Photovoltaic field slope deformation real-time monitoring and early warning system

Abstract

The invention discloses a photovoltaic field slope deformation real-time monitoring and early warning system, and particularly relates to the technical field of photovoltaic power station safety monitoring. The system comprises a data acquisition module, a data processing and analysis module, an early warning judgment module and an early warning information management module. The data acquisition module acquires multi-source monitoring data through a tilt angle sensor, a GNSS displacement monitoring station and an environment sensor which are deployed on site; the data processing and analysis module adopts a Kalman filtering and machine learning fusion algorithm to carry out fusion processing and trend prediction on the data; the early warning judgment module performs intelligent judgment based on a dynamic multi-level threshold value and a rule; and the early warning information management module executes a graded and channel-divided information issuing strategy according to the early warning level. According to the invention, real-time accurate sensing, intelligent analysis, research and judgment and efficient early warning information management of the deformation state of the photovoltaic field slope are realized, and the active protection capability of power station safety management is improved.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power plant safety monitoring technology, and more specifically, to a real-time monitoring and early warning system for slope deformation at photovoltaic sites. Background Technology

[0002] As an important clean energy source, photovoltaic power generation is expanding in scale, inevitably involving complex terrains such as hills and mountains. Therefore, photovoltaic arrays are often built on slopes, and the long-term stability of their supporting structures and the slopes directly affects the safe operation of the entire power station. In recent years, safety monitoring technologies for slopes and structures have continuously developed, evolving from traditional manual periodic surveys to automated monitoring based on sensor networks. Monitoring indicators have also expanded from single surface displacement data to multi-dimensional data including tilt and environmental factors, showing a clear trend towards automation and multi-source monitoring.

[0003] However, existing monitoring technologies still have significant shortcomings. First, most systems only focus on data acquisition and simple display, lacking effective fusion and in-depth analysis of multi-source heterogeneous monitoring data (such as displacement, tilt, rainfall, and vibration). This results in information silos between data points, making it difficult to comprehensively reflect the overall deformation trend of the slope. Second, the early warning mechanisms of existing systems are mostly based on simple single-point threshold judgments, failing to combine temporal evolution trends for dynamic and forward-looking predictions. This leads to high false alarm and false negative rates, and early warning thresholds are usually statically set, unable to adaptively adjust according to changes in geological conditions and the external environment. Finally, the release and management of early warning information lack intelligent strategies, often failing to achieve precise, tiered, and channel-specific delivery based on risk levels, resulting in critical information not reaching relevant responsible parties in a timely manner.

[0004] Therefore, a real-time monitoring and early warning system for slope deformation in photovoltaic sites is proposed to address the aforementioned issues. The system aims to overcome key deficiencies in existing technologies, primarily focusing on how to achieve real-time, accurate perception and trend prediction of slope deformation in photovoltaic sites, how to establish an intelligent early warning criterion that integrates multi-source information and dynamically adjusts it, and how to ensure that early warning information is efficiently and reliably transmitted to the operation and maintenance team, thereby comprehensively improving the proactive protection capabilities of photovoltaic power station slope safety. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a real-time monitoring and early warning system for slope deformation in photovoltaic sites, thereby addressing the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a real-time monitoring and early warning system for slope deformation in photovoltaic sites, the system comprising:

[0007] The data acquisition module is used to acquire multi-source monitoring data related to slope stability through a sensor array deployed in the photovoltaic site area;

[0008] The data processing and analysis module is communicatively connected to the data acquisition module. It is used to receive the multi-source monitoring data and perform time-series alignment and feature extraction on the data using a preset fusion algorithm. Based on this, it calculates the slope deformation displacement and deformation trend data through a built-in deformation analysis model.

[0009] The early warning determination module, coupled to the data processing and analysis module, is used to compare the deformation displacement and deformation trend data with the preset multi-level early warning thresholds in real time, and generate an early warning signal containing early warning level and location information when any early warning threshold is determined to be triggered.

[0010] The early warning information management module is communicatively connected to the early warning determination module. It is used to receive the early warning signal and distribute the early warning information to designated user terminals or external systems based on a predefined distribution strategy.

[0011] Alternatively, the multi-source monitoring data acquired by the data acquisition module includes at least two of the following: support structure tilt angle data measured by tilt sensors, three-dimensional surface coordinate data acquired by GNSS displacement monitoring stations, and rainfall and vibration data acquired by environmental sensors.

