interference features.

[0008] Furthermore, a construction energy consumption node network is constructed based on the interference pattern summary. Specifically, each construction stage in the interference pattern summary is treated as an energy consumption node; the energy consumption feature vector of each construction stage in the interference pattern summary is extracted and used as the initial state value of the corresponding energy consumption node; the edge weights between energy consumption nodes are calculated based on the interference strength between different construction stages in the interference pattern summary; and a directed weighted network is constructed using energy consumption nodes as vertices and edge weights as edge attributes to serve as the construction energy consumption node network. This intuitively expresses the energy consumption transmission and influence relationships between construction stages, making the energy consumption structure of the entire construction process clear and readable.

[0009] In a preferred embodiment of the present invention, a fuzzy interval decision tree is used to traverse and decide on interference patterns, determine construction potential energy consumption nodes, and generate a dynamic evolution sequence of the construction energy consumption node network. Specifically, this involves: establishing a fuzzy interval mapping table to divide the numerical range of the interference patterns into multiple fuzzy sub-intervals; using the fuzzy sub-intervals as branching conditions for the decision tree to construct a tree structure; inputting the interference pattern summary into the fuzzy interval decision tree, traversing layer by layer according to the fuzzy sub-intervals, and outputting the potential probability value of each construction stage; marking construction stages with potential energy consumption nodes whose potential probability values ​​exceed a preset probability threshold as construction potential energy consumption nodes; and recording the occurrence sequence of construction potential energy consumption nodes in chronological order to generate a dynamic evolution sequence. By using fuzzy interval division to handle the uncertainty of energy consumption data, the gradual change trend of construction energy consumption can be captured more robustly, forming a development trajectory with temporal depth.

[0010] Based on this, a composite discrimination mechanism is performed on candidate latent energy consumption nodes according to the dynamic evolution sequence and incremental thresholds to identify energy consumption anomaly nodes. Specifically: a first-level threshold and a second-level threshold are set, where the second-level threshold is greater than the first-level threshold; the cumulative latent frequency of each candidate latent energy consumption node in the dynamic evolution sequence within the current time window is extracted; the cumulative latent frequency is compared with the first-level threshold, and if the cumulative latent frequency exceeds the first-level threshold, the candidate latent energy consumption node is marked as a node to be reviewed; the latent probability value of the node to be reviewed is jointly compared with the latent probability values ​​of adjacent nodes, and if the latent probability value of the node to be reviewed is greater than that of adjacent nodes and the continuous residence time of the node to be reviewed in the dynamic evolution sequence exceeds a preset duration, the node to be reviewed is determined to be an energy consumption anomaly node. This composite discrimination mechanism integrates frequency statistics, probability comparison, and residence time verification, which can effectively filter out occasional fluctuations during construction and significantly reduce the false alarm rate of energy consumption anomalies.

[0011] Preferably, the energy consumption time-series data corresponding to the energy consumption anomaly nodes is vector-quantized to generate a multi-dimensional energy consumption anomaly feature vector. This includes: dividing the energy consumption time-series data corresponding to the energy consumption anomaly nodes into equal-length segments to obtain multiple data segments; performing vector quantization on each data segment to generate a codebook index for that data segment; concatenating all codebook indices in chronological order to generate an initial feature index sequence; and assigning the energy amplitude of its corresponding data segment as a weighting coefficient to each index in the initial feature index sequence to obtain a multi-dimensional energy consumption anomaly feature vector. Through vector quantization and energy weighting, the time-series pattern of abnormal energy consumption is condensed into a compact and recognizable feature vector, facilitating rapid matching and localization.

[0012] Furthermore, based on the multidimensional energy consumption anomaly feature vector, spatial localization and stage matching are performed for construction energy consumption anomalies. Specifically, a three-dimensional spatial coordinate mapping table is constructed for each construction stage, recording the coordinate interval of each stage. The Euclidean distance between the multidimensional energy consumption anomaly feature vector and the feature template of each construction stage in the three-dimensional spatial coordinate mapping table is calculated. The coordinate interval corresponding to the feature template of the construction stage with the smallest Euclidean distance is used as the spatial localization interval of the energy consumption anomaly node. The names of the construction stages within the localization interval are associated with the energy consumption anomaly node, and the stage matching result is output. This localization method does not rely on the location of a single sensor, but rather achieves spatial source tracing of abnormal energy consumption based on pattern matching, thereby accurately locating the construction stage and specific area causing the energy consumption anomaly.

[0013] This application also provides an IoT-based differential pressure power generation engineering construction energy consumption monitoring system, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the aforementioned IoT-based differential pressure power generation engineering construction energy consumption monitoring method. This system integrates multi-source sensor data acquisition and intelligent analysis functions, and can automatically complete the entire process from interferometric mode mining to spatial location of abnormal nodes, providing reliable support for refined monitoring of energy consumption during differential pressure power generation engineering construction.

[0014] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This approach integrates energy consumption time-series data with construction machinery operating status time-series data into a multi-dimensional interferometric pattern mining framework. A correlation coefficient matrix of symbol sequences is calculated between the energy consumption feature space and the operating status feature space. Interfering symbol sequence pairs are extracted through local extremum detection and compressed into an interferometric pattern summary. This scheme establishes a dynamic correlation between previously independently analyzed energy consumption data and operating status data such as machinery vibration, speed, and angle. It can capture the interference effect on energy consumption from minute changes in machinery operating parameters, thus exposing abnormal precursors hidden in normal energy consumption fluctuations in advance. A fuzzy interval mapping table is used to divide the numerical range of the interferometric pattern into multiple fuzzy sub-intervals and construct a decision tree. The interferometric pattern summary is traversed layer by layer to output the latent probability value of each construction stage, marking latent energy consumption nodes and generating a dynamic evolution sequence in chronological order. Based on a set first-level threshold, the cumulative latent frequency of candidate nodes is filtered. Then, combined with the joint comparison of probability values ​​of adjacent nodes and the determination of continuous dwell time, a composite discrimination is implemented to identify energy consumption anomaly nodes. By employing fuzzy interval decision-making, misclassification of energy consumption patterns under critical conditions by hard boundaries is avoided, and the continuous evolution process from gradual degradation to sudden anomalies is fully presented. Incremental composite discrimination, based on frequency statistics, incorporates correlation node comparison and duration constraints, filtering out false alarms caused by instantaneous fluctuations in operating conditions. It only makes final judgments on latent change nodes whose probability values ​​are consistently higher than adjacent links and remain stable over time, achieving accurate identification of abnormal construction energy consumption nodes and reducing false alarms. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 This is a flowchart of an IoT-based method for monitoring energy consumption during the construction of differential pressure power generation projects. Figure 2This is a flowchart of multidimensional interferometric pattern mining and interferometric pattern summary generation; Figure 3 It is a multi-dimensional time-series data curve of energy consumption of key equipment in differential pressure power generation engineering construction; Figure 4 It is a correlation coefficient between energy consumption characteristics and working status symbol sequence, and an interference point detection map. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] See Figure 1 This invention provides an IoT-based method for monitoring energy consumption during the construction of differential pressure power generation projects, comprising the following steps: acquiring time-series energy consumption data throughout the entire construction cycle of the differential pressure power generation project; acquiring time-series data of the working status of construction machinery and construction progress scenarios during the construction process; performing multi-dimensional interference pattern mining on the energy consumption time-series data and working status time-series data to generate an interference pattern summary; constructing a construction energy consumption node network based on the interference pattern summary; performing traversal decision-making on the interference patterns using a fuzzy interval decision tree to determine potential energy consumption nodes and generate a dynamic evolution sequence of the construction energy consumption node network; performing composite discrimination on candidate potential energy consumption nodes based on the dynamic evolution sequence and incremental thresholds to obtain energy consumption anomaly nodes; performing vector quantization on the energy consumption time-series data corresponding to the energy consumption anomaly nodes to generate a multi-dimensional energy consumption anomaly feature vector; and performing spatial localization and stage matching of construction energy consumption anomalies based on the multi-dimensional energy consumption anomaly feature vector.

