A method and system for assessing the operational risk of sodium-ion battery energy storage devices
By using multimodal image processing and cross-modal analysis, the dust shading rate was extracted and a spatial compensation mask was constructed. Combined with the logical fusion of visible light and infrared gradient fields, the problem of false alarms in risk assessment of sodium-ion battery energy storage devices under extreme environments was solved, improving the accuracy of assessment and the safety of power plants.
Patent Information
- Application Number
- CN202611267317.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-20
- Publication Date
- 2026-11-17
AI Technical Summary
Existing battery risk assessment technologies cannot effectively overcome the impact of external environmental interference and non-uniform dust cover on sodium-ion battery energy storage devices, leading to false risk alarms or missed thermal faults, and making it difficult to accurately assess the uniformity of heat generation.
By acquiring multimodal images, the dust occlusion rate of the ventral region is extracted, a spatial compensation mask is constructed to filter the infrared image, and cross-modal analysis of the visible light illumination gradient and infrared thermal conduction gradient field is combined to perform logical fusion and cross-validation, and the final operational risk score is output.
It enables accurate risk assessment of sodium-ion battery energy storage devices in extreme environments, reduces false alarm rates, and improves the availability and safety of power plants.
Smart Images

Figure CN123115656A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of battery energy storage status monitoring and safety assessment technology, and more specifically, to a method and system for scoring the operational risk of sodium-ion battery energy storage devices. Background Technology
[0002] Sodium-ion batteries, due to their abundant raw material reserves, excellent low-temperature performance, high-rate charge and discharge capabilities, and safety characteristics such as allowing deep discharge to zero volts (0V) without the risk of current collector dissolution, are gradually becoming an important technological route in the field of energy storage for new power systems. With the continuous advancement of the national new energy strategy, a large number of integrated photovoltaic and energy storage projects are being deployed in arid regions such as deserts, Gobi, and wastelands. In these application scenarios, sodium-ion battery energy storage devices typically need to perform specific high-power day and night peak-shaving tasks and maintain a long-term 0V deep static state during non-photovoltaic power generation periods to minimize power loss and standby safety risks.
[0003] However, the unique natural environment of the Gobi Desert and the inherent physicochemical properties of sodium-ion batteries present significant challenges to the monitoring and safety assessment of energy storage equipment. On one hand, the Gobi Desert region experiences extreme diurnal temperature variations and intense solar radiation, and frequent sandstorms easily lead to random and uneven dust accumulation in the cooling ducts and vents of the energy storage container and battery pack. On the other hand, due to their large ionic radius, sodium ions, after a long period of 0V resting, experience a rapid increase in internal charge transfer resistance and lattice stress when suddenly awakened by high power and subjected to charge / discharge peak regulation, resulting in substantial transient heat generation. When this transient high heat generation is superimposed on the uneven external sandstorm environment, localized heat accumulation easily forms inside the battery equipment, leading to extreme non-uniformity in heat generation and dissipation across different areas and exhaust surfaces.
[0004] Traditional risk assessment methods for energy storage devices primarily rely on voltage and current parameters collected by the battery management system (BMS) and data from contact temperature sensors placed on the cell surfaces. This assessment mechanism, based on limited point-based electrical and thermal parameters, can provide basic early warning in typical indoor constant-temperature or well-conditioned environments, but it has significant limitations under complex, extreme desert conditions. Point-based temperature sensors can only reflect the thermodynamic state of local contact points and cannot macroscopically acquire the spatial heat generation distribution and heat escape characteristics of the vents on multiple surfaces of the battery pack. When the ventilation channels on one side of the battery device fail due to external dust blockage, the internal heat flow transmission path will undergo spatial distortion, making it difficult to achieve early and accurate identification of spatial heat accumulation faults using only conventional threshold judgment logic.
[0005] Furthermore, existing risk scoring mechanisms struggle to effectively decouple external environmental disturbances from actual internal faults. In desert environments, under intense direct sunlight during the day, the surface temperature and thermal radiation characteristics of the battery pack's sun-exposed surface significantly increase. If conventional thermal assessment logic is relied upon, the system is highly susceptible to misinterpreting unilateral temperature rises caused by external solar radiation as internal battery heat generation anomalies, triggering numerous false risk alarms and severely impacting the availability of peak-shaving power plants. Conversely, blindly raising the risk scoring trigger threshold may result in missed risks before actual localized asymmetric overheating and structural anomalies occur within the battery.
[0006] In summary, existing battery risk assessment technologies fail to fully consider the operational characteristics of sodium-ion batteries under 0V high-power wake-up conditions, as well as the environmental characteristics of deserts with high dust levels and intense solar and thermal radiation. How to effectively overcome the effects of strong solar and thermal interference and non-uniform dust cover from the external environment, accurately assess the heat generation uniformity of sodium-ion battery energy storage devices across multiple spatial surfaces, and thus achieve a highly reliable, low-false-alarm operational risk score, is a pressing technical challenge in the field of safety technology for new energy storage systems. Summary of the Invention
[0007] This invention provides a method for assessing the operational risk of sodium-ion battery energy storage devices, the method comprising: Acquire the initial visible light image and initial infrared image of the sodium-ion battery energy storage device at the reference time, as well as the dynamic visible light video stream and real-time infrared image during the high-power wake-up operation phase, and perform image normalization processing on each. Image detection is performed on the standardized initial visible light image to extract the dust shading rate of the vent areas on each surface; A spatial compensation mask is constructed based on the dust cover rate. The spatial compensation mask is used to filter the standardized real-time infrared image to obtain the compensated infrared grayscale matrix, and the grayscale extreme value surface is selected accordingly. Based on the spatial coordinates of the gray-scale extreme surface, the cross-modal region of interest is extracted. The visible light illumination gradient vector field and the infrared heat conduction gradient vector field are obtained respectively. The cross-modal gradient similarity between the two is calculated to extract the confidence features of the internal heat source. The infrared grayscale features extracted based on the compensated infrared grayscale matrix are logically fused and cross-validated with the confidence features of the internal heat source to output the final operational risk score.
[0008] Image detection is performed on the standardized initial visible light image to extract the dust occlusion rate of the vent regions on each surface. This includes: cropping the region of interest (ROI) containing the vents and their surrounding background; performing a morphological undercap transformation on the image within the ROI to enhance the image and obtain the undercap grayscale value. ; Calculate the local grayscale standard deviation of a local window within the region of interest. ; Set the grayscale value of the bottom cap Local grayscale standard deviation Nonlinear fusion is performed to generate a feature matrix of permeability of holes. The calculation formula is as follows: in, For pixel coordinates, It is a positive smoothing constant.