[0012] Alternatively, the built-in deformation analysis model in the data processing and analysis module is a combination model integrating time series analysis algorithm and machine learning regression algorithm. This model takes the multi-source monitoring data as input and outputs the predicted value of slope deformation displacement and its changing trend within a set time period in the future.

[0013] Multiple options are available. The multi-level early warning thresholds preset in the early warning determination module include at least a blue early warning threshold, a yellow early warning threshold, and a red early warning threshold. The thresholds at each level are set differently according to the geological conditions and slope design parameters of the photovoltaic site. The early warning determination module is configured to support the dynamic adjustment and updating of the multi-level early warning thresholds via remote commands.

[0014] Multiple options are available. The release strategy of the early warning information management module supports selecting different information push frequencies and at least one push channel according to the level of the early warning. The push channel includes sending data packets to the external management system via SMS, email, or API interface.

[0015] Alternatively, the system may also include a three-dimensional visualization module, which is connected to the data processing and analysis module and the early warning information management module. This module is used to render and display the deformation state of the slope and the spatial location of the early warning area in the three-dimensional terrain model in real time based on the deformation displacement, deformation trend data and early warning signals.

[0016] Alternatively, the machine learning regression algorithm is one of the random forest algorithm or gradient boosting decision tree algorithm, used to establish a nonlinear mapping relationship between monitoring data and deformation displacement.

[0017] Alternatively, the system also includes a data storage and management submodule, which is used to structure and store historical monitoring data, analysis result records, and early warning event logs, and provides data query and export functions based on time range or early warning level.

[0018] Alternatively, the system may interact with the sensor group and user terminal via a wired or wireless communication network, wherein the wireless communication network may include one of a fourth-generation mobile communication technology network, a fifth-generation mobile communication technology network, or a LoRa wireless network.

[0019] The technical effects and advantages of this invention are as follows:

[0020] Compared to existing technologies, this invention achieves deep fusion and intelligent analysis of multi-source heterogeneous data such as tilt angle, displacement, rainfall, and vibration by introducing a combined model of Kalman filtering and machine learning. This method first performs time-series alignment and feature extraction on various sensor data, then utilizes a trained model to uncover the inherent correlations and nonlinear patterns between the data, ultimately outputting high-precision deformation displacement and trend prediction values. This effectively breaks down data silos, improves the comprehensiveness of state perception and the accuracy of prediction, and provides more reliable data support for early warning decision-making.

[0021] Compared to existing technologies, this invention achieves intelligent early warning judgment by constructing a multi-level early warning threshold system and fuzzy inference rule base that supports remote dynamic adjustment. This system not only relies on real-time displacement but also integrates displacement rate, acceleration trend, and environmental factors for fusion judgment, and can adaptively optimize thresholds based on geological conditions and historical data. This overcomes the shortcomings of traditional fixed threshold methods, such as slow response or susceptibility to false alarms, improving the accuracy and timeliness of early warnings, and realizing a shift from "static alarm" to "dynamic early warning."

[0022] Compared to existing technologies, this invention achieves efficient and accurate delivery of early warning information by formulating a differentiated information dissemination strategy based on warning levels. This mechanism automatically matches different push frequencies (e.g., real-time, 10-minute, 30-minute) and selects the optimal push channel (e.g., SMS, email, API interface) according to the severity of the warning, ensuring that critical information is delivered to designated responsible persons in the most effective way at the first opportunity. This advantage optimizes information flow efficiency, avoids warning information being overwhelmed or delayed, and ensures the timely activation of emergency response procedures. Attached Figure Description

[0023] Figure 1 This is a system framework diagram of the present invention.

[0024] Figure 2 This is a flowchart of the process of the present invention. Detailed Implementation

[0025] 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.

[0026] Example 1:

[0027] As attached Figure 1 The system shown is a real-time monitoring and early warning system for slope deformation in a photovoltaic site. The data acquisition module consists of a distributed monitoring network composed of multiple sensing units, and each sensing unit includes a data acquisition node and a signal conditioning circuit.

[0028] The sensing unit is packaged in an industrial-grade protective housing and integrates a microprocessor and communication unit. The data acquisition node supports multi-channel synchronous sampling with a sampling accuracy of at least 16 bits and a configurable sampling frequency range of 1 Hz to 100 Hz. The signal conditioning circuit includes analog filtering, amplification, and analog-to-digital conversion functions, and uses a programmable gain amplifier to achieve adaptive signal amplitude adjustment.