[0019] Example 1 In the specific implementation, the energy consumption time series data of the differential pressure power generation project throughout the entire construction cycle is obtained, and the specific process is as follows.

[0020] Three-phase power sensors and flow sensors were deployed at several key equipment locations during the construction of the differential pressure power generation project. Key equipment included differential pressure generator sets, high-pressure water pumps, air compressors, lifting equipment, and welding equipment. The three-phase power sensors were installed on the busbars between the output terminals of the distribution cabinet and the input terminals of each key equipment. Flow sensors were installed at key nodes in the cooling water circulation pipeline, steam transmission pipeline, compressed air pipeline, and fuel supply pipeline. Voltage, current, power factor, and medium flow data were collected during the construction process using the three-phase power sensors and flow sensors at a fixed sampling frequency of 50 Hz. The voltage and current transformers inside the three-phase power sensors acquired the instantaneous values ​​of the three-phase voltage and current in real time, and the embedded processing unit calculated the power factor of each phase. The flow sensors acquired the instantaneous flow rate of the medium in the pipeline through differential pressure measuring elements. The voltage, current, power factor, and medium flow data were synchronized and aligned according to timestamps to generate a multi-dimensional synchronous data stream. The time synchronization alignment process uses the Network Time Protocol (NTP) to calibrate the clocks of the acquisition terminals of each three-phase power sensor and each flow sensor, ensuring that the time deviation of all acquisition terminals is less than 1 millisecond. For data points with inconsistent timestamps from different acquisition terminals, linear interpolation is used for resampling on a unified time axis to form a time-aligned multidimensional synchronous data stream. The multidimensional synchronous data stream is an M-row, N-column matrix, where M corresponds to the total number of sampling times and N corresponds to the combination of four dimensions: voltage, current, power factor, and medium flow rate.

[0021] A sliding window segmentation process is applied to the multidimensional synchronous data stream to obtain multiple energy consumption time-series data segments. The window length of the sliding window is set to... There are 1 sampling point, with a step size of 1. There are sampling points, and the overlap ratio factor is . ,but .in, This indicates the number of consecutive sampling points contained in each energy consumption time-series data segment. The value is 2000, corresponding to a time span of 40 seconds; This indicates the percentage of overlapping sampling points between adjacent sliding windows. The proportion; The value range is from 0 to 0.5. In this embodiment... Set to 0.2; This represents the difference between the starting sampling point indices of adjacent sliding windows. The setting of 0.2 is based on the relationship between the typical duration of a construction phase and a fixed sampling frequency: the shortest duration of a typical construction phase is approximately 10 seconds, corresponding to 500 sampling points. To ensure that energy consumption events with any duration exceeding 500 sampling points are covered by at least two consecutive sliding windows, the following conditions must be met: , combined ,Depend on Calculation The condition is not met, therefore choose Make This ensures that events are captured repeatedly by multiple windows without generating excessive computational redundancy. (Based on the settings...) and The multidimensional synchronous data stream is slidably intercepted from the start time, resulting in K energy consumption time-series data segments. Each energy consumption time-series data segment is... The matrix is ​​denoted by D, where D is the number of dimensions of the multidimensional synchronous data stream.

[0022] In practice, the following process is used to obtain the working status time sequence data of construction machinery and the construction progress scenario data during the construction process.

[0023] Vibration sensors, speed sensors, and angle sensors are deployed in the hydraulic system, power output shaft, and slewing mechanism of the construction machinery. The vibration sensors are triaxial microelectromechanical system accelerometers, mounted on the outer shell of the hydraulic pump and the outer wall of the hydraulic cylinder. The speed sensors are Hall effect sensors, installed on the bearing end cover of the power output shaft, with the sensing gear coaxially mounted with the power output shaft. The angle sensors are absolute photoelectric encoders, installed at the meshing point of the external gear ring of the slewing bearing in the slewing mechanism. The vibration, speed, and angle sensors collect time-series data of the construction machinery's operating parameters. The vibration sensors output vibration acceleration values ​​in three axes; the speed sensors output pulse signals per revolution of the power output shaft, which are converted into speed values ​​by a signal conditioning circuit; and the angle sensors output the absolute angular position value of the slewing mechanism. All these sensors are synchronously collected at a fixed sampling frequency of 50 Hz, the same as the energy consumption data, and timestamps are added to form time-series data of operating parameters including vibration intensity, speed value, and angular position.

[0024] Video stream data of the construction area is acquired through multiple fixed cameras deployed at the construction site. The fixed cameras are network high-definition dome cameras, installed at the top of monitoring columns at the four corners of the construction area and at the top of the tower crane. The combined field of view of all fixed cameras covers the entire construction work area. The video stream is transmitted in real-time to the edge computing server at a frame rate of 25 frames per second and a resolution of 1920×1080.

[0025] The start and end time markers of each construction stage are extracted from the video stream data to generate construction progress scene data. An object detection algorithm is used to identify construction objects and personnel movements within each frame of the video stream data. An action recognition algorithm is used to classify construction stages in continuous frame sequences, and the actual start and end frames of each construction stage are determined through time series labeling, converting the data into time interval data with stage type labels. The object detection algorithm is based on the YOLOv5 model. The core architecture of the YOLOv5 model consists of three parts: the backbone network CSPDarknet, the neck network PANet, and the detection head. The backbone network CSPDarknet uses a cross-stage local network structure for feature extraction, the neck network PANet aggregates multi-scale features through top-down and bottom-up paths, and the detection head outputs the target bounding box coordinates and class confidence. The training process of the YOLOv5 model is as follows: 15,000 images of targets including excavators, loaders, welding equipment, steel bar bundles, construction workers, and formwork were collected at the differential pressure power generation construction site. Bounding boxes and class labels were created for each target in each image, with 8 classes. Data augmentation was performed using mosaic data augmentation, randomized affine transformation, and HSV color dithering. The loss function used during training was a weighted combination of CIoU loss and binary cross-entropy loss. The optimizer used stochastic gradient descent with momentum. The initial learning rate was 0.01, the momentum factor was 0.937, and the weight decay coefficient was 0.0005. Training was conducted on 4 GPUs with a total batch size of 64 for 300 rounds. The weights with the highest average accuracy on the validation set were selected as the final model parameters. The trained YOLOv5 model takes a 3×640×640 frame as input tensor for each video image and outputs a tensor containing the predicted results for a preset number of anchor boxes. After non-maximum suppression, the detected target bounding boxes and classes are obtained.

[0026] The action recognition algorithm is based on the SlowFast network. The core architecture of the SlowFast network consists of a slow path and a fast path. The slow path operates at low temporal resolution with a large network depth to capture spatial semantic information; the fast path operates at high temporal resolution with a shallow network depth to capture dynamic temporal changes. At multiple stages, the feature maps of the fast path are transformed and concatenated into the slow path through lateral connections. The training process of the SlowFast network is as follows: a dataset of construction stages containing multiple video clips is constructed. Each video clip has a uniform duration of 3 seconds, corresponding to 75 frames of images, and is labeled with one of five construction stage categories: foundation excavation, rebar tying, formwork support, equipment hoisting, and pipe welding. A total of 12,000 clips are collected. The input of the SlowFast network is a spatiotemporal block composed of T consecutive stacked frames centered on the current time, with T set to 32. The input step size of the slow path is 8 frames, i.e., 4 frames are actually processed; the step size of the fast path is 2 frames, i.e., 16 frames are actually processed. The output of the network is the probability vector of the five construction stages. The training loss function used was cross-entropy loss, the optimizer was Adam, the initial learning rate was 0.001, and the weight decay was 0.0001. Training was performed for 200 epochs on 4 GPUs with a total batch size of 16. When the trained SlowFast network was used for video stream processing, it slid along the video stream in a fixed time window. For each spatiotemporal block processed, it output a probability for a construction stage category, and the construction stage with the highest probability was taken as the category label at the center of that window.

