Intelligent construction site construction power load prediction method and system

Abstract

The present application relates to the technical field of construction power operation and maintenance management, and particularly relates to a smart construction site construction power load prediction method and system, comprising: establishing an electricity label system associated with construction conditions and giving an initial weight value; obtaining instantaneous electricity characteristics and current condition information of the work surface, matching the label and calling the corresponding weight value; based on the reference electricity characteristics, combining the current weight value to adaptively denoise and align the instantaneous electricity characteristics, and extracting the standardized electricity characteristics; according to the dynamic deviation degree of the standardized characteristics relative to the reference characteristics, combining the condition dimension label weight value, generating a sensitivity control parameter; based on the parameter, dynamically configuring the weight parameters of the space-time coupling deduction network, and taking the initial weight value of the electricity label system as the network initial constraint, and finally outputting the load prediction result. Through the present application, the dynamic change of the construction condition and the multi-device coupling interference can be adapted, and the construction power load prediction accuracy can be improved.

Description

Technical Field

[0001] This invention relates to the technical field of construction power operation and maintenance management, and in particular to a method and system for predicting construction power load at smart construction sites. Background Technology

[0002] With the deepening of digital transformation in the construction industry, smart construction site development has become an important means to improve the efficiency and safety of construction site management. As a major component of energy consumption on construction sites, accurate load forecasting of construction electricity is of great significance for optimizing temporary power supply configuration, preventing power distribution system overload, and reducing construction energy costs.

[0003] Most existing construction power load forecasting schemes adopt a general time-series forecasting architecture, which trains a forecasting model with a fixed structure based on historical power consumption data, and outputs the forecasting results after inputting the collected real-time power consumption data.

[0004] However, due to the inherent characteristics of the construction phase of smart construction sites, the construction process dynamically changes as the project progresses, and the combinations of electrical equipment vary greatly at different construction phases. Simultaneous operation of multiple high-power devices leads to strong coupling interference and high noise in power consumption data. Furthermore, extreme conditions such as temporary access to high-capacity equipment and sudden weather changes cause abrupt changes in power consumption patterns. Existing general-purpose time-series prediction architectures, employing fixed parameter configurations and generic feature processing logic, cannot adapt to the unique characteristics of these construction site scenarios. This results in significant prediction errors in practical applications, making it difficult to meet the power scheduling and safety management requirements of smart construction sites. Summary of the Invention

[0005] This invention provides a method and system for predicting the power load of smart construction sites, which can effectively solve the problems in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for predicting construction power load at smart construction sites includes: Establish an electricity consumption tagging system associated with construction conditions and assign initial weights to it in order to structurally represent the electricity consumption characteristics of construction scenarios. Obtain instantaneous power consumption characteristics and current construction conditions information of the work site; perform tag matching on the current construction conditions based on the power consumption tag system; and call the initial weight corresponding to the conditions. Based on preset benchmark power consumption characteristic information, combined with the initial weights of the current operating conditions, the instantaneous power consumption characteristic information is subjected to adaptive noise reduction and alignment processing in order to extract standardized power consumption characteristic information. Based on the dynamic offset of standardized electricity consumption characteristics relative to the baseline characteristics, and combined with the initial weights of the operating condition dimension labels, sensitivity control parameters are generated. The weight parameters of the spatiotemporal coupled simulation network are dynamically configured based on the sensitivity control parameters. At the same time, the initial weights corresponding to the electricity label system are used as the initial weight constraints of the network. The configured spatiotemporal coupled simulation network is used to output the construction electricity load prediction results.

[0007] Furthermore, in the electricity labeling system associated with construction conditions, construction conditions include construction nodes, equipment combinations, and environmental parameters; the process of assigning initial weights includes: calculating the correlation coefficient between the corresponding dimension labels of each construction condition and the fluctuation of electricity load based on historical construction records, and quantifying and assigning weights to various dimension labels based on the correlation coefficients.

[0008] Furthermore, the electricity consumption characteristic information includes multi-dimensional electrical parameters that reflect the operating status of electrical equipment. These multi-dimensional electrical parameters include at least two of the following: effective voltage value, current waveform distortion rate, fundamental power factor, and specific harmonic content.

[0009] Furthermore, adaptive noise reduction and alignment processing is performed on the instantaneous power consumption characteristic information, including: By utilizing the attention mechanism, the matching weights of real-time feature information and preset benchmark power consumption feature information in the time axis and feature dimension are corrected by combining the initial weights of the current operating conditions. Based on the corrected matching weights, the spatiotemporal misalignment noise caused by environmental electromagnetic interference and sensor sampling jitter is weighted and suppressed.

[0010] Furthermore, based on the dynamic offset of standardized electricity consumption characteristics relative to baseline characteristics, and combined with the initial weights of the operating condition dimension labels, sensitivity control parameters are generated, including: The feature divergence between standardized electricity consumption characteristics and preset benchmark electricity consumption characteristics under corresponding operating conditions is calculated. Based on the rate of change of feature divergence, the initial weights of the corresponding operating condition dimension labels are combined for weighted correction to generate a sensitivity coefficient for adjusting the learning step size of the spatiotemporal coupling inference network.