[0029] The sensor suite includes a resistive strain gauge tilt sensor, a global navigation satellite system receiver, a piezoelectric vibration sensor, and a capacitive rain gauge. The tilt sensor has a measurement range of ±30 degrees and a linearity error of no more than 0.5% of full scale.

[0030] The global navigation satellite system receiver supports dual-mode positioning using both BeiDou and Global Positioning System (GPS), employing carrier phase differential technology to achieve millimeter-level positioning accuracy. The vibration sensor has a frequency response range of 0.5 Hz to 200 Hz and a dynamic range of no less than 90 dB. The rain gauge uses a double-tipping bucket structure with a resolution of 0.1 mm.

[0031] All sensors are connected to the data acquisition module via an industrial fieldbus and wireless transmission network. Wired transmission uses shielded twisted-pair cable to transmit RS-485 signals, while wireless transmission uses LoRa spread spectrum communication technology, with the communication protocol conforming to the Modbus RTU standard. The data acquisition module has a built-in clock synchronization unit, employing a precision clock protocol to achieve microsecond-level time synchronization, ensuring the consistency of timing among multi-source data. The data acquisition module also includes a data buffer unit, using a ring buffer structure to store the raw data of the most recent 24 hours, supporting functions such as resume transmission after disconnection and data retransmission.

[0032] The tilt sensor employs a MEMS inertial measurement unit, integrating a three-axis accelerometer and a three-axis gyroscope, and outputs attitude angle data through a sensor fusion algorithm. The sensor utilizes a temperature compensation algorithm, employing polynomial fitting to correct for temperature drift, and operates within a temperature range of -40°C to 85°C. The GNSS displacement monitoring station includes a high-precision receiver and antenna. The receiver supports multi-frequency, multi-system signal tracking and employs a solution mode combining precise single-point positioning technology and real-time dynamic differential technology.

[0033] The calculation software uses a Kalman filter algorithm to process carrier phase observations, outputting a three-dimensional coordinate sequence with a horizontal accuracy better than 3 mm and an elevation accuracy better than 5 mm. Among the environmental sensors, rainfall monitoring employs a tipping-bucketraingauge mechanism with a water inlet area of ​​200 square centimeters and a measurement error not exceeding 4%. Vibration monitoring uses a triaxial accelerometer with selectable ranges of ±2g or ±4g and a noise density below 100 micrograms per square root of hertz.

[0034] All sensor data are timestamped and quality-identified, and the data format conforms to the ISO 8601 time standard and the OGC sensor observation standard. The data acquisition module performs preliminary quality checks on the raw data, including range checks, jump checks, and consistency checks, and adds quality flags to abnormal data. Multi-source data are aligned according to a unified time base, and interpolation algorithms are used to process data with different sampling rates to generate a time-synchronized multimodal dataset.

[0035] The data processing and analysis module is deployed on a server platform, employing a microservice architecture to achieve distributed computing. The data preprocessing stage begins with outlier detection and missing value imputation. Outlier detection combines the statistical 3σ criterion with a distance-based isolated forest algorithm, while missing value imputation uses time-series linear interpolation and multiple interpolation methods.

[0036] Multi-source data fusion employs a Kalman filter framework to establish system state and observation equations. State variables include displacement, velocity, and acceleration, while observation variables are the measurements from various sensors. Time series analysis combines an autoregressive integral moving average model with seasonal decomposition. Model parameters are determined using maximum likelihood estimation, and the model order is optimized using the Akaike information criterion.

[0037] Wherein, the state equation is: x k ·=Ax k-1 +Bu k-1 +w k-1 ; Observation equation: z k =Hx k +v k ;

[0038] x k and x k-1 These represent the state vectors of the system at time k and time k-1, respectively, and typically include physical quantities such as displacement, velocity, and acceleration.

[0039] A is the state transition matrix, which describes how the system state evolves from time k-1 to time k.

[0040] B is the control input matrix.

[0041] u k-1 It is the control input vector at time k-1.

[0042] w k-1 It is the process noise at time k-1, which is assumed to be white noise with zero mean.

[0043] z k It is the observation vector at time k, that is, the actual measurement value of various sensors.

[0044] H is the observation matrix, which maps the system state vector to the observation vector.