[0027] The specific process of time series labeling is as follows: the output category label sequence of the SlowFast network on the continuous window is smoothed by median filtering to filter out isolated labels with a duration of less than 0.5 seconds. Then, continuous segments with the same category label are merged, and the start frame index and end frame index of each continuous segment are extracted. Based on the video frame rate, they are converted into start timestamps and end timestamps to form time interval data corresponding to construction stages such as foundation excavation, rebar binding, formwork support, equipment hoisting, and pipeline welding, i.e., construction progress scene data.

[0028] See Figure 3 The figure shows the changes in multidimensional energy consumption time series data of key equipment during the construction of the differential pressure power generation project. The horizontal axis represents time in seconds, the left side of the vertical axis represents active power and reactive power in kilowatts (kW) and kilovars (kvar) respectively, and the right side of the vertical axis represents power factor and medium flow rate in dimensionless factor and cubic meters per hour (m³ / h) respectively. The figure contains four curves: active power (blue solid line), reactive power (red dashed line), power factor (green dotted line), and medium flow rate (purple dotted line).

[0029] Observing the trend of the curve, during the construction time from 0 seconds to about 4 seconds, the active power fluctuated between approximately 110 and 130 kW, the reactive power fluctuated slightly between 35 and 45 kvar, the power factor remained relatively stable at around 0.9, and the medium flow rate remained between approximately 14.5 and 15 m³ / h, indicating that the equipment was operating relatively smoothly during this stage. At approximately 4 seconds, the active power experienced a significant jump, rapidly increasing from approximately 110 kW to over 140 kW, while the reactive power simultaneously rose to around 50 kvar, and the medium flow rate also increased accordingly to approximately 15.5 m³ / h. The power factor fluctuated slightly but remained stable, indicating a sudden increase in the load on the construction machinery and a corresponding increase in the medium supply, possibly corresponding to the start of a certain construction phase. Subsequently, the active power remained at a high level (approximately 120 to 145 kW) between 4 and 7 seconds, the reactive power fluctuated synchronously between 40 and 55 kvar, the medium flow rate fluctuated slightly but remained generally stable, and the power factor showed no significant abnormalities, indicating that this construction phase continued to operate.

[0030] At approximately 7 seconds, the active power began to decrease, gradually dropping from about 130kW to around 100kW, while the reactive power followed suit, decreasing to approximately 30-40kvar. The medium flow rate also slightly decreased to approximately 14m³ / h. The power factor remained stable, indicating a reduction in the load on the construction machinery, with the medium flow rate adjusting accordingly. This likely corresponds to a transition or pause in the construction process. After approximately 8 seconds, both active and reactive power began to slowly recover, and the medium flow rate rose to around 15m³ / h. The power factor showed no significant fluctuations, indicating that the equipment had resumed operation and was ready to proceed to the next construction phase.

[0031] Example 2: In practice, multi-dimensional interference pattern mining is performed on energy consumption time-series data and operating status time-series data to generate interference pattern summaries. The specific process is as follows. (See reference...) Figure 2 This paper defines an energy consumption feature space and an operating state feature space, and symbolically represents the energy consumption time-series data and operating state time-series data respectively. The energy consumption feature space includes four dimensions: mean active power, standard deviation of reactive power, power factor skewness, and slope of the medium flow trend. The mean active power is defined as the arithmetic mean of the instantaneous active power at each sampling point within a single energy consumption time-series data segment. The instantaneous active power is calculated from the instantaneous voltage, instantaneous current, and instantaneous power factor values. The standard deviation of reactive power is defined as the standard deviation of the instantaneous reactive power at each sampling point within the same energy consumption time-series data segment. The instantaneous reactive power is calculated from the instantaneous voltage, instantaneous current, and instantaneous power factor values. The power factor skewness is defined as the third-order normalized moment of the power factor sampling sequence within the same energy consumption time-series data segment. The slope of the medium flow trend is defined as the slope of the line obtained by least-squares linear fitting of the medium flow sampling sequence within the same energy consumption time-series data segment.

[0032] The operating state feature space comprises four dimensions: vibration intensity, rotational speed fluctuation rate, hydraulic pressure change rate, and rotational angle oscillation amplitude. Vibration intensity is defined as the root mean square value of the composite triaxial vibration acceleration within a single operating state time-series data segment. Rotational speed fluctuation rate is defined as the ratio of the standard deviation of the rotational speed sampling sequence to the average rotational speed within the same operating state time-series data segment. Hydraulic pressure change rate is defined as the difference between the maximum and minimum values ​​of the hydraulic pressure sampling sequence within the same operating state time-series data segment, divided by the sampling time span. Rotational angle oscillation amplitude is defined as the difference between the maximum and minimum values ​​of the angular position sampling sequence within the same operating state time-series data segment.

[0033] The method for symbolic representation of energy consumption time series data and operating status time series data is as follows: The time series of each dimension in the energy consumption feature space and the time series of each dimension in the operating status feature space are discretized with equal width, mapping continuous values ​​to symbols in a finite set of letters. During the equal-width discretization process, the global minimum and global maximum values ​​of each dimension's time series across all energy consumption time series data segments are first obtained. The interval from the global minimum to the global maximum value is then divided into S equal-width sub-intervals, where S is set to 8, corresponding to the 8 letters in the letter set {a, b, c, d, e, f, g, h}. The value of S as 8 is based on the commonly used alphabet size setting in the SAX representation of time series data. Eight letters achieve a balance between compression ratio and information preservation, and are consistent with the eight-band division method commonly used in energy spectrum analysis, facilitating interferometric pattern recognition. For the first energy consumption time series data segment... For each segment, the sequence values ​​of the active power mean dimension in the energy consumption feature space are mapped to corresponding letters according to their sub-interval affiliation, resulting in a symbolic sequence for the active power mean dimension. Symbolic sequences are then generated for the reactive power standard deviation dimension, power factor skewness dimension, and medium flow rate trend slope dimension in the same manner. For the first segment in the working state time series data... Each segment is used to discretize the sequence values ​​in the vibration intensity dimension, rotational speed fluctuation rate dimension, hydraulic pressure change rate dimension, and rotation angle swing amplitude dimension of the working state feature space into corresponding letters, thus obtaining the symbol sequence of each dimension.

[0034] Calculate the correlation coefficient matrix between each symbol sequence in the energy consumption feature space and the operating state feature space. Let the nth symbol sequence in the energy consumption feature space be... The symbol vector formed by the symbol sequence of each dimension over all K energy consumption time series data segments is: The first in the working state feature space The symbol vector formed by the symbol sequence of each dimension over all K working state time-series data segments is: ,in The values ​​are 1, 2, 3, and 4, which correspond to the active power mean dimension, reactive power standard deviation dimension, power factor skewness dimension, and medium flow rate trend slope dimension, respectively. The values ​​1, 2, 3, and 4 correspond to the vibration intensity dimension, rotational speed fluctuation dimension, hydraulic pressure change rate dimension, and rotation angle swing amplitude dimension, respectively. After converting the symbols {a→1, b→2, ..., h→8} into numerical values, the calculation is performed. and The Pearson correlation coefficient between them is calculated using the following formula: in, This represents the total number of energy consumption time-series data segments. Represents the first in the energy consumption characteristic space The dimensional symbol sequence in the th dimension is at the th dimensional symbol sequence in the th dimensional symbol sequence The value converted from a time-series energy consumption data segment to a numerical value. express In all The average value over a time-series energy consumption data segment Represents the first in the feature space of working states The dimensional symbol sequence in the th dimension is at the th dimensional symbol sequence in the th dimensional symbol sequence The values ​​converted from time-series data segments of the working state to numerical values. express In all The average value over a time series data segment of each working state. Represents the energy consumption characteristic space. The first dimension symbol sequence and the working state feature space Pearson correlation coefficients between symbol sequences of each dimension. As described above... and After calculating all combinations, we get The correlation coefficient matrix, the matrix in which the first... Line 1 The elements of the column are .

