Water conservancy construction risk dynamic analysis and identification system based on multi-source data

Abstract

The application relates to the technical field of water conservancy construction management, and particularly discloses a water conservancy construction risk dynamic analysis and identification system based on multi-source data, which synchronously collects and standardizes multi-source dynamic data of a construction site to form a standardized construction data sequence; state transition abnormal characteristic values and collaborative correlation deviation characteristic values are calculated and generated from two dimensions of internal continuity of construction activities and collaborative evolution of multi-source data; the two characteristic values are dynamically fused, and a judgment standard containing a risk level and an evolution mode is generated through pattern matching; and according to the standard and in combination with real-time working condition constraints, differentiated regulation instructions are generated and verified to drive on-site dynamic adjustment; the application realizes multi-dimensional and forward-looking accurate perception of construction risks, forms a closed-loop intelligent management and control, and improves the comprehensiveness, accuracy of risk identification and the timeliness and effectiveness of management and control.

Description

Technical Field

[0001] This invention relates to the field of water conservancy project construction management technology, specifically to a dynamic analysis and identification system for water conservancy project construction risks based on multi-source data. Background Technology

[0002] Water conservancy project construction typically faces severe challenges such as variable geological conditions, complex hydrological and meteorological environments, interactions between large structures, and multi-disciplinary, multi-faceted operations. Typical risks during construction include progressive slope instability, temperature and shrinkage cracks in large-volume concrete structures, seepage stability of high cofferdams, and the safety of heavy machinery operating in complex environments. These risks are characterized by their hidden nature, coupled nature, and dynamic evolution. These risks not only directly threaten the safety of personnel and the project itself, but the resulting quality defects, delays, and cost overruns can also have a profound impact on the overall project benefits. Therefore, shifting from static, post-hoc experience-based judgment of construction risks to dynamic, pre-emptive, and in-process accurate identification and early warning has become a key requirement for improving the intelligent level of modern water conservancy project construction management and ensuring the safety of the entire project lifecycle.

[0003] Existing technologies have the following shortcomings: traditional methods cannot deeply integrate and analyze massive, multi-source, and heterogeneous dynamic data during construction, making it difficult to quantify and characterize the two types of hidden risks: the inherent disruption of continuity in construction activities and the imbalance of multi-source collaborative relationships. Furthermore, they lack the closed-loop capability to dynamically generate precise and executable control strategies based on real-time risk evolution patterns. As a result, risk identification is incomplete, early warnings are not timely, and intervention measures are not targeted enough. They cannot effectively adapt to the dynamic changes in complex construction environments, thus hindering the fundamental improvement of engineering safety management from passive response to proactive prevention. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic analysis and identification system for construction risks in water conservancy projects based on multi-source data, so as to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions: A dynamic analysis and identification system for construction risks in water conservancy projects based on multi-source data includes: The multi-source data synchronous acquisition and standardization module synchronously acquires multi-source dynamic data during the construction process of water conservancy projects. The multi-source dynamic data includes construction process data that characterizes construction activities and resource status. The acquired multi-source dynamic data is then standardized to form a standardized construction data sequence with spatiotemporal alignment. The construction continuity state analysis module, based on standardized construction data sequences, constructs a continuous state vector of the construction process from the perspective of the inherent continuity of construction activities; based on the continuous state vector of the construction process, it calculates the state transition anomaly degree of each time window in the standardized construction data sequence and generates state transition anomaly feature values. The multi-source data collaborative analysis module, based on standardized construction data sequences, constructs a collaborative state matrix of the construction process from the perspective of collaborative evolution among multi-source data; based on the collaborative state matrix of the construction process, it calculates the dynamic correlation deviation of different data sources in the time dimension and generates collaborative correlation deviation feature values. The dynamic risk comprehensive analysis module integrates state transition anomaly characteristic values ​​and collaborative correlation deviation characteristic values ​​in time and space, inputs them into the comprehensive analysis process, and generates dynamic construction risk judgment criteria through weighted aggregation and pattern matching. The judgment criteria include risk level and risk evolution mode. The risk warning and control execution module generates differentiated warning information or control instructions in real time based on the risk level and risk evolution mode in the construction risk assessment criteria, and dynamically adjusts the work process or resource allocation at the construction site.

[0006] As a further aspect of the present invention: the formation of a standardized construction data sequence with spatiotemporal alignment specifically includes: Based on the original timestamps and collection frequencies of each data source, a unified benchmark timeline is determined, and alignment time windows are dynamically divided according to the construction phase. Within each alignment time window, time position interpolation is performed based on the change characteristics of various physical quantities of data at different acquisition times, and uniformly mapped to a specified time point on the reference time axis. Based on the spatial location information of the mapped data at the construction site, coordinate transformation is performed to unify it under the same engineering coordinate system, and the data format and dimensions are normalized to generate a standardized construction data sequence.

[0007] As a further aspect of the present invention: the generation of state transition anomaly feature values ​​specifically includes: For each time window, extract the ordered sequence of continuous state vectors in the construction process within the window; based on the ordered sequence, calculate the actual transition change between adjacent state vectors; Obtain historical normal construction modes that match the current construction conditions, determine the expected transfer benchmark of the state vector from them, compare the actual transfer change with the expected transfer benchmark, and obtain the original deviation measure. The original deviation metrics of multiple consecutive time windows are aggregated by sliding window and the probability distribution is estimated. By calculating the degree of deviation between the current window aggregated value and the tail fraction of the historical distribution, the final state transition anomaly feature value is generated.