[0045] v k It is the observation noise at time k, assumed to be white noise with zero mean.

[0046] This set of equations forms the core framework of the Kalman filter. The state equation is used to predict the system state at the current time k, while the observation equation describes the relationship between the system's observations and state values. By recursively executing the "prediction" and "update" steps, the algorithm can effectively suppress noise interference, fuse multi-source heterogeneous data, and obtain a more accurate and reliable system state estimate than that of a single sensor, i.e., a high-precision deformation displacement.

[0047] in,

[0048] y t This represents the observed value at time point t, i.e., the deformation displacement.

[0049] L is a lag operator, for example L yt =y t -1.

[0050] p is the order of the autoregressive term.

[0051] φ i The coefficient of the i-th autoregressive term.

[0052] d is the minimum difference order required to make the sequence stationary.

[0053] q is the order of the moving average term.

[0054] θj It is the coefficient of the j-th moving average term.

[0055] ∈ t It is a white noise error term with a mean of zero.

[0056] The ARIMA model stationaries non-stationary time series by performing d-order differencing, and then uses autoregressive (AR) and moving average (MA) terms to model and predict future deformation values. This model effectively handles the memory and random fluctuations in displacement data, providing a reliable mathematical tool for deformation trend analysis.

[0057] The machine learning component uses the gradient boosting decision tree algorithm, and the feature engineering includes time-domain feature extraction (mean, variance, peak factor, etc.) and frequency-domain feature extraction (fast Fourier transform to obtain spectral features).

[0058] The model training employs a sliding window cross-validation strategy, using the Adam optimizer to minimize the mean squared error loss function. The deformation analysis model output includes predicted displacement values ​​and confidence intervals, with a confidence level set at 95%. The model update mechanism utilizes online learning, retraining the model parameters every 24 hours using the latest data. All algorithms are implemented based on the Python scientific computing ecosystem, using NumPy for numerical computation, Pandas for data processing, Scikit-learn for machine learning algorithms, and Statsmodels for time series analysis.

[0059] The early warning determination module adopts a rule engine architecture, which includes a threshold management unit and a logic judgment unit. The threshold management unit stores multi-level early warning threshold parameters: the displacement rate threshold for blue early warning is 0.1 mm to 0.3 mm per minute, the displacement rate threshold for yellow early warning is 0.3 mm to 0.5 mm per minute, and the displacement rate threshold for red early warning is 0.5 mm to 1.0 mm per minute.

[0060] The threshold setting is based on a geotechnical mechanics calculation model, considering geological parameters such as soil internal friction angle, cohesion, slope angle, and slope height. The safety factor is calculated using the limit equilibrium method, and the displacement warning threshold is derived from the safety factor. The threshold update mechanism supports both manual and automatic modes. In manual mode, a new threshold is input through a graphical user interface. In automatic mode, the threshold is dynamically adjusted based on historical data statistical analysis. The mean and standard deviation of the displacement data are calculated using a sliding window statistical method, and an adaptive threshold is set according to the 3σ principle.

[0061] The logic judgment unit implements a multi-condition fusion judgment algorithm, which not only considers the instantaneous displacement value, but also integrates the displacement acceleration trend, duration and environmental factors (such as automatically reducing the threshold by 20% when the cumulative rainfall exceeds 50 mm).

[0062] The judgment logic employs a fuzzy inference system. Input variables are displacement, displacement rate, and acceleration; the output variable is the warning level. Membership functions utilize trigonometric and Gaussian functions, and the inference rule base contains 25 IF-THEN rules. All threshold parameters and judgment rules are stored in a relational database, supporting transactional updates and version management.

[0063] The early warning information management module adopts a message middleware architecture, using a publish-subscribe pattern to distribute information. The message format follows the JSON standard and includes fields such as early warning number, timestamp, location coordinates, early warning level, displacement data, and confidence level. The push strategy configuration library stores the publishing parameters corresponding to different early warning levels: blue warnings are pushed every 30 minutes, yellow warnings every 10 minutes, and red warnings are pushed continuously in real time.

[0064] The channel selector automatically chooses the transmission method based on the alert level and the type of recipient: the SMS channel uses a GSM modem pool or a third-party SMS gateway interface, supporting Chinese SMS encoding; the email channel uses the SMTP protocol, and the email template uses HTML format to include data tables and trend charts; the API interface follows the RESTful specification, the data encapsulation adopts the OGC sensor alarm standard format, and supports HTTPS transmission and OAuth2.0 authentication. The message queue is implemented using RabbitMQ to ensure reliable message transmission and flow control.