[0035] Local extrema are detected in the correlation coefficient matrix using a sliding time window to extract symbol sequence pairs exhibiting interference. Along the time dimension, K energy consumption time-series data segments and K operating status time-series data segments are arranged chronologically, with each segment corresponding to a time index. , The value can range from 1 to K. Construct a time-varying index. The sequence of correlation coefficients of change: for each pair of energy consumption feature space dimensions and working state feature space dimension The combination of dimensions, indexed by time Centered on a time window, H segments are selected before and after the time frame to form a window. The Pearson correlation coefficient is calculated within this window, with H set to 5. The reason for choosing H as 5 is to ensure the window contains a total of 11 segments, corresponding to a time span of approximately 440 seconds. This time span roughly covers the switching cycle between two adjacent construction phases and the duration of the stable transition process of construction machinery, thus filtering out transient noise fluctuations while maintaining local detection sensitivity. The sliding time window moves along the time index in steps of 1. Move, index at each time A correlation coefficient value is obtained. This forms a correlation coefficient sequence.

[0036] Local extrema detection is performed on each correlation coefficient sequence: if a certain time index absolute value of the correlation coefficient at the location satisfy and ,and Exceeding the interference discrimination threshold Then the time index The corresponding energy consumption feature space dimension Symbol sequence and working state feature space dimension The symbol sequence is recorded as the pair of symbols that exhibit interference. Interference discrimination threshold. The threshold is set to 0.7, based on the convention of setting a strong correlation threshold for the Pearson correlation coefficient. When the absolute value of the correlation coefficient is greater than 0.7, a strong linear correlation is considered to exist, representing a significant interference between energy consumption and operating status. Simultaneously, the temporal synchronicity of the two symbol sequences is also detected when determining interference: the time-delay cross-correlation coefficient of the two symbol sequences within the window is calculated. If the time delay corresponding to the maximum cross-correlation coefficient is less than or equal to two time index steps, the symbol sequence pair exhibiting interference is considered to have a synchronous or fixed time-delay relationship and is recorded; otherwise, it is not recorded.

[0037] All interfering symbol sequence pairs are compressed and encoded to generate an interference pattern summary. All recorded interfering symbol sequence pairs are sorted in ascending order by time index to obtain the original sequence of interfering symbol sequence pairs. Run-length encoding is performed on interfering symbol sequence pairs with the same dimension combination and consecutive time indices in the original sequence: T consecutive interfering symbol sequence pairs with the same dimension combination are merged into one encoding entry. The encoding entry format is "[Energy Consumption Feature Space Dimension Code, Working State Feature Space Dimension Code, Interfering Pattern Identifier] – Number of Continuous Segments," where the combination of the energy consumption feature space dimension code and the working state feature space dimension code constitutes the interference pattern identifier. For example, the combination of energy consumption feature space dimension 1 and working state feature space dimension 3 constitutes an interference pattern identifier, indicating an interference between the average active power and the rate of change of hydraulic pressure. The number of continuous segments records the number of consecutive segments. After run-length encoding compression, all encoding entries are summarized in chronological order to form the interference pattern summary. Each entry in the interference pattern summary corresponds to a specific interference pattern and the duration of that interference pattern over time.

[0038] Example 3: In practical implementation, a construction energy consumption node network is constructed based on the interference pattern summary, as follows: Each construction stage in the interference pattern summary is treated as an energy consumption node. Each coded entry in the interference pattern summary is associated with a construction stage type, which includes foundation excavation, rebar tying, formwork support, equipment hoisting, and pipe welding. All occurrences of construction stage types are extracted from all coded entries in the interference pattern summary. The extracted construction stage types are then deduplicated, and each deduplicated construction stage type is identified as a separate energy consumption node. During the deduplication process, multiple instances of the same construction stage type occurring at different time periods are merged into the same energy consumption node, and each energy consumption node is identified by the corresponding construction stage type name.

[0039] Energy consumption feature vectors are extracted from each construction stage in the interferometric mode summary, and these feature vectors are used as the initial state values ​​of the corresponding energy consumption nodes. For a given energy consumption node, all coded entries containing the construction stage type corresponding to that energy consumption node are retrieved from the interferometric mode summary, and the energy consumption time-series data segments associated with these coded entries are obtained. Each energy consumption time-series data segment has four dimensions in the energy consumption feature space: mean active power, standard deviation of reactive power, power factor skewness, and slope of medium flow trend. For all energy consumption time-series data segments associated with that energy consumption node, the arithmetic mean of all values ​​in the mean active power dimension, the arithmetic mean of all values ​​in the standard deviation of reactive power dimension, the arithmetic mean of all values ​​in the power factor skewness dimension, and the arithmetic mean of all values ​​in the slope of medium flow trend dimension are calculated. The calculated arithmetic means of the four dimensions are arranged in order to form a four-dimensional vector, which is used as the initial state value of that energy consumption node. Perform the above operation on all energy consumption nodes, and assign a four-dimensional energy consumption feature vector as the initial state value to each energy consumption node.

[0040] The edge weights between energy consumption nodes are calculated based on the interference strength between different construction stages in the interference pattern summary. Interference strength is defined as the product of the temporal overlap of the interference patterns corresponding to two construction stages and the absolute value of their correlation coefficients. For any two different energy consumption nodes, let the construction stage type corresponding to energy consumption node u be U, and the construction stage type corresponding to energy consumption node v be V. All time segments in which construction stage types U and V co-occur in interference coding entries are obtained from the interference pattern summary. The set of interference time intervals for construction stage type U consists of the start and end times of construction stage type U appearing in each interference coding entry, and the set of interference time intervals for construction stage type V consists of the start and end times of construction stage type V appearing in each interference coding entry. The temporal overlap between the set of interference time intervals for construction stage type U and the set of interference time intervals for construction stage type V is calculated. The temporal overlap is defined as the ratio of the total duration of the intersection of the two time interval sets to the total duration of the union. The total duration of the intersection is the sum of the time lengths of the overlapping portions of the time intervals in the two sets, and the total duration of the union is the minimum continuous time span covering all time intervals in both sets. The absolute value of the average interference correlation coefficient between construction phase type U and construction phase type V is calculated as follows: extract all interference coding entries that simultaneously contain construction phase type U and construction phase type V from the interference pattern summary, obtain the absolute value of the correlation coefficient corresponding to each coding entry in the correlation coefficient matrix, and calculate the arithmetic mean of these absolute values ​​of correlation coefficients.

[0041] The edge weight between energy-consuming node u and energy-consuming node v Calculate using the following formula: in, The time overlap represents the set of interference time intervals between construction phase type U corresponding to energy consumption node u and construction phase type V corresponding to energy consumption node v. The time overlap is a dimensionless ratio with a value ranging from 0 to 1. The value represents the arithmetic mean of the absolute values ​​of the correlation coefficients of all common interference coding entries between construction stage type U corresponding to energy consumption node u and construction stage type V corresponding to energy consumption node v, with a value range from 0 to 1. This represents the weight of the directed edge from energy consumption node u to energy consumption node v. The dimensions of the edge weight are the same as those of the other edge. and Consistent, and is a dimensionless numerical value.

