Flood control facility operation state remote monitoring and fault early warning method and system

Abstract

The invention discloses a flood control facility operation state remote monitoring and fault early warning method and system, and relates to the technical field of hydraulic engineering facility safety monitoring. The remote monitoring and fault early warning method for the running state of the flood control facility comprises the following steps: acquiring running time sequence data of a target flood control facility in a set time period, and preprocessing the running time sequence data; inputting into a pre-trained deep learning model, and extracting facility fault state features of the target flood control facility; based on the facility fault state characteristics of the target flood control facility, analyzing an operation fault index of the target flood control facility; according to the operation fault index of the target flood control facility, a corresponding fault early warning strategy is triggered, the operation time sequence data of the target flood control facility is analyzed based on the deep learning model, the facility fault state characteristics of the target flood control facility are extracted, and the comprehensive operation fault index is generated, so that the crossing from passive alarm to intelligent early warning is realized; the fault identification accuracy and the early warning time efficiency are greatly improved, and a powerful guarantee is provided for safe operation of flood control facilities.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring technology for water conservancy engineering facilities, specifically to a method and system for remote monitoring and fault early warning of the operational status of flood control facilities. Background Technology

[0002] Flood control facilities (such as gates, pumping stations, and dikes) are crucial infrastructure for ensuring flood control safety in river basins. In recent years, with the development of IoT and sensor technologies, remote monitoring of the operational status of flood control facilities has become an industry trend.

[0003] Currently, in the field of flood control facility monitoring, various sensors, such as vibration sensors, stress strain gauges, and displacement sensors, are typically installed at key locations of the facilities to collect operational parameters. These sensors continuously generate operational sequence data, providing a data foundation for facility condition assessment. Simultaneously, advancements in communication technology, particularly the application of 5G and NB-IoT, have provided favorable technical conditions for the remote transmission of monitoring data.

[0004] In terms of data analysis, existing technologies primarily involve storing and managing the collected monitoring data and using basic data processing methods to identify obvious anomalies. These technologies provide some information support for the operation and maintenance of flood control facilities, helping managers understand the basic operational status of the facilities.

[0005] The limitations of existing technologies include at least the following problems: Currently, remote monitoring and fault early warning of flood control facility operation mainly rely on traditional threshold judgments or single-dimensional data analysis, which has systemic shortcomings. Existing technologies typically set fixed thresholds for only a few key parameters, triggering an alarm when data exceeds these limits. This method is neither able to deeply mine the fault correlation information hidden in time-series data nor can it leverage deep learning to extract multi-dimensional fault features, making it difficult to form a fault index that comprehensively reflects the health of the facilities.

[0006] In practical applications, this limitation can easily lead to two types of problems: First, false alarms are frequent, as momentary anomalies caused by environmental fluctuations are misjudged as faults, interfering with normal operation and maintenance; second, false alarms are serious, as potential faults with multiple abnormal trends but no single parameter exceeding the threshold are difficult to identify, and are often only discovered after the fault has worsened, missing the opportunity to deal with it, creating hidden dangers for safety accidents, and seriously affecting the safe operation of flood control facilities. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method and system for remote monitoring and fault early warning of flood control facilities, which solves the problems of high false alarm rate, high missed alarm rate, and poor early warning timeliness caused by existing technologies relying on fixed thresholds and single-dimensional monitoring.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for remote monitoring and fault early warning of the operational status of flood control facilities, comprising the following steps: acquiring operational sequence data of the target flood control facility within a set time period and preprocessing it; inputting the preprocessed operational sequence data of the target flood control facility into a pre-trained deep learning model to extract facility fault status features of the target flood control facility; analyzing the operational fault index of the target flood control facility based on the facility fault status features; and triggering a corresponding fault early warning strategy according to the operational fault index of the target flood control facility.

[0009] Furthermore, the runtime sequence data includes gate vibration acceleration values, hydraulic cylinder piston rod stress values, hydraulic pump station motor current values, and hydraulic cylinder piston displacement values ​​at several time points.

[0010] Furthermore, deep learning models include one-dimensional convolutional layers, long short-term memory networks, statistical analysis layers, splicing layers, fully connected layers, and output layers.