[0009] Extracting the dust shading rate of the vent areas on each surface also includes: the permeability feature matrix of the pores. Binarization is performed to obtain the initial set of high-response pixels, and connected component analysis and geometric filtering are carried out. The total number of pixels in the effective transparent connected components residing within the projected outline of the vent is counted and denoted as the effective transparent area. ; Calculate the first Dust occlusion rate of the vents on each surface : in, This represents the standard pixel area of the projected outline without any occlusion. This is the lower limit clamping function.
[0010] A spatial compensation mask is constructed based on the dust occlusion rate. This mask is then used to filter the standardized real-time infrared image to obtain the compensated infrared grayscale matrix. This process includes: performing a reference difference between the real-time infrared image frame and the standardized initial infrared image at the reference time, and extracting the dynamic temperature rise image. Based on the first On the surface, the first Dust shading rate of each air vent Generate a two-dimensional spatial compensation mask : Dynamic temperature rise image With two-dimensional spatial compensation mask A pixel-by-pixel weighted multiplication is performed as image filtering to obtain the compensated infrared grayscale matrix. ;in, For the first The total number of air vents present on each surface. pixel coordinates To the The Euclidean distance between the geometric centers of the air vents This is the thermal coupling penalty coefficient. This is a parameter representing the thermal diffusion scale.
[0011] Based on this, grayscale extreme surfaces are selected, including: calculating the average compensated grayscale value of all pixels on each surface, and the global mean and inter-surface variance of the average compensated grayscale value of all surfaces; if the inter-surface variance is greater than the preset uniformity benchmark threshold, the surface number where the average compensated grayscale value reaches the maximum value is located and recorded as the grayscale extreme surface.
[0012] The visible light illumination gradient vector field and the infrared thermal conduction gradient vector field are obtained separately, including: extracting the brightness channel matrix from the visible light region of interest corresponding to the gray-level extreme surface; performing Gaussian smoothing on the brightness channel matrix to perform image filtering to obtain the illumination basis matrix; calculating the first-order partial derivative of the illumination basis matrix to generate the visible light illumination gradient vector field to indicate the normal direction and intensity of the external sunlight light-dark boundary line; extracting the unmasked dynamic temperature rise image corresponding to the infrared region of interest, calculating the two-dimensional spatial gradient, and generating the infrared thermal conduction gradient vector field to indicate the direction of the steepest increase in infrared gray level.
[0013] Calculating the cross-modal gradient similarity between the two to extract the confidence features of the internal heat source includes: calculating the visible light illumination gradient vector field. With infrared thermal conduction gradient vector field Cross-modal gradient collinearity is analyzed, and the absolute values are taken within the infrared region of interest. Then, weighted aggregation based on thermal conductivity intensity is performed to extract the confidence features of internal heat sources. : in, Region of Interest (ROI) in infrared This represents the magnitude of the vector gradient. and All are positive smoothing constants.
[0014] The infrared grayscale features extracted based on the compensated infrared grayscale matrix are logically fused and cross-validated with the confidence features of the internal heat source to output the final operational risk score, including: calculating the spatial imbalance index based on the inter-surface variance of the compensated infrared grayscale matrix. The absolute calorific index is calculated based on the highest average compensated gray level of the gray-scale extreme surface. Calculate the feature coordinates. The Euclidean distance to the origin reduces the nonlinearity of the two-dimensional features to a single basic thermal risk index. : Using the confidence characteristics of internal heat sources Basic thermal risk index Perform multiplicative modulation and output a comprehensive operational risk score. : .
[0015] The method also includes linkage control based on the comprehensive operation risk score for the wake-up peak shaving condition: if the comprehensive operation risk score is less than the first-level warning threshold, the equipment is determined to be in a safe balance state and continues to perform high-power wake-up charging and discharging; if the comprehensive operation risk score is greater than or equal to the first-level warning threshold and less than the second-level fuse threshold, a power clamping command is issued to dynamically reduce the current ramp-up slope in the initial stage of the current wake-up; if the comprehensive operation risk score is greater than or equal to the second-level fuse threshold, the physical high-voltage electrical circuit of the battery module is cut off and the fire-fighting medium spraying action of the corresponding explosion-proof zone is driven.
[0016] The present invention also provides a risk assessment system for the operation of sodium-ion battery energy storage devices, the system comprising: The image acquisition and standardization module is used to acquire the initial visible light image and initial infrared image of the sodium-ion battery energy storage device at the reference time, as well as the dynamic visible light video stream and real-time infrared image during the high-power wake-up operation phase, and perform image standardization processing respectively. The image detection and occlusion extraction module is used to perform image detection on the standardized initial visible light image and extract the dust occlusion rate of the vent area of each surface. The spatial mask and filtering compensation module is used to construct a spatial compensation mask based on the dust occlusion rate, and to perform image filtering on the standardized real-time infrared image using the spatial compensation mask to obtain the compensated infrared grayscale matrix, and to select the grayscale extreme value surface accordingly. The cross-modal gradient analysis module is used to extract cross-modal regions of interest based on the spatial coordinates of the gray-scale extreme value surface, obtain the visible light illumination gradient vector field and the infrared heat conduction gradient vector field respectively, and calculate the cross-modal gradient similarity between the two to extract the confidence features of the internal heat source. The logic fusion and scoring output module is used to logically fuse and cross-validate the infrared grayscale features extracted based on the compensated infrared grayscale matrix with the confidence features of the internal heat source, and output the final operational risk score.
[0017] This invention provides a method and system for risk assessment of sodium-ion battery energy storage devices, and addresses the extreme environmental characteristics of desert regions with high light and heat and strong winds and sandstorms. It represents a leap from traditional point-based electrical parameter monitoring to multimodal visual spatial feature decoupling. Through the cascaded processing and constraint of multi-source visual sensors, this invention assesses the actual heat generation status of battery devices under specific operating conditions, significantly improving the safety of novel energy storage systems and the availability of power stations.
[0018] To address the technical challenge of localized heat dissipation channel failure caused by frequent sandstorms during desertification periods, this invention utilizes image detection techniques to extract sand shading features from vent areas and transforms these features into spatial constraints to compensate for real-time infrared thermal field images. This cross-modal spatial constraint mechanism adaptively adjusts the localized abnormally high grayscale response caused by sand blockage. Compared to traditional methods relying on a single absolute temperature threshold for alarms, this mechanism accurately eliminates false hotspots caused by deteriorating external ventilation. This ensures that high-heat areas, even after compensation, accurately reflect heating faults within sodium-ion batteries caused by high-rate charge transfer or abnormal structural stress, thus resolving misjudgments of spatial heat accumulation caused by non-uniform sand shading.
