A fire warning method for energy storage equipment based on neural network

Abstract

The present invention discloses a neural network-based fire warning method for energy storage equipment, belonging to the field of energy storage fire protection technology. The method includes: obtaining historical overheating risk points within each battery cluster in the energy storage device; collecting temperature change data for any historical overheating risk point within the collection time period to construct a temperature change curve; analyzing the temperature change curve to extract slope characteristics, inflection point information, and statistically related features, integrating them to form a feature vector; inputting the feature vector into a pre-trained neural network-based overheating warning model to output a warning information label; and determining whether an alarm is needed based on the warning information label. The present invention utilizes the predictive capabilities of neural networks to improve the accuracy of fire warnings.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy storage fire protection, and in particular relates to a fire protection early warning method for energy storage equipment based on a neural network. Background Art

[0002] With the rapid development of the new energy and energy storage industries, energy storage fire protection solutions have become a crucial component in ensuring the safe operation of energy storage systems. Existing energy storage equipment primarily utilizes lithium-ion batteries, which are compact and offer excellent cycle performance. However, they release significant amounts of heat during the charging and discharging process. Excessive heat can easily cause combustion or even explosion, and even after extinguishing a fire, there is still the risk of a secondary fire reigniting. Therefore, implementing comprehensive fire protection measures for energy storage equipment is crucial.

[0003] At present, existing technologies usually use battery clusters as basic units and set temperature sensors, smoke sensors, and voltage and current detection equipment to monitor the working status of the battery cluster in real time. When the temperature, smoke value, voltage or current reaches an abnormal value, the fire protection facilities are automatically activated. However, this method only uses fixed abnormal values ​​to provide fire warnings for battery clusters, which has large errors in accuracy. Therefore, there is an urgent need for a fire warning method for energy storage equipment based on neural networks, which uses the predictive ability of neural networks to improve the accuracy of fire warnings. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a fire warning method for energy storage equipment based on neural network to solve the above technical problems.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A fire warning method for energy storage equipment based on a neural network, comprising:

[0007] Obtain historical overheating risk points within each battery cluster in the energy storage device;

[0008] Collect temperature change data of any historical overheating risk point within the collection period to construct a temperature change curve;

[0009] Analyze the temperature change curve to extract slope characteristics, inflection point information, and statistically relevant features to form a feature vector. Input the feature vector into a pre-trained neural network-based overheating warning model, and output the warning information label.

[0010] Determine whether an alarm is needed based on the warning information label.

[0011] Furthermore, historical overheating risk points within each battery cluster in the energy storage device are obtained, including:

[0012] Obtain the battery cluster thermal management history information from the thermal management system database corresponding to the energy storage device;

[0013] The single cells with overheating risk in the battery cluster thermal management history information are extracted and recorded as historical overheating risk points.

[0014] Furthermore, the temperature change data of any historical overheating risk point within a preset time period is collected to construct a temperature change curve, including:

[0015] Determine the collection time period based on user-set parameters, and identify any historical overheating risk point as a point to be collected;

[0016] When the energy storage device is in the startup state, the temperature change data of the points to be collected are periodically collected based on the collection time period to construct a temperature change curve that changes over time.

[0017] Furthermore, the temperature change curve is analyzed to extract slope characteristics, inflection point information, and statistically relevant features to form a feature vector, including:

[0018] Obtain all temperature change curves and perform feature analysis on any temperature change curve based on a preset feature extraction method. The slope characteristics, inflection point information, and statistically relevant features of the temperature change curve are extracted and recorded as features to be fused. Statistically relevant features include the temperature mean and temperature variance.

[0019] Based on the feature fusion method, the features to be fused are fused and normalized to obtain a feature vector.