[0011] Further, the specific steps for extracting the facility failure state features of the target flood control facility are as follows: In the one-dimensional convolutional layer of the deep learning model, the gate vibration acceleration values ​​and hydraulic pump station motor current values ​​at several time points are convolved to obtain local feature maps; in the long short-term memory network of the deep learning model, the hydraulic cylinder piston rod stress values ​​and hydraulic cylinder piston displacement values ​​at several time points are sequentially modeled to obtain temporal feature vectors; in the statistical analysis layer of the deep learning model, the runtime sequence data is statistically analyzed to obtain basic statistical features; in the concatenation layer of the deep learning model, the local feature maps, temporal feature vectors, and basic statistical features are concatenated to obtain fused feature vectors; in the fully connected layer of the deep learning model, the fused feature vectors are nonlinearly transformed to obtain deep features; in the output layer of the deep learning model, the deep features are linearly regressed to output the facility failure state features of the flood control facility.

[0012] Furthermore, the fault state characteristics of the facility include the effective value of the vibration signal, peak stress, current waveform distortion, displacement steady-state error, and energy-to-work ratio.

[0013] Furthermore, the specific steps for analyzing the operational failure index of the target flood control facility are as follows: perform preliminary processing on the facility failure status characteristics of the target flood control facility; input the preliminary processed operational failure index of the target flood control facility into a preset early warning analysis model for fusion analysis, and output the operational failure index of the target flood control facility.

[0014] Furthermore, the specific steps for preliminary processing of the facility fault state characteristics of the target flood control facility are as follows: read the effective value of the vibration signal, peak stress, current waveform distortion, displacement steady-state error, and energy-power ratio of the target flood control facility; and normalize the effective value of the vibration signal, peak stress, current waveform distortion, displacement steady-state error, and energy-power ratio of the target flood control facility respectively.

[0015] Furthermore, the pre-defined early warning analysis model is as follows: ;in, The operational failure index of the target flood control facilities. , , , , The following are the normalized values ​​of the target flood control facility's vibration signal RMS value, peak stress, current waveform distortion, displacement steady-state error, and energy-to-work ratio. , , , , , The weighting coefficients are, in order, vibration weighting coefficient, stress weighting coefficient, current weighting coefficient, displacement weighting coefficient, efficiency weighting coefficient, and synergistic amplification weighting coefficient stored in the database.

[0016] Furthermore, the specific steps for triggering the corresponding fault warning strategy based on the operational fault index of the target flood control facility are as follows: the operational fault index of the target flood control facility is matched and analyzed with several preset operational fault intervals, and each operational fault interval corresponds to a fault warning level; based on the fault warning level corresponding to a preset operational fault interval where the operational fault index is located, the corresponding combined warning strategy is triggered.

[0017] A remote monitoring and fault early warning system for the operational status of flood control facilities includes: a runtime sequence data acquisition unit, used to acquire runtime sequence data of a target flood control facility within a set time period and perform preprocessing; a fault status feature analysis unit, used to input the preprocessed runtime sequence data of the target flood control facility into a pre-trained deep learning model to extract facility fault status features of the target flood control facility; an operational fault analysis unit, used to analyze the operational fault index of the target flood control facility based on the facility fault status features; and a fault early warning execution unit, used to trigger a corresponding fault early warning strategy according to the operational fault index of the target flood control facility.

[0018] The present invention has the following beneficial effects:

[0019] (1) The remote monitoring and fault early warning method for the operation status of flood control facilities collects the operation sequence data consisting of gate vibration acceleration and hydraulic cylinder piston rod stress, and combines it with a multi-structure deep learning model to accurately extract fault state features. This allows the focus on the key operation parameters of the hydraulically operated flood gate, ensuring a high correlation between the collected information and facility faults. In the design of the deep learning model, a one-dimensional convolutional layer is used to process vibration, current and other data to obtain local features, and a long short-term memory network is used to model stress, displacement and other time-series data. Then, a statistical analysis layer is used to extract basic features. After splicing, fusion and nonlinear transformation, the facility fault state features of the target flood control facility are accurately output, effectively solving the problems of single dimension of traditional monitoring data and one-sided feature extraction.