[0019] To address the problem of strong direct sunlight in desert regions easily generating localized high temperatures on metal casings and triggering false alarms, this invention proposes a cross-modal vector consistency verification method based on visible light illumination gradient fields and infrared thermal conduction gradient fields. This method explains that external photothermal radiation exhibits strict light and shadow boundary following, while the actual micro-short circuit or fault-induced heating inside the battery exhibits isotropic conduction from the inside out. Through rigorous image gradient direction decorrelation verification, the system can extremely accurately distinguish whether abnormally high heat originates from photothermal absorption by external container edges or dirt patches, or from a genuine short circuit inside the battery. This invention eliminates misjudgments caused by extreme lighting conditions, establishing extremely high reliability for thermal runaway early warning in complex outdoor environments.
[0020] In the final risk scoring output stage, this solution integrates features characterizing the degree of spatial heat distribution imbalance with features characterizing local absolute heat intensity. It utilizes the heat source authenticity features extracted through cross-modal consistency verification to modulate the basic thermal risk, further cutting off the risk transmission link of environmental illusions. While ensuring the safe operation of the Shagohuang sodium-ion battery energy storage power station to the greatest extent, it effectively avoids unplanned shutdowns caused by frequent false alarms, and has extremely high engineering implementation value and industrial economic benefits. Attached Figure Description
[0021] Figure 1 This is a flowchart of the operational risk assessment process for the sodium-ion battery energy storage device of the present invention. Figure 2 The image shows the results of identifying the permeable area of the vent and extracting the dust shading rate in this invention. Among them, (a) is the original visible light image, (b) is the image enhancement result after the bottom cap transformation, (c) is the permeability feature map of the hole formed by the fusion of grayscale and local texture, and (d) is the result of the effective permeable area and shading boundary after the connected component screening. Figure 3The images show the comparison of infrared thermal fields before and after spatial compensation for sand and dust, where (a) is the dynamic temperature rise image after baseline difference, (b) is the two-dimensional spatial compensation mask image, (c) is the infrared grayscale image after spatial compensation, and (d) is the extraction result of residual high grayscale areas. Figure 4 This is a cross-modal comparison experiment diagram of the external photothermal spot and the internal real heat source. Detailed Implementation
[0022] Example 1 This embodiment provides a method for assessing the operational risk of sodium-ion battery energy storage devices. This method evaluates the heat generation uniformity of the battery device under specific operating conditions through cascaded processing and feature analysis of multimodal images. The overall scheme of this method includes the following steps: S100: Acquire the baseline image and dynamic image sequence, and perform image normalization processing.
[0023] At the initial moment when the sodium-ion battery is at the end of its 0V deep resting period and is ready for high-power wake-up, initial visible light and infrared images of multiple surfaces of the energy storage device are simultaneously acquired. During the high-power wake-up operation, dynamic visible light video streams and real-time infrared images are continuously acquired. Image correction and standardization are performed on the acquired multi-source images to eliminate lens distortion and environmental noise introduced by different lighting conditions. Due to the extremely strong natural light and drastic temperature differences in the desert landscape, directly acquired raw images often suffer from overexposure or sensor thermal noise. By introducing pre-processing image correction and standardization, images from different time points and perspectives can be unified to the same data distribution benchmark. Choosing the moment at the end of the 0V resting period as the initial acquisition point aims to obtain an environmental film unaffected by the thermal effects of the device's internal current, thus removing environmental temperature interference and providing basic reference data.
[0024] S200: Perform image detection on the visible light image and extract the occlusion features of the pore region.
[0025] For the initial visible light image processed as described above, image enhancement is applied to highlight the structural edges of the device casing. Subsequently, image detection and recognition algorithms are run to locate the corresponding vent areas on each surface and calculate the pixel occlusion rate caused by sand and dust accumulation in each area. External sand and dust blockage is a direct cause of heat dissipation failure. Directly calculating a three-dimensional fluid dynamics model is computationally intensive and prone to distortion. Using image recognition to directly detect the occlusion area percentage from a two-dimensional image provides the most intuitive and lightweight visual representation of heat dissipation capacity degradation. The introduction of image enhancement technology aims to accurately sharpen the true physical boundaries of the vents in low-contrast images covered by sand and dust, ensuring the accuracy of the occlusion rate calculation.
[0026] S300 constructs a spatial mask based on occlusion features to perform image filtering and compensation on real-time infrared images.
[0027] The vent occlusion rates extracted in the preceding steps are converted into spatial compensation weights to construct corresponding image masks. These masks are then used to filter real-time infrared images during high-power operation, dynamically adjusting the image contrast and grayscale distribution of specific regions in the infrared image to obtain a compensated infrared grayscale matrix. Based on this matrix, grayscale extreme values are selected. If a vent on a certain surface is severely blocked, its infrared image will inevitably exhibit abnormally high grayscale, indicating high heat accumulation. Without decoupling this phenomenon, the system will frequently generate false alarms. By using prior knowledge of occlusion obtained in the visible light domain, targeted filtering and contrast reduction are applied to the infrared image. Essentially, this algorithmically smooths out false hotspots caused by external environmental blockages, allowing the high grayscale areas that remain prominent after filtering to truly reflect the abnormal heat generation inside the battery.
[0028] S400: Based on the spatial coordinates of the gray-scale extreme surface, guide the extraction of the dynamic region of interest and perform gradient vector field analysis of the cross-modal image.
[0029] The spatial coordinates of the grayscale extrema locked in the above steps are mapped onto the real-time visible light image and the compensated infrared image, and spatially strictly aligned regions of interest (ROIs) for both visible light and infrared ROIs are simultaneously cropped. Within these ROIs, image filtering and gradient feature extraction operations are performed to obtain the visible light illumination gradient vector field and the infrared heat conduction gradient vector field. By calculating the cross-modal gradient similarity between the two, the confidence feature of the internal heat source characterizing the authenticity of the battery's internal heating is extracted. External solar radiation heating has a strong light and shadow following property, and the edge of the hot spot formed on the infrared image must be highly consistent with the edge of the bright and dark shadows in the visible light image. However, the heating caused by a real internal battery fault is an isotropic conduction of heat flow from the inside to the outside through the metal casing, completely unconstrained by the geometry of external visible light and shadow. Panoramic dual-modal pixel-by-pixel comparison is extremely computationally intensive. Therefore, the extreme value region obtained from the front infrared analysis is used as the confirmation point to guide the accurate cropping of cross-modal regions of interest, which greatly reduces the data processing dimension. At the same time, by introducing the vector similarity calculation of visible light gradient and infrared gradient, external light and heat interference and internal real heat source are successfully extracted at the image processing level.
