Photovoltaic module logistics intelligent monitoring method and system based on multiple sensors

Abstract

The invention relates to the technical field of logistics intelligent monitoring, and discloses a photovoltaic module logistics intelligent monitoring method and system based on multiple sensors. The method comprises the steps that multi-source data streams such as temperature and humidity, vibration and position are collected in real time, abnormal path segments are marked, environment abnormal parameters are screened, and an initial data set is formed through integration; de-noising and screening are carried out through edge calculation local preprocessing to obtain a refined data set; extracting feature modeling to capture a data change trend, activating an alarm, generating a label, enhancing a weak signal and positioning damage when the data change trend is abnormal; after a target image is matched, obtaining an enhanced feature map through a convolutional neural network and an abnormal mask, and clustering to confirm an abnormal event; and finally, constructing a tracing data chain, and generating monitoring output in combination with a vehicle running log. According to the method, the monitoring real-time performance and the abnormity identification precision are improved, full-link tracing is realized, and the transportation safety of the photovoltaic module is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of intelligent logistics monitoring technology, and in particular to a method and system for intelligent monitoring of photovoltaic module logistics based on multiple sensors. Background Technology

[0002] Against the backdrop of accelerated global energy transition, the photovoltaic industry, as a core pillar of clean energy, faces higher demands on the efficiency and security of all links in the industrial chain due to its large-scale development. As the core component of photovoltaic systems, photovoltaic modules are directly related to product quality stability and project progress efficiency in their logistics and transportation, making them a crucial link in ensuring the sustainable development of the photovoltaic industry. With the improvement of logistics informatization, real-time monitoring, anomaly warning, and traceability of photovoltaic module transportation through technological means have become an urgent need within the industry.

[0003] Currently, existing technologies in the logistics monitoring field mostly employ sensor networks to collect environmental and location data during transportation, combined with big data collection and preprocessing techniques for preliminary analysis to achieve basic monitoring of cargo status. For example, some solutions use temperature and humidity sensors and vibration sensors to collect environmental parameters, and use simple data filtering algorithms to filter out abnormal data exceeding thresholds; other solutions use GPS positioning to monitor transportation routes, and complete basic traceability through cloud data storage and backtracking. These technologies have been applied to some extent in conventional logistics scenarios, but their design ideas mostly focus on the collection and simple processing of single-dimensional data, lacking dynamic adaptability to complex transportation environments.

[0004] During the transportation of photovoltaic modules, environmental factors are complex and constantly changing. The multi-source data streams collected by sensors are not only large in volume and multi-dimensional, but also contain a large amount of noise data caused by equipment vibration and road condition fluctuations. In existing technologies, the big data collection and preprocessing stages mostly rely on centralized cloud processing, resulting in high data transmission latency, which makes it difficult to meet the needs of real-time monitoring. At the same time, preprocessing methods are mostly simple threshold screening or noise filtering, failing to deeply explore the correlation characteristics between multi-source data and unable to effectively distinguish between valid abnormal signals and noise interference.

[0005] This also makes it difficult for existing technologies to accurately capture weak abnormal signals hidden in massive amounts of noise data, and to identify potential damage during the transportation of photovoltaic modules in a timely and accurate manner. As a result, the rate of missed and false alarms of abnormal events is high, and it is difficult to trace the responsibility in the future. Therefore, existing technologies have the problem of low accuracy in identifying abnormal events. Summary of the Invention

[0006] This invention provides a method and system for intelligent monitoring of photovoltaic module logistics based on multiple sensors, which solves the problems of poor real-time performance, low accuracy of anomaly identification, and difficulty in tracing responsibility caused by the reliance on centralized cloud processing for big data collection and preprocessing in the prior art. At the same time, it realizes local real-time processing of multi-source data of photovoltaic module transportation, accurate capture of weak abnormal signals, and full-link traceability, so as to ensure the safety of photovoltaic module transportation.

[0007] In a first aspect, to address the aforementioned technical problems, this invention provides a multi-sensor-based intelligent monitoring method for photovoltaic module logistics, comprising: Real-time acquisition of multi-source data streams and screening of abnormal environmental parameters and abnormal path segments; integration of the abnormal environmental parameters and abnormal path segments to form an initial dataset. Based on the initial dataset, a refined dataset is obtained through local preprocessing using edge computing. Key environmental features are extracted from the refined dataset to form a feature vector set. Based on the feature vector set, the data change trend under sudden situations is analyzed to obtain a sequence pattern. The sequence pattern is compared with a preset normal pattern, and combined with the location data in the multi-source data stream, potential damage locations are filtered and marked to obtain the final anomaly label; The target image captured by the high-definition camera is acquired and matched with the final anomaly label. After the match is successful, the anomaly region is enhanced using a convolutional neural network to obtain an enhanced feature map. Cluster analysis is performed on the enhanced feature map to screen preliminary abnormal events and fuse them with the sudden pattern sequence in the transportation process to obtain confirmed abnormal events; Based on the confirmed abnormal events, transportation routes and cargo status information are extracted, a traceability data chain is constructed, high-risk routes are identified, and abnormal details are integrated with vehicle operation logs to generate the final monitoring output.

[0008] Secondly, the present invention provides a multi-sensor-based intelligent monitoring system for photovoltaic module logistics, comprising: The data acquisition module is used to collect multi-source data streams in real time and filter out abnormal environmental parameters and abnormal path segments from them, and integrate the abnormal environmental parameters and abnormal path segments to form an initial dataset; The data refining module is used to perform local preprocessing on the initial dataset through edge computing to obtain a refined dataset; The sequence modeling module is used to extract key environmental features from the refined dataset to form a feature vector set, and to analyze the data change trend under sudden situations based on the feature vector set to obtain the sequence pattern. The label generation module is used to compare the sequence pattern with a preset normal pattern, combine the location data in the multi-source data stream, filter and mark potential damage locations, and obtain the final abnormal label. The feature extraction module is used to acquire the target image captured by the high-definition camera and match it with the final anomaly label. After the match is successful, the convolutional neural network is used to perform feature enhancement processing on the anomaly region to obtain an enhanced feature map. The cluster confirmation module is used to perform cluster analysis on the enhanced feature map, screen preliminary abnormal events, and fuse them with the sudden pattern sequence in the transportation process to obtain confirmed abnormal events. The traceability output module is used to extract transportation route and cargo status information based on the confirmed abnormal events, construct a traceability data chain and determine high-risk routes, integrate abnormal details with vehicle operation logs, and generate the final monitoring output.

[0009] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention performs local preprocessing of the initial dataset through edge computing, and combines Kalman filtering for noise reduction and support vector machine for filtering synchronous abnormal data. This avoids the transmission delay of centralized processing in the cloud, while filtering out invalid noise and retaining key features, effectively improving the real-time performance of data processing and laying the foundation for accurate identification of anomalies in the future.