[0020] (2) The remote monitoring and fault early warning method for the operation status of flood control facilities achieves scientific and controllable early warning judgment through a standardized fault index analysis process and a multi-weight fusion early warning model. First, the fault status characteristics of the facilities are normalized to eliminate the dimensional differences between different parameters and ensure that the data can be directly compared and fused. The early warning analysis model adopts multi-weight coefficient combination calculation, which considers the independent influence of core parameters such as vibration, stress, and current, and also takes into account the correlation effect between parameters by synergistically amplifying the weight coefficients, avoiding false alarms or missed alarms caused by single parameter judgment. This design gets rid of the rigid limitations of traditional fixed threshold judgment, and makes the degree of fault measurable through quantitative calculation.

[0021] (3) The remote monitoring and fault early warning method for the operation status of flood control facilities adopts interval matching corresponding early warning levels in the early warning stage to trigger a combined early warning strategy. It can accurately allocate operation and maintenance resources according to the severity of the fault. Minor faults can remind personnel to pay close attention, and serious faults can be promptly initiated for emergency response, avoiding blind operation and maintenance. At the same time, the remote monitoring mode does not require on-site personnel to be on duty. It can not only keep track of the facility status in real time, but also reduce the risk of manual inspection in harsh environments, and comprehensively improve the intelligent level and safety guarantee capability of flood control facility operation and maintenance.

[0022] (4) The remote monitoring and fault early warning system for the operation status of the flood control facility, by constructing a runtime sequence data acquisition unit, a fault status feature analysis unit, an operation fault analysis unit and a fault early warning execution unit, realizes modular control and flexible expansion of the entire monitoring and early warning process, adapts to the application needs of hydraulic opening and closing flood gates of different scales. For example, for the gate monitoring needs in different scenarios, the acquisition frequency of the runtime sequence data acquisition unit can be optimized separately, or the parameters of the deep learning model in the fault status feature analysis unit can be upgraded without changing the overall system architecture. At the same time, the modular design reduces the difficulty of system maintenance. When an abnormality occurs in a certain unit, it can be quickly located and investigated without affecting the operation of other links. Combined with the remote monitoring characteristics, it can realize centralized control of multiple gates across regions.

[0023] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0024] Figure 1 This is a flowchart of a method for remote monitoring and fault early warning of the operation status of flood control facilities according to the present invention.

[0025] Figure 2 This is a flowchart illustrating the specific steps involved in analyzing the operational fault index of a target flood control facility in a remote monitoring and fault early warning method for the operational status of flood control facilities according to the present invention.

[0026] Figure 3 This is a block diagram of a remote monitoring and fault early warning system for the operation status of flood control facilities according to the present invention. Detailed Implementation

[0027] Please see Figure 1 This invention provides a technical solution: a method for remote monitoring and fault early warning of the operational status of flood control facilities, comprising the following steps: acquiring the operational sequence data of a target flood control facility (in this embodiment, a hydraulically operated floodgate) within a set time period and preprocessing it; inputting the preprocessed operational sequence data of the target flood control facility into a pre-trained deep learning model to extract the facility fault status features of the target flood control facility; analyzing the operational fault index of the target flood control facility based on the facility fault status features; and triggering a corresponding fault early warning strategy according to the operational fault index of the target flood control facility.

[0028] Specifically, data preprocessing includes:

[0029] Missing data were filled using a sliding window mean method, with the window size set to 64 sampling points;

[0030] Use the 3σ rule to identify and remove outlier data points;

[0031] The vibration signal was denoising by wavelet transform, with the db4 wavelet basis function selected and a decomposition layer of 5 layers.

[0032] The PTP precision clock protocol is used to ensure that the time synchronization accuracy of multi-source sensor data is within 1ms.

[0033] The runtime sequence data includes gate vibration acceleration values, hydraulic cylinder piston rod stress values, hydraulic pump station motor current values, and hydraulic cylinder piston displacement values ​​at several time points.

[0034] The vibration acceleration value of the gate can be measured and obtained by a piezoelectric accelerometer, which is installed in the gate body (or at the gate frame connection).