[0008] As a further aspect of the present invention: the calculation of the actual transition change between adjacent state vectors based on the ordered sequence specifically includes: Extract the corresponding state components from the previous state vector and the next state vector respectively, and obtain the statistical stability of each state component in the historical normal construction process. Weights are dynamically assigned to each state component based on statistical stability. Based on dynamically allocated weights, the absolute values ​​of the differences between corresponding components of the state vectors before and after are summed in a weighted manner, and the summation result is used as the actual transition change.

[0009] As a further aspect of the present invention: the construction of the collaborative state matrix for the construction process specifically includes: The values ​​from different data sources at the same time in the standardized construction data sequence are combined to form a time-coordinated state vector; Based on a preset time window, the collaborative state vectors of multiple consecutive time points within the time window are arranged in rows. Perform pairwise operations on the row vectors of the arrangement, and fill the intermediate product of the symmetric structure with the scalar output of the operation result of each pair of row vectors. The intermediate product is the collaborative state matrix of the construction process.

[0010] As a further aspect of the present invention: the generation of collaborative correlation deviation feature values ​​specifically includes: Obtain the historical construction process coordination state matrix corresponding to the current construction stage, and use the historical construction process coordination state matrix as the reference matrix; The quotient of corresponding elements in the construction process coordination state matrix of the current time window and the baseline matrix is ​​calculated, and the structure of the calculated matrix is ​​analyzed to identify the regions where the element values ​​exceed the preset stable range. Aggregate the deviation intensity of all identified abnormal regions within the current time window, and after standardizing this aggregated value, output it as the co-correlation deviation feature value of the current window.

[0011] As a further aspect of the present invention: the generation of dynamic construction risk assessment criteria specifically includes: Based on the time series change rates of state transition anomaly eigenvalues ​​and co-association deviation eigenvalues, the fusion ratio of state transition anomaly eigenvalues ​​and co-association deviation eigenvalues ​​is determined; The two feature values ​​of the current time window are integrated according to the fusion ratio to form a comprehensive risk vector. The comprehensive risk vector is then matched with the risk evolution sequence recorded in the preset historical risk pattern library for morphological similarity. Based on the historical risk pattern with the highest similarity in the matching results, the current risk evolution pattern is directly mapped to determine the current risk evolution pattern; and based on the magnitude of the comprehensive risk vector, a joint judgment is made in conjunction with the determined risk evolution pattern to output the current risk level.

[0012] As a further aspect of the present invention: the morphological similarity matching between the comprehensive risk vector and the risk evolution sequences recorded in the preset historical risk pattern library specifically includes: Multi-scale decomposition was performed on the time series and risk evolution sequence of the comprehensive risk vector to obtain the components of each sequence at different time granularities; At different time granularities, the components of the comprehensive risk vector and the components of the risk evolution sequence are compared in segments, and the morphological distance is calculated segment by segment. We aggregate the weighted morphological distances calculated at all time granularities, and determine the historical risk evolution sequence with the highest matching degree through the distance minimization principle, which is then used as the result of morphological similarity matching.

[0013] As a further aspect of the present invention: the dynamic adjustment of the work process or resource allocation at the construction site specifically includes: Based on the risk evolution model, the corresponding basic instruction set is selected from the preset early warning and control strategy matrix. The control strategy matrix defines the mapping relationship between different risk evolution models and preliminary control actions. Based on the risk level, the intensity of each action parameter in the basic instruction set is calibrated to generate preliminary control instructions with specific execution parameters; The initial control instructions are combined with key construction process data acquired in real time. The instruction parameters are then fine-tuned and verified through decision optimization rules, and executable differentiated control instructions are output and issued.

[0014] As a further aspect of the present invention: the method of combining risk levels to perform intensity calibration on the action parameters of each action in the basic instruction set, and generating preliminary control instructions with specific execution parameters, specifically includes: Based on the risk level, the preset baseline mapping is queried to obtain the baseline intensity value of each action parameter; Real-time acquisition of status data of key constraints related to various actions during the current construction process; Based on the state data of key constraints, the baseline strength value is dynamically corrected, and the corrected strength value is assigned to the corresponding action parameters to generate preliminary control commands.

[0015] The beneficial effects of this invention are: (1) By synchronously collecting and standardizing multi-source construction data, and extracting state transition anomaly features and collaborative correlation deviation features from the two dimensions of the inherent continuity of construction activities and multi-source collaborative evolution, a multi-angle and in-depth perception of construction risks is achieved. This method can effectively identify hidden risks caused by the coupling of temporal misalignment and collaborative failure, improve the comprehensiveness and accuracy of risk identification, and reduce the possibility of safety accidents caused by missed or misjudged risks.

[0016] (2) Based on the dynamically fused risk level and risk evolution model, and combined with real-time construction constraints, the command optimization and verification were carried out, realizing the accurate generation and rapid execution of risk warning and control strategies. This closed-loop management mechanism ensures timely response from risk perception to on-site intervention, enabling the construction process to make adaptive adjustments in the early stages of risk, thereby effectively curbing risk deterioration, reducing project rework, resource waste and unplanned shutdowns, and improving the overall construction safety management level and project economic benefits. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0019] 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, and 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.