[0010] (2) This invention performs multi-layer convolution and pooling on the matched standardized image through a convolutional neural network, and then integrates anomaly mask to enhance the features of the abnormal region, highlighting the features of the component’s subtle damage. This solves the problem that the existing technology is difficult to capture weak abnormal signals, significantly improves the sensitivity of identifying subtle damage and potential risks on the surface of photovoltaic modules, reduces the false alarm and false alarm rates, and realizes early warning of hidden damage.

[0011] (3) By confirming abnormal events and constructing a traceability data chain, this invention integrates abnormal details with vehicle operation logs, which can clarify the time, location and cause of the abnormality, and also identify high-risk paths, realize full-link traceability of abnormalities, and provide a reliable basis for subsequent transportation route optimization and responsibility definition. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of a method for intelligent monitoring of photovoltaic module logistics based on multiple sensors, provided in the first embodiment of the present invention. Figure 2 This is a schematic diagram of a photovoltaic module logistics intelligent monitoring system based on multiple sensors, provided in the second embodiment of the present invention. Detailed Implementation

[0013] 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.

[0014] Reference Figure 1 The first embodiment of the present invention provides a method for intelligent monitoring of photovoltaic module logistics based on multiple sensors, including the following steps: S11, collect multi-source data streams in real time and filter out abnormal environmental parameters and abnormal path segments from them, and integrate the abnormal environmental parameters and the abnormal path segments to form an initial dataset; S12, Based on the initial dataset, perform local preprocessing through edge computing to obtain a refined dataset; S13, extract key environmental features from the refined dataset to form a feature vector set, and analyze the data change trend under the sudden situation based on the feature vector set to obtain the sequence pattern; S14, compare the sequence pattern with the preset normal pattern, combine the location data in the multi-source data stream, filter and mark potential damage locations, and obtain the final abnormal label; S15, acquire the target image captured by the high-definition camera and match it with the final anomaly label. After the match is successful, use a convolutional neural network to perform feature enhancement processing on the anomaly region to obtain an enhanced feature map. S16, perform cluster analysis on the enhanced feature map, screen preliminary abnormal events and fuse them with the sudden pattern sequence in the transportation process to obtain confirmed abnormal events; S17. Based on the confirmed abnormal events, extract the transportation route and cargo status information, construct a traceability data chain and determine high-risk routes, integrate abnormal details with vehicle operation logs, and generate the final monitoring output.

[0015] In step S11, multi-source data streams are collected in real time, and environmental anomaly parameters are filtered from them. The environmental anomaly parameters are integrated with the anomaly path segments to form an initial dataset, including: Temperature and humidity data, vibration data, and location data are acquired in real time through a sensor network on the transport vehicle, forming a multi-source data stream; wherein the sensor network includes temperature and humidity sensors, vibration sensors, and GPS positioning devices; If the deviation exceeds the preset path threshold, the corresponding road segment is marked as an abnormal road segment. Temperature and humidity data and vibration data corresponding to the photovoltaic modules are extracted from the abnormal path segments to obtain the environmental parameter change sequence. Data exceeding a preset environmental parameter change threshold in the environmental parameter change sequence are selected as environmental anomaly parameters. The environmental anomaly parameters are then integrated with the anomaly path segments to obtain an initial dataset.

[0016] It is worth noting that, firstly, a multi-source data stream is formed by acquiring real-time temperature and humidity data, vibration data, and location data through a sensor network deployed on the transport vehicles. This sensor network includes an SHT35 temperature and humidity sensor, a DS-200 vibration sensor, and a Ublox NEO-7M GPS positioning device. The SHT35 temperature and humidity sensor has a measurement range of -40℃ to 125℃ and a humidity range of 0 to 100%RH, with a sampling frequency set to once every 5 seconds to ensure the capture of subtle environmental fluctuations. The DS-200 vibration sensor has a measurement range of 0 to 10 m / s², an accuracy of ±0.05 m / s², and a sampling frequency of once every 2 seconds. The vibration data refers to the acceleration signals collected by the vibration sensor. Physically, acceleration is a physical quantity that describes the rate of change of an object's velocity. Its magnitude directly reflects the severity of the impact force on the photovoltaic modules during transportation. By measuring the instantaneous acceleration changes generated by the vehicle during transportation, the vibration sensor can capture the mechanical impact on the photovoltaic modules caused by events such as road bumps, sudden braking, and crossing speed bumps. The Ublox NEO-7M GPS positioning device achieves a positioning accuracy within 1 meter, updates every 1 second, and synchronously records vehicle location information. All sensors are equipped with high-precision hardware clocks, which are calibrated every 100 milliseconds via clock synchronization commands from the onboard main control unit, ensuring that the timestamp deviation of multi-source data is ≤1 millisecond.

[0017] Based on a comparison of location data from multi-source data streams with a preset transportation route, the preset route threshold is set at 500 meters. This threshold is based on the precision transportation route deviation standard set in the electronic logistics industry. If the deviation exceeds 500 meters, the corresponding road segment is marked as an abnormal route segment. For example, if a transport vehicle is originally scheduled to travel from point A to point B along a highway, and GPS data shows that the vehicle deviates 600 meters from the preset route at a certain moment, the system automatically marks the road segment corresponding to that time period from 10:10 to 10:15 as an abnormal route segment.

[0018] Next, the InfluxDB time series database was used to store and extract data. This database supports fast querying of high-frequency time series data and can accurately extract corresponding data according to the time interval of abnormal path segments. For example, for the abnormal path segment from 10:10 to 10:15, a continuous change sequence of temperature rising from 25℃ to 30℃, humidity decreasing from 60% to 45%, and vibration peak increasing from 3m / s² to 4.5m / s² was extracted.

[0019] Then, data exceeding preset environmental parameter change thresholds in the environmental parameter change sequence are selected as environmental anomaly parameters. The preset temperature threshold is 20 to 30℃, the humidity threshold is 40 to 70%RH, and the vibration peak threshold is 4 m / s². These threshold ranges are determined based on impact resistance tests of photovoltaic module glass cover plates and temperature and humidity tolerance tests of encapsulation materials. For example, if the vibration peak of 4.5 m / s² exceeds 4 m / s² and the temperature of 30℃ reaches the upper limit of the threshold in the above sequence, these two parameters are recorded as environmental anomaly parameters.

[0020] Finally, the environmental anomaly parameters were integrated with the time and location information of the anomaly path segment to obtain the initial dataset. This dataset includes the anomaly path segment from 10:10 to 10:15, located at 120.5 degrees east longitude and 30.2 degrees north latitude, with anomaly parameters of vibration of 4.5 m / s² and temperature of 30℃.

[0021] In step S12, the step of performing local preprocessing based on the initial dataset using edge computing to obtain a refined dataset includes: Based on the initial dataset, the original vibration signal and temperature and humidity signal are extracted from it through edge nodes to form an original signal set; The original signal set is filtered by Kalman filtering to obtain a preliminary filtered signal. If the vibration peak value in the preliminary filtered signal exceeds a preset vibration threshold, the corresponding vibration peak value sequence is extracted; if it does not exceed the threshold, the corresponding preliminary filtered signal is discarded to obtain a vibration feature sequence. The synchronization integrity between the vibration feature sequence and temperature and humidity fluctuations is analyzed by support vector machine, and data with synchronization anomalies are retained to obtain a refined dataset.