[0065] The retransmission mechanism employs an exponential backoff algorithm, with a maximum of 5 retries. The information management module also includes a user management subsystem, maintaining the list of recipients and their contact information. It supports group management by organizational structure and allows setting different receiving strategies for different time slots. All sending records are stored in the database, including fields such as sending time, recipient, message content, and sending status, providing message tracking and status query functions.

[0066] The 3D visualization module is implemented based on WebGL technology and uses the Cesium open-source framework to build the 3D scene. Terrain data uses a digital elevation model with an accuracy of no less than 5 meters, overlaid with high-resolution orthophoto maps. The 3D model of the photovoltaic site is constructed using BIM technology, including detailed models of the support foundation, supporting structure, photovoltaic modules, etc., and the model format is glTF standard. Deformation data visualization uses a color-coding scheme: blue indicates normal status, yellow indicates blue warning, orange indicates yellow warning, and red indicates red warning.

[0067] The rendering engine enables real-time exaggerated deformation display, amplifying deformation displacement by 10 to 50 times through vertex offset while preserving the original coordinate mesh reference system. The spatiotemporal data playback control supports retracing historical deformation processes, with an adjustable speed range from 0.1x to 10x real-time speed. The warning area display uses a semi-transparent color block overlay technique, generating a continuous color patch map from discrete point displacement data based on the Kriging interpolation algorithm.

[0068] Interactive features include view control, layer management, attribute querying, and profile analysis. The profile analysis tool can generate displacement curves along arbitrary profile lines and supports multi-period data comparison. The system provides a real-time data push interface based on the WebSocket protocol, ensuring data update latency in the 3D scene is less than 1 second. All visualization components adopt a responsive design, supporting access from desktop browsers and mobile devices, with cross-platform compatibility including major browsers such as Chrome, Firefox, and Safari.

[0069] The random forest algorithm is implemented using the Scikit-learn machine learning library and contains 100 decision trees, each with a maximum depth of 20. The node splitting criterion adopts the mean squared error reduction criterion. Input features include 28 dimensions such as tilt change rate, first and second difference of GNSS three-dimensional coordinate time series, amplitude of the five main frequency components of vibration spectrum, and cumulative rainfall.

[0070] Feature importance was evaluated using the Gini importance score. Each tree was trained using 66% of the samples sampled by Bootstrap and 70% of the features randomly selected. Gradient boosting decision trees were implemented using XGBoost with a learning rate of 0.1, a maximum tree depth of 6, a subsampling ratio of 0.8, and a regularization parameter λ of 1.

[0071] The objective function includes a mean squared error loss term and an L2 regularization term. The optimization process employs a greedy algorithm to construct an additive model, adding a regression tree in each iteration to fit the negative gradient of the current model. Model validation uses 10-fold time series cross-validation, with evaluation metrics including mean absolute error, root mean square error, and coefficient of determination. Hyperparameter optimization uses a Bayesian optimization method, iterating 50 times to find the optimal parameter combination. Model interpretability analysis uses the SHAP value method to calculate the contribution of each feature to the prediction result. Model deployment uses the ONNX open neural network exchange format to ensure cross-platform inference consistency. The inference engine supports CPU and GPU acceleration, with a single prediction time not exceeding 10 milliseconds.

[0072] The data storage and management submodule adopts a hybrid database architecture. Time-series data is stored in the InfluxDB time-series database, with a retention policy of storing raw data for 180 days and downsampled data for 5 years. Relational data uses a PostgreSQL database to store structured data such as device information, alert rules, and user data. Document data uses MongoDB to store unstructured reports and configuration files. The data schema design follows time-series data standards, with each data point containing a timestamp, measurement value, quality indicator, and device identifier.

[0073] Data compression employs a lossy compression algorithm, with a revolving door compression algorithm threshold set to 0.5%. Data indexing utilizes a multi-level indexing strategy, including time range indexes, device identifier indexes, and alert level indexes. The query interface provides a RESTful API, supporting queries based on multiple conditions such as time range, device ID, and alert level. Return formats include JSON, CSV, and Excel. Data export functionality includes batch export and scheduled automatic export. The automatic export task can be configured to execute daily at midnight, exporting the full amount of data from the previous day.