[0042] Using energy consumption nodes as vertices and edge weights as edge attributes, a directed weighted network is constructed as the construction energy consumption node network. The direction of the directed edges is determined as follows: For two construction stage types U and V that have an interference relationship, the earliest timestamp of construction stage type U in the interference pattern summary is compared with the earliest timestamp of construction stage type V in the interference pattern summary. The energy consumption node corresponding to the construction stage type with the earlier timestamp is used as the starting node of the directed edge, and the energy consumption node corresponding to the construction stage type with the later timestamp is used as the pointing node of the directed edge, forming a directed edge from the prior energy consumption node to the subsequent energy consumption node. When the earliest timestamps of the two construction stage types are the same, directed edges are established in both directions, and the weights of each edge are calculated independently according to the formula above. All deduplicated energy consumption nodes are used as the vertex set, and all directed edges with edge weights are used as the edge set. The edge weights are assigned to the edge attributes of the corresponding directed edges, forming a directed weighted network, which is the construction energy consumption node network. The number of vertices in the construction energy consumption node network is equal to the total number of construction process types, and the number of edges depends on the number of construction process types with interference relationships.

[0043] See Figure 4 In the figure, the horizontal axis represents the index of the energy consumption time series data segment, ranging from 1 to 500, and the vertical axis represents the absolute value of the correlation coefficient of the corresponding segment, ranging from 0 to 1. The blue solid curve reflects the trend of the absolute value of the Pearson correlation coefficient between the symbol sequences of energy consumption characteristic space and working state characteristic space during construction, changing with time segments. The curve fluctuates frequently overall, with most values ​​concentrated between 0.1 and 0.6, indicating that the correlation between energy consumption and mechanical working state is at a low to medium level for most time segments.

[0044] The red dashed line in the figure represents the interference discrimination threshold. This threshold is used to determine whether there is a significant interference relationship between symbol sequence pairs. Multiple blue curves exceeding this threshold indicate a strong correlation coefficient at the corresponding segment index.

[0045] The interference points marked with purple triangles correspond to local maxima on the blue curve, and the absolute values ​​of their correlation coefficients all exceed the interference discrimination threshold of 0.7. These interference points are scattered on the horizontal axis, appearing at approximately indices 50, 120, 125, 210, 320, 440, and 480, reflecting a significant interference effect between energy consumption and mechanical state symbol sequences within each key time segment.

[0046] Example 4: In practice, the interference mode is traversed and decided by a fuzzy interval decision tree to determine the construction potential energy consumption nodes and generate a dynamic evolution sequence of the construction energy consumption node network. The specific process is as follows.

[0047] A fuzzy interval mapping table is established to divide the numerical range of the interferometric mode into multiple fuzzy sub-intervals. The numerical range of the interferometric mode includes the interference intensity range and the range of each dimension of the energy consumption feature vector. Each dimension of the energy consumption feature vector includes the active power mean dimension, reactive power standard deviation dimension, power factor skewness dimension, and medium flow rate trend slope dimension. The interference intensity range is determined by the edge weights of all directed edges in the construction energy consumption node network, and the minimum and maximum values ​​of all edge weights are taken as the upper and lower bounds of the interference intensity range. The range of the active power mean dimension is determined by the minimum and maximum values ​​of the active power mean dimension in the initial state values ​​of all energy consumption nodes. The range of the reactive power standard deviation dimension is determined by the minimum and maximum values ​​of the reactive power standard deviation dimension in the initial state values ​​of all energy consumption nodes. The range of the power factor skewness dimension is determined by the minimum and maximum values ​​of the power factor skewness dimension in the initial state values ​​of all energy consumption nodes. The range of the medium flow rate trend slope dimension is determined by the minimum and maximum values ​​of the medium flow rate trend slope dimension in the initial state values ​​of all energy consumption nodes.

[0048] For each value range, a triangular membership function is used to divide it into three fuzzy sub-intervals, labeled as the low-value fuzzy sub-interval, the mid-value fuzzy sub-interval, and the high-value fuzzy sub-interval, respectively. For any given value range, let the range of the value range be... ,in, This represents the minimum value in the range. This represents the maximum value in the range. Define the center point positions of the three triangular membership functions: the center point of the low-value fuzzy sub-interval. The center point of the median fuzzy subinterval The center point of the high-value fuzzy sub-interval For any value within the range Its membership degree belongs to the low-value fuzzy subinterval The calculation method is as follows: when hour, ;when hour, ;when hour, Numerical value Membership degree of the median fuzzy subinterval The calculation method is as follows: when or hour, ;when hour, ;when hour, Numerical value Membership degree of high-value fuzzy subinterval The calculation method is as follows: when hour, ;when hour, ;when hour, Following the above method, fuzzy interval mapping tables are established for the four dimensions of the interference intensity range and the energy consumption feature vector. Each entry in the fuzzy interval mapping table records the center points of the three fuzzy sub-intervals of a range and the membership calculation function type.

[0049] A tree structure for a fuzzy interval decision tree is constructed using fuzzy sub-intervals as branching conditions. The root node of the fuzzy interval decision tree is set to interference intensity as its partitioning attribute, and it branches into three branches, corresponding to low-value, medium-value, and high-value fuzzy sub-intervals within the interference intensity value range, respectively. The first internal node layer is set to power factor skewness as its partitioning attribute, connecting to each branch of the root node. Each node in the first internal node layer also branches into three branches, corresponding to low-value, medium-value, and high-value fuzzy sub-intervals within the power factor skewness value range, respectively. The second internal node layer is set to medium flow rate trend slope as its partitioning attribute, connecting to each branch of the first internal node layer. Each node in the second internal node layer branches into three branches, corresponding to low-value, medium-value, and high-value fuzzy sub-intervals within the medium flow rate trend slope value range, respectively. The third internal node layer is defined by the average active power value, connecting to each branch of the second internal node layer. Each third internal node layer node branches into three branches, corresponding to the low-value fuzzy sub-interval, medium-value fuzzy sub-interval, and high-value fuzzy sub-interval of the average active power value range. All paths passing through the root node, the first internal node layer, the second internal node layer, and the third internal node layer terminate at leaf nodes, with the total number of leaf nodes equal to the total number of path combinations. Each leaf node stores a preset latent probability baseline value. The latent probability baseline value is set by statistically analyzing the actual frequency of energy consumption anomalies in each construction stage of historical construction projects. These actual frequencies are grouped according to combinations of interference intensity, power factor skewness, medium flow rate trend slope, and average active power value. The frequency of anomalies in each group is calculated, and the normalized frequency of anomalies in each group is used as the latent probability baseline value for the corresponding leaf node. For leaf nodes without corresponding historical data, the latent probability baseline value is the interpolation result of the latent probability baseline values ​​of adjacent combination intervals.