[0022] It should be noted that, firstly, based on the initial dataset, raw vibration and temperature / humidity signals are extracted using an edge gateway deployed on the transport vehicle to form the raw signal set. The edge gateway is equipped with an ARM Cortex-A9 processor, supporting local real-time data processing, which avoids the latency issues of data transmission to the cloud, with a processing response time of ≤50 milliseconds.

[0023] Next, the original signal set was filtered using Kalman filtering to obtain a preliminary filtered signal. The process noise covariance Q of the Kalman filter was set to 0.01, and the measurement noise covariance R was set to 0.1. This parameter combination was tested with 100 sets of noise data and can effectively filter out high-frequency noise generated by sensor jitter and vehicle engine vibration. For example, the original vibration signal contained high-frequency noise of 1.2 m / s², and after filtering, the signal was smoothed, and the vibration peak value was stabilized at 4.5 m / s². Fluctuations in temperature and humidity signals of ±0.5℃ and ±2%RH were also eliminated.

[0024] It should be noted that the preset vibration threshold is 4 m / s², which is determined based on the impact resistance test of the core structure of the photovoltaic module. If the vibration peak value in the preliminary filtered signal exceeds 4 m / s², the corresponding vibration peak value sequence is extracted; otherwise, the corresponding preliminary filtered signal is discarded, resulting in a vibration feature sequence. For example, if the peak value of the filtered vibration signal is 4.5 m / s², exceeding the threshold, the vibration peak values ​​every 2 seconds within that time period are extracted to form the sequence 4.0 m / s², 4.2 m / s², 4.5 m / s², 4.3 m / s², and 4.4 m / s². If the vibration peak value of a certain segment of the preliminary filtered signal is all 3.8 m / s², which does not exceed the threshold, that segment of signal is directly discarded.

[0025] Then, the synchronization integrity of vibration feature sequences and temperature and humidity fluctuations was analyzed using Support Vector Machines (SVMs), retaining data with synchronization anomalies to obtain a refined dataset. The core hyperparameters of the SVM were determined through a combination of grid search and 5-fold cross-validation, commonly used in the field. Considering the multi-dimensional parameter characteristics of photovoltaic module transportation scenarios, a Radial Basis Function (RBF) kernel was chosen to efficiently handle the nonlinear correlation between vibration and temperature / humidity data. The penalty coefficient C was set to 10, a value that balances overfitting and underfitting, ensuring a synchronization anomaly detection accuracy of no less than 90%. The gamma value of the RBF kernel was set to 1, accurately distinguishing between valid anomalies of synchronized vibration and temperature / humidity fluctuations and asynchronous interference from isolated vibration exceeding limits. The training data for the SVM came from 1000 sets of historical data from different photovoltaic module transportation conditions, with 800 sets used as the training set and 200 sets as the validation set. Under the above hyperparameter configuration, the model achieved a fitting accuracy ≥95%, a synchronization anomaly detection accuracy ≥94%, and an asynchronous anomaly removal rate ≥92% after training, accurately determining the synchronization between parameters. For example, if the temperature rises from 25℃ to 30℃ and the humidity drops from 60% to 45% in the aforementioned vibration characteristic sequence, and the temperature and humidity change synchronously with the vibration, this is identified as a synchronization anomaly, and the data set is retained. If the peak value of a vibration sequence exceeds the standard but the temperature and humidity do not change significantly, it is identified as a non-synchronization anomaly and is discarded. The final refined dataset contains the vibration, temperature, humidity, location, and time information of the synchronization anomalies.

[0026] In step S13, key environmental features are extracted from the refined dataset to form a feature vector set. Based on the feature vector set, the data change trend under sudden situations is analyzed to obtain a sequence pattern, including: The feature vector set is divided into a continuous time series segment set by a sliding window, and the duration of the sliding window can be dynamically adjusted according to the transportation conditions and data collection frequency. Calculate the rate of change of vibration peak value for each time series segment set, and extract characteristic correlation patterns from the vibration data of the time series segment sets; When the rate of change of the vibration peak exceeds a preset rate of change threshold, the time series segment set is input into a long short-term memory network, and a predicted sequence is output. If the vibration peak change rate does not exceed the preset change rate threshold, the analysis is carried out directly based on the temperature and humidity fluctuation data of the time series segment set to obtain real-time data characteristics; By combining the predicted sequence or real-time data features with the correlation patterns of the features, the intrinsic relationship between temperature and humidity fluctuations and vibrations is analyzed to form a sequence pattern that reflects the trend of data changes under sudden situations.

[0027] Vibration peak sequence and temperature and humidity fluctuation sequence were extracted from the refined dataset to form a feature vector set. Each feature vector contains three dimensions of data: vibration peak, temperature value, and humidity value. For example, the feature vector for the time 10:10 is 4.0 m / s², 25℃, and 60%.

[0028] It's worth noting that, firstly, when dividing the feature vector set into continuous time series segments using a sliding window, the duration of the sliding window can be dynamically adjusted based on road conditions and data collection frequency. Under smooth highway conditions, the window duration is set to 10 minutes; under bumpy mountain road conditions, it's set to 5 minutes. The data collection frequency is once every 2 seconds, and each window contains 150 to 300 feature vectors. For example, on a bumpy road segment from 10:10 to 10:15, a 5-minute window is used to extract all feature vectors within that time period, forming a time series segment set.

[0029] Next, it is necessary to calculate the rate of change of vibration peak value for each time series segment, and simultaneously extract characteristic correlation patterns from the vibration data. The rate of change of vibration peak value is calculated by dividing the difference in vibration peak values ​​between two adjacent characteristic vectors by the time interval. For example, if the vibration peak value changes from 4.0 m / s² to 4.2 m / s² between 10:10 and 10:12, with a time interval of 2 seconds, the calculated rate of change is 0.1 m / s² / second. The characteristic correlation patterns are obtained through analysis of historical data, showing a correlation between an average temperature increase of 2°C and an average humidity decrease of 5% for every 1 m / s² increase in vibration peak value.

[0030] It should be noted that the preset peak vibration rate of change threshold is 0.2 m / s² / second. This threshold is determined based on the photovoltaic module transportation vibration risk test. Through 100 sets of module transportation tests under different road conditions, when the peak vibration rate of change is ≤0.2 m / s² / second, the maximum deformation of the module aluminum frame is ≤0.1 mm, and no cracks are generated in the glass cover, with a damage probability of <5%. When this threshold is exceeded, the probability of aluminum frame deformation exceeding 0.1 mm or microcracks appearing in the glass cover increases sharply to over 85%, which meets the safety limit for vibration rate of change.