[0011] Furthermore, the spatiotemporal coupled inference network is a hybrid neural network that includes temporal convolutional modules and graph convolutional modules; Among them, the temporal convolution module is used to capture the long-range dependency features of the load sequence of a single device, and the graph convolution module is used to construct the implicit electrical association topology between construction equipment based on the data-driven approach, which is used to characterize the coupling interference features when multiple devices are running in parallel. In the process of constructing the association topology of the graph convolution module, the initial allocation of the association weight between devices is optimized by combining the initial weight of the device combination dimension in the power consumption label system.

[0012] Furthermore, the benchmark electricity consumption characteristic information is classified according to the tag combination of the electricity consumption tag system. Each tag combination corresponds to a unique set of benchmark electricity consumption characteristic information, which comes from the historical high-quality electricity consumption data of completed projects that are the same as or similar to the current smart construction site in terms of construction type, scale, and process.

[0013] Furthermore, when generating sensitivity control parameters, the characteristic divergence sequence is first smoothed by moving average to eliminate divergence abrupt changes caused by random factors, and then the rate of change of characteristic divergence is calculated.

[0014] Furthermore, in the spatiotemporal coupled inference network, the temporal convolution module adopts a dilated convolution structure and sets a normalization layer, and the initial association weight allocation of the graph convolution module is based on the initial weight of the device combination dimension.

[0015] On the other hand, the present invention also provides a smart construction site power load prediction system, comprising: The tag building module is used to establish an electricity tag system associated with construction conditions and assign initial weights to it, so as to structurally represent the electricity characteristics of the construction scenario. The feature acquisition module is used to acquire instantaneous power consumption feature information and current construction condition information of the work surface. Based on the power consumption tag system, it performs tag matching on the current construction condition and calls the initial weight corresponding to the condition. The feature standardization module is used to perform adaptive noise reduction and alignment processing on instantaneous power consumption feature information based on preset benchmark power consumption feature information and combined with the initial weights of the current operating conditions, so as to extract standardized power consumption feature information. The sensitivity generation module is used to generate sensitivity control parameters based on the dynamic offset of standardized electricity consumption characteristic information relative to the benchmark characteristic, combined with the initial weights of the operating condition dimension labels. The load forecasting module is used to dynamically configure the weight parameters of the spatiotemporal coupled simulation network based on the sensitivity control parameters. At the same time, the initial weights corresponding to the electricity consumption tag system are used as the initial weight constraints of the network. The configured spatiotemporal coupled simulation network is used to output the construction electricity load forecasting results.

[0016] The technical solution of this invention can achieve the following technical effects: By establishing an electricity consumption tagging system associated with construction conditions and assigning quantified initial weights, a structured representation of construction electricity consumption characteristics is achieved. Combined with condition tag matching, the initial weights are ensured to be accurately adapted to the current construction scenario. An attention mechanism is used in conjunction with the initial condition weights to adaptively denoise and align instantaneous electricity consumption characteristics, suppressing noise caused by environmental electromagnetic interference and sensor sampling jitter, and extracting accurate standardized electricity consumption characteristics. Based on feature divergence and rate of change combined with the initial condition weights, sensitivity control parameters are generated to achieve dynamic configuration of the spatiotemporal coupling inference network weight parameters. At the same time, the initial weights of the electricity consumption tags are used as the initial constraints of the network. Combined with a hybrid structure of temporal convolution and graph convolution, the long-range dependency characteristics of single devices and the coupling interference characteristics of multiple devices are captured, adapting to dynamic changes in construction conditions, multi-device coupling interference, and extreme conditions, thereby improving the accuracy and stability of construction electricity load prediction.

[0017] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating a smart construction site power load prediction method according to the present invention. Figure 2 This is a schematic diagram of the structure of a smart construction site power load prediction system according to the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0022] like Figure 1 As shown, the present invention provides a method for predicting the power load of a smart construction site, which specifically includes the following steps: Step S1: Establish an electricity consumption tagging system associated with construction conditions and assign initial weights to it in order to structurally represent the electricity consumption characteristics of the construction scenario. Step S2: Obtain the instantaneous power consumption characteristics and current construction conditions of the work surface. Based on the power consumption tag system, perform tag matching on the current construction conditions and call the initial weight corresponding to the conditions. Step S3: Based on the preset benchmark power consumption feature information and combined with the called current operating condition initial weight, perform adaptive noise reduction and alignment processing on the instantaneous power consumption feature information to extract standardized power consumption feature information. Step S4: Based on the dynamic offset of the standardized electricity consumption characteristic information relative to the benchmark characteristic, and combined with the initial weights of the operating condition dimension labels, generate sensitivity control parameters. Step S5: Dynamically configure the weight parameters of the spatiotemporal coupling simulation network based on the sensitivity control parameters, and use the initial weights corresponding to the electricity label system as the initial weight constraints of the network. Then, use the configured spatiotemporal coupling simulation network to output the construction electricity load prediction results.

[0023] In this embodiment, by combining the power consumption tag system associated with the working conditions and the initial quantified weights with working condition tag matching, adaptive noise reduction and alignment, sensitivity control parameter generation, and spatiotemporal coupling inference network dynamic configuration and initial constraint depth, a complete working condition adaptation-feature optimization-network regulation closed loop is formed. Through the initial weights of the tag system running through the entire process, noise reduction and alignment can accurately adapt to the equipment combination and environmental interference of the current construction working conditions, sensitivity parameters can accurately reflect the fluctuation of power consumption characteristics under the guidance of working conditions, and network configuration can respond to changes in working conditions in real time. The three work together to achieve a synchronous improvement in prediction accuracy and working condition adaptability.