[0035] The stress value of the hydraulic cylinder piston rod can be obtained by measuring a strain gauge stress sensor, which is installed on the surface of the hydraulic cylinder piston rod (attached and fixed).

[0036] The current value of the hydraulic pump station motor can be measured and obtained by a Hall effect current sensor, which is installed on the power supply line of the hydraulic pump station motor.

[0037] The displacement value of the hydraulic cylinder piston can be measured and obtained by a magnetostrictive displacement sensor, which is installed inside the hydraulic cylinder barrel (or at the end of the piston rod).

[0038] Specifically, the characteristics of facility failure status include the effective value of vibration signal, peak stress, current waveform distortion, steady-state displacement error, and energy-to-work ratio.

[0039] Deep learning models include one-dimensional convolutional layers, long short-term memory networks, statistical analysis layers, splicing layers, fully connected layers, and output layers.

[0040] The specific steps for pre-training a deep learning model are as follows:

[0041] Collect historical operation sequence data of the target flood control facilities during a continuous operation period of no less than twelve months, covering normal, abnormal and typical fault conditions; the data must completely include four types of parameters: gate vibration acceleration value, hydraulic cylinder piston rod stress value, hydraulic pump station motor current value, and hydraulic cylinder piston displacement value, and all data channels must maintain strict time synchronization, with a sampling frequency of no less than 1kHz.

[0042] For each complete operating cycle (including one gate opening and closing cycle) in historical data, an expert group consisting of three or more engineers with over five years of experience in hydraulic gate operation and maintenance, combined with the corresponding equipment maintenance records, fault repair reports, and video surveillance records, jointly determines the true health status of the facilities within that cycle, and manually labels the true values ​​of five facility fault state characteristics based on this; among which:

[0043] The true effective value of the vibration signal is obtained by calculating the root mean square of the vibration acceleration sequence within that period.

[0044] The true value of the peak stress is obtained by finding the maximum value of the stress sequence within that period;

[0045] The true value of the current waveform distortion is obtained by calculating the total harmonic distortion rate of the current sequence within that period;

[0046] The true value of the steady-state displacement error is obtained by calculating the average absolute deviation between the displacement sensor reading and the target command value during the time period after the gate reaches the target opening and stabilizes.

[0047] The true value of the active power ratio is obtained by integrating the active electrical energy consumed by the motor and the mechanical work output by the hydraulic cylinder in the cycle.

[0048] Finally, a dataset containing no less than 5,000 labeled samples was constructed and randomly divided into training set, validation set and test set in a ratio of 7:2:1.

[0049] The original time-series data of all samples in the dataset are standardized by calculating the mean and standard deviation of each data channel on the training set, and then subtracting the mean and dividing by the standard deviation from all data points of that channel so that the processed data conforms to a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0050] Data augmentation is performed using a sliding window technique with a fixed window length of 1024 time points and a sliding step of 256 time points. This operation expands the number of training set samples to more than three times the original number, thereby improving the generalization ability of the model training.

[0051] The specific implementation of building a deep learning model is as follows:

[0052] One-dimensional convolutional layers are implemented using the Conv1d module in the PyTorch framework, where the convolutional kernel parameters of the vibration channel are initialized to a Kaiming normal distribution, and the convolutional kernel parameters of the current channel are initialized to a Xavier uniform distribution.

[0053] The Long Short-Term Memory network is implemented using the LSTM module in the PyTorch framework. Its forget gate bias is initialized to 1.0, and the other parameters are initialized orthogonally.

[0054] The statistical analysis layer is implemented through a custom Python class, which calculates the root mean square, peak value, crest factor, and standard deviation in real time within the forward propagation function;

[0055] The weight matrix of the fully connected layer is initialized to a Kaiming normal distribution, and the bias term is initialized to zero.

[0056] On a compute server equipped with at least two NVIDIA RTX 3090 graphics processors, the model was trained end-to-end using the training set data, with the following hyperparameter configurations:

[0057] The loss function is defined as the weighted mean square error of the five output features, where the loss weights for peak stress and energy-to-work ratio are set to 2.0, and the weights for the other features are 1.0.