[0031] When the rate of change of vibration peak exceeds the threshold, the predicted sequence is obtained by training the time series segments using a Long Short-Term Memory (LSTM) network. The training data for the LSTM network comes from 800 sets of photovoltaic module transportation time-series data, covering 12 typical operating conditions such as highways, mountain roads, and rainy weather transportation. Each set of data contains a continuous 10-minute time-series sequence of vibration peak, temperature, and humidity. After denoising and normalization preprocessing, the data is divided into a training subset and a validation subset in a 7:3 ratio, determined based on industry practice for time-series data modeling and the amount of data in this invention. The LSTM network consists of 8 neurons in the input layer, 16 neurons in the hidden layer, and 1 neuron in the output layer. Mean squared error is used as the loss function, and the model is trained iteratively for 500 rounds, resulting in a prediction error ≤3% after training. For example, if the rate of change of vibration peak in a certain time series segment is 0.25 m / s² / second, exceeding the threshold, the model predicts that the vibration peak will rise to 4.6 m / s² and the temperature will rise to 31°C within the next 10 seconds.

[0032] If the rate of change of the vibration peak does not exceed the preset threshold, the analysis is directly carried out based on the temperature and humidity fluctuation data of the time series segment set to obtain real-time data characteristics. For example, the rate of change of the vibration peak in the period from 10:10 to 10:15 is 0.1 m / s² / second, which does not exceed the threshold. The analysis of temperature and humidity fluctuations shows that the temperature increases by 1°C every 2 minutes and the humidity decreases by 3% every 2 minutes, forming real-time data characteristics.

[0033] Then, by combining the characteristics of the predicted sequence or real-time data with the correlation patterns of these characteristics, the intrinsic relationship between temperature and humidity fluctuations and vibration is analyzed to form a sequence pattern. For example, by combining the characteristics of real-time data with the correlation between vibration, temperature, and humidity, the resulting sequence pattern shows that the vibration peak rises slowly, accompanied by a gradual increase in temperature and a gradual decrease in humidity. This pattern reflects the trend of environmental data under smooth and bumpy road conditions.

[0034] In step S14, comparing the sequence pattern with a preset normal pattern, and combining the location data in the multi-source data stream to filter and mark potential damage locations to obtain the final anomaly label includes: Calculate the deviation value between the sequence pattern and the preset normal pattern. When the deviation value exceeds the preset deviation threshold, activate the alarm mechanism and generate a preliminary abnormal label containing the timestamp of the abnormality. Based on the time interval corresponding to the preliminary anomaly label, weak signals are extracted from massive amounts of raw data, and the weak signals are enhanced through signal filtering to obtain enhanced signals; The enhanced signal is aligned with the location data collected at the same time by using a timestamp to determine the transportation location of the photovoltaic module corresponding to the enhanced signal. Based on the transportation location of the photovoltaic modules, the potential damage coordinates of the photovoltaic modules are marked by location mapping. The timestamp of the initial anomaly label, the feature information of the enhanced signal, and the coordinates of the potential damage are integrated to form a final anomaly label containing the anomaly time, location, and signal features.

[0035] It is worth noting that, firstly, the deviation between the sequence mode and the preset normal mode is calculated. The preset normal mode has a vibration peak of 2 to 3 m / s², a temperature of 23 to 27°C, and a humidity of 50 to 60%RH. When calculating the deviation, the three parameters are first subjected to Min-Max normalization to eliminate the difference in dimensions, and then the deviation is calculated using the Euclidean distance formula.

[0036] It should be noted that the preset deviation threshold is 0.5, which was determined statistically from 500 sets of sequence patterns in normal transportation scenarios. By collecting normal transportation sequence patterns under different seasons and road conditions, the Euclidean distance between each pattern and the standard normal template was calculated, and the 90th percentile was set to 0.5. That is, 90% of the normal transportation sequence patterns have a deviation value ≤ 0.5. When the deviation value exceeds this value, the probability of the sequence pattern corresponding to an abnormal transportation state is ≥ 92%, which can effectively balance the accuracy of anomaly identification and the false alarm rate.

[0037] When the deviation exceeds 0.5, the alarm mechanism is activated. This mechanism consists of two parts: an in-vehicle audible and visual alarm and a remote push of alarm information to the monitoring center. The audible and visual alarm loudness is ≥80 dB, and the remote push response time is ≤10 seconds. Simultaneously, a preliminary anomaly tag with a timestamp of the anomaly occurrence is generated. For example, if the deviation between the above sequence pattern and the normal pattern is 0.8, exceeding the threshold, the alarm is activated, and a preliminary anomaly tag with a timestamp of 10:15 is generated.

[0038] Next, based on the time interval of 10:05 to 10:25 corresponding to the initial anomaly label, weak signals were extracted from the massive amount of raw data. The massive amount of raw data included all raw data collected by the sensor during this period, totaling 3000 sets. The weak signals were enhanced by wavelet filtering, with the wavelet basis function selected as db4 and the decomposition level set to 3. This enhanced the original weak vibration signal of 0.2 m / s² to 0.5 m / s², and the weak fluctuations of 0.3℃ and 0.5%RH in the temperature and humidity signals were also clearly extracted, resulting in enhanced signals.

[0039] Then, the enhanced signal is aligned with the location data collected at the same time using timestamps to establish a spatiotemporal correlation. For example, the enhanced signal at 10:15 corresponds to a GPS location of 120.5 degrees east longitude and 30.2 degrees north latitude, and the photovoltaic module transportation location corresponding to this enhanced signal is clearly identified as this coordinate point.

[0040] Based on spatiotemporal correlation, the potential damage coordinates of photovoltaic modules can be marked by location mapping. First, the transport compartment is divided into grids according to the photovoltaic module loading specifications. The photovoltaic modules are arranged closely in 3 columns horizontally and 5 rows vertically within the transport compartment. Each module measures 1.6m × 1m. The loading area within the compartment is 8m long and 3m wide. Therefore, the loading area is divided into 5 rows (1.6m long) and 3 columns (1m wide) horizontally, forming 15 grid areas (5 rows, 3 columns). Each grid area has a physical size of 1.6m × 1m, perfectly matching the size of a single photovoltaic module. Next, one vibration sensor is deployed at the center of each grid area on the bottom of the compartment, for a total of 15 sensors. The sensor numbers correspond one-to-one with the grid row and column identifiers; for example, the sensor number S2-3 corresponds to the 2nd row, 3rd column grid. Finally, by combining the vibration intensity data in the enhanced signal and comparing the amplitude of the enhanced vibration signals of 15 sensors, the sensor with the strongest vibration, such as S2-3, was identified. The corresponding grid area, namely the second row and third column, is the potential damage area. At the same time, the physical coordinates of this grid area were recorded. With the left corner of the front of the carriage as the origin, the area in the second row and third column corresponds to the range of 3.2-4.8m on the X-axis and 2-3m on the Y-axis. The fixed number of the photovoltaic module in this area, such as A0203, was also associated with it. Finally, the coordinates of the potential damage were determined to be the grid in the second row and third column, the physical coordinates X 3.2-4.8m / Y 2-3m, and the module number A0203.