[0050] The interferometric pattern summary is input into a fuzzy interval decision tree, and traversed layer by layer according to the fuzzy sub-intervals, outputting the latent probability value of each construction stage. Each record in the interferometric pattern summary contains values ​​for four dimensions: construction stage type, interference intensity, and the energy consumption feature vector corresponding to that construction stage. For the first record in the interferometric pattern summary... For each record, extract the values ​​of the four dimensions of the interference intensity value and energy consumption feature vector, and calculate the membership degree of the interference intensity value to the three fuzzy sub-intervals of the interference intensity value range. Calculate the membership degree of the power factor skewness value to the three fuzzy subintervals of the power factor skewness range. Calculate the membership degree of the slope value of the medium flow rate trend to the three fuzzy sub-intervals of the slope value range of the medium flow rate trend. Calculate the membership degree of the mean active power value to the three fuzzy subintervals of the mean active power value range. , where superscript , , These represent the low-value fuzzy sub-interval, the medium-value fuzzy sub-interval, and the high-value fuzzy sub-interval, respectively. Subscript 1 indicates the interference intensity of the root node layer classification attribute, subscript 2 indicates the power factor skewness of the first internal node layer classification attribute, subscript 3 indicates the slope of the medium flow trend of the second internal node layer classification attribute, and subscript 4 indicates the average active power of the third internal node layer classification attribute.

[0051] Starting from the root node, for the three branches under the root node, respectively, based on their membership degree... , , These are used as weights for entering the low-value fuzzy sub-interval branch, the medium-value fuzzy sub-interval branch, and the high-value fuzzy sub-interval branch. In the first internal node layer, for paths entering from the low-value fuzzy sub-interval branch of the root node, the weights are respectively... , , As the weight coefficients for entering the three branches of the next level; for paths entering from the root node's median fuzzy sub-interval branch, the same applies. , , As the weight coefficients for entering the three branches of the next level; for paths entering from the high-value fuzzy sub-interval branch of the root node, the same applies. , , These serve as weight coefficients for entering the next level's three branches. The branch weights for the second and third inner node levels are assigned in the same way. After four levels of traversal, a total of... There are 3 paths, each corresponding to a leaf node with a baseline latent probability. Latent probability value of a record Calculate using the following formula: in, Indices representing the indices of the ambiguous sub-intervals of interference intensity. Corresponding to low-value fuzzy sub-intervals, Corresponding to the median fuzzy sub-interval, Corresponding to high-value fuzzy sub-intervals; Indicates the index of the fuzzy subinterval of power factor skewness. Corresponding to low-value fuzzy sub-intervals, Corresponding to the median fuzzy sub-interval, Corresponding to high-value fuzzy sub-intervals; The index represents the fuzzy sub-interval representing the slope of the medium flow trend. Corresponding to low-value fuzzy sub-intervals, Corresponding to the median fuzzy sub-interval, Corresponding to high-value fuzzy sub-intervals; Indicates the index of the fuzzy subinterval of the mean active power. Corresponding to low-value fuzzy sub-intervals, Corresponding to the median fuzzy sub-interval, Corresponding to high-value fuzzy sub-intervals; This indicates that the numerical value of the interference intensity corresponds to the index in the range of interference intensity values. The membership values ​​of the fuzzy subintervals; This indicates the power factor skewness value relative to the index in the power factor skewness range. The membership values ​​of the fuzzy subintervals; This indicates the value of the slope of the medium flow trend relative to the index of the medium flow trend slope value range. The membership values ​​of the fuzzy subintervals; This indicates the index of the average active power value within the average active power value range. The membership values ​​of the fuzzy subintervals; Representing a path The path corresponding to the latent probability baseline value stored in the leaf node indicates that the interference intensity falls into the index. The fuzzy sub-intervals and power factor skewness fall into the index. The fuzzy sub-intervals and the slope of the medium flow rate trend fall into the index. The fuzzy sub-intervals and the mean active power fall into the index. The leaf node reached when the fuzzy sub-interval is reached; The first in the summary of interferometric modes Each record corresponds to a latent probability value for a construction stage. The latent probability value is a dimensionless value ranging from 0 to 1. The latent probability value is calculated for each record in the interferometric mode summary. Then, the arithmetic mean of the latent probability values ​​of multiple records belonging to the same construction stage is taken as the latent probability value of that construction stage within the time period covered by the interferometric mode summary.

[0052] Construction stages with latent probability values ​​exceeding a preset probability threshold are marked as latent energy consumption nodes. The preset probability threshold is set to 0.7. The rationale for setting the preset probability threshold to 0.7 is as follows: the latent probability value represents the likelihood of a construction stage evolving into an energy consumption anomaly node under interference conditions. According to the binary classification criterion, when the probability value is greater than 0.5, it indicates that the probability exceeds the random level. 0.7 corresponds to a discrimination threshold approximately twice the random level, which can filter out most construction stages in the fuzzy boundary range, retaining construction stages with significant latent trends, while avoiding missed detections due to excessively high thresholds. The latent probability value of each construction stage appearing in the interference pattern summary is compared with 0.7. If the latent probability value of a construction stage is greater than 0.7, the construction stage is marked as a latent energy consumption node, and the construction stage type and latent probability value of the latent energy consumption node are recorded.

[0053] The occurrence sequence of construction latent energy consumption nodes is recorded chronologically to generate a dynamic evolution sequence. For each construction stage marked as a construction latent energy consumption node in the interferometric mode summary, the time tags of all records corresponding to that construction stage in the interferometric mode summary are extracted. The minimum time tag value is taken as the first occurrence time of the construction latent energy consumption node, and the maximum time tag value is taken as the last occurrence time of the construction latent energy consumption node. All construction latent energy consumption nodes are arranged in ascending order of their first occurrence time to generate a dynamic evolution sequence. The storage content of each element in the dynamic evolution sequence includes the node identifier, construction stage type, latent probability value, first occurrence time, and last occurrence time of the construction latent energy consumption node. The node identifier uses a combination string of construction stage type and first occurrence time as a unique identifier. When multiple construction latent energy consumption nodes have the same first occurrence time, they are arranged in lexicographical order according to the construction stage type. The dynamic evolution sequence records the temporal emergence relationship of latent nodes in the construction energy consumption node network.

[0054] Example 5: In practice, candidate potential energy consumption nodes are judged by a combination of dynamic evolution sequence and incremental threshold to obtain energy consumption abnormal nodes. The specific process is as follows.

[0055] A first-level threshold and a second-level threshold are set, with the second-level threshold being greater than the first-level threshold. The first-level threshold is set to 3, and the second-level threshold is set to 5. The first-level threshold is set to 3 because the cumulative latent frequency reflects the density of repeated marking of candidate latent energy consumption nodes in the dynamic evolution sequence. According to the statistics of continuous marking frequency when typical construction anomalies occur, when a single construction link is continuously marked 3 times in the dynamic evolution sequence, it indicates that the latent trend has been initially established rather than random fluctuations. Therefore, the first-level threshold is set to 3. The second-level threshold is set to 5 because when the cumulative latent frequency reaches 5 times, the anomaly indication of the candidate latent energy consumption node has a high degree of confidence, and the time span corresponding to 5 markings usually covers the replacement cycle of at least two adjacent construction links, which can effectively eliminate false alarms caused by transient changes in the operating conditions of a single link.

[0056] Extract the cumulative latent frequency of each candidate latent energy consumption node in the dynamic evolution sequence within the current time window. The current time window is defined as the entire time range from the start time of the dynamic evolution sequence to the current judgment time. This current time window continuously increases over time and is an incremental window. The cumulative latent frequency is the total number of times the candidate latent energy consumption node appears within the current time window. Each time a candidate latent energy consumption node is recorded in the dynamic evolution sequence, the cumulative latent frequency increases by 1.

[0057] The cumulative latent frequency is compared with the first-level threshold. If the cumulative latent frequency of a candidate latent energy consumption node exceeds the first-level threshold of 3, the candidate latent energy consumption node is marked as a node to be reviewed. The comparison process is performed independently for each candidate latent energy consumption node in the dynamic evolution sequence. Candidate latent energy consumption nodes with a cumulative latent frequency of less than 3 are not marked.