[0074] Data backup employs a combined strategy of full and incremental backups. Full backups are performed weekly, while incremental backups are performed every 6 hours. Backup data is stored off-site. Data security management includes authentication, access control, and operation auditing functions, and database operation logs are retained for 365 days.

[0075] Wired communication adopts an industrial Ethernet architecture with a star topology. The core switch is a gigabit Ethernet switch supporting the IEEE 1588 precision clock protocol. Armored optical cable is used as the transmission medium, with an LC-LC single-mode fiber interface, and a transmission distance of up to 20 kilometers. In wireless communication, the 4G network uses LTE Cat-1 modules, supporting both FDD-LTE and TDD-LTE dual-mode, with a maximum downlink speed of 10 Mbps and an uplink speed of 5 Mbps. The 5G network uses industrial-grade 5G modules, supporting both NSA and SA networking, operating in frequency bands including n1 / n3 / n28 / n41 / n78, and equipped with high-gain omnidirectional antennas. The LoRa network uses a star topology, with an 8-channel receiver at the gateway, supporting adaptive adjustment of the SF7 to SF12 spreading factor, and a transmission distance of up to 5 kilometers.

[0076] The network communication protocol at the application layer uses the MQTT protocol with a QoS level of 1 and a 60-second heartbeat interval to maintain the connection. Data transmission encryption uses the TLS 1.2 protocol, and the certificate system adopts the X.509 standard. Network management functions include link quality monitoring, traffic statistics, fault alarms, and real-time monitoring of network latency, packet loss rate, signal strength, and other indicators. Remote device management supports firmware upgrades and parameter configuration via the LwM2M protocol, with support for resumeable interruptions and version rollback during the upgrade process. The communication module has a watchdog mechanism that automatically restarts when communication is interrupted, with a maximum of 10 reconnections. The reconnection interval uses an exponential backoff algorithm, gradually increasing from 1 second to 60 seconds.

[0077] Example 2

[0078] The detailed workflow of the real-time monitoring and early warning system for slope deformation at photovoltaic sites described in this invention is as follows.

[0079] The workflow of this invention begins with a multi-source sensor network deployed on the slope of a photovoltaic site. This network consists of tilt sensors, GNSS displacement monitoring stations, rain gauges, and vibration sensors, and is responsible for continuously collecting raw data such as the tilt angle of the supporting structure, the three-dimensional coordinates of the ground surface, rainfall, and environmental vibration.

[0080] The data acquisition module synchronously receives these heterogeneous data via wired industrial Ethernet or wireless LoRa / 4G / 5G networks and immediately performs preprocessing, including outlier detection based on statistical criteria, missing value imputation using linear interpolation, and microsecond-level timing alignment of all data streams using a precision clock protocol to form a multimodal dataset with a unified time reference.

[0081] Subsequently, the preprocessed high-quality data is transmitted to the data processing and analysis module. This module first applies the Kalman filter algorithm to fuse multi-source data. Its state equation and observation equation model the evolution of the system state (displacement, velocity, acceleration) and the sensor observation relationship, respectively. Noise is filtered out through the recursive prediction and update process, and high-precision deformation displacement estimates are output.

[0082] Next, the module calls the built-in deformation analysis model, which integrates the ARIMA time series algorithm (used to capture the long-term trend and seasonality of displacement data) and the XGBoost machine learning algorithm (whose objective function controls the model complexity through regularization terms to establish a nonlinear mapping between environmental factors and deformation). The model trains and predicts the fused displacement sequence, and finally generates the predicted value of deformation displacement and the confidence interval of the trend in the future.

[0083] These real-time and predicted deformation data are then sent to the early warning judgment module. This module has a built-in multi-level threshold rule base that supports dynamic adjustment (the initial threshold value is calculated based on geotechnical parameters and can be automatically optimized according to factors such as cumulative rainfall). It uses a fuzzy inference system to comprehensively evaluate instantaneous displacement value, displacement rate, acceleration and environmental factors. If any early warning threshold (such as blue, yellow or red) is triggered, a structured early warning signal is immediately generated.

[0084] The signal is received by the early warning information management module, which, according to a predefined release strategy (e.g., real-time SMS push and API call for red alerts, and emails sent every 10 minutes for yellow alerts), reliably distributes the early warning information to the terminal devices of designated users, 3D visualization platforms, and external management systems through message middleware in a hierarchical and channel-specific manner.