[0041] Finally, the timestamps of the initial anomaly labels, the characteristic information of the enhanced signals, and the coordinates of potential damage are integrated to form the final anomaly labels. The final anomaly label includes the anomaly time 10:15, the anomaly location 120.5 degrees east longitude and 30.2 degrees north latitude, the component number A0203, the characteristic vibration of the enhanced signal 0.5 m / s², the temperature 30℃, the humidity 45%, and the region in the 2nd row and 3rd column of the potential damage coordinates.

[0042] In step S15, the target image captured by the high-definition camera is acquired and matched with the final anomaly label. After successful matching, a convolutional neural network is used to perform feature enhancement processing on the anomaly region to obtain an enhanced feature map, including: Acquire target images captured by high-definition cameras, and then adjust the image resolution to obtain standardized images; If the normalized image matches the anomaly type and damage coordinates of the final anomaly label, then a multi-layer convolution operation is performed on the normalized image through a convolutional neural network to obtain a preliminary feature map; The preliminary feature map is downsampled using pooling filtering to obtain a compressed feature map. An anomaly mask corresponding to the final anomaly label is generated, and the compressed feature map is fused with the anomaly mask to obtain an enhanced feature map.

[0043] It should be noted that the sensor network and the high-definition camera are synchronized using the same high-precision clock, ensuring precise alignment of the sensor data stream and image data stream on the timeline. The target image is captured by a high-definition camera installed inside the vehicle. This camera has a resolution of 1920×1080 and a capture frequency of 1 frame / second, capable of clearly capturing the surface condition of the photovoltaic modules. For module number A0203, the image at 10:15 is extracted as the target image. Then, the image resolution is standardized using a resizing device, adjusting the target image to a standardized 1024×768 image. During the adjustment process, the aspect ratio of the image is kept constant, and a bilinear interpolation algorithm is used to ensure that the image is free from stretching and distortion, and that the details of the module surface are not distorted.

[0044] Next, it is determined whether the anomaly type and damage coordinates of the standardized image match those of the final anomaly label. For example, if the anomaly type is a potential surface crack and the damage coordinates are the region in the 2nd row and 3rd column, the presence of crack features in this region is observed in the standardized image. If fine lines appear in the corner area of ​​component A0203 in the image, a match is determined. If there are no crack features in this region in the image, or if the anomaly type does not match the label description, a match is determined to be unsuccessful. In this case, other target images within the time interval corresponding to the anomaly label need to be extracted again from the refined dataset, prioritizing adjacent frame images within 10 seconds before and after the label timestamp, and the resolution unification and matching operations are performed again. If the images extracted three times consecutively fail to match, the anomaly label is marked as pending verification, and the corresponding data is pushed to the monitoring center for manual verification of the label accuracy and image acquisition integrity to avoid subsequent feature extraction deviations due to single image blurring, camera angle deviation, or label errors.

[0045] During matching, a convolutional neural network (CNN) performs multi-layer convolution operations on the standardized image to obtain preliminary feature maps. During training, a dataset and training strategy adapted to the surface feature extraction of photovoltaic (PV) modules are first constructed. The training data consists of 1500 sets of PV module surface images, covering five common anomaly types (cracks, deformation, stains, etc.) and normal surface images. Each set of images is labeled with the anomaly type and corresponding region coordinates, and divided into training, validation, and test sets in a 7:2:1 ratio, determined based on the amount of PV module surface image data and model training requirements. All images are preprocessed to a 1024×768 resolution to adapt to the network input. The CNN contains three convolutional layers and two pooling layers. The first convolutional layer has a 3×3 kernel size and 64 kernels, with ReLU activation. The second convolutional layer has a 3×3 kernel size and 128 kernels, with ReLU activation. The third convolutional layer has a 5×5 kernel size and 256 kernels, with ReLU activation. The Adam optimizer was used during training, with an initial learning rate of 0.001, decreasing by a factor of 0.1 every 20 epochs. Cross-entropy loss was used as the loss function, and training was iterated for 100 epochs. An early stopping mechanism was introduced: training stopped when the validation set loss did not decrease for five consecutive epochs to avoid overfitting. Batch normalization was performed after each convolutional layer to further improve training stability and model generalization ability. Testing showed that the network achieved an accuracy of ≥93% in extracting abnormal region features on the test set, accurately capturing subtle surface features of components. For example, after the first convolutional layer, the normalized image extracted component edge features; the second convolutional layer extracted surface texture features; and the third convolutional layer extracted subtle texture features in the corner areas, forming a preliminary feature map.

[0046] Next, the initial feature map is downsampled using pooling filtering to obtain a compressed feature map. The pooling filter uses a 2×2 max-pooling window with a stride of 2, which reduces the amount of data while preserving key features. For example, if the initial feature map size is 512×384, the compressed feature map size after downsampling is 256×192, reducing the data volume by 75%, while still clearly preserving edge and texture features.

[0047] Then, an anomaly mask corresponding to the final anomaly label is generated. Anomaly mask size and compression features. Figure 1 For example, with a resolution of 256×192, the region in the 2nd row and 3rd column of the damage coordinates in the final anomaly label is marked as white pixels in the mask, and the remaining regions are marked as black pixels, forming an anomaly mask that only highlights the damage region.

[0048] Finally, pixel-level weighted summation was used in the fusion process, with weights set at 0.6 for the compressed feature map and 0.4 for the anomaly mask. Experimental data showed that at this ratio, the feature signal intensity of the anomaly region was increased by 30% compared to a single feature map, and the probability of misidentifying background noise as an anomaly was controlled within 5%. If the weight of the compressed feature map was lower than 0.6, the original feature details would become blurred; if it was higher than 0.6, the enhancement effect on the anomaly region would be insufficient. If the weight of the anomaly mask was higher than 0.4, it would easily amplify local noise; if it was lower than 0.4, it would fail to highlight the damage location. After fusion, the feature signal intensity of the damaged area was increased by 30%, the fine texture features at the edges and corners were more prominent, and the features of the non-damaged area were suppressed, ultimately resulting in an enhanced feature map that highlighted the anomaly pattern.

[0049] In step S16, the enhanced feature map is subjected to cluster analysis to screen preliminary abnormal events and fused with the sudden pattern sequence during transportation to obtain confirmed abnormal events, including: A set of feature points is extracted from the enhanced feature map, and a density clustering operation is performed on the feature point set to obtain the clustering result; Based on the clustering results, the density peak and boundary size of each cluster are calculated. When the density peak is higher than a preset density threshold and the boundary size exceeds a preset range threshold, an abnormal cluster is determined to be formed, and an abnormal cluster set is formed. Based on real-time vehicle operation data and road condition monitoring data collected during transportation, dynamic adaptation parameters are determined, and the abnormal cluster set and the dynamic adaptation parameters are integrated to obtain preliminary abnormal events. The sudden pattern sequence during transportation is obtained, and the confirmed abnormal event is obtained by combining the preliminary abnormal event with the sudden pattern sequence through data fusion processing.