[0030] S500 performs logical fusion and cross-validation of multi-source visual features and outputs the final operational risk score.
[0031] The filtered and compensated infrared grayscale features are logically matched and cross-validated with the extracted internal heat source confidence features: it is determined whether the heat conduction gradient field of the high infrared grayscale region is spatially decorrelated with the visible light illumination gradient field, thereby eliminating image acquisition and image synthesis artifacts caused by external environmental light and heat radiation; finally, based on the verification results, the final operational risk score for uneven heat generation on the extreme surface is output.
[0032] Example 2 As a preferred embodiment of the present invention, this embodiment further details step S100 based on Embodiment 1. This embodiment specifically illustrates how to acquire a reference image and a dynamic image sequence under conditions of desert environment and sodium-ion battery 0V static peak adjustment, and how to perform image normalization processing to eliminate interference from the external extreme environment on subsequent visual analysis.
[0033] Step S100 in this embodiment specifically includes the following process: S110. Determine the visual inspection reference time and collect initial image data.
[0034] The system receives operating status commands from the battery management system and determines whether the sodium-ion battery energy storage device meets the baseline triggering conditions. Specifically, sodium-ion batteries have the characteristic of no risk of copper current collector dissolution. During non-photovoltaic power generation periods at the Shagohuang photovoltaic-storage integrated base, they are often deeply discharged and kept in a zero-volt state to ensure ultimate safety. The moment when the device ends this zero-volt deep resting period and is about to inject a high-rate current to execute the wake-up peak-shaving command is determined as the baseline moment. At the reference time Since the battery has not yet generated the thermal effect of high-rate charging and discharging, the thermodynamic performance and visual appearance of the device casing are entirely determined by the natural environment of the desert (such as temperature and sunlight). At the aforementioned reference time... The dual-modal vision sensors deployed on five surfaces of the energy storage device (excluding the grounded bottom surface) are activated to simultaneously acquire initial visible light images of each surface. and initial infrared image It was used as a background contrast film.
[0035] S120: During the high-power wake-up cycle, continuously acquire dynamic multi-source image sequences.
[0036] From the reference time Afterwards, the sodium-ion battery energy storage device enters a high-power wake-up operation cycle. During the wake-up cycle Inside, sodium ions shuttle at high frequency within the layered structure of the electrode, generating significant transient internal resistance heating. During this period, the dual-modal vision sensor is controlled to continuously acquire dynamic visible light video streams from each surface according to a preset sampling frequency. and real-time infrared image sequences The acquired dynamic visible light video stream The system recorded the dynamic process of continuous light and shadow distribution on various surfaces of the energy storage device as the solar azimuth and irradiance changed, providing a visible light reference for subsequent imaging; real-time infrared image sequences were also recorded. The dynamic process of heat dissipation in space generated by current injection on various surfaces was recorded. Real-time infrared image sequence. The dynamic process of spatial heat distribution generated by current injection on each surface was recorded.
[0037] S130. Perform image correction processing on multi-source images using preset camera parameters.
[0038] In the extreme natural environment of the Gobi Desert, the alternating high surface temperatures during the day and extreme cold at night cause minute thermal expansion and contraction deformation of the mechanical mounting bracket of the visual sensor, resulting in a shift in the lens's field of view. Simultaneously, wide-angle lenses inherently suffer from barrel distortion. To ensure absolute accuracy of the image spatial coordinates, the pre-calibrated camera intrinsic and extrinsic parameter matrices are retrieved and applied to the acquired initial visible light image. Initial infrared image Dynamic visible light video stream Video frames and real-time infrared image sequences Image correction is performed. This image correction process uses perspective transformation and distortion elimination algorithms to remap the offset or deformed pixel coordinates onto a standard orthogonal projection plane, so that the image coordinates of the five surfaces after correction can be strictly aligned with the physical position of the three-dimensional structural model of the energy storage device casing and vent.
[0039] S140, Perform image normalization processing.
[0040] After image correction, further noise interference from natural lighting and ambient temperature is needed. The extremely strong solar radiation during the day in the Shago Desert region can cause localized overexposure of illuminated surfaces and deep shadows on shaded surfaces in visible light images. This requires addressing the dynamic visible light video stream. Perform the same histogram equalization or grayscale stretching to obtain the corresponding sampling time. Standardized real-time visible light image frames , to convert the original visible light pixel values The pixels are uniformly mapped to a preset standard pixel range, thereby compressing the overflow values in extreme highlight areas and improving the contrast in shadow areas, ensuring that the physical outlines of the pores can be clearly identified subsequently. Similarly, the extremely high temperatures of the desert environment cause severe thermal drift noise in the infrared sensor itself; directly using the raw response signal would result in an extremely high false alarm rate. This applies to the initial infrared image. and real-time infrared image sequences Image normalization is performed to remove the global DC component response caused by the ambient base temperature, and the original sensor analog-to-digital conversion signal is mapped to normalized infrared grayscale values. The normalized infrared grayscale value It only represents the relative thermal differences between different areas on the device surface, not the absolute physical temperature. Through image standardization processing of the dual-modal image data, a high-quality, uniformly distributed data source is provided for the system to accurately identify fault features under strong light, high temperature, and non-uniform dust obstruction interference.
[0041] Example 3 As another preferred embodiment of the present invention, this embodiment further details step S200 based on Embodiments 1 and 2. This embodiment specifically illustrates how to perform image detection on visible light images and extract the occlusion features of pore regions in a desert environment.
[0042] Energy storage containers and battery packs are typically covered in dust that closely resembles the color of sand and dust. Furthermore, the intense direct sunlight in desert regions creates deep light and shadow boundaries—structural shadows or scratches—on the raised metal edges of the equipment casing. This low contrast between the sand and the metal casing, along with the edge shadows / scratches masquerading as holes, causes traditional global thresholding segmentation algorithms to fail.
[0043] S210, Delineate the region of interest and contour for image detection.
[0044] Obtain the standardized initial visible light image output after image standardization processing in Example 2. Using a pre-calibrated camera intrinsic and extrinsic parameter matrix, the preset vent edge coordinates in three-dimensional space are projected onto a two-dimensional standardized initial visible light image. On the pixel plane where it is located, generate the first Projected outline of the vent holes on each surface ( ). With this projected outline Using the geometric center as a reference, a preset pixel margin is extended outward to capture the region of interest containing the vent and its surrounding background. To prevent subsequent morphological and neighborhood convolution operations from exceeding the image boundary or causing boundary effects, the width of the pixel margin is strictly limited to be no less than the radius of the largest neighborhood operator in subsequent steps.