[0024] In a specific implementation, as one example, given the strong correlation between construction power load fluctuations and construction node progress, equipment combination switching, and environmental parameter changes, and the inability of general time series prediction models to distinguish the respective contributions of these different operating conditions to the load when directly processing raw power consumption data, leading to significant prediction errors when operating conditions change abruptly, it is necessary to establish a scenario description vector that can be numerically calculated before load prediction. Initial weights need to be assigned to the three dimensions of construction nodes, equipment combinations, and environmental parameters included in the construction conditions. This allows the subsequent network to selectively utilize operating condition information based on the actual impact of each dimension label on load fluctuations. This embodiment calculates the correlation coefficient between each dimension label and power load fluctuations through statistical correlation analysis based on historical construction records, and quantifies and weights them based on the correlation coefficients to achieve a structured representation of the power consumption characteristics of the construction scenario, as detailed below: Step S11: Construct the dimensional structure of the electricity labeling system, with each dimension corresponding to a type of construction condition; The first dimension is the construction node label. Based on the sub-item division of the project in the construction organization design document, the entire construction process is discretized into several nodes, and each construction node corresponds to a discrete label value. For example, the foundation pit support stage, basement floor slab pouring, first floor wall column reinforcement binding of the main structure, and first floor roof slab concrete pouring of the main structure are sequentially coded as 01, 02, 03, 04, and so on. This coding method gives each construction stage a unique identifier. The change of construction node will directly lead to a fundamental change in the combination of electrical equipment, which is one of the driving factors of power load fluctuation. Through this type of label, the distinguishable characterization of power consumption characteristics at different construction stages can be achieved. The second dimension is the equipment combination label, with each equipment combination state corresponding to an independent vector label. By collecting all electrical equipment sets on the current work surface, this set is encoded into a fixed-length vector, with each bit representing a type of equipment, such as tower crane, construction elevator, welding machine, and rebar bending machine. The length of this vector does not change with the increase or decrease of equipment used in different construction stages. Unconfigured equipment bits are always 0, while the values ​​of configured equipment bits correspond to the operating status and quantity combination of the equipment. For example, a value of 1 represents a single piece of equipment of this type operating without load, a value of 2 represents a single piece of equipment of this type operating with light load, a value of 3 represents a single piece of equipment of this type operating with full load, a value of 4 represents two pieces of equipment of this type operating without load, and so on. Differences in equipment combinations directly lead to different magnitudes and fluctuation patterns of electrical load. Among them, changes in the operating status of high-power equipment have a significantly greater impact on load fluctuations than those of conventional power equipment. Through this vector label, the equipment composition and its corresponding electrical load characteristics under different working conditions can be comprehensively characterized. The third dimension is the environmental parameter label. Changes in environmental parameters affect the operating efficiency of equipment, which in turn leads to fluctuations in electrical load. Relevant environmental parameters include temperature, humidity, illuminance, wind speed, and thunderstorm conditions. Among them, excessively high temperatures will cause the equipment's heat dissipation efficiency to decrease, the operating current to increase, and the electrical load to rise. Thunderstorms will cause some outdoor operating equipment to shut down, resulting in a sharp drop in electrical load. Excessive wind will limit the operating power of large equipment such as tower cranes, resulting in a decrease in electrical load. Therefore, the environmental parameter dimension labels are set as high temperature label, high humidity label, low illuminance label, strong wind label, and thunderstorm label. Each label corresponds to a specific range of environmental parameters, and the label switches accordingly when the environment changes abruptly, in order to characterize the electrical characteristics under extreme or special environmental conditions. Step S12: Obtain historical construction records. The scope of obtaining historical construction records includes completed projects with the same or similar construction type, scale, and process as the current smart construction site, ensuring the relevance and reference value of historical data. Historical construction records include: Construction log: Includes start and end times, construction procedures, and work area distribution for each construction node, which are used to label the corresponding construction node dimensions; Equipment operation records include information such as equipment type, quantity in operation, and operating status at each construction stage, which are used for corresponding equipment combination dimension tags; Environmental monitoring data: Parameters such as temperature, humidity, illuminance, wind speed and thunderstorm conditions are obtained from the construction site meteorological station or environmental monitoring sensors and used for corresponding environmental parameter dimension labels; Electricity monitoring data: collected by electricity meters deployed at the construction site, including electricity load data for each time period; The aforementioned historical construction records need to be preprocessed to remove abnormal data and supplement missing content to ensure the authenticity and completeness of the data; Step S13: Based on the preprocessed historical construction records, calculate the correlation between each dimension label and the electricity load: For each label in the construction node dimension, extract all the power load data of the construction stage corresponding to that label, calculate the fluctuation value and frequency of the power load in that stage, and then calculate the correlation between the construction stage duration corresponding to that label and the power load fluctuation value. This can be calculated by "load fluctuation value × fluctuation frequency ÷ construction stage duration". For each label in the equipment combination dimension, extract all power load data of the corresponding equipment combination during operation, calculate the power load fluctuation value during the operation of the equipment combination, and then calculate the correlation between the number of equipment combinations in operation, the operating status and the power load fluctuation value. This can be calculated by "load fluctuation value × number of equipment in operation × operating status weight ÷ running time". Among them, the operating status weight is preset according to the load impact of no-load, light-load and full-load to ensure the rationality of quantification. For each label in the environmental parameter dimension, extract all electricity load data under the corresponding environmental conditions, calculate the electricity load fluctuation value under the environmental conditions, and then calculate the correlation between the environmental parameter and the electricity load fluctuation value. This can be calculated by "load fluctuation value × parameter change amplitude ÷ duration of environmental parameter". Step S14: Normalize the correlation quantification values ​​of all tags. The normalized values ​​are the correlation coefficients between each tag and the fluctuation of electricity load. Each coefficient is the initial weight of the corresponding dimension tag.