[0050] It is worth noting that, firstly, when extracting the feature point set from the enhanced feature map, the SIFT algorithm is used to extract the feature points. Each feature point contains three dimensions of information: position, scale, and orientation, covering the key areas in the enhanced feature map.

[0051] Next, density-based clustering is performed on the feature point set to obtain the clustering results. The clustering algorithm used is DBSCAN, with a cluster radius set to 2 pixels and a minimum number of points set to 5. These parameters were tested with 50 sets of feature point data and can accurately cluster adjacent abnormal feature points. For example, 1000 feature points are clustered into 3 clusters, one of which contains 85 feature points concentrated in the corner areas of the enhanced feature map.

[0052] Then, based on the clustering results, the density peak and boundary size of each cluster are calculated. The density peak is the number of feature points per unit area within the cluster, calculated using a grid counting method. For example, the density peak of the corner region cluster mentioned above is 60 feature points per pixel, while the preset density threshold is 48 feature points per pixel. This threshold is derived from the clustering analysis of 100 sets of abnormal photovoltaic module images. It was found that when the density threshold is 48 feature points per square centimeter, noisy clusters can be accurately filtered out, retaining valid abnormal features.

[0053] The boundary size is the proportion of the area occupied by the clustered region in the enhanced feature map. The clustered region accounts for 4% of the total image area, and its preset range threshold is 2%. This threshold is combined with the performance impact judgment of the surface damage of photovoltaic modules. When the proportion of surface damage area is <2%, the efficiency decay is ≤1%, which can be ignored; when it exceeds 2%, the efficiency decay is ≥3%, which affects the normal use of the module. Therefore, 2% is set as the boundary size threshold.

[0054] If the density peak value is higher than 48 and the boundary size exceeds 2%, an abnormal cluster is determined to be formed, and an abnormal cluster set is created. The density peak value of the aforementioned corner region clusters is 60, and the boundary size is 4%, both meeting the threshold, thus determining that an abnormal cluster is formed, and an abnormal cluster set is created.

[0055] Subsequently, dynamic adaptation parameters were determined based on real-time vehicle operation data and road condition monitoring data collected during transportation. The core function of the dynamic adaptation parameters is to adjust the anomaly judgment criteria according to the actual transportation conditions. The vibration and impact intensity experienced by the components varies under different road conditions. Fixed thresholds can easily lead to misjudgment of low-risk conditions or missed judgment of high-risk conditions. Therefore, the thresholds need to be dynamically optimized through coefficients. The vehicle operation data includes a vehicle speed of 50 km / h and a braking frequency of 2 times / minute. The road condition monitoring data shows a level 3 bumpy road condition, which is divided into 1-5 levels according to the degree of bumpiness. Level 3 and above belong to medium-risk conditions. Referring to a database of 1000 sets of historical transportation conditions, under this type of medium-risk condition, the probability of surface damage to the components due to vibration and impact is 30% higher than under stable road conditions. To ensure that no anomaly is missed and that noise is not amplified, the density adjustment coefficient needs to be set to 1.2. This coefficient can improve the sensitivity of density threshold judgment and avoid missing minor damage under high vibration. The boundary scale adjustment coefficient should be set to 1.1. This coefficient combination has been tested under 50 sets of three-level bumpy working conditions, and the anomaly identification accuracy rate reaches 94%, with the false alarm rate controlled within 5%.

[0056] Then, by integrating the abnormal cluster set with the dynamic adaptation parameters through threshold judgment logic, a preliminary abnormal event is obtained. The density peak of the abnormal cluster, 60, is multiplied by the adjustment factor 1.2 to obtain 72, and the boundary size of 4% is multiplied by the adjustment factor 1.1 to obtain 4.4%. Both meet the adjusted threshold and are judged as a preliminary abnormal event. The event content is that there is a potential crack in the corner area of ​​component A0203.

[0057] Next, it is necessary to obtain the sudden pattern sequence during the transportation process. The sudden pattern sequence is the abnormal signal record between 10:05 and 10:25, which includes two vibration peak exceeding the standard and one temperature exceeding the standard, forming a continuous abnormal sudden pattern sequence.

[0058] Finally, when combining preliminary abnormal events and sudden pattern sequences through data fusion processing, a weighted voting method was adopted. The weight of the preliminary abnormal event was 0.6, and the weight of the sudden pattern sequence was 0.4. When the comprehensive judgment component A0203 was located at 120.5 degrees east longitude and 30.2 degrees north latitude at 10:15, a surface crack appeared at the corner due to a level 3 bumpy road condition. This event was confirmed as an abnormal event.

[0059] It is worth noting that the weight ratio was determined through verification of 100 cases of abnormal photovoltaic module transportation. Under this ratio, the accuracy of anomaly detection after integration is improved to 95%, and the probability of misjudging normal fluctuations as abnormalities is reduced to below 3%. If the weight of the initial abnormal event is adjusted to be lower than 0.6, the influence of physical characteristics will be reduced, and if it is higher than 0.6, the early warning value of continuous anomalies in time series will be easily ignored. If the weight of the sudden mode sequence is higher than 0.4, it will amplify noise interference, and if it is lower than 0.4, it cannot effectively assist in verification.

[0060] In step S17, the process of extracting transportation route and cargo status information based on the confirmed abnormal event, constructing a traceability data chain and identifying high-risk routes, integrating abnormal details with vehicle operation logs, and generating final monitoring output includes: From the confirmed abnormal events, the transportation route sequence and cargo status record are extracted. Delay timestamps and location deviation data in the transportation route sequence are extracted through sequence parsing to construct a traceability data chain. Based on the traceability data chain, the cumulative delay time and the set of affected goods are calculated. When the cumulative delay time is higher than a preset delay threshold, the corresponding path is determined to be a high-risk path and listed as a priority traceability item. By combining the priority traceability items with the vehicle operation logs from the on-board monitoring system of the transport vehicle through fusion processing, a complete traceability framework is formed. Based on the complete traceability framework, standardized final monitoring output is generated by integrating the time, location, cargo impact range, route risk level, and cause analysis of abnormal events.

[0061] It is worth noting that, firstly, it is necessary to extract the transportation route sequence and cargo status record from the confirmed abnormal events. The transportation route sequence contains all location coordinates from 10:00 to 10:30, and the cargo status record contains the surface condition, vibration tolerance, and temperature and humidity exposure of component A0203.

[0062] Next, delay timestamps and location deviation data are extracted from the transportation route sequence through sequence parsing to construct a traceability data chain. For example, the delay timestamp is from 10:10 to 10:15, during which the vehicle's speed drops to 30 km / h due to bumpy road conditions, a delay of 5 minutes from the planned speed of 40 km / h; the location deviation data is 600 meters off the preset route at 10:15. These data are integrated in chronological order to obtain the traceability data chain.