[0085] Meanwhile, the 3D visualization platform uses WebGL technology to render real-time deformation data and early warning information on the oblique photogrammetry model. It intuitively presents the risk area through color coding and displacement exaggeration display technology, and supports historical data backtracking analysis.

[0086] The entire process is repeated cyclically, forming a closed-loop automated system from physical data perception, intelligent analysis and judgment to precise information feedback, realizing continuous monitoring and forward-looking early warning of the stability status of photovoltaic slopes.

[0087] Finally, the following points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms "installation", "connection", and "linkage" should be interpreted broadly, and can be mechanical or electrical connections, or internal connections between two components, or direct connections. "Up", "down", "left", "right", etc. are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may change.

[0088] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0089] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A real-time monitoring and early warning system for slope deformation in photovoltaic sites, characterized in that, The system includes: The data acquisition module is used to acquire multi-source monitoring data related to slope stability through a sensor array deployed in the photovoltaic site area; The data processing and analysis module is communicatively connected to the data acquisition module. It is used to receive the multi-source monitoring data and perform time-series alignment and feature extraction on the data using a preset fusion algorithm. Based on this, it calculates the slope deformation displacement and deformation trend data through a built-in deformation analysis model. The early warning determination module, coupled to the data processing and analysis module, is used to compare the deformation displacement and deformation trend data with the preset multi-level early warning thresholds in real time, and generate an early warning signal containing early warning level and location information when any early warning threshold is determined to be triggered. The early warning information management module is communicatively connected to the early warning determination module. It is used to receive the early warning signal and distribute the early warning information to designated user terminals or external systems based on a predefined distribution strategy.

2. The real-time monitoring and early warning system for slope deformation in photovoltaic sites according to claim 1, characterized in that, The multi-source monitoring data acquired by the data acquisition module includes at least two of the following: tilt angle data of the support structure measured by the tilt sensor, three-dimensional coordinate data of the ground surface collected by the GNSS displacement monitoring station, and rainfall and vibration data collected by the environmental sensor.

3. The real-time monitoring and early warning system for slope deformation in photovoltaic sites according to claim 1, characterized in that, The built-in deformation analysis model in the data processing and analysis module is a combination model that integrates time series analysis algorithm and machine learning regression algorithm. This model takes the multi-source monitoring data as input and outputs the predicted value of slope deformation displacement and its changing trend within a set time period in the future.

4. The real-time monitoring and early warning system for slope deformation in photovoltaic sites according to claim 1, characterized in that, The pre-set multi-level early warning thresholds in the early warning determination module include at least a blue early warning threshold, a yellow early warning threshold, and a red early warning threshold. Each threshold is set differently according to the geological conditions and slope design parameters of the photovoltaic site. The early warning determination module is configured to support dynamic adjustment and updating of the multi-level early warning thresholds via remote commands.

5. The real-time monitoring and early warning system for slope deformation in photovoltaic sites according to claim 1, characterized in that, The release strategy of the early warning information management module supports selecting different information push frequencies and at least one push channel according to the level of the early warning. The push channels include sending data packets to the external management system via SMS, email, or API interface.

6. The real-time monitoring and early warning system for slope deformation at photovoltaic sites according to claim 1, characterized in that, The system also includes a three-dimensional visualization module, which is connected to the data processing and analysis module and the early warning information management module. This module is used to render and display the deformation state of the slope and the spatial location of the early warning area in the three-dimensional terrain model in real time based on the deformation displacement, deformation trend data and early warning signals.

7. The real-time monitoring and early warning system for slope deformation in photovoltaic sites according to claim 3, characterized in that, The machine learning regression algorithm is one of the random forest algorithm or gradient boosting decision tree algorithm, used to establish a nonlinear mapping relationship between monitoring data and deformation displacement.

8. The real-time monitoring and early warning system for slope deformation at photovoltaic sites according to claim 1, characterized in that, The system also includes a data storage and management submodule, which is used to structure and store historical monitoring data, analysis results records and early warning event logs, and provides data query and export functions based on time range or early warning level.

9. The real-time monitoring and early warning system for slope deformation in photovoltaic sites according to claim 1, characterized in that, The system interacts with the sensor group and user terminal via a wired or wireless communication network, wherein the wireless communication network includes one of the following: fourth-generation mobile communication technology network, fifth-generation mobile communication technology network, or LoRa wireless network.

Images