[0025] In this embodiment, each tag in the electricity tagging system is directly associated with the construction conditions, enabling precise and structured characterization of the electricity consumption characteristics of the construction scenario. The correlation coefficient quantifies the strength of the association between the tag and the fluctuation of electricity load, and this association has been calibrated using measured data. Normalization only adjusts the numerical scale without changing the essential strength of the association. Therefore, the normalization result can be directly used as the correlation coefficient of each tag to the fluctuation of electricity load, preserving objectivity and providing a unified comparison standard. The initial weight is used to characterize the influence weight of the tag on the prediction of electricity load. The stronger the association between the tag and the load fluctuation, the greater the influence on the prediction result, and the higher the weight should be, and vice versa. Consistent with the quantification logic of the correlation coefficient, the correlation coefficient is directly used as the initial weight, which not only ensures that the weight matches the actual influence of the tag, but also simplifies the weighting process and avoids errors introduced by secondary calculations.

[0026] In some embodiments of the present invention, existing methods for acquiring instantaneous power consumption characteristics of the work surface typically only collect a single electrical parameter, which cannot distinguish the differences in power consumption characteristics generated by different equipment types under the same power level. For example, a 50-kilowatt tower crane and a 50-kilowatt welding machine have the same effective voltage value, but their current waveforms are completely different. If only the effective value is input, the prediction model will treat them as the same load, leading to prediction deviation. This embodiment achieves the acquisition and matching of instantaneous power consumption characteristics of the work surface and current construction condition information through targeted collection, matching, and weighted call logic. Specifically, the following operations are performed: Step S21: Combine the existing monitoring system of the smart construction site to obtain the instantaneous power consumption characteristics of the work surface, including at least two of the following: effective voltage value, current waveform distortion rate, fundamental power factor, and specific harmonic content. The effective voltage value is collected by an AC voltage sensor, which is deployed at the incoming end of the main distribution cabinet on the work site and at the power supply end of high-power equipment. The effective voltage value is used to capture the power supply stability of the temporary construction power distribution system, reflect voltage fluctuations caused by line loss, simultaneous start-up of multiple devices, etc., and thus reflect the changes in the operating power of the equipment. The current waveform distortion rate is obtained through the collaborative acquisition of an AC current sensor and a waveform analysis module. The current sensor and voltage sensor are deployed synchronously. After acquiring the real-time current waveform, the distortion rate is obtained by comparing it with the standard sine waveform corresponding to the rated frequency of the construction power supply and calculating the degree of deviation. The current waveform distortion rate is used to characterize the degree of harmonic interference generated by the operation of nonlinear equipment in the construction scenario, reflect the ineffective loss of the line, and help distinguish the differences in the electrical characteristics of the equipment operation. The fundamental power factor is collected by a power factor sensor, which is deployed at the incoming line of the main distribution cabinet on the work surface. The sensor simultaneously collects voltage and current signals, separates the fundamental active power and the fundamental reactive power, and calculates the ratio. The fundamental power factor is used to reflect the active power utilization efficiency of the equipment, distinguish between effective electrical load and redundant losses caused by reactive power, and avoid ineffective loads interfering with the prediction results. The specific harmonic content rate is collected by a harmonic analyzer and works in conjunction with voltage and current sensors. The 3rd, 5th, and 7th harmonic components are separated by Fourier transform, and the ratio of the harmonic content rate to the fundamental component is calculated. The specific harmonic content rate is used to distinguish the operating status of different types of nonlinear equipment, further refine the power consumption characteristics, and improve the pertinence of the characteristic characterization. The four electrical parameters mentioned above each have their own focus. By selecting at least two parameters, the complementary nature of power consumption characteristics in various dimensions can be achieved, ensuring that instantaneous power consumption characteristics can accurately reflect the equipment's operating status and power quality. Step S22: Combine the existing management system and monitoring equipment of the smart construction site to obtain the current construction status information. The status information includes three dimensions: construction nodes, equipment combination and environmental parameters, which correspond to the three dimensions of the electricity label system. Step S23: Construct a tag matching matrix. The row dimensions are the three dimensions of the electricity tag system: construction node, equipment combination, and environmental parameters. The column dimensions are the three dimensions of the current construction condition information. Matching is performed sequentially according to the dimensions. When matching construction node tags, the current construction node code is compared with all tag codes in the construction node dimension of the electricity tag system. Tags with completely identical codes are matched, and the corresponding initial weights are called. When matching equipment combination tags, the current equipment combination vector tag is compared with all vector tags in the equipment combination dimension of the electricity tag system bit by bit. Tags with completely identical values ​​for each bit are matched, and the corresponding initial weights are called. When matching environmental parameter tags, the current environmental parameter tag is compared with all tags in the environmental parameter dimension of the electricity tag system one by one. Tags corresponding to all current environmental parameters are matched to form environmental parameter tag combinations, and the corresponding initial weights are called.