[0063] Then, based on the traceability data chain, the cumulative delay time and the affected goods set are calculated. For example, if the cumulative delay time is 5 minutes and the preset delay threshold is 2 hours, statistical analysis of 1000 sets of transportation data reveals that when the delay is ≤2 hours, the performance degradation rate of the components due to environmental exposure is ≤2%; after 2 hours, the degradation rate increases to over 8%, and it is prone to problems such as packaging dampness and condensation on the component surface. Therefore, 2 hours is set as the delay threshold. The affected goods set is component A0203 and adjacent components A0202 and A0204. Since the cumulative delay time does not exceed the threshold, the corresponding path is determined to be a general concern path and listed as a regular traceability item.

[0064] Subsequently, a complete traceability framework was formed by integrating regular traceability items with vehicle operation logs from the onboard monitoring system of the transport vehicles. The vehicle operation logs included speed and braking records from 10:10 to 10:15, and driver operation records showed that the driver slowed down and braked during this period to avoid obstacles, resulting in increased bumps. Correlation analysis between regular traceability items and these log data clarified the relationship between abnormal events and driver operations and road conditions.

[0065] Finally, based on the complete traceability framework, standardized final monitoring output is generated by integrating the time, location, cargo impact range, route risk level, and causal analysis of the abnormal event. The final monitoring output includes the abnormal time of 10:15, the abnormal location at 120.5 degrees east longitude and 30.2 degrees north latitude, the affected cargo components A0203, A0202, and A0204, the route risk level of "general concern," and the cause of excessive vibration at the component edges and corners due to a combination of level three bumpy road conditions and driver swerving, leading to potential cracks. It is recommended that subsequent transportation avoid this route and that surface crack detection be performed on component A0203.

[0066] In summary, this invention discloses a multi-sensor-based intelligent monitoring method for photovoltaic module logistics. The method includes real-time acquisition of multi-source data streams, screening and labeling abnormal path segments, extracting environmental anomaly parameters, and integrating them to form an initial dataset. A refined dataset is obtained through local preprocessing using edge computing. Key features are extracted from the refined dataset to model sequence patterns; if deviations exceed a threshold, an alarm is activated, labels are generated, and damage locations are marked. After target image matching, enhanced feature maps are obtained through convolutional neural networks and anomaly masks. Anomaly clusters are identified by clustering the enhanced feature maps, and dynamic parameters and burst pattern sequences are integrated to confirm abnormal events. A traceability data chain is constructed based on the confirmed abnormal events, and the final monitoring output is generated by combining vehicle operation logs. This invention, through multi-stage collaboration, improves the real-time performance of monitoring and the accuracy of anomaly identification, achieving full-chain traceability and ensuring the safety of photovoltaic module transportation.

[0067] Reference Figure 2 The second embodiment of the present invention provides a photovoltaic module logistics intelligent monitoring system based on multiple sensors, comprising: The data acquisition module is used to collect multi-source data streams in real time and filter out abnormal environmental parameters and abnormal path segments from them, and integrate the abnormal environmental parameters and abnormal path segments to form an initial dataset; The data refining module is used to perform local preprocessing on the initial dataset through edge computing to obtain a refined dataset; The sequence modeling module is used to extract key environmental features from the refined dataset to form a feature vector set, and to analyze the data change trend under sudden situations based on the feature vector set to obtain the sequence pattern. The label generation module is used to compare the sequence pattern with a preset normal pattern, combine the location data in the multi-source data stream, filter and mark potential damage locations, and obtain the final abnormal label. The feature extraction module is used to acquire the target image captured by the high-definition camera and match it with the final anomaly label. After the match is successful, the convolutional neural network is used to perform feature enhancement processing on the anomaly region to obtain an enhanced feature map. The cluster confirmation module is used to perform cluster analysis on the enhanced feature map, screen preliminary abnormal events, and fuse them with the sudden pattern sequence in the transportation process to obtain confirmed abnormal events. The traceability output module is used to extract transportation route and cargo status information based on the confirmed abnormal events, construct a traceability data chain and determine high-risk routes, integrate abnormal details with vehicle operation logs, and generate the final monitoring output.

[0068] It should be noted that the photovoltaic module logistics intelligent monitoring system based on multiple sensors provided in this embodiment of the invention is used to execute all the process steps of the photovoltaic module logistics intelligent monitoring method based on multiple sensors in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0069] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps described in the various embodiments of the multi-sensor-based intelligent monitoring method for photovoltaic module logistics, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above system embodiments, such as the data acquisition module.

[0070] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0071] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0072] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0073] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0074] If the modules / units integrated into the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or system capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0075] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0076] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for intelligent monitoring of photovoltaic module logistics based on multiple sensors, characterized in that, include: Real-time acquisition of multi-source data streams and screening of abnormal environmental parameters and abnormal path segments; integration of the abnormal environmental parameters and abnormal path segments to form an initial dataset. Based on the initial dataset, a refined dataset is obtained through local preprocessing using edge computing. Extract key environmental features from the refined dataset. A feature vector set is constructed, and the data change trend under sudden situations is analyzed based on the feature vector set to obtain a sequence pattern; The sequence pattern is compared with a preset normal pattern, and combined with the location data in the multi-source data stream, potential damage locations are filtered and marked to obtain the final anomaly label; The target image captured by the high-definition camera is acquired and matched with the final anomaly label. After the match is successful, the anomaly region is enhanced using a convolutional neural network to obtain an enhanced feature map. Cluster analysis is performed on the enhanced feature map to screen preliminary abnormal events and fuse them with the sudden pattern sequence in the transportation process to obtain confirmed abnormal events; Based on the confirmed abnormal events, transportation routes and cargo status information are extracted, a traceability data chain is constructed, high-risk routes are identified, and abnormal details are integrated with vehicle operation logs to generate the final monitoring output.

2. The intelligent monitoring method for photovoltaic module logistics based on multiple sensors according to claim 1, characterized in that, The process involves real-time acquisition of multi-source data streams, filtering out abnormal environmental parameters and abnormal path segments, and integrating the abnormal environmental parameters and abnormal path segments to form an initial dataset, including: Temperature and humidity data, vibration data, and location data are acquired in real time through a sensor network on the transport vehicle, forming a multi-source data stream; wherein the sensor network includes temperature and humidity sensors, vibration sensors, and GPS positioning devices; If the deviation exceeds the preset path threshold, the corresponding road segment is marked as an abnormal road segment. Temperature and humidity data and vibration data corresponding to the photovoltaic modules are extracted from the abnormal path segments to obtain the environmental parameter change sequence. Data exceeding a preset environmental parameter change threshold in the environmental parameter change sequence are selected as environmental anomaly parameters. The environmental anomaly parameters are then integrated with the anomaly path segments to obtain an initial dataset.