[0020] Please see Figure 1 As shown, this invention is a dynamic analysis and identification system for construction risks in water conservancy projects based on multi-source data, comprising: The multi-source data synchronous acquisition and standardization module synchronously acquires multi-source dynamic data during the construction process of water conservancy projects. The multi-source dynamic data includes construction process data that characterizes construction activities and resource status. The acquired multi-source dynamic data is then standardized to form a standardized construction data sequence with spatiotemporal alignment. The construction continuity state analysis module, based on standardized construction data sequences, constructs a continuous state vector of the construction process from the perspective of the inherent continuity of construction activities; based on the continuous state vector of the construction process, it calculates the state transition anomaly degree of each time window in the standardized construction data sequence and generates state transition anomaly feature values. The multi-source data collaborative analysis module, based on standardized construction data sequences, constructs a collaborative state matrix of the construction process from the perspective of collaborative evolution among multi-source data; based on the collaborative state matrix of the construction process, it calculates the dynamic correlation deviation of different data sources in the time dimension and generates collaborative correlation deviation feature values. The dynamic risk comprehensive analysis module integrates state transition anomaly characteristic values ​​and collaborative correlation deviation characteristic values ​​in time and space, inputs them into the comprehensive analysis process, and generates dynamic construction risk judgment criteria through weighted aggregation and pattern matching. The judgment criteria include risk level and risk evolution mode. The risk warning and control execution module generates differentiated warning information or control instructions in real time based on the risk level and risk evolution mode in the construction risk assessment criteria, and dynamically adjusts the work process or resource allocation at the construction site.

[0021] In the multi-source data synchronous acquisition and standardization module, multi-source dynamic data during the construction process of water conservancy projects are collected synchronously. This multi-source dynamic data includes construction process data characterizing construction activities and resource status. The collected multi-source dynamic data is then standardized to form a spatiotemporally aligned standardized construction data sequence, specifically including: Data acquisition is accomplished through various sensing and recording devices deployed at the construction site. For construction machinery, GPS receivers and tilt sensors are installed at key locations to collect real-time 3D coordinates, travel speed, and the angle between the robotic arm and the horizontal plane. This data characterizes the machinery's spatial position and attitude. In the concrete pouring area, distributed temperature sensors and strain gauges are pre-embedded to collect data on changes in the internal temperature and stress fields of the concrete at set time intervals (e.g., every 5 minutes). Environmental data is collected through a weather station on-site, which continuously records precipitation, wind speed, ambient temperature, and relative humidity. Furthermore, high-definition cameras are deployed at key work areas to record the operations of construction workers and the movement trajectories of machinery in continuous image frames. All raw data collected by these devices is appended with timestamps generated by the devices themselves and spatial location identifiers, and transmitted in real-time to the data processing center via an industrial IoT gateway.

[0022] The standardization process begins with establishing a unified reference timeline and dividing it into alignment time windows. The reference timeline is based on Coordinated Universal Time (UTC), and its time resolution is determined by the fastest acquisition frequency. The alignment time windows are not fixed in duration but are dynamically determined based on the construction logs: a new time window is initiated when the construction log records the start of a major process (e.g., "concrete pouring"); this window lasts until the process is completed or paused. Each window contains all data points from all data sources whose timestamps fall within that time period.

[0023] The second step involves time position interpolation within each defined alignment time window. This process is for data that is within the window but not sampled directly at the precise moment on the reference time axis. Different interpolation strategies are employed based on the changing characteristics of the physical quantities in the data: for rapidly changing data such as vibration and displacement, linear interpolation is used to calculate the estimated value at the reference moment, i.e., a linear calculation is performed based on the values ​​of two actual sampling points before and after the data, proportional to time; for data with continuous and smooth changes such as temperature and stress, cubic spline interpolation is used, fitting a smooth curve based on multiple adjacent actual sampling points, and reading the corresponding value at the reference moment from this curve. After this step, all data is mapped to a unified set of equally spaced reference time points.

[0024] The third step is to perform spatial coordinate transformation and data normalization. Coordinate transformation applies to all data with spatial location attributes. First, a global engineering coordinate system is defined based on the overall engineering control network. The original local coordinates of each data point (such as the coordinate system relative to a certain machine) are calculated to obtain its coordinate value in the global engineering coordinate system through a transformation relationship composed of rotation and translation parameters. This transformation relationship is determined through a calibration measurement after equipment installation. Data normalization includes format unification and dimension unification. Format unification refers to converting all numerical data into double-precision floating-point format. Dimension unification adopts the maximum-minimum value normalization method: for each type of data (such as concrete temperature), its possible maximum and minimum values ​​are statistically analyzed from historical normal construction data; when processing the current data, the original value of the data is subtracted from the statistically obtained minimum value, and then divided by the difference between the statistically obtained maximum and minimum values, finally obtaining a dimensionless value between 0 and 1. After completing all the above steps, a standardized construction data sequence with a regular structure is output. Each data entry in this sequence contains a uniform time, global spatial coordinates, and dimensionless physical quantity values.