[0027] In this embodiment, by using distributed data acquisition in collaboration with a smart construction site system, at least two types of instantaneous power consumption characteristic information of work surfaces are obtained, which makes up for the deficiency that a single electrical parameter cannot distinguish between different types of equipment with the same power. At the same time, three aspects of working condition information, namely construction nodes, equipment combinations, and environmental parameters, are collected to construct a three-dimensional label matching matrix. According to the logic of complete matching of construction node codes, bit-by-bit matching of equipment combination vectors, and one-by-one matching of environmental parameter labels, the corresponding initial weights are called respectively to achieve efficient matching of power consumption characteristics and working condition information, thereby improving prediction accuracy.

[0028] In a specific implementation, as one example, existing noise reduction and alignment processing for instantaneous power consumption feature information typically employs fixed thresholds or general filtering algorithms. In construction scenarios, environmental electromagnetic interference and sensor sampling jitter inevitably lead to misalignment between instantaneous power consumption features and baseline features on the time axis and deviations in feature dimensions. Without using an attention mechanism to correct the matching weights, it is impossible to accurately distinguish between effective features and noise, easily resulting in the loss of effective features or incomplete noise suppression. Furthermore, without combining the initial weights of the current operating condition, it is impossible to adjust the processing focus according to the degree of influence of the operating condition on power load fluctuations, leading to insufficient noise reduction and alignment accuracy under different operating conditions, and a significant deviation between the standardized power consumption feature information and the actual power consumption state. This embodiment utilizes an attention mechanism and combines the initial weights of the operating condition to correct the matching weights, achieving weighted suppression of spatiotemporal misalignment noise. The specific implementation steps are as follows: Step S31: Obtain benchmark power consumption characteristic information; the benchmark power consumption characteristic information is classified according to the power consumption tag system, based on the tag combination of construction node, equipment combination, and environmental parameters. Each tag combination corresponds to a unique set of benchmark power consumption characteristic information; the acquisition scope is limited to historical high-quality power consumption data of completed projects that are the same or similar to the current smart construction site construction type, scale, and process. The screening criteria are clearly defined as normal and stable operation of construction equipment, no environmental electromagnetic interference, no jitter in sensor sampling, and power load fluctuation within a reasonable range; through statistical analysis of the screened historical high-quality power consumption data, the stable characteristic values ​​and characteristic change patterns of electrical parameters corresponding to the acquired instantaneous power consumption characteristic information under each working condition tag combination are extracted to form benchmark power consumption characteristic information that includes the characteristic time sequence pattern of the time axis dimension and the characteristic standard range of the characteristic dimension. Step S32: Using an attention mechanism, combined with the initial weights of the current operating conditions, correct the matching weights of instantaneous power consumption feature information and preset benchmark power consumption feature information in the time axis and feature dimension; the matching weight is used to characterize the degree of correlation between instantaneous power consumption feature and benchmark power consumption feature. The higher the degree of correlation, the larger the weight, and the higher the degree of preservation of this part of the feature during the noise reduction and alignment process; the lower the degree of correlation, the smaller the weight, and the higher the degree of suppression of this part of the noise during the noise reduction and alignment process. The weighting adjustment along the time axis involves first dividing the electricity consumption sequence into several continuous intervals based on the duration of each process within the construction node, with each interval corresponding to a specific process. Then, the initial weights of the construction nodes are proportionally allocated to each time interval; the higher the weight, the higher the basic matching weight. Next, the initial weights of environmental parameters are dynamically adjusted: if the environmental weight is higher than a set threshold, indicating strong interference, the matching weights for all time intervals are proportionally reduced to suppress sampling jitter noise; if it is lower than the threshold, the basic weights remain unchanged. The weighting of feature dimensions is first adjusted by determining the core electrical parameters based on the initial weights of the equipment combination: the higher the weight, the more high-power equipment there are, and the core parameters are the current waveform distortion rate and the specific harmonic content rate; for conventional equipment combinations, the core parameters are the effective voltage value and the fundamental power factor. The basic matching weight of the core parameters is set as a reasonable multiple of the non-core parameters. Then, the initial weights of environmental parameters are adjusted in a targeted manner: for parameters susceptible to electromagnetic interference, such as the effective voltage value and the current waveform distortion rate, the matching weight is reduced according to the interference intensity when the environmental weight is higher than the threshold; for parameters with strong anti-interference capabilities, such as the fundamental power factor and the specific harmonic content rate, the basic weight remains unchanged. During the weight correction process, upper and lower limits for weight correction are set to limit the reasonable range of weight values, thereby avoiding overcorrection that leads to the loss of effective features or undercorrection that leads to uncontrollable noise. Step S33: For spatiotemporal misalignment noise in the time axis dimension, the timing of instantaneous power consumption feature information is aligned and adjusted by combining the corrected time axis matching weights: for time intervals with high matching weights, the original timing information of instantaneous features is retained, and only obvious sampling jitter noise is removed; for time intervals with low matching weights, the timing pattern of the baseline power consumption feature information is used to correct the instantaneous features and correct time axis misalignment problems such as sampling delay and sampling jump. To address spatiotemporal misalignment noise in the feature dimensions, noise reduction is performed on the instantaneous power consumption feature information by combining the corrected feature dimension matching weights: for electrical parameters with high matching weights, an adaptive filtering method is used to remove noise, and the filtering intensity is dynamically adjusted according to the noise intensity of the parameter. The noise intensity is determined by the degree of deviation between the instantaneous feature and the reference feature; for electrical parameters with low matching weights, the instantaneous feature is corrected using the standard range of the reference feature, and outliers exceeding the reasonable range are removed. These outliers are caused by environmental electromagnetic interference and are corrected using the reasonable range of the reference feature. The processed instantaneous power consumption characteristics are standardized by normalization, mapping the values ​​of each electrical parameter to a unified standard range, eliminating differences between different parameter dimensions, and ensuring that the standardized power consumption characteristics have a unified comparison benchmark.