3. The intelligent monitoring method for photovoltaic module logistics based on multiple sensors according to claim 1, characterized in that, The step of obtaining a refined dataset by performing local preprocessing through edge computing based on the initial dataset includes: Based on the initial dataset, the original vibration signal and temperature and humidity signal are extracted from it through edge nodes to form an original signal set; The original signal set is filtered by Kalman filtering to obtain a preliminary filtered signal. If the vibration peak value in the preliminary filtered signal exceeds a preset vibration threshold, the corresponding vibration peak value sequence is extracted; if it does not exceed the threshold, the corresponding preliminary filtered signal is discarded to obtain a vibration feature sequence. The synchronization integrity between the vibration feature sequence and temperature and humidity fluctuations is analyzed by support vector machine, and data with synchronization anomalies are retained to obtain a refined dataset.

4. The intelligent monitoring method for photovoltaic module logistics based on multiple sensors according to claim 1, characterized in that, The step of analyzing data change trends under sudden situations based on the feature vector set to obtain sequence patterns includes: The feature vector set is divided into a continuous time series segment set by a sliding window, and the duration of the sliding window can be dynamically adjusted according to transportation conditions and data collection frequency. Calculate the rate of change of vibration peak value for each time series segment set, and extract characteristic correlation patterns from the vibration data of the time series segment sets; When the rate of change of the vibration peak exceeds a preset rate of change threshold, the time series segment set is input into a long short-term memory network, and a predicted sequence is output. If the vibration peak change rate does not exceed the preset change rate threshold, the analysis is carried out directly based on the temperature and humidity fluctuation data of the time series segment set to obtain real-time data characteristics; By combining the predicted sequence or real-time data features with the correlation patterns of the features, the intrinsic relationship between temperature and humidity fluctuations and vibrations is analyzed to form a sequence pattern that reflects the trend of data changes under sudden situations.

5. The intelligent monitoring method for photovoltaic module logistics based on multiple sensors according to claim 1, characterized in that, The step of comparing the sequence pattern with a preset normal pattern, combining the location data in the multi-source data stream, filtering and marking potential damage locations, and obtaining the final anomaly label includes: Calculate the deviation value between the sequence pattern and the preset normal pattern. When the deviation value exceeds the preset deviation threshold, activate the alarm mechanism and generate a preliminary abnormal label containing the timestamp of the abnormality. Based on the time interval corresponding to the preliminary anomaly label, weak signals are extracted from massive amounts of raw data, and the weak signals are enhanced through signal filtering to obtain enhanced signals; The enhanced signal is aligned with the location data collected at the same time by using a timestamp to determine the transportation location of the photovoltaic module corresponding to the enhanced signal. Based on the transportation location of the photovoltaic modules, the potential damage coordinates of the photovoltaic modules are marked by location mapping. The timestamp of the initial anomaly label, the feature information of the enhanced signal, and the coordinates of the potential damage are integrated to form a final anomaly label containing the anomaly time, location, and signal features.

6. The intelligent monitoring method for photovoltaic module logistics based on multiple sensors according to claim 2, characterized in that, The process involves acquiring the target image captured by a high-definition camera and matching it with the final anomaly label. After successful matching, a convolutional neural network is used to perform feature enhancement processing on the anomaly region to obtain an enhanced feature map, including: Acquire target images captured by high-definition cameras, and then adjust the image resolution to obtain standardized images; If the normalized image matches the anomaly type and damage coordinates of the final anomaly label, then a multi-layer convolution operation is performed on the normalized image through a convolutional neural network to obtain a preliminary feature map; The preliminary feature map is downsampled using pooling filtering to obtain a compressed feature map. An anomaly mask corresponding to the final anomaly label is generated, and the compressed feature map is fused with the anomaly mask to obtain an enhanced feature map.

7. The intelligent monitoring method for photovoltaic module logistics based on multiple sensors according to claim 1, characterized in that, The enhanced feature map is subjected to cluster analysis to screen preliminary abnormal events and fused with sudden pattern sequences during transportation to obtain confirmed abnormal events, including: A set of feature points is extracted from the enhanced feature map, and a density clustering operation is performed on the set of feature points to obtain the clustering result; Based on the clustering results, the density peak and boundary size of each cluster are calculated. When the density peak is higher than a preset density threshold and the boundary size exceeds a preset range threshold, an abnormal cluster is determined to be formed, and an abnormal cluster set is formed. Based on real-time vehicle operation data and road condition monitoring data collected during transportation, dynamic adaptation parameters are determined, and the abnormal cluster set and the dynamic adaptation parameters are integrated to obtain preliminary abnormal events. The sudden pattern sequence during transportation is obtained, and the confirmed abnormal event is obtained by combining the preliminary abnormal event with the sudden pattern sequence through data fusion processing.

8. The intelligent monitoring method for photovoltaic module logistics based on multiple sensors according to claim 1, characterized in that, The process involves extracting transportation routes and cargo status information based on the confirmed abnormal events, constructing a traceability data chain and identifying high-risk routes, integrating abnormal details with vehicle operation logs, and generating final monitoring output, including: From the confirmed abnormal events, the transportation route sequence and cargo status record are extracted. Delay timestamps and location deviation data in the transportation route sequence are extracted through sequence parsing to construct a traceability data chain. Based on the traceability data chain, the cumulative delay time and the set of affected goods are calculated. When the cumulative delay time is higher than a preset delay threshold, the corresponding path is determined to be a high-risk path and listed as a priority traceability item. By combining the priority traceability items with the vehicle operation logs from the on-board monitoring system of the transport vehicle through fusion processing, a complete traceability framework is formed. Based on the complete traceability framework, standardized final monitoring output is generated by integrating the time, location, cargo impact range, route risk level, and cause analysis of abnormal events.

9. A multi-sensor-based intelligent monitoring system for photovoltaic module logistics, characterized in that, include: The data acquisition module is used to collect multi-source data streams in real time and filter out abnormal environmental parameters and abnormal path segments from them, and integrate the abnormal environmental parameters and abnormal path segments to form an initial dataset; The data refining module is used to perform local preprocessing on the initial dataset through edge computing to obtain a refined dataset; The sequence modeling module is used to extract key environmental features from the refined dataset. A feature vector set is constructed, and the data change trend under sudden situations is analyzed based on the feature vector set to obtain a sequence pattern; The label generation module is used to compare the sequence pattern with a preset normal pattern, combine the location data in the multi-source data stream, filter and mark potential damage locations, and obtain the final abnormal label. The feature extraction module is used to acquire the target image captured by the high-definition camera and match it with the final anomaly label. After the match is successful, the convolutional neural network is used to perform feature enhancement processing on the anomaly region to obtain an enhanced feature map. The cluster confirmation module is used to perform cluster analysis on the enhanced feature map, screen preliminary abnormal events, and fuse them with the sudden pattern sequence in the transportation process to obtain confirmed abnormal events. The traceability output module is used to extract transportation route and cargo status information based on the confirmed abnormal events, construct a traceability data chain and determine high-risk routes, integrate abnormal details with vehicle operation logs, and generate the final monitoring output.

Images