[0058] The optimizer used is AdamW, with an initial learning rate of 0.001 and a weight decay coefficient of 0.01.

[0059] The batch size is fixed at 32, and the total number of training rounds is set to 200.

[0060] The learning rate scheduling strategy uses ReduceLROnPlateau, where the learning rate is halved if the validation set loss does not decrease for five consecutive rounds.

[0061] During training, after each training round, the loss value is calculated using the validation set and monitored through an early stopping mechanism: when the validation set loss does not break the lowest record for 15 consecutive rounds, training is automatically terminated and rolled back to the model parameters with the best performance on the validation set.

[0062] After training, the final model's performance is evaluated using a reserved test set. Evaluation metrics include:

[0063] The mean absolute error must be less than 3% for each characteristic range;

[0064] The coefficient of determination must be higher than 0.85;

[0065] Only models that meet all of the above requirements are considered qualified and usable pre-trained models. The model's structure definition and the trained parameter weights are then serialized into a standard PKL file.

[0066] The specific steps for extracting the fault status characteristics of the target flood control facility are as follows:

[0067] In the one-dimensional convolutional layer of the deep learning model, the gate vibration acceleration values ​​and hydraulic pump station motor current values ​​at several time points are convolved to obtain local feature maps. Specifically, the vibration acceleration sequence and the motor current sequence are input into two independent convolutional channels. For the vibration channel, thirty-two convolutional kernels with a width of five are used for one-dimensional convolution with a stride of one, and a linear rectified function is used as the activation function to extract the high-frequency impact components in the vibration signal. For the current channel, sixteen convolutional kernels with a width of eleven are used for one-dimensional convolution with a stride of two, and a linear rectified function is also used for activation to capture transient distortions and harmonic modes in the current waveform. Subsequently, the feature maps output from the two channels are concatenated along the channel dimension to form a joint local feature map, which is then downsampled through a max pooling layer with a kernel size of two to finally output the local feature map.

[0068] In the Long Short-Term Memory (LSTM) network of a deep learning model, the stress values ​​of the hydraulic cylinder piston rod and the displacement values ​​of the hydraulic cylinder piston at several time points are sequentially modeled to obtain a temporal feature vector. Specifically, the stress sequence and the displacement sequence are concatenated into a two-dimensional input vector sequence according to the corresponding time steps. This sequence is input into a bidirectional LSM network with 64 units, which includes forward and backward propagation directions. For each time point in the sequence, the forward network processes the sequence from left to right, learning the influence of historical information on the current state; the backward network processes the sequence from right to left, learning the supplementary information of future information to the current state. Finally, the hidden state vector generated by forward propagation and the hidden state vector generated by backward propagation at the last time point of the sequence are concatenated to obtain a 128-dimensional temporal feature vector. This vector fully encodes the dynamic coupling relationship and contextual information of stress and displacement within a complete working cycle.

[0069] In the statistical analysis layer of the deep learning model, statistical analysis is performed on the runtime sequence data to obtain basic statistical features. Specifically, real-time statistical calculations are performed in parallel on four raw time-series data within a set time window. Specifically: for the vibration acceleration sequence, its root mean square value is calculated; for the piston rod stress sequence, the maximum value within the time window is found and recorded as the stress peak value; for the motor current sequence, the ratio of the maximum value in its absolute value sequence to the root mean square value of the current, i.e., the crest factor, is calculated; for the piston displacement sequence, the dispersion of its data points relative to its average value, i.e., the standard deviation, is calculated. Finally, the four statistical measures obtained, namely the vibration root mean square value, stress peak value, current crest factor, and displacement standard deviation, are combined into a four-dimensional basic statistical feature vector.

[0070] In the concatenation layer of the deep learning model, local feature maps, temporal feature vectors, and basic statistical features are concatenated to obtain a fused feature vector. Specifically, the two-dimensional local feature map structure is first flattened into a one-dimensional vector. Then, this flattened one-dimensional vector, the 128-dimensional temporal feature vector output by the long short-term memory network, and the four-dimensional basic statistical feature vector output by the statistical analysis layer are concatenated end-to-end along the feature dimensions to form a high-dimensional fused feature vector that integrates local signal details, temporal dynamic context, and basic physical statistics.