[0029] In this embodiment, a reference power consumption feature information that corresponds one-to-one with the working condition label combination is adopted, which can fit the actual power consumption patterns of different construction scenarios and avoid the alignment deviation caused by the general reference. Relying on the attention mechanism combined with the initial weight of the working condition to dynamically correct the dual-dimensional matching weight, it can specifically distinguish effective features and spatiotemporal misalignment noise according to the differences in construction working conditions and the intensity of environmental interference. It retains the core power consumption features and suppresses the noise caused by environmental electromagnetic interference and sensor sampling jitter, avoiding the problem of loss of effective features or incomplete noise suppression.

[0030] In a specific implementation, as one example, in a construction scenario, the degree and trend of deviation between standardized electricity consumption characteristics and baseline characteristics directly determine the reasonable adjustment direction of the prediction network's learning step size. Adjusting the prediction network's learning rate requires capturing this dynamic characteristic to avoid the learning rhythm becoming disconnected from actual electricity consumption changes. This example quantifies the degree of deviation through feature divergence and combines it with working condition weight correction optimization to generate targeted sensitivity control parameters, achieving precise adjustment of the prediction network's learning step size. The specific implementation steps are as follows: Step S41: Based on the continuously collected standardized electricity consumption characteristic information, calculate the feature divergence between each group and the corresponding benchmark feature to form a feature divergence sequence; the feature divergence is used to quantify the dynamic offset between the two, and is obtained by extracting the distribution features of each of the two types of features based on their respective electrical parameters, and calculating the difference in distribution. Step S42: The rate of change is obtained by the ratio of the difference between two adjacent divergence sets to the acquisition time interval. The acquisition time interval is kept consistent with the acquisition frequency of electrical parameters to ensure accurate capture of real-time change trends. During the calculation process, the divergence sequence is smoothed by the moving average method to eliminate divergence abrupt changes caused by accidental factors. The characteristic divergence rate of change is used to reflect the dynamic change trend of the degree of offset and to provide a dynamic basis for adjusting the learning step size. Step S43: Set correction weights based on the quantitative characteristics of each dimension's weights, with the equipment combination dimension having the highest weight, followed by the environmental parameter dimension, and the construction node dimension serving as an auxiliary dimension; weight and fuse each dimension's weights with the characteristic divergence change rate to obtain the corrected change rate. The correction process does not change the change trend, but only adjusts the numerical value to ensure that it reflects the actual fluctuation trend under the current working conditions. Step S44: Map the corrected rate of change to a preset value range to generate a sensitivity coefficient. The sensitivity coefficient is a sensitivity control parameter used to directly regulate the learning step size of the spatiotemporal coupled inference network. The mapping logic corresponds to the trend and value of the rate of change: if the rate of change is positive and the value is larger, the sensitivity coefficient is larger and the network learning step size is larger to adapt to drastic fluctuations; if the rate of change is negative and the absolute value is larger, the sensitivity coefficient is smaller and the learning step size is smaller to ensure prediction stability; if the rate of change approaches zero, the sensitivity coefficient is maintained in a reasonable intermediate range to ensure stable network learning. The preset value range is determined based on the network structure characteristics and the historical patterns of construction power load. The upper limit avoids excessively large learning steps that may cause oscillations, and the lower limit avoids excessively small step sizes that cannot adapt to sudden load changes.

[0031] In this embodiment, the sensitivity control parameters are accurately generated step by step, and the benchmark and standardized features adapted to the working conditions are invoked to ensure the accuracy of the calculation basis. Among them, feature divergence and rate of change are used to quantify the degree of deviation and fluctuation trend of electricity consumption characteristics, providing data support for learning step size adjustment. Combined with the weighted correction of the initial weight of the working conditions, the sensitivity coefficient is made to fit the current construction conditions, improving the targeting of the adjustment. The generated sensitivity coefficient can be directly adapted to the spatiotemporal coupled inference network to realize dynamic adjustment of the learning step size, making the learning rate adjustment targeted and matching the network learning rhythm with the fluctuation trend of electricity load.