[0058] The latent probability value of the node to be audited is jointly compared with the latent probability values ​​of its neighboring nodes. Neighboring nodes refer to other energy consumption nodes in the construction energy consumption node network that have a direct directed edge connection with the node to be audited, including all predecessor neighboring nodes with directed edges pointing to the node to be audited and successor neighboring nodes pointed to by directed edges originating from the node to be audited. The joint comparison process is as follows: Obtain the latent probability value of the node to be audited, and simultaneously obtain the most recently occurring latent probability value of each neighboring node in the dynamic evolution sequence. If no neighboring node is recorded in the dynamic evolution sequence, its latent probability value is set to the default value of 0. Compare whether the latent probability value of the node to be audited is greater than the latent probability values ​​of each neighboring node. If the latent probability value of the node to be audited is greater than the latent probability values ​​of all neighboring nodes, then the continuous residence time determination condition is entered. If the latent probability value of the node to be audited is not greater than the latent probability value of any neighboring node, then the node to be audited is restored to a normal candidate latent energy consumption node and is not judged as an energy consumption anomaly node.

[0059] The continuous dwell time determination criterion is as follows: extract the duration span of the node to be audited that is continuously marked as a construction latent energy consumption node in the dynamic evolution sequence. The continuous dwell time is equal to the difference between the end timestamp and the start timestamp of the most recent continuous occurrence of the node to be audited. The continuous dwell time is compared with a preset time, which is set to 600 seconds. The basis for setting the preset time to 600 seconds is that the average duration of a typical construction link in differential pressure power generation engineering is about 1200 to 1800 seconds. A continuous dwell time of 600 seconds covers about one-third to one-half of the construction link, indicating that the latent state has formed a continuous impact within the construction link rather than a transient occurrence. When both conditions are met simultaneously, the node to be audited is determined to be an energy consumption abnormal node, and the continuous dwell time exceeds the preset time of 600 seconds and the latent probability value of the node to be audited is greater than the latent probability values ​​of all adjacent nodes.

[0060] In the specific implementation, the energy consumption time series data corresponding to the energy consumption anomaly nodes are vector-quantized and multi-dimensional energy consumption anomaly feature vectors are generated. The specific process is as follows.

[0061] The energy consumption time-series data corresponding to the energy consumption anomaly nodes are segmented into equal-length segments to obtain multiple data segments. The energy consumption time-series data corresponding to the energy consumption anomaly node is a continuous time series spliced ​​in chronological order from all energy consumption time-series data segments associated with that energy consumption anomaly node in the interferometric mode summary. The process of equal-length segmentation is as follows: the segment length is set to Q sampling points, and Q is set to 1000. The basis for setting Q to 1000 is: the sampling frequency is fixed at 50 Hz, and 1000 sampling points correspond to a time span of 20 seconds. This time span matches the minimum cycle for construction machinery to complete a complete operation, ensuring that each data segment can contain a complete energy consumption change cycle. The energy consumption time-series data is non-overlapped and segmented with a step size of Q, resulting in R data segments. Each data segment is a Q-row, D-column matrix, where D is the number of dimensions of the energy consumption feature space, and D is set to 4, corresponding to the dimensions of the mean active power, the standard deviation of reactive power, the power factor skewness, and the slope of the medium flow trend, respectively.

[0062] Each data segment undergoes vector quantization, generating a codebook index for that segment. The codebook used for vector quantization is pre-constructed using the LBG algorithm on historical normal construction energy consumption data. The number of code vectors M in the codebook is set to 256. The reason for setting M to 256 is that 256 corresponds to an 8-bit encoding length, which is universal in vector quantization, achieving an engineering-acceptable balance between compression efficiency and quantization error, while also facilitating computer storage and indexing operations. For the x-th data segment, where x takes values ​​from 1 to R, this data segment is expanded row-wise into a vector of length Q×D, denoted as . Calculate vectors Find the Euclidean distance between the M code vectors in the codebook and all M code vectors, and find the corresponding code vector among the M code vectors. The code vector with the smallest Euclidean distance is indexed in the codebook. As the codebook index of the x-th data segment, The value of is an integer ranging from 0 to 255.

[0063] All codebook indices are concatenated in chronological order to generate an initial feature index sequence. The concatenation process uses the R codebook indices corresponding to the R data segments. Arrange the data segments sequentially according to their time order to form an initial feature index sequence of length R. .

[0064] Each index in the initial feature index sequence is assigned the energy amplitude of its corresponding data segment as a weighting coefficient, resulting in a multidimensional energy consumption anomaly feature vector. For the x-th data segment, the energy amplitude... The calculation method involves taking the root mean square value of the instantaneous power at all sampling points within the data segment. The instantaneous power is obtained by multiplying the instantaneous voltage value, the instantaneous current value, and the instantaneous power factor value. Each index in the initial feature index sequence... Multiply by energy amplitude To obtain the weighted value Arrange all R weighted values ​​in chronological order to form a multidimensional energy consumption anomaly feature vector. The vector dimension is equal to R.

[0065] In practice, the construction energy consumption anomalies are spatially located and matched to specific links based on the multidimensional energy consumption anomaly feature vector. The specific process is as follows.

[0066] A 3D spatial coordinate mapping table is constructed for each construction stage, recording the coordinate range for each stage. The method for constructing the 3D spatial coordinate mapping table is as follows: acquire 3D point cloud data or Building Information Model (BIM) of the construction area, divide the construction area into multiple sub-areas according to the construction stage type, with each sub-area corresponding to one construction stage. For each construction stage, calculate the minimum X-axis coordinate of all spatial points within that sub-area. and maximum value Minimum value of Y-axis coordinate and maximum value Minimum Z-axis coordinate and maximum value Minimum bounding box Define the coordinate range for this construction stage. Store the coordinate ranges of all construction stages, along with the names of the construction stages, in a three-dimensional spatial coordinate mapping table.

[0067] Calculate the Euclidean distance between the multidimensional energy consumption anomaly feature vector and the feature template of each construction stage in the three-dimensional spatial coordinate mapping table. The feature template of each construction stage is the mean vector of the multidimensional energy consumption anomaly feature vectors extracted from each construction stage under multiple normal construction conditions. For the h-th construction stage in the three-dimensional spatial coordinate mapping table, its feature template is denoted as... ,in, This represents the component value of the feature template of the h-th construction stage in the x-th dimension. This represents the weighted value of the x-th data segment under normal conditions during the historical construction phase. The average of multiple measurements. Multidimensional energy consumption anomaly feature vector. Feature template of the h-th construction stage Euclidean distance between Calculate using the following formula: in, This represents the number of dimensions of the multidimensional energy consumption anomaly feature vector, i.e., the number of data segments. The first element in the multidimensional energy consumption anomaly feature vector represents the... The index of each component; Represents a multidimensional energy consumption anomaly feature vector In the The value of each component; Indicates the first Characteristic templates for each construction stage In the The value of each component; Represents a multidimensional energy consumption anomaly feature vector With the Characteristic templates for each construction stage The Euclidean distance between them, the dimensions of the Euclidean distance and and They have the same dimensions.

[0068] The coordinate interval corresponding to the feature template of the construction stage with the smallest Euclidean distance is used as the spatial location interval of the energy consumption anomaly node. Calculate the multidimensional energy consumption anomaly feature vector. The Euclidean distance between the feature template of each construction stage and the coordinate mapping table in the three-dimensional space is used to obtain the distance set. Where H represents the total number of construction stages. Find the minimum value from the distance set. ,Will The corresponding number The coordinate intervals of each construction stage are extracted from the three-dimensional spatial coordinate mapping table and used as the spatial location intervals for energy consumption anomaly nodes. These spatial location intervals are: .