[0071] In the fully connected layer of the deep learning model, a nonlinear transformation is performed on the fused feature vector to obtain deep features. Specifically, the high-dimensional fused feature vector is input into a first fully connected layer containing 256 neurons, and a nonlinear activation transformation is performed using a linear rectified function. Subsequently, the output vector of the first fully connected layer is input into a second fully connected layer containing 64 neurons, and a linear rectified function is used again for activation to further compress and refine the features, ultimately outputting a 64-dimensional, highly refined deep feature vector.

[0072] In the output layer of the deep learning model, linear regression is performed on the deep features to output the fault state features of the flood control facilities. Specifically, a 64-dimensional deep feature vector is input into an output layer containing five neurons without using any activation function. Each neuron in this layer performs a linear weighted summation operation on the input deep feature vector through its internal preset weights and bias parameters. Finally, the five values ​​output by this layer correspond to the specific values ​​of the five fault state features of the flood control facilities, namely: effective value of vibration signal, peak stress, current waveform distortion, displacement steady-state error, and energy-power ratio.

[0073] In this implementation scheme, a convolutional neural network (CNN) is used to extract local features of gate vibration acceleration and hydraulic pump motor current, which can accurately capture the dynamic changes of high-frequency impact and current waveform. Then, a long short-term memory network (LSTM) is used to model the stress and displacement sequence of the hydraulic cylinder, further capturing the dynamic coupling relationship and enhancing the system's ability to perceive time-series changes. In addition, a statistical analysis layer is used to perform real-time calculations on the original time-series data, extracting basic statistical features such as the root mean square value of vibration, peak stress, and current crest factor, which provides strong support for subsequent fault assessment. Finally, these features are fused through a splicing layer and fed into a fully connected layer for nonlinear transformation and deep feature extraction, outputting accurate fault state features. This method comprehensively considers signal details, time-series dynamics, and physical statistical characteristics, and can provide multi-dimensional fault warning information, thereby significantly improving the fault monitoring and early warning capabilities of flood control facilities.

[0074] Specifically, such as Figure 2 As shown, the specific steps for analyzing the operational failure index of the target flood control facility are as follows: perform preliminary processing on the facility failure status characteristics of the target flood control facility; input the preliminary processed operational failure index of the target flood control facility into the preset early warning analysis model for fusion analysis, and output the operational failure index of the target flood control facility.

[0075] The specific steps for preliminary processing of the fault state characteristics of the target flood control facility are as follows: Read the effective value of the vibration signal, peak stress, current waveform distortion, displacement steady-state error, and energy-power ratio of the target flood control facility; Normalize the effective value of the vibration signal, peak stress, current waveform distortion, displacement steady-state error, and energy-power ratio of the target flood control facility respectively (using the min-max normalization method, that is, for each feature value, calculate by (actual value - feature minimum value) / (feature maximum value - feature minimum value), and map the result to the [0,1] interval. The maximum and minimum values ​​of the feature are determined statistically based on the historical dataset used by the pre-trained model).

[0076] The pre-set early warning analysis model is as follows: ;in, The operational failure index of the target flood control facilities. , , , , The following are the normalized values ​​of the target flood control facility's vibration signal RMS value, peak stress, current waveform distortion, displacement steady-state error, and energy-to-work ratio. , , , , , The values ​​are, in order, the vibration weight coefficient, stress weight coefficient, current weight coefficient, displacement weight coefficient, efficiency weight coefficient, and synergistic amplification weight coefficient stored in the database, and in this embodiment, they are taken as follows: 0.25, 0.25, 0.15, 0.15, 0.10, and 0.15.

[0077] The vibration weighting coefficient stored in the database has a value range of [0, 1].

[0078] The stress weighting coefficients stored in the database have a range of values: [0, 1].

[0079] The current weighting coefficients stored in the database have a range of values: [0, 1].

[0080] The displacement weight coefficients stored in the database have a range of values: [0, 1].

[0081] The performance weight coefficients stored in the database have a range of values: [0, 1].

[0082] The range of values ​​for the collaborative amplification weight coefficients stored in the database is [0.05, 0.3].