[0025] In the construction continuity state analysis module, based on standardized construction data sequences, a continuous state vector of the construction process is constructed from the perspective of the inherent continuity of construction activities. Based on the continuous state vector of the construction process, the state transition anomaly degree of each time window in the standardized construction data sequence is calculated, and state transition anomaly feature values ​​are generated, specifically including: The method for constructing the continuous state vector of the construction process is as follows: From the standardized construction data sequence, a set of core physical quantities that can directly and critically characterize the continuous evolution of construction activities are selected. These physical quantities typically include: the average travel speed of the main construction machinery within a reference time interval, the rate of change of concrete temperature at key measuring points within a reference time interval, and the average distribution density of on-site workers within a reference time interval. At the end of each reference time interval, the latest values ​​of each of the selected core physical quantities are arranged in a fixed order to form a multidimensional array, which is defined as the continuous state vector of the construction process at that moment. The fixed position of each physical quantity in the array is called a state component. Therefore, a state vector containing three core physical quantities is an ordered array with three state components.

[0026] For a given analysis time window, all continuous state vectors of the construction process arranged chronologically within that window are extracted to form the ordered sequence. The specific process for calculating the actual transition changes between adjacent state vectors is as follows: First, for a pair of state vectors (i.e., the vector at the previous time step and the vector at the next time step), the state component values ​​at the same position are extracted. Second, for each state component, its statistical stability during historical normal construction processes needs to be calculated. This stability is quantified by analyzing the fluctuations of the component in long-term data under historical normal conditions. The specific calculation method is: calculate the standard deviation of all values ​​of the component in the historical data, then divide it by the absolute value of the average of all values ​​of the component in the historical data; the quotient is the statistical stability measure of the component; the smaller this value, the more stable the component is historically. Subsequently, weights are dynamically assigned to each state component based on this statistical stability. The allocation principle is: the reciprocal of the statistical stability is used as the basis for the weights; that is, the worse the historical stability of a component (the larger its statistical stability value), the smaller its reciprocal, and the lower the weight assigned accordingly. Then, the basic weights of all components are normalized so that their sum is 1, thus obtaining the final dynamic weight for each state component. Finally, the actual transition change of the state vector pair is calculated: for each pair of corresponding state components, the absolute value of the difference between the component value at the next time step and the component value at the previous time step is calculated; this absolute value is multiplied by the previously calculated dynamic weight of that component; the weighted absolute values ​​of all state components are summed, and the sum is the actual transition change between these two adjacent state vectors.

[0027] The next step is to obtain the expected transition benchmark and calculate the original deviation metric. The historical normal construction mode is a pre-established database that stores the standard range of changes in state vectors between adjacent moments under various typical construction conditions (such as "excavation," "main structure pouring," and "curing period"). By comparing with the current construction log, the current construction condition is determined, and the corresponding historical data is retrieved from this database. The expected transition benchmark is determined by statistically analyzing the median of changes in all adjacent state vectors in the historical data for that construction condition. The actual transition change calculated in the previous step is compared with this expected transition benchmark (median). Specifically, the expected transition benchmark is subtracted from the actual transition change, and the absolute value is taken. The result is the original deviation metric for this state transition.

[0028] The final step in generating the final state transition anomaly feature value is to perform sliding window aggregation and probability distribution estimation. An aggregation window containing multiple consecutive time windows is set, for example, five consecutive windows. The average of all original deviation measures within the current aggregation window is calculated to obtain the current window's aggregation value. Simultaneously, based on historical normal construction data, the probability distribution followed by all historical aggregation values ​​under the same aggregation window length is calculated. This distribution is obtained using kernel density estimation. This distribution allows the determination of a high percentile threshold, such as the 95th percentile. The final state transition anomaly feature value is generated by the following calculation: the difference between the current window aggregation value and the historical distribution threshold (95th percentile) is calculated, and then this difference is divided by the historical distribution threshold itself. The resulting ratio is the final state transition anomaly feature value. If this ratio is zero or negative, it indicates that the current state transition is within the normal fluctuation range; if the ratio is positive, its magnitude directly represents the degree of anomaly.

[0029] In the multi-source data collaborative analysis module, based on standardized construction data sequences, a collaborative state matrix of the construction process is constructed from the perspective of collaborative evolution among multi-source data. Based on the collaborative state matrix of the construction process, the dynamic correlation deviation of different data sources in the time dimension is calculated, and collaborative correlation deviation feature values ​​are generated, specifically including: The construction of the collaborative state matrix for the construction process begins with the formation of the time-series collaborative state vector. From the standardized construction data sequence, multiple data sources with potential collaborative relationships in a physical sense are selected. These data sources include, but are not limited to: the three-dimensional vibration acceleration of major construction machinery, the internal temperature of concrete in key pouring sections, the relative humidity of the construction area, and the infrared thermal imaging density of personnel on the work surface. At each unified reference time point, the standardized dimensionless values ​​of all the selected data sources at that time are arranged in a predefined fixed order to form a combined array, which is defined as the collaborative state vector for that time. For example, if four types of data are selected, the vector is an ordered list containing four numerical components.

[0030] Next, a matrix is ​​constructed based on a preset time series analysis window. The length of this time series window is typically set to cover a relatively complete construction microcycle, such as 60 consecutive reference time points. Within this window, the cooperative state vector of each time point arranged in chronological order is used as a row of the matrix. If the time series window contains 60 time points and the cooperative state vector has 4 components, a temporary matrix with 60 rows and 4 columns will be obtained.