[0032] In a specific implementation, as one example, to achieve accurate prediction of construction power load, a spatiotemporal coupled simulation network needs to be constructed. The established power consumption tagging system and the invoked initial weights of the operating conditions are used as the initial weight constraints of the network. At the same time, the network weight parameters are dynamically configured based on the generated sensitivity control parameters. The construction power load prediction is completed through the configured network. The specific implementation steps are as follows: Step S51: Construct a spatiotemporal coupled inference network. This network is a hybrid neural network including a temporal convolutional module and a graph convolutional module, using a serial design. The output of the temporal convolutional module serves as the input of the graph convolutional module, and the output of the graph convolutional module is connected to a fully connected layer. The fully connected layer serves as the output of the network, responsible for converting the fused feature vector into the predicted value of construction power load, realizing the orderly fusion of the two types of features and the output of the prediction result. Among them, the temporal convolutional module is used to capture the long-range dependency features of the load sequence of a single device. It adopts a dilated convolutional structure and sets a normalization layer. The input is the standardized power consumption feature sequence of a single device, and the output is the long-range dependency feature vector of a single device. The graph convolutional module is used to construct the implicit electrical association topology between devices based on data-driven methods, representing the coupling interference of multiple devices. The input is the output vector of the temporal convolutional module, and the output is the overall power consumption feature vector that fuses the coupling interference. When constructing the association topology, the initial weight of the current working condition device combination dimension is called. This weight is directly used as the initial allocation benchmark for the association weight between devices. The higher the weight of the corresponding device in the device combination, the greater the initial association weight allocation between it and other devices, thereby optimizing the initial allocation of the association weight between devices. Step S52: Use the initial weights of the three dimensions of construction nodes, equipment combinations, and environmental parameters as the initial weight constraints of the spatiotemporal coupled inference network to complete the network initialization. Among them, the weights of the equipment combination dimension constrain the associated weights of the graph convolution module, the weights of the construction node dimension constrain the convolution kernel weights of the temporal convolution module, and the weights of the environmental parameter dimension constrain the parameters of the network normalization layer. During the constraint process, set the weight constraint threshold to ensure that the initial weights are within a reasonable range, and avoid excessive constraints that lead to insufficient network training flexibility, or insufficient constraints that fail to play the role of working condition guidance. Step S53: Based on the sensitivity control parameters, the network weight parameters are dynamically configured in real time. This is achieved by mapping the sensitivity coefficient to the adjustment range of the network weight parameters. The mapping logic corresponds to the value and trend of the sensitivity coefficient. A larger sensitivity coefficient indicates more severe fluctuations in the power load, resulting in a larger adjustment range for the weight parameters, accelerating the network weight update speed and enabling the network to quickly adapt to fluctuations. A smaller sensitivity coefficient indicates a more stable power load, resulting in a smaller adjustment range for the weight parameters, slowing down the weight update speed and ensuring the stability of the network prediction. When the sensitivity coefficient approaches zero, the current weight parameters are maintained unchanged to keep the network in a stable learning state. The configuration range covers the temporal convolution module, the graph convolution module, and the core weights of the fully connected layer. The core weights of the fully connected layer are used to optimize the output accuracy of the prediction results. The adjustment range of the graph convolution module is greater than that of the temporal convolution module. The adjustment range of the fully connected layer is combined with the initial weight correction under the operating conditions; that is, the higher the weight of the device combination dimension, the larger the adjustment range of the fully connected layer weights, ensuring that the output prediction results fit the current operating conditions. Each generation of a set of sensitivity coefficients completes one configuration. Step S54: Input the standardized power consumption characteristic information into the initialized and dynamically configured network. The two main modules capture and fuse the long-range dependency and coupling interference characteristics respectively. After processing by the fully connected layer, the construction power load prediction result is output. The electrical parameters with the prediction time granularity are collected at the same frequency.

[0033] In this embodiment, a hybrid spatiotemporal coupled inference network composed of a temporal convolutional module and a graph convolutional module is constructed to achieve the collaborative capture of long-range dependence features of construction power load in the time dimension and coupling interference features in the equipment dimension. Specifically, by using the initial weights of the power consumption label system as the initial weight constraints of the network, the network initialization is made to fit the current working conditions, reducing training bias and improving network adaptability. The network weight parameters are dynamically configured based on sensitivity control parameters, so that the network learning rhythm matches the power load fluctuations in real time. When the graph convolutional module constructs the associated topology, it combines the initial weights of the equipment combination dimension to optimize the initial association weight allocation and improve the accuracy of the coupling interference feature representation.

[0034] Based on the same inventive concept as the smart construction site power load prediction method in the foregoing embodiments, this invention also provides a smart construction site power load prediction system, such as... Figure 2 As shown, the system includes: The tag building module is used to establish an electricity tag system associated with construction conditions and assign initial weights to it, so as to structurally represent the electricity characteristics of the construction scenario. The feature acquisition module is used to acquire instantaneous power consumption feature information and current construction condition information of the work surface. Based on the power consumption tag system, it performs tag matching on the current construction condition and calls the initial weight corresponding to the condition. The feature standardization module is used to perform adaptive noise reduction and alignment processing on instantaneous power consumption feature information based on preset benchmark power consumption feature information and combined with the initial weights of the current operating conditions, so as to extract standardized power consumption feature information. The sensitivity generation module is used to generate sensitivity control parameters based on the dynamic offset of standardized electricity consumption characteristic information relative to the benchmark characteristic, combined with the initial weights of the operating condition dimension labels. The load forecasting module is used to dynamically configure the weight parameters of the spatiotemporal coupled simulation network based on the sensitivity control parameters. At the same time, the initial weights corresponding to the electricity consumption tag system are used as the initial weight constraints of the network. The configured spatiotemporal coupled simulation network is used to output the construction electricity load forecasting results.

[0035] The system described above in this invention can effectively realize a method for predicting the power load of smart construction sites, and the technical effects it can achieve are as described in the above embodiments, which will not be repeated here.