[0069] The system associates the names of construction stages within the positioning interval with energy consumption anomaly nodes, outputting the stage matching results. These results include the node identifier of the energy consumption anomaly node, the X-axis, Y-axis, and Z-axis coordinate ranges of the spatial positioning interval, the name of the matched construction stage, and its corresponding Euclidean distance value. This matching result indicates the specific location of the energy consumption anomaly node in the construction physical space, as well as the type of construction process corresponding to that anomaly location.

[0070] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A differential pressure power generation engineering construction energy consumption monitoring method based on the Internet of Things, characterized by, Includes the following steps: Obtain energy consumption time-series data throughout the entire construction cycle of differential pressure power generation projects; Acquire time-series data on the working status of construction machinery and construction progress scenario data during the construction process; Multidimensional interference pattern mining is performed on the energy consumption time-series data and the operating status time-series data to generate an interference pattern summary; Construct a construction energy consumption node network based on the described interference mode summary; The interference pattern is traversed and decided by a fuzzy interval decision tree to determine the construction potential energy consumption nodes and generate a dynamic evolution sequence of the construction energy consumption node network. Based on the dynamic evolution sequence and incremental threshold, candidate potential energy consumption nodes are combined for discrimination to obtain energy consumption abnormal nodes; The energy consumption time series data corresponding to the energy consumption anomaly nodes are vector-quantized to generate a multi-dimensional energy consumption anomaly feature vector. Based on the multidimensional energy consumption anomaly feature vector, spatial location and process matching of construction energy consumption anomalies are performed.

2. The differential pressure power generation engineering construction energy consumption monitoring method based on the Internet of Things according to claim 1, characterized in that, The acquisition of energy consumption time-series data for the entire construction cycle of the differential pressure power generation project specifically involves: Three-phase power sensors and flow sensors were deployed at several key equipment locations during the construction of the differential pressure power generation project; Voltage, current, power factor, and medium flow data during the construction process are collected at a fixed sampling frequency using the three-phase power sensor and the flow sensor. The voltage, current, power factor, and medium flow rate data are synchronized and aligned according to timestamps to generate a multi-dimensional synchronized data stream; The multidimensional synchronous data stream is segmented by a sliding window to obtain multiple energy consumption time-series data segments.

3. The differential pressure power generation engineering construction energy consumption monitoring method based on the Internet of Things according to claim 1, characterized in that, The acquisition of the working status time-series data of construction machinery and the construction progress scenario data during the construction process specifically includes: Vibration sensors, speed sensors, and angle sensors are respectively deployed in the hydraulic system, power output shaft, and slewing mechanism of the construction machinery; The vibration sensor, the speed sensor, and the angle sensor collect the timing data of the operating parameters of the construction machinery. Video stream data of the construction area is acquired by deploying multiple fixed cameras at the construction site; Extract the start and end time markers of each construction stage from the video stream data to generate construction progress scene data.

4. The differential pressure power generation engineering construction energy consumption monitoring method based on the Internet of Things according to claim 1, characterized in that, The process of performing multi-dimensional interference pattern mining on the energy consumption time-series data and the operating status time-series data to generate an interference pattern summary is as follows: Define an energy consumption feature space and an operating state feature space, and symbolically represent the energy consumption time series data and the operating state time series data respectively; Calculate the correlation coefficient matrix between each symbol sequence in the energy consumption feature space and the operating state feature space; Local extrema are detected in the correlation coefficient matrix by using a sliding time window, and symbol sequence pairs that exhibit interference are extracted. All interfering symbol sequences are compressed and encoded to generate an interferometric mode summary.

5. The differential pressure power generation engineering construction energy consumption monitoring method based on the Internet of Things according to claim 1, characterized in that, The construction of the construction energy consumption node network based on the interference mode summary is specifically as follows: Each construction stage in the interference mode summary is considered as an energy consumption node; Extract the energy consumption feature vector of each construction stage from the interference mode summary, and use the energy consumption feature vector as the initial state value of the corresponding energy consumption node; Calculate the edge weights between energy-consuming nodes based on the interference intensity between different construction stages in the interference mode summary; Using the energy consumption nodes as vertices and the edge weights as edge attributes, a directed weighted network is constructed as the construction energy consumption node network.

6. The differential pressure power generation engineering construction energy consumption monitoring method based on the Internet of Things according to claim 1, characterized in that, The step of traversing the interference pattern using a fuzzy interval decision tree to determine the construction potential energy consumption nodes and generate a dynamic evolution sequence of the construction energy consumption node network specifically involves: A fuzzy interval mapping table is established to divide the numerical range of the interference mode into multiple fuzzy sub-intervals; Using the fuzzy sub-intervals as branching conditions for the decision tree, a tree structure of the fuzzy interval decision tree is constructed; The interference mode summary is input into the fuzzy interval decision tree, and the tree is traversed layer by layer according to the fuzzy sub-intervals to output the latent probability value of each construction stage. The construction stage where the latent probability value exceeds the preset probability threshold is marked as a construction latent energy consumption node. Record the occurrence sequence of the construction potential energy consumption nodes in chronological order to generate a dynamic evolution sequence.

7. The method for monitoring energy consumption during construction of differential pressure power generation projects based on the Internet of Things, as described in claim 6, is characterized in that... The step of performing a combined discrimination of candidate latent energy consumption nodes based on the dynamic evolution sequence and incremental threshold to obtain energy consumption anomaly nodes is as follows: Set a first-level threshold and a second-level threshold, wherein the second-level threshold is greater than the first-level threshold; Extract the cumulative latent frequency of each candidate latent energy consumption node in the dynamic evolution sequence within the current time window; The cumulative latent frequency is compared with the first-level threshold. If the cumulative latent frequency exceeds the first-level threshold, the candidate latent energy consumption node is marked as a node to be reviewed. The latent probability value of the node to be audited is compared with the latent probability values ​​of the adjacent nodes. If the latent probability value of the node to be audited is greater than the latent probability value of the adjacent nodes and the continuous residence time of the node to be audited in the dynamic evolution sequence exceeds a preset time, then the node to be audited is determined to be an energy consumption abnormal node.

8. The method for monitoring energy consumption during construction of differential pressure power generation projects based on the Internet of Things, as described in claim 7, is characterized in that... The step of vector quantizing the energy consumption time-series data corresponding to the energy consumption anomaly nodes and generating a multi-dimensional energy consumption anomaly feature vector is as follows: The energy consumption time-series data corresponding to the energy consumption anomaly nodes are divided into equal-length segments to obtain multiple data segments; Perform vector quantization on each data segment to generate the codebook index for that data segment; All codebook indices are concatenated in chronological order to generate an initial feature index sequence; Each index in the initial feature index sequence is assigned the energy amplitude of its corresponding data segment as a weighting coefficient to obtain a multidimensional energy consumption anomaly feature vector.

9. The method for monitoring energy consumption during construction of differential pressure power generation projects based on the Internet of Things, as described in claim 8, is characterized in that... The step of spatially locating and matching construction energy consumption anomalies based on the multidimensional energy consumption anomaly feature vector specifically involves: A three-dimensional spatial coordinate mapping table is constructed for each construction stage, and the mapping table records the coordinate range of each construction stage; Calculate the Euclidean distance between the multidimensional energy consumption anomaly feature vector and the feature template of each construction stage in the three-dimensional spatial coordinate mapping table; The coordinate interval corresponding to the construction segment feature template with the smallest Euclidean distance is used as the spatial positioning interval of the energy consumption anomaly node. Associate the names of construction stages within the positioning range with the energy consumption anomaly nodes, and output the stage matching results.

10. An Internet of Things-based differential pressure power generation engineering construction energy consumption monitoring system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for monitoring energy consumption during construction of a differential pressure power generation project based on the Internet of Things, as described in any one of claims 1 to 9.

Images