[0045] S220 applies morphological algorithms to perform image enhancement, reducing background noise caused by shell reflection and uneven lighting.
[0046] The actual vent, which is not obscured by sand and dust, is a cavity extending into the device. It has an extremely high absorption rate of visible light and appears as a dark area with very low grayscale in the image. The area where sand and dust have accumulated has a significantly higher grayscale than the actual vent due to the diffuse reflection of silicate sand particles, but lower grayscale than the metal casing exposed to direct sunlight.
[0047] To accurately separate the real dark holes, the region of interest was... The image is enhanced by performing a morphological under-hat transformation. The under-hat transformation uses a closing operation to fill in dark structures smaller than a given operator in the image, and then subtracts these dark areas from the original image to highlight them. Its calculation is expressed as follows: Region of Interest Pixel coordinates within; This is the grayscale value of the bottom cap after image enhancement. This represents the morphological closing operation; This is a two-dimensional disk-shaped morphological structural element. To ensure that the closing operation can completely fill the actual vent cavity, the disk-shaped morphological structural element... The pixel diameter is set to be larger than the projected outline. The maximum pixel inner diameter (preferably 1.2 to 1.5 times) is significantly larger than the pixel size of the largest adhering dust cluster in the desert scene. After this image enhancement step, the large area of bright metallic shell background is effectively suppressed, while the deep vent cavity and the smaller size are more prominent. Dark scratches or shadows are all in The value in the middle is reversed and highlighted as a high intensity value.
[0048] S230: Perform image recognition by fusing grayscale and local texture features to construct a hole transparency feature map.
[0049] Bottom cap transformation alone cannot distinguish between the pore cavity and the dark scratches on the metal shell. Considering the significant particle undulations on the surface of sand and dust accumulation, which manifest as local high-frequency grayscale fluctuations in imaging (i.e., coarse texture), while the actual cavity interior has an extremely smooth pixel distribution due to the lack of light reflection, image recognition is performed by combining local standard deviation texture. Within the region of interest... Inside, in each pixel Centered on, take a side length of Calculate the local grayscale mean within a square local window. and local grayscale standard deviation : The number must be a positive odd number to ensure the symmetry of the center pixel; The value must be smaller than the pixel radius of the vent, but larger than the texture period of a typical sand grain.
[0050] Characterizing darkness With characterization of smoothness Nonlinear fusion is performed to generate a feature matrix of permeability of holes. : To prevent positive smoothing constants with a denominator of zero, this characteristic matrix... Only pixel regions that are both extremely dark relative to their surroundings and absolutely smooth internally will produce a very large response peak.
[0051] S240. Perform image detection and connected component geometric filtering to extract the pore dust occlusion rate.
[0052] To further eliminate interference from thin, elongated scratches that also exhibit smooth, dark characteristics, and to define the occlusion range, based on the feature matrix... Perform image detection with geometric constraints.
[0053] The transparency threshold is generated using the Otsu method or a preset empirical threshold. For the characteristic matrix Binarization is performed to obtain an initial set of high-response pixels. Eight-neighbor connected component analysis is then performed on this initial set to extract several independent connected component branches. For each connected component, its pixel area is calculated. With roundness ( ,in (This refers to the perimeter of the connected region). By setting minimum area thresholds and minimum circularity thresholds, interfering connected regions belonging to discrete isolated noise and thin, dark scratches are eliminated.
[0054] The valid connected components retained after geometric filtering are compared with the projected contours obtained in step S210. Perform a spatial logical AND operation to ensure that only transparent pixels located within the physical area of the vent are counted. The final count resides on... The total number of pixels in the effective transparent connected domains inside is denoted as the effective transparent area. .
[0055] Based on the projection calibration results, obtain the projection contour. Standard pixel area without any obstruction Based on this, the first... Dust occlusion rate of the vents on each surface : Introduction The clamping operation is designed to eliminate minute out-of-bounds pixel errors caused by extreme light diffraction, strictly ensuring... The dust cover rate This constitutes the distribution vector of sand and dust cover rate. This vector quantifies the degree of damage caused by external dust accumulation in the desert environment to the physical heat dissipation channels of each gas outlet surface of the sodium-ion battery, and serves as a constraint in the separation of infrared thermal field features.
[0056] Example 4 As another preferred embodiment of the present invention, this embodiment further details step S300 based on the foregoing embodiments. Specifically, this embodiment describes how, in a high-power peak-shaving scenario in a desert region, a two-dimensional spatial mask is constructed based on sand and dust occlusion features extracted in the visible light domain, and weighted filtering and compensation are applied to real-time infrared images, thereby achieving mechanism decoupling between external environmental interference and actual internal thermal faults at the algorithm level.
[0057] In the barren desert environment, the thermal interference experienced by energy storage devices mainly comes from two aspects: first, direct sunlight from one side causes a large initial temperature difference on various surfaces of the device; second, non-uniform sand and dust obstructs the heat dissipation channels on specific surfaces. This embodiment employs cross-modal feature cascading and spatially constrained image processing techniques, with the specific steps as follows: S310, Perform benchmark differential to eliminate static initial ambient light and heat noise.
[0058] Obtain the standardized initial infrared image after image standardization processing in Example 2. (Data collected at the reference time before high-power wake-up) ) and real-time infrared image frames during the wake-up operation phase. ,in Belongs to real-time infrared image sequence any sampling time In the system preset, the five infrared sensors have undergone unified radiation cross-calibration, and their output signals are all mapped to a globally unified normalized grayscale scale. Let... In pixel coordinates The normalized infrared grayscale value at that location is ,make In the same coordinate The normalized infrared grayscale value at that location is To separate the static environment of the desert (initial temperature and...) Extracting the dynamic temperature rise image induced by high-power current injection from the thermal noise caused by constant solar radiation. : This is a lower bound clamping function used to filter out the cooling effect caused by the sudden drop in ambient temperature at night in the desert. To normalize the gray levels. Through a differential step, the device surface due to... The artificially high grayscale caused by uneven heating at the initial moment is canceled out. (Regarding the wake-up cycle...) The residual environmental temperature rise interference introduced by the change in the internal solar azimuth and the dynamic surge in irradiance will be eliminated by the cross-modal cross-validation in the subsequent step S500 of this application.
[0059] S320, Construct an adaptive two-dimensional spatial compensation mask.