[0083] In this implementation plan, the various fault state characteristics of the facilities (such as the effective value of vibration signals, peak stress, etc.) are normalized to ensure that different characteristic data can be scaled uniformly, facilitating subsequent analysis. This step ensures the comparability of the data and avoids the influence of different dimensions of certain characteristics on the overall evaluation results. Next, the processed characteristic data are fused and calculated through an early warning analysis model to obtain a comprehensive operational fault index. This index combines information from multiple dimensions such as vibration, stress, current, and displacement, and has high accuracy and reliability. In particular, the weighting coefficients used in the model (such as vibration, stress, current, etc.) are stored in a database and dynamically adjusted, which can optimize the evaluation results according to the actual situation and flexibly adapt to the operating status of different facilities. Through this method, the system can monitor the operating status of the facilities in real time and accurately, promptly detect potential fault risks, provide early warnings, and avoid safety hazards caused by equipment failure.

[0084] Specifically, the steps for triggering a corresponding fault warning strategy based on the operational fault index of the target flood control facility are as follows: The operational fault index of the target flood control facility is matched with several preset operational fault intervals, with each operational fault interval corresponding to a fault warning level; based on the fault warning level corresponding to a preset operational fault interval, a corresponding combined warning strategy is triggered, the strategy including at least one of the following:

[0085] Primary alert: Generate an alert work order and push it to the mobile terminal of maintenance personnel;

[0086] Intermediate warning: Based on the implementation of the primary warning, automatically increase the data collection and reporting frequency of the relevant sensors of the facility;

[0087] Advanced warning: Based on the execution of intermediate warning, send instructions to the facility's own control system to limit frequency operation or lock out safety.

[0088] In this implementation plan, by matching and analyzing the facility's operational failure index with preset failure ranges, the system achieves accurate determination of failure levels, ensuring the timeliness and accuracy of early warnings. When the failure index reaches the primary warning range, the system generates a work order and pushes it to the mobile terminal of maintenance personnel, reminding them to pay attention to the problem. When the failure index is in the intermediate warning range, the system not only triggers a work order but also automatically increases the data acquisition frequency of the sensors, thereby enhancing monitoring accuracy and detecting potential risks in advance. The advanced warning, based on the first two warnings, sends frequency limiting or safety interlock commands to the facility control system to ensure that the equipment can be effectively protected in the event of a serious failure, preventing the accident from escalating. This tiered early warning strategy can better protect the operational safety of flood control facilities, reduce response time when failures occur, improve the system's adaptability and emergency response efficiency, and effectively ensure the stable operation of flood control facilities.

[0089] Please see Figure 3 This invention provides a technical solution: a remote monitoring and fault early warning system for the operational status of flood control facilities, comprising: a runtime sequence data acquisition unit, used to acquire runtime sequence data of a target flood control facility within a set time period and perform preprocessing; a fault status feature analysis unit, used to input the preprocessed runtime sequence data of the target flood control facility into a pre-trained deep learning model to extract facility fault status features of the target flood control facility; an operational fault analysis unit, used to analyze the operational fault index of the target flood control facility based on the facility fault status features; and a fault early warning execution unit, used to trigger corresponding fault early warning strategies according to the operational fault index of the target flood control facility.

[0090] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0091] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for remote monitoring and fault early warning of the operational status of flood control facilities, characterized in that, Includes the following steps: Within a set time period, acquire the operational sequence data of the target flood control facilities and perform preprocessing. The preprocessed runtime sequence data of the target flood control facilities is input into a pre-trained deep learning model to extract the facility failure state features of the target flood control facilities. Based on the facility failure status characteristics of the target flood control facilities, the operational failure index of the target flood control facilities is analyzed; Based on the operational failure index of the target flood control facility, a corresponding failure early warning strategy is triggered.

2. The method for remote monitoring and fault early warning of flood control facility operation status according to claim 1, characterized in that, The runtime sequence data includes gate vibration acceleration values, hydraulic cylinder piston rod stress values, hydraulic pump station motor current values, and hydraulic cylinder piston displacement values ​​at several time points.