[0036] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Thus, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for predicting the power load of a smart construction site, characterized in that, include: Establish an electricity consumption tagging system associated with construction conditions and assign initial weights to it in order to structurally represent the electricity consumption characteristics of construction scenarios. Obtain instantaneous power consumption characteristics and current construction conditions information of the work site; perform tag matching on the current construction conditions based on the power consumption tag system; and call the initial weight corresponding to the conditions. Based on preset benchmark power consumption feature information, and combined with the initial weights of the current operating conditions, the instantaneous power consumption feature information is subjected to adaptive noise reduction and alignment processing to extract standardized power consumption feature information. Based on the dynamic offset of the standardized electricity consumption characteristic information relative to the benchmark characteristic, and combined with the initial weights of the operating condition dimension labels, sensitivity control parameters are generated. The weight parameters of the spatiotemporal coupling simulation network are dynamically configured based on the sensitivity control parameters. At the same time, the initial weights corresponding to the electricity label system are used as the initial weight constraints of the network. The configured spatiotemporal coupling simulation network is used to output the construction electricity load prediction results. In the electricity labeling system associated with construction conditions, the construction conditions include construction nodes, equipment combinations, and environmental parameters; the process of assigning initial weights includes: calculating the correlation coefficient between the corresponding dimension labels of each construction condition and the fluctuation of electricity load based on historical construction records, and quantifying and assigning weights to various dimension labels according to the correlation coefficients.

2. The method for predicting construction power load at a smart construction site according to claim 1, characterized in that, The electricity consumption characteristic information includes multi-dimensional electrical parameters that reflect the operating status of electrical equipment. These multi-dimensional electrical parameters include at least two of the following: effective voltage value, current waveform distortion rate, fundamental power factor, and specific harmonic content.

3. The method for predicting construction power load at a smart construction site according to claim 1, characterized in that, Adaptive noise reduction and alignment processing is performed on the instantaneous power consumption characteristic information, including: By utilizing the attention mechanism, the matching weights of instantaneous power consumption feature information and the reference power consumption feature information in the time axis and feature dimension are corrected by combining the initial weights of the current operating condition. Based on the corrected matching weights, the spatiotemporal misalignment noise caused by environmental electromagnetic interference and sensor sampling jitter is weighted and suppressed.

4. The method for predicting construction power load at a smart construction site according to claim 1, characterized in that, Based on the dynamic offset of the standardized electricity consumption characteristic information relative to the baseline characteristics, and combined with the initial weights of the operating condition dimension labels, sensitivity control parameters are generated, including: Calculate the feature divergence between the standardized electricity consumption feature information and the baseline electricity consumption feature information under the corresponding operating conditions. Based on the rate of change of the feature divergence, perform weighted correction in combination with the initial weights of the corresponding operating condition dimension labels to generate a sensitivity coefficient for adjusting the learning step size of the spatiotemporal coupling inference network.

5. The method for predicting construction power load at a smart construction site according to claim 1, characterized in that, The spatiotemporal coupling inference network is a hybrid neural network including a temporal convolution module and a graph convolution module; The temporal convolution module is used to capture the long-range dependency features of the load sequence of a single device, and the graph convolution module is used to construct the implicit electrical association topology between construction devices based on a data-driven approach, which is used to characterize the coupling interference features when multiple devices are running in parallel. In the process of constructing the association topology of the graph convolution module, the initial allocation of the association weights between devices is optimized by combining the initial weights of the device combination dimension in the power consumption label system.

6. The method for predicting construction power load at a smart construction site according to claim 1, characterized in that, The benchmark electricity consumption characteristic information is classified according to the tag combination of the electricity consumption tag system. Each tag combination corresponds to a unique set of benchmark electricity consumption characteristic information, which comes from the historical electricity consumption data of completed projects with the same construction type, scale and process as the current smart construction site.

7. The method for predicting construction power load at a smart construction site according to claim 4, characterized in that, When generating the sensitivity control parameters, the characteristic divergence sequence is first smoothed by moving average to eliminate divergence abrupt changes caused by random factors, and then the rate of change of characteristic divergence is calculated.

8. The method for predicting construction power load at a smart construction site according to claim 5, characterized in that, In the spatiotemporal coupled inference network, the temporal convolution module adopts a dilated convolution structure and sets a normalization layer, and the initial association weight allocation of the graph convolution module is based on the initial weight of the device combination dimension.

9. A smart construction site power load prediction system, wherein the system is applied to the smart construction site power load prediction method as described in claim 1, characterized in that, The system includes: The tag building module is used to establish an electricity tag system associated with construction conditions and assign initial weights to it, so as to structurally represent the electricity characteristics of the construction scenario. The feature acquisition module is used to acquire instantaneous power consumption feature information and current construction condition information of the work surface, and perform tag matching on the current construction condition based on the power consumption tag system, and call the initial weight corresponding to the condition. The feature standardization module is used to perform adaptive noise reduction and alignment processing on the instantaneous power consumption feature information based on preset benchmark power consumption feature information and combined with the initial weight of the current operating condition, so as to extract standardized power consumption feature information. The sensitivity generation module is used to generate sensitivity control parameters based on the dynamic offset of the standardized electricity consumption characteristic information relative to the reference characteristic, combined with the initial weight of the operating condition dimension label. The load prediction module is used to dynamically configure the weight parameters of the spatiotemporal coupling simulation network based on the sensitivity control parameters, and at the same time use the initial weights corresponding to the electricity consumption tag system as the initial weight constraints of the network, and output the construction electricity load prediction results using the configured spatiotemporal coupling simulation network.

Images