[0060] To address the "environmental pseudo-hot spots caused by vent blockage," the facet-by-face dust scalar prior extracted in Example 3 needs to be mapped abbreviated to a two-dimensional spatial weight distribution. Considering that heat accumulation typically diffuses non-linearly from the vent center outwards in a Gaussian manner, for the first... Each surface ( ), assuming that the surface coexists For each air outlet, obtain the geometric center coordinates of the projected contour of each air outlet. and the corresponding sand and dust cover rate ( ). Regarding the first Any pixel point within a surface Calculate its up to the th Euclidean distance between the geometric centers of the air vents : For the first Each surface generates a corresponding two-dimensional spatial compensation mask. : This is the thermal coupling penalty coefficient. For thermal diffusion scale parameters, mask The range of values always belongs to If wake-up cycle If the system detects a sandstorm weather event during a prolonged period, the S200 update can be retried. To ensure the timeliness of the mask.
[0061] S330 uses a spatial compensation mask to perform weighted filtering and reconstruct infrared contrast.
[0062] Extracted dynamic temperature rise images With the generated two-dimensional spatial compensation mask Pixel-wise weighted multiplication is performed, followed by spatially adaptive image filtering (i.e., pixel-wise gain modulation) to obtain the compensated infrared grayscale matrix. : The response of localized high-brightness infrared areas caused by dust blockage was dynamically reduced. Meanwhile, for areas that still exhibit extremely high grayscale anomalies after mask suppression, whether due to extreme environmental heat buildup exceeding the mask suppression limit or a genuine micro-short circuit fault inside the battery, it indicates a very high safety risk. These residual high-grayscale areas will directly enter the subsequent S400–S500 dynamic video cross-verification process.
[0063] S340. Evaluate the grayscale uniformity of multiple surfaces, and screen and lock the grayscale extreme surfaces.
[0064] Based on the compensated infrared grayscale matrix To evaluate the overall heat generation uniformity of the battery device across multiple outlet surfaces, the following calculations were performed: The average compensated gray value of all pixels on the surface : For the first A set of pixel coordinates of a surface. This represents the total number of pixels within the set. Since the five sensors use a unified calibration scale, It can characterize the heat generation intensity of the surface after environmental decoupling.
[0065] Calculate the global mean of the average compensated grayscale of the five surfaces. Inter-surface variance : Obtain the preset uniformity benchmark threshold If the variance between surfaces This indicates that the sodium-ion battery may have a severe spatial heat generation imbalance or structural anomaly during the current wake-up peak-shaving cycle. At this time, positioning... The surface number where the maximum value is located is obtained and denoted as the gray-scale extreme value surface. : Locked grayscale extremum surface The spatial coordinates of the residual high grayscale region will be directly passed to the dynamic video stream processing module as spatial priors, and will be verified in multiple dimensions through the cross-modal gradient vector field analysis in step S400.
[0066] Example 5 As another preferred embodiment of the present invention, this embodiment further details step S400 based on the foregoing embodiments. This embodiment specifically illustrates how the extraction of a dynamically generated region of interest is guided by the grayscale extremum surface coordinates locked in the preceding steps, and how visual features characterizing the actual internal heating state of the device are extracted through gradient vector field analysis of cross-modal images.
[0067] In the Shagohuang energy storage power station, extreme surface The residual high infrared gray level may have two physical causes: one is outward heat conduction caused by a genuine micro-short circuit or structural anomaly inside the battery; the other is direct sunlight from a strong directional light spot at a specific angle or from high-absorption dirt patches, such as sharp shadows formed by the corners of a container or ventilation louvers, or dark oil and dust accumulation. These two heat sources are fundamentally different in mechanism: the temperature rise of the casing caused by the external environment strictly follows optical geometry and material absorptivity distribution, and its spatial distribution boundary of infrared heat must be highly consistent with the light and shadow boundaries (hot in bright areas or hot in dark areas) in the visible light image; while the heat generated by an internal battery fault is heat conduction from the inside out on the metal panel, and its infrared heat gradient exhibits a radial distribution centered on the fault point. Based on the above mechanism, this embodiment uses cross-modal vector consistency analysis of the visible light illumination gradient field and the infrared heat conduction gradient field to decouple the internal heat source from external photothermal interference. The specific steps are as follows: S410. Based on the residual high grayscale coordinates of the extreme surface, extract and resample the cross-modal dynamic region of interest.
[0068] Obtain the locked grayscale extremum surface in Example 4 During the wake-up runtime cycle Any sampling time within Simultaneously acquire the standardized real-time visible light image frames from Example 2. Dynamic temperature rise image without masking in Example 4 and the compensated infrared grayscale matrix .
[0069] After compensation, the infrared grayscale matrix extreme surfaces Above, extract the residual high-grayscale connected regions exceeding the threshold, and calculate the pixel centroid coordinates of these connected regions. Using this centroid as an anchor point, expand outwards by a preset pixel span (preferably covering the residual high-grayscale region and its complete spatial attenuation edge), within the real-time visible light image frame. and dynamic temperature rise images In the process, local regions are cropped out. Based on the camera extrinsic correction matrix obtained in Example 2, the extracted dual-modal local regions are resampled to the same pixel grid resolution to achieve strict spatial alignment pixel by pixel, generating visible light regions of interest. With infrared region of interest .
[0070] S420: Perform image smoothing and gradient calculations to extract the visible light illumination distribution vector field.
[0071] Region of interest for visible light Extract its luminance channel matrix (Acquired directly through a grayscale sensor, or using RGB images weighted according to standard) (Conversion). Due to the fine particle texture noise on the surface of sand and dust, direct differentiation would result in a chaotic gradient field. To extract the illumination boundary, the brightness channel matrix is... Gaussian smoothing is performed on the image for filtering to obtain the illumination basis matrix. : In the formula, This represents a two-dimensional linear convolution operation. This is a normalized Gaussian low-pass filter kernel. The kernel size (pixels) is set to be larger than the typical sand grain texture period and smaller than the feature scale of the light and shadow boundary to achieve frequency band separation.
[0072] Using the central difference algorithm, the illumination basis matrix is calculated in... and The first-order partial derivatives in two directions generate the visible light illumination gradient vector field. : Visible light illumination gradient vector It accurately indicates the normal direction and intensity of the boundary between light and shadow in the external sunlight.
[0073] S430: Extract the true heat conduction vector field from the dynamic temperature rise image without masking.
[0074] For infrared regions of interest Extract the corresponding unmasked dynamic temperature rise image. The compensated matrix is not used here. Calculate the gradient and avoid introducing spatial masks. The artificial radial pseudo-gradient interference generated by the entity itself. Calculation The two-dimensional spatial gradient is used to generate an infrared thermal conduction gradient vector field. : Infrared thermal conduction gradient vector It accurately indicates the direction of the steepest rise in infrared grayscale (i.e., the direction that strictly points to the center of the real heat source).