[0020] Furthermore, the training process of the neural network-based overheating warning model includes:

[0021] Obtain temperature change curves of several historical overheating risk points, and use a preset feature extraction method to extract the resulting feature vectors as first training data; wherein the first training data corresponds to a warning information label indicating safety;

[0022] Acquire experimental data from multiple overheating risk points with fire hazards to construct a temperature change curve, and use a preset feature extraction method to extract the resulting feature vector as second training data; wherein the second training data corresponds to a warning information label indicating danger;

[0023] Acquire multiple groups of feature vectors from the first training data and the second training data as test data;

[0024] An initial neural network prediction model is constructed, and the first training data and the second training data are input into the initial neural network prediction model for training. The trained initial neural network prediction model is verified using test data, and an overheating warning model is generated after the training conditions are met. The result output by the overheating warning model is a warning information label.

[0025] Furthermore, a neural network-based energy storage device fire warning method further includes: obtaining the locations of all energy storage devices in a target area and corresponding warning information tags, and generating a monitoring image in real time; wherein the monitoring image includes a plurality of warning information tags and their corresponding locations in the target area;

[0026] Obtain a preset time interval, extract image frames from the surveillance image, obtain several key image frames corresponding to different time nodes, and calculate the similarity between each pair of key image frames based on the extraction order; where the image feature vectors corresponding to different warning information labels in the key image frames are different, and the warning information labels include safety labels and danger labels;

[0027] According to the similarity calculation results, combined with the preset emergency ladder diagram, warning information corresponding to the warning level is generated for warning, and real-time monitoring images are output accordingly.

[0028] Furthermore, a fire warning method for energy storage equipment based on a neural network also includes: if the number of danger tags in the initial key image frame is greater than a preset danger threshold, issuing a large-scale fire warning information.

[0029] Furthermore, based on the similarity calculation results and the preset emergency ladder diagram, warning information corresponding to the warning level is generated, including:

[0030] Obtaining a preset emergency ladder diagram; wherein the preset emergency ladder diagram includes multiple warning levels, each warning level corresponding to a similarity sudden change range or a similarity change rate range;

[0031] According to the key frame extraction order, the nth and n+1th key image frames are sequentially obtained for similarity calculation to obtain the nth similarity calculation result; wherein n is an integer with an initial value of 1;

[0032] Determine whether the similarity difference between the nth similarity calculation result and the n+1th similarity calculation result falls within the similarity sudden change range. If so, generate warning information corresponding to the warning level according to the preset emergency level ladder diagram; if not, obtain the similarity difference and store it; when the number of similarity difference values ​​obtained is greater than the preset judgment number threshold, construct a similarity change trend curve diagram based on all similarity differences, and calculate the similarity slope of the similarity change trend curve diagram;

[0033] Determine whether the similarity slope falls within the similarity change rate range. If so, generate warning information corresponding to the warning level according to the preset urgency ladder diagram; if not, clear the currently stored similarity difference and re-acquire a new similarity difference for storage.

[0034] Furthermore, a fire warning method for energy storage equipment based on a neural network also includes: when performing similarity calculation; if multiple danger labels appear at the same energy storage device location, their corresponding image feature vectors are superimposed to form a new image feature vector; if multiple safety labels appear at the same energy storage device location, their corresponding image feature vectors are deduplicated to retain only one image feature vector.

[0035] The beneficial effects of the present invention are:

[0036] This paper proposes a neural network-based fire warning method for energy storage equipment. Compared with traditional methods that generally rely on fixed temperature thresholds or other fixed parameters to determine whether anomalies exist, this paper utilizes the learning, recognition, and prediction capabilities of neural networks to construct an overheating warning model for fire warning. The neural network can identify complex nonlinear relationships by learning from large amounts of historical data and make high-precision predictions of future temperature trends. This data-driven dynamic warning method can effectively improve the accuracy of fire warnings and reduce the risk of false alarms and missed alarms.

[0037] At the same time, since the emission of harmful gases by the battery and the voltage and current fluctuations of the battery are related to the battery temperature, that is, the emission of harmful gases by the battery and