[0031] Finally, the final construction process coordination state matrix is ​​generated by analyzing the relationships between the row vectors in the temporary matrix. Specifically, the similarity between each row vector in the temporary matrix and all other row vectors is calculated. This similarity is obtained by calculating the cosine similarity between two row vectors. The calculation process is as follows: first, multiply the corresponding components of the two row vectors, then add all the products to obtain the numerator; next, calculate the square root of the sum of the squares of each component of the two row vectors, multiply these two square root values ​​to obtain the denominator; finally, divide the numerator by the denominator. The result is a scalar value between -1 and +1, where the value closer to +1 indicates that the construction coordination states represented by the two vectors at different times are more similar. This calculation process is applied to each pair of row vectors in the temporary matrix, and the calculation result of each pair of vectors (e.g., row i and row j) is filled into the position of row i and column j of a new square matrix. Since cosine similarity is symmetric (the result of row i and row j is equal to the result of row j and row i), the generated new matrix is ​​a symmetric matrix. This symmetric matrix is ​​the final construction process collaborative state matrix, which describes the similarity pattern of the multi-source collaborative state of construction at different times within the time window.

[0032] The first step in generating collaborative correlation deviation eigenvalues ​​is to obtain a baseline matrix. This baseline matrix is ​​obtained by analyzing a large amount of historical normal construction data. For different major construction stages (such as "earthwork excavation", "reinforcement binding", and "concrete pouring"), collaborative state matrices of the construction process calculated under the same time window length are extracted from historical data. The average value of the values ​​at each same row and column position of all historical matrices is then calculated to form a baseline matrix representing the normal collaborative state of that stage.

[0033] Subsequently, quotient calculation and structural analysis are performed. The collaborative state matrix of the construction process calculated in real time is compared with the baseline matrix of the same stage, and the corresponding position elements are compared. The comparison method is as follows: the value at each position in the current matrix is ​​divided by the value at the corresponding position in the baseline matrix to obtain a quotient matrix. The preset stable interval is defined as: the baseline value multiplied by 0.95 as the lower limit and the baseline value multiplied by 1.05 as the upper limit, forming a numerical range. The quotient matrix is ​​scanned to identify the positions of all matrix elements with a quotient value less than 0.95 or greater than 1.05. The set of these positions is the "abnormal region". The deviation intensity of each abnormal region is defined as the absolute value of the difference between the quotient value at that position and the value 1.

[0034] Finally, the deviation intensities are aggregated and the feature values ​​are output. The aggregation process involves summing the deviation intensity values ​​of all abnormal regions identified in the current time window. The standardization process involves dividing this sum by the theoretically maximum possible sum of deviation intensities. The theoretical maximum sum is estimated by assuming that the quotient values ​​at all positions in the current matrix reach the boundaries of a preset stable interval (i.e., 0.95 or 1.05) and calculating the sum of their absolute deviations from 1. Finally, the standardized ratio is output as the collaborative correlation deviation feature value for the current time window. This value is between 0 and 1; a larger value indicates a higher overall deviation between the current multi-source collaborative state of the construction process and the historical normal pattern.

[0035] In the dynamic risk comprehensive analysis module, state transition anomaly characteristic values ​​and collaborative correlation deviation characteristic values ​​are spatiotemporally fused and input into the comprehensive analysis process. Through weighted aggregation and pattern matching, dynamic construction risk assessment criteria are generated. The assessment criteria include risk level and risk evolution mode, specifically including: The process of generating dynamic construction risk assessment criteria begins by determining the fusion ratio between state transition anomaly characteristic values ​​and co-correlation deviation characteristic values. This fusion ratio is not a fixed value, but is dynamically adjusted based on the change activity exhibited by both in recent time series. Specifically, the standard deviations of the first-order differences (i.e., the difference between the value at the next time step and the value at the previous time step) of the state transition anomaly characteristic values ​​and co-correlation deviation characteristic values ​​over the past ten consecutive time windows are calculated, and these are used as quantitative indicators of their respective rates of change. Let the standard deviation of the rate of change of the state transition anomaly characteristic value be denoted as... The standard deviation of the rate of change of the co-correlation eigenvalue is denoted as . The fusion weights of the state transition anomaly features. Determined by the following mathematical formula: ; Correspondingly, the fusion weights of collaborative associations deviate from the eigenvalues That is The logic behind this calculation is to assign higher weight to features that have shown more significant recent volatility, as these may indicate the more dominant risk drivers at present. and The calculations are all based on the baseline range obtained from the statistical data under historical normal operating conditions, and normalization processing is performed in the actual calculation to ensure the rationality of the weight allocation.

[0036] Next, based on the aforementioned dynamic weights, the two feature values ​​of the current time window are integrated to form a comprehensive risk vector. Let the anomaly feature value of the state transition in the current time window be... The deviation eigenvalue of the collaborative association is Then the comprehensive risk vector In a two-dimensional feature space, it is represented as This vector not only contains comprehensive information on the risk level, but its direction also implies whether the risk primarily stems from disruptions in temporal continuity or a tendency towards multi-source dysregulation. To facilitate comparison with historical patterns, this vector is extended into a short-time series: taking the current time window and the four immediately preceding historical windows, the comprehensive risk vector for each window is calculated and arranged chronologically, forming a time series containing comprehensive risk vectors at five consecutive moments, denoted as [vector name missing]. This sequence represents the current comprehensive risk evolution trajectory to be matched.

[0037] The historical risk pattern database was constructed through retrospective analysis of a large number of historical construction cases (including normal operating conditions and various risk event cases). For each identified typical risk evolution pattern (such as "gradual accumulation", "sudden impact", and "oscillating repetitive"), the comprehensive risk vector time series during the risk manifestation period was extracted from the case. After alignment and normalization, this time series was stored in the database as a standard template sequence for that pattern. The length of each template sequence is consistent with the current sequence to be matched.