[0075] S440. Calculate cross-modal gradient collinearity, extract internal heat source confidence features, and output risk score.
[0076] Based on the aforementioned mechanism, if the hot spot is caused by external photothermal coupling, whether it is "bright and hot" due to illumination boundaries or "dark and hot" due to strong absorption by dark dust accumulation, then... The vector will be with The vectors exhibit a highly collinear state in space: parallel in the same direction or parallel in opposite directions. The cosine similarity of the two gradient vector fields is calculated pixel by pixel, and the absolute value is taken within the region of interest before weighted aggregation based on thermal conductivity.
[0077] Define the confidence features of internal heat sources as follows: By absolute value Placed within the summation, this ensures that both parallel (shadow heating) and antiparallel (dark high-absorption heating) directions contribute positively, avoiding cancellation of positive and negative effects in mixed scenes. As a weighting coefficient, the algorithm automatically focuses on the steep regions with the strongest correlation at the hotspot edges. Furthermore, since cosine similarity possesses amplitude scale invariance, the absolute numerical differences caused by different normalization benchmarks in the two modes do not affect the determination of directional consistency. As a preferred implementation, the region of interest can also be divided into sub-blocks for separate calculations. And take the maximum value to enhance the sensitivity to local real heat sources. The range of values is strictly defined. When the infrared heat source is constrained by external light and shadow, the collinearity weighting term approaches 1, leading to... This indicates that the hot spot is caused by external environmental interference. When heat conduction diffuses isotropically, it exhibits spatial decorrelation with the visible light gradient, and the collinearity term approaches zero, leading to... This indicates that the problem is caused by a fault in the internal heat source.
[0078] Example 6 As another preferred embodiment of the present invention, this embodiment further details step S500 based on the foregoing embodiments. This embodiment specifically describes how to perform logical fusion and cross-validation of multi-source visual features, output the final operational risk score, and trigger linkage control for the sodium-ion battery 0V wake-up peak-shaving condition based on the score.
[0079] S510, Construct mutually orthogonal risk components.
[0080] Obtain the inter-surface variance, which characterizes the overall heat generation uniformity of the device surface, as output in Example 4. And the highest average compensated gray level of the locked gray-scale extreme surface. Obtain the preset uniformity benchmark threshold. Computational spatial imbalance index : Obtain the preset limit of grayscale heating This threshold characterizes the maximum normalized grayscale allowed under the design heat dissipation limit. Calculate the absolute heat index. : Independently characterizes the degree of anomaly in the spatial distribution of heat, while The two independently characterize the intensity of heating of local heat sources, forming two mutually orthogonal feature dimensions for describing thermodynamic anomalies.
[0081] S520: Based on nonlinear spatial distance mapping, extract the basic thermal risk index.
[0082] Build with and Let the coordinate axes be a two-dimensional thermal risk phase space. The current characteristic state is mapped to a coordinate point within this phase space. Calculate the Euclidean distance from the feature coordinate point to the origin, thus nonlinearly reducing the two-dimensional feature to a single basic thermal risk index. : Using Euclidean distance ensures that regardless of extreme space heat imbalances ( Is it dominated by ) or caused by extreme local overheating () Dominated by the basic thermal risk index All of these can rapidly amplify and generate high-value risk responses.
[0083] S530 introduces the confidence level of internal heat sources to perform cross-validation and outputs a comprehensive operational risk score.
[0084] although The apparent degree of thermal anomaly can be quantified, even though it may still be caused by extreme external strong directional radiation, as mentioned in Example 5, such as severe localized heat accumulation in light and shadow. To eliminate this environmental artifact, the confidence feature of the internal heat source based on the cross-modal gradient vector field consistency analysis output in Example 5 is introduced. .
[0085] use Basic thermal risk index Perform multiplicative modulation to output the final comprehensive operational risk score. : In the formula, The range of values for is strictly defined as follows: When abnormally high temperatures are confirmed to originate from external light and shadow interference, the visible light and infrared gradient fields exhibit a high degree of collinearity. Multiplicative modulation forcibly cuts off the risk transmission link, making Completely suppress false positive alarms caused by environmental factors; if and only if the abnormally high temperature exhibits isotropic internal heat conduction characteristics ( )hour, The system accurately outputs a high-risk score caused by internal short circuits or structural deformation.
[0086] S540, based on the comprehensive operational risk score, executes linkage control for wake-up peak shaving conditions.
[0087] The output comprehensive operational risk score Compared with the system's preset Level 1 warning threshold and secondary circuit breaker threshold Compare: If If the equipment is determined to be in a safe balance between heat generation and dissipation, continue to execute the high-power wake-up charging and discharging from 0V to the rated voltage according to the original instructions; if The system determined that asymmetric thermal stress had accumulated inside the equipment. In response, the system issued a power clamping command to dynamically reduce the current ramp-up rate during the initial wake-up phase. This slows down the lattice expansion rate caused by the insertion and extraction of large sodium ions, thus providing a time buffer for heat dissipation; if This confirms that the equipment has experienced an irreversible internal micro-short circuit or severe localized thermal runaway. The system immediately disconnects the physical high-voltage electrical circuit of the battery module and activates the fire-fighting spray system in the corresponding explosion-proof zone.
[0088] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0089] In this specification, the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the descriptions of the embodiments described later are relatively simple, and relevant parts can be referred to the descriptions of the foregoing embodiments.
[0090] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for assessing the operational risk of sodium-ion battery energy storage devices, characterized in that, The method includes: Acquire the initial visible light image and initial infrared image of the sodium-ion battery energy storage device at the reference time, as well as the dynamic visible light video stream and real-time infrared image during the high-power wake-up operation phase, and perform image normalization processing on each. Image detection is performed on the standardized initial visible light image to extract the dust shading rate of the vent areas on each surface; A spatial compensation mask is constructed based on the dust cover rate. The spatial compensation mask is used to filter the standardized real-time infrared image to obtain the compensated infrared grayscale matrix, and the grayscale extreme value surface is selected accordingly. Based on the spatial coordinates of the gray-scale extreme surface, the cross-modal region of interest is extracted. The visible light illumination gradient vector field and the infrared heat conduction gradient vector field are obtained respectively. The cross-modal gradient similarity between the two is calculated to extract the confidence features of the internal heat source. The infrared grayscale features extracted based on the compensated infrared grayscale matrix are logically fused and cross-validated with the confidence features of the internal heat source to output the final operational risk score.