3. The method for remote monitoring and fault early warning of flood control facility operation status according to claim 2, characterized in that, Deep learning models include one-dimensional convolutional layers, long short-term memory networks, statistical analysis layers, splicing layers, fully connected layers, and output layers.

4. The method for remote monitoring and fault early warning of flood control facility operation status according to claim 3, characterized in that, The specific steps for extracting the fault status characteristics of the target flood control facility are as follows: In the one-dimensional convolutional layer of the deep learning model, the gate vibration acceleration values ​​and hydraulic pump station motor current values ​​at several time points are convolved to obtain local feature maps; In the long short-term memory network of the deep learning model, the stress values ​​of the hydraulic cylinder piston rod and the displacement values ​​of the hydraulic cylinder piston at several times are sequentially modeled to obtain a temporal feature vector; In the statistical analysis layer of the deep learning model, statistical analysis is performed on the runtime sequence data to obtain basic statistical features; In the concatenation layer of the deep learning model, local feature maps, temporal feature vectors, and basic statistical features are concatenated to obtain a fused feature vector. In the fully connected layer of the deep learning model, a nonlinear transformation is performed on the fused feature vector to obtain deep features; In the output layer of the deep learning model, linear regression is performed on the deep features to output the fault status features of the flood control facilities.

5. The method for remote monitoring and fault early warning of flood control facility operation status according to claim 1, characterized in that, The fault state characteristics of the facility include the effective value of the vibration signal, the peak stress, the distortion of the current waveform, the steady-state error of the displacement, and the energy-to-work ratio.

6. The method for remote monitoring and fault early warning of flood control facility operation status according to claim 5, characterized in that, The specific steps for analyzing the operational failure index of the target flood control facilities are as follows: Preliminary processing is performed on the facility failure status characteristics of the target flood control facility; The operational failure index of the target flood control facilities after preliminary processing is input into the preset early warning analysis model for fusion analysis, and the operational failure index of the target flood control facilities is output.

7. The method for remote monitoring and fault early warning of flood control facility operation status according to claim 6, characterized in that, The specific steps for preliminary processing of the facility failure status characteristics of the target flood control facility are as follows: Read the effective value of vibration signal, peak stress, current waveform distortion, displacement steady-state error, and energy-power ratio of the target flood control facility; The effective value of the vibration signal, peak stress, current waveform distortion, displacement steady-state error, and dynamic power ratio of the target flood control facility are normalized respectively.

8. The method for remote monitoring and fault early warning of flood control facility operation status according to claim 7, characterized in that, The preset early warning analysis model is as follows: ;in, The operational failure index of the target flood control facilities. , , , , The following are the normalized values ​​of the target flood control facility's vibration signal RMS value, peak stress, current waveform distortion, displacement steady-state error, and energy-to-work ratio. , , , , , The weighting coefficients are, in order, vibration weighting coefficient, stress weighting coefficient, current weighting coefficient, displacement weighting coefficient, efficiency weighting coefficient, and synergistic amplification weighting coefficient stored in the database.

9. The method for remote monitoring and fault early warning of flood control facility operation status according to claim 1, characterized in that, The specific steps for triggering the corresponding fault early warning strategy based on the operational fault index of the target flood control facility are as follows: The operational failure index of the target flood control facility is matched and analyzed with several preset operational failure intervals, and each operational failure interval corresponds to a failure warning level. Based on the fault warning level corresponding to the operation fault index falling within a preset operation fault range, a corresponding combined warning strategy is triggered.

10. A remote monitoring and fault early warning system for the operational status of flood control facilities, employing the remote monitoring and fault early warning method for the operational status of flood control facilities as described in any one of claims 1-9, characterized in that, include: The runtime sequence data acquisition unit is used to acquire the runtime sequence data of the target flood control facility within a set time period and perform preprocessing. The fault state feature analysis unit is used to input the preprocessed runtime sequence data of the target flood control facility into a pre-trained deep learning model to extract the facility fault state features of the target flood control facility. The operation failure analysis unit is used to analyze the operation failure index of the target flood control facility based on the facility failure status characteristics of the target flood control facility; The fault early warning execution unit is used to trigger the corresponding fault early warning strategy based on the operational fault index of the target flood control facility.

Images