[0038] Subsequently, the current comprehensive risk vector time series is matched with various template sequences in the historical pattern library for morphological similarity. This process begins with multi-scale decomposition. Discrete wavelet transform is used to process the current sequence and each template sequence, decomposing them into three layers of approximate and detail components at different time granularities. The first layer reflects short-term fluctuations (approximately 1-2 window periods), the second layer reflects medium-term trends (approximately 3-4 window periods), and the third layer reflects the long-term profile (the entire sequence length). Through multi-scale decomposition, the morphological features of the sequence can be captured at different time resolutions, avoiding interference from local noise or global offset at a single scale on the matching results.

[0039] Next, segmented comparisons are performed at different time granularities, and the morphological distance is calculated segment by segment. Taking the third layer (long-term contour) component as an example, the component of the current sequence and the corresponding component of a template sequence are normalized to the [0,1] interval and then divided into three equal segments. For each segment, the sum of the Euclidean distances between the corresponding data points of the two sequences within that segment is calculated. Simultaneously, a segment saliency coefficient is introduced, which is determined based on the variance of the variation in the data points within that segment. The larger the variance, the more significant the morphological features of the segment, and the higher the weight assigned in the comparison. Assuming that for the third... The first layer component The Euclidean distance between this segment of the current sequence and the template sequence is... The significance coefficient of this segment is Then the weighted morphological distance at this granularity The following second mathematical formula is given: ; in, Indicates the first The total number of segments in the layer component division. For the first and second layer components, a similar segmented weighting method is used to calculate their respective weighted morphological distances. and Paragraph significance coefficient The specific calculation method is as follows: first, calculate the variance of all data points in the paragraph, and then divide the variance by the maximum value of the variance of all paragraphs in the layer component to obtain a relative significance index between 0 and 1.

[0040] Then, the calculation results at all time granularities are aggregated. The final overall dissimilarity between the current sequence and a template sequence is then calculated. It is obtained by synthesizing the weighted distances of the three scales, and the synthesis formula is as follows: ,in For the preset scale weights, , and This indicates the weighted pattern distance, assigning higher weight to the second-layer components that reflect the medium-term trend (e.g., setting...). This is because it best reflects the evolutionary trend of risk. Traverse all template sequences in the historical pattern library and calculate the overall dissimilarity with the current sequence for each.

[0041] Finally, the matching results are determined using the distance minimization principle. The matching results are selected based on the overall dissimilarity. The risk evolution pattern corresponding to the historical template sequence with the smallest value is directly mapped to determine the evolution pattern to which the current risk belongs.

[0042] After determining the risk evolution pattern, the final risk level is jointly determined. The risk level is primarily based on the magnitude of the comprehensive risk vector within the current time window. The determination is made by taking the square root of the sum of the squares of the two components of the vector. Simultaneously, historical case statistics of matched risk evolution patterns are referenced. For example, if the matching pattern is "gradual accumulation," its risk level threshold is usually set more conservatively than "sudden impact," because the former indicates that the risk is continuously accumulating. In the specific output, three risk levels are set: "low," "medium," and "high." First, based on the current construction stage, two modulus thresholds are preset for each risk evolution pattern. and .like If it is, then it is judged as "low" risk; if If it is, it is judged as "medium" risk; if If a risk level is not met, it is classified as "high" risk. These thresholds are derived from the statistical quantiles of the comprehensive risk vector magnitude of a large number of cases under corresponding historical patterns (e.g., the "low-medium" threshold corresponds to the 70th quantile, and the "medium-high" threshold corresponds to the 90th quantile), and can be fine-tuned by domain experts according to the specific characteristics of the current project. Finally, a binary tuple containing the specific risk level (e.g., "medium") and the risk evolution pattern (e.g., "gradually accumulating") is output as a dynamic construction risk assessment standard, providing a precise basis for subsequent early warning and control.

[0043] In the risk warning and control execution module, differentiated warning information or control instructions are generated in real time based on the risk level and risk evolution mode in the construction risk assessment criteria. The module also dynamically adjusts the work processes or resource allocation at the construction site, specifically including: The early warning and control strategy matrix is ​​a predefined data structure. Its rows correspond to different risk evolution modes (such as "gradual accumulation," "sudden impact," and "oscillating repetitive"), and its columns correspond to different categories of controllable construction elements (such as "mechanical operation intensity," "monitoring frequency," "personnel evacuation range," and "auxiliary cooling measures"). Each cell in the matrix contains a set of basic control action descriptions. For example, for the "gradual accumulation" risk and the "monitoring frequency" element, the basic action might be described as "increasing the frequency of sensor data acquisition and transmission in the relevant area." This matrix is ​​constructed based on experience summaries of historical risk event handling, construction safety regulations, and domain expert knowledge. Once the current risk evolution mode is determined, all basic control action descriptions corresponding to the mode row are selected from the matrix to form a preliminary set of instructions to be executed.

[0044] To generate preliminary control instructions with specific execution parameters, the intensity of basic actions must first be calibrated based on the risk level. A pre-defined level-intensity benchmark mapping table is used, which defines a benchmark intensity value for each risk level (low, medium, high) and each type of control action. This value is a dimensionless percentage or multiplier, representing the adjustment range relative to the standard execution intensity under normal operating conditions. For example, for the action of "increasing monitoring frequency" under the "medium" risk level, the benchmark intensity value might be set to 150%, meaning increasing the frequency to 1.5 times the standard value. This mapping table is obtained by analyzing historical data on the minimum intervention intensity required for effective risk control at different risk levels.