2. The method for assessing the operational risk of sodium-ion battery energy storage devices as described in claim 1, characterized in that, Image detection is performed on the standardized initial visible light image to extract the dust occlusion rate of the vent regions on each surface. This includes: cropping the region of interest (ROI) containing the vents and their surrounding background; performing a morphological undercap transformation on the image within the ROI to enhance the image and obtain the undercap grayscale value. ; Calculate the local grayscale standard deviation of a local window within the region of interest. ; Set the grayscale value of the bottom cap Local grayscale standard deviation Nonlinear fusion is performed to generate a feature matrix of permeability of holes. The calculation formula is as follows: ; in, For pixel coordinates, It is a positive smoothing constant.
3. The method for assessing the operational risk of sodium-ion battery energy storage devices as described in claim 2, characterized in that, Extracting the dust shading rate of the vent areas on each surface also includes: the permeability feature matrix of the pores. Binarization is performed to obtain the initial set of high-response pixels, and connected component analysis and geometric filtering are carried out. The total number of pixels in the effective transparent connected components residing within the projected outline of the vent is counted and denoted as the effective transparent area. ; Calculate the first Dust occlusion rate of the vents on each surface : ; in, This represents the standard pixel area of the projected outline without any occlusion. This is the lower limit clamping function.
4. The method for assessing the operational risk of sodium-ion battery energy storage devices as described in claim 1, characterized in that, A spatial compensation mask is constructed based on the dust occlusion rate. This mask is then used to filter the standardized real-time infrared image to obtain the compensated infrared grayscale matrix. This process includes: performing a reference difference between the real-time infrared image frame and the standardized initial infrared image at the reference time, and extracting the dynamic temperature rise image. Based on the first On the surface, the first Dust shading rate of each air vent Generate a two-dimensional spatial compensation mask : ; Dynamic temperature rise image With two-dimensional spatial compensation mask A pixel-by-pixel weighted multiplication is performed as image filtering to obtain the compensated infrared grayscale matrix. ;in, For the first The total number of air vents present on each surface. pixel coordinates To the The Euclidean distance between the geometric centers of the air vents This is the thermal coupling penalty coefficient. This is a parameter representing the thermal diffusion scale.
5. The method for assessing the operational risk of sodium-ion battery energy storage devices as described in claim 4, characterized in that, Based on this, grayscale extreme surfaces are selected, including: calculating the average compensated grayscale value of all pixels on each surface, and the global mean and inter-surface variance of the average compensated grayscale value of all surfaces; if the inter-surface variance is greater than the preset uniformity benchmark threshold, the surface number where the average compensated grayscale value reaches the maximum value is located and recorded as the grayscale extreme surface.
6. The method for assessing the operational risk of sodium-ion battery energy storage devices as described in claim 1, characterized in that, The visible light illumination gradient vector field and the infrared thermal conduction gradient vector field are obtained separately, including: extracting the brightness channel matrix from the visible light region of interest corresponding to the gray-level extreme surface; performing Gaussian smoothing on the brightness channel matrix to perform image filtering to obtain the illumination basis matrix; calculating the first-order partial derivative of the illumination basis matrix to generate the visible light illumination gradient vector field to indicate the normal direction and intensity of the external sunlight light-dark boundary line; extracting the unmasked dynamic temperature rise image corresponding to the infrared region of interest, calculating the two-dimensional spatial gradient, and generating the infrared thermal conduction gradient vector field to indicate the direction of the steepest increase in infrared gray level.
7. The method for assessing the operational risk of sodium-ion battery energy storage devices as described in claim 6, characterized in that, Calculating the cross-modal gradient similarity between the two to extract the confidence features of the internal heat source includes: calculating the visible light illumination gradient vector field. With infrared thermal conduction gradient vector field Cross-modal gradient collinearity is analyzed, and the absolute values are taken within the infrared region of interest. Then, weighted aggregation based on thermal conductivity intensity is performed to extract the confidence features of internal heat sources. : ; in, Region of Interest (ROI) in infrared This represents the magnitude of the vector gradient. and All are positive smoothing constants.
8. The method for assessing the operational risk of sodium-ion battery energy storage devices as described in claim 1, characterized in that, The infrared grayscale features extracted based on the compensated infrared grayscale matrix are logically fused and cross-validated with the confidence features of the internal heat source to output the final operational risk score, including: calculating the spatial imbalance index based on the inter-surface variance of the compensated infrared grayscale matrix. The absolute calorific index is calculated based on the highest average compensated gray level of the gray-scale extreme surface. Calculate the feature coordinates. The Euclidean distance to the origin reduces the nonlinearity of the two-dimensional features to a single basic thermal risk index. : ; Using the confidence characteristics of internal heat sources Basic thermal risk index Perform multiplicative modulation and output a comprehensive operational risk score. : 。 9. The method for assessing the operational risk of sodium-ion battery energy storage devices as described in claim 8, characterized in that, The method also includes implementing linkage control for wake-up peak shaving based on the comprehensive operation risk score: if the comprehensive operation risk score is less than the first-level warning threshold, the equipment is determined to be in a safe balance state and high-power wake-up charging and discharging continues; If the comprehensive operational risk score is greater than or equal to the first-level warning threshold and less than the second-level fuse threshold, a power clamping command will be issued to dynamically reduce the current ramp-up slope in the initial wake-up phase; if the comprehensive operational risk score is greater than or equal to the second-level fuse threshold, the physical high-voltage electrical circuit of the battery module will be cut off, and the fire-fighting media spraying action of the corresponding explosion-proof zone will be activated.
10. A risk assessment system for the operation of sodium-ion battery energy storage devices, characterized in that, The system includes: The image acquisition and standardization module is used to acquire the initial visible light image and initial infrared image of the sodium-ion battery energy storage device at the reference time, as well as the dynamic visible light video stream and real-time infrared image during the high-power wake-up operation phase, and perform image standardization processing respectively. The image detection and occlusion extraction module is used to perform image detection on the standardized initial visible light image and extract the dust occlusion rate of the vent area of each surface. The spatial mask and filtering compensation module is used to construct a spatial compensation mask based on the dust occlusion rate, and to perform image filtering on the standardized real-time infrared image using the spatial compensation mask to obtain the compensated infrared grayscale matrix, and to select the grayscale extreme value surface accordingly. The cross-modal gradient analysis module is used to extract cross-modal regions of interest based on the spatial coordinates of the gray-scale extreme value surface, obtain the visible light illumination gradient vector field and the infrared heat conduction gradient vector field respectively, and calculate the cross-modal gradient similarity between the two to extract the confidence features of the internal heat source. The logic fusion and scoring output module is used to logically fuse and cross-validate the infrared grayscale features extracted based on the compensated infrared grayscale matrix with the confidence features of the internal heat source, and output the final operational risk score.