[0045] To ensure that control commands align with real-time on-site conditions, the baseline strength value needs dynamic correction. This process requires real-time acquisition of status data for key constraints directly related to each control action. This data originates directly from real-time monitoring during construction. For example, if the control action involves adjusting the concrete cooling water flow rate, the key constraint status data includes the current cooling water inlet pressure and total available flow rate; if it involves adjusting the intensity of mechanical operations, the status data may include the machine's current fuel reserves, operator continuous working hours, and lighting conditions at the work surface. The correction method is as follows: For each control action, the ideal resource requirement or operating condition corresponding to its baseline strength value is compared with the actual status of the key constraints acquired in real-time. The specific correction coefficient is obtained by calculating the percentage of the actual status value relative to the ideal status value. If the actual status is better than the ideal status, the correction coefficient is greater than 1, and the control intensity can be appropriately increased; if the actual status is worse than the ideal status, the correction coefficient is less than 1, and the control intensity needs to be appropriately reduced to avoid secondary problems. The baseline strength value is multiplied by this correction coefficient to obtain the specific execution parameter strength value adapted to the operating conditions. This value replaces the abstract strength identifier in the basic action description, generating the initial control command.

[0046] Finally, the initial control commands are fine-tuned and verified using decision optimization rules. These rules are based on a set of constraints and optimization objectives. Constraints include, but are not limited to: the total resource consumption after adjusting all control commands must not exceed the upper limit of available resources on site in real time; and there must be no logical or safety conflicts between multiple commands. The optimization objective is to maximize the expected reduction in overall risk. The fine-tuning process is an iterative verification process: First, the initial control commands and their parameters are substituted into a simplified construction process impact simulation model. This model, based on statistical relationships established from historical data, quickly predicts the possible changing trends of key risk indicators (such as stress and deformation rate) after the commands are executed. Next, it is checked whether the simulation results meet the objective of "expected risk reduction" and all constraints. If not, the parameters of the relevant commands are slightly adjusted according to a preset priority rule (such as "safety constraints take precedence over resource constraints"), and the simulation verification is repeated until a set of command parameter combinations that satisfies all constraints and is closest to the optimization objective is found. Once the verified combination of command parameters is locked, the output becomes the final executable differentiated control command, which is immediately sent through the construction management network to the corresponding mechanical controller, personnel handheld terminal or monitoring system configuration interface to complete the dynamic adjustment of the work process or resource allocation.

[0047] The working principle of this invention is as follows: Multi-source dynamic data from the construction site are collected synchronously and standardized to form a spatiotemporally aligned standardized construction data sequence. From the perspective of the inherent continuity of construction activities, a continuous state vector of the construction process is constructed and its state transition anomaly degree is calculated, generating state transition anomaly feature values. Simultaneously, from the perspective of multi-source data collaborative evolution, a collaborative state matrix of the construction process is constructed and its dynamic correlation deviation degree is calculated, generating collaborative correlation deviation feature values. These two feature values ​​are spatiotemporally fused, and through weighted aggregation and pattern matching, a dynamic construction risk judgment standard containing risk level and risk evolution pattern is generated. Based on this judgment standard, differentiated early warning information or control instructions are generated in real time to dynamically adjust the work process or resource allocation at the construction site, thereby achieving dynamic, accurate identification and closed-loop management of construction risks.

[0048] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A dynamic analysis and identification system for construction risks in water conservancy projects based on multi-source data, characterized in that, include: The multi-source data synchronous acquisition and standardization module synchronously acquires multi-source dynamic data during the construction process of water conservancy projects. The multi-source dynamic data includes construction process data that characterizes construction activities and resource status. The acquired multi-source dynamic data is then standardized to form a standardized construction data sequence with spatiotemporal alignment. The construction continuity state analysis module, based on standardized construction data sequences, constructs a continuous state vector of the construction process from the perspective of the inherent continuity of construction activities. Based on this continuous state vector, it calculates the state transition anomaly degree for each time window in the standardized construction data sequence, generating state transition anomaly feature values. Specifically, generating these feature values ​​includes: For each time window, extract the ordered sequence of continuous state vectors in the construction process within the window; based on the ordered sequence, calculate the actual transition change between adjacent state vectors; Obtain historical normal construction modes that match the current construction conditions, determine the expected transfer benchmark of the state vector from them, compare the actual transfer change with the expected transfer benchmark, and obtain the original deviation measure. The original deviation measures of multiple consecutive time windows are aggregated by sliding window and the probability distribution is estimated. By calculating the degree of deviation between the current window aggregated value and the tail fraction of the historical distribution, the final state transition anomaly feature value is generated. The multi-source data collaborative analysis module, based on standardized construction data sequences, constructs a collaborative state matrix of the construction process from the perspective of collaborative evolution among multi-source data. Based on this collaborative state matrix, it calculates the dynamic correlation deviation of different data sources over time, generating collaborative correlation deviation feature values. The construction of the collaborative state matrix specifically includes: The values ​​from different data sources at the same time in the standardized construction data sequence are combined to form a time-coordinated state vector; Based on a preset time window, the collaborative state vectors of multiple consecutive time points within the time window are arranged in rows. Perform pairwise operations on the row vectors of the arrangement, and fill the intermediate product of the symmetrical structure with the scalar output of the operation result of each pair of row vectors. The intermediate product is the collaborative state matrix of the construction process. The generation of collaborative association deviation feature values ​​specifically includes: Obtain the historical construction process coordination state matrix corresponding to the current construction stage, and use the historical construction process coordination state matrix as the reference matrix; The quotient of corresponding elements in the construction process coordination state matrix of the current time window and the baseline matrix is ​​calculated, and the structure of the calculated matrix is ​​analyzed to identify the regions where the element values ​​exceed the preset stable range. Aggregate the deviation intensity of all identified abnormal regions within the current time window, and after standardizing this aggregated value, output it as the collaborative correlation deviation feature value of the current window; The dynamic risk comprehensive analysis module integrates state transition anomaly characteristic values ​​and collaborative correlation deviation characteristic values ​​in time and space, inputs them into the comprehensive analysis process, and generates dynamic construction risk judgment criteria through weighted aggregation and pattern matching. The judgment criteria include risk level and risk evolution mode. The risk warning and control execution module generates differentiated warning information or control instructions in real time based on the risk level and risk evolution mode in the construction risk assessment criteria, and dynamically adjusts the work process or resource allocation at the construction site.

2. The system for dynamic analysis and identification of construction risks in water conservancy projects based on multi-source data as described in claim 1, characterized in that, The standardized construction data sequence that forms a spatiotemporally aligned sequence specifically includes: Based on the original timestamps and collection frequency of each data source, a unified benchmark timeline is determined, and the alignment time window is dynamically divided according to the construction stage. Within each alignment time window, time position interpolation is performed based on the change characteristics of various physical quantities of data at different acquisition times, and uniformly mapped to a specified time point on the reference time axis. Based on the spatial location information of the mapped data at the construction site, coordinate transformation is performed to unify it under the same engineering coordinate system, and the data format and dimensions are normalized to generate a standardized construction data sequence.

3. The system for dynamic analysis and identification of construction risks in water conservancy projects based on multi-source data as described in claim 1, characterized in that, The calculation of the actual transition change between adjacent state vectors based on the ordered sequence specifically includes: Extract the corresponding state components from the previous state vector and the next state vector respectively, and obtain the statistical stability of each state component in the historical normal construction process. Weights are dynamically assigned to each state component based on statistical stability. Based on dynamically allocated weights, the absolute values ​​of the differences between corresponding components of the state vectors before and after are summed in a weighted manner, and the summation result is used as the actual transition change.

4. The system for dynamic analysis and identification of construction risks in water conservancy projects based on multi-source data as described in claim 1, characterized in that, The generated dynamic construction risk assessment criteria specifically include: Based on the time series change rates of state transition anomaly eigenvalues ​​and co-association deviation eigenvalues, the fusion ratio of state transition anomaly eigenvalues ​​and co-association deviation eigenvalues ​​is determined; The two feature values ​​of the current time window are integrated according to the fusion ratio to form a comprehensive risk vector. The comprehensive risk vector is then matched with the risk evolution sequence recorded in the preset historical risk pattern library for morphological similarity. Based on the historical risk pattern with the highest similarity in the matching results, the current risk evolution pattern is directly mapped to determine the current risk evolution pattern; and based on the magnitude of the comprehensive risk vector, a joint judgment is made in conjunction with the determined risk evolution pattern to output the current risk level.

5. The system for dynamic analysis and identification of construction risks in water conservancy projects based on multi-source data according to claim 4, characterized in that, The step of performing morphological similarity matching between the comprehensive risk vector and the risk evolution sequences recorded in a preset historical risk pattern library specifically includes: Multi-scale decomposition was performed on the time series and risk evolution sequence of the comprehensive risk vector to obtain the components of each sequence at different time granularities; At different time granularities, the components of the comprehensive risk vector and the components of the risk evolution sequence are compared in segments, and the morphological distance is calculated segment by segment. We aggregate the weighted morphological distances calculated at all time granularities, and determine the historical risk evolution sequence with the highest matching degree through the distance minimization principle, which is then used as the result of morphological similarity matching.

6. The system for dynamic analysis and identification of construction risks in water conservancy projects based on multi-source data according to claim 1, characterized in that, The aforementioned dynamic adjustment of the work process or resource allocation at the construction site specifically includes: Based on the risk evolution model, the corresponding basic instruction set is selected from the preset early warning and control strategy matrix. The control strategy matrix defines the mapping relationship between different risk evolution models and preliminary control actions. Based on the risk level, the intensity of each action parameter in the basic instruction set is calibrated to generate preliminary control instructions with specific execution parameters; The initial control instructions are combined with key construction process data acquired in real time. The instruction parameters are then fine-tuned and verified through decision optimization rules, and executable differentiated control instructions are output and issued.

7. The system for dynamic analysis and identification of construction risks in water conservancy projects based on multi-source data according to claim 6, characterized in that, The process involves combining risk levels to calibrate the intensity of each action parameter in the basic instruction set, generating preliminary control instructions with specific execution parameters, including: Based on the risk level, the preset baseline mapping is queried to obtain the baseline intensity value of each action parameter; Real-time acquisition of status data of key constraints related to various actions during the current construction process; Based on the state data of key constraints, the baseline strength value is dynamically corrected, and the corrected strength value is assigned to the corresponding action parameters to generate preliminary control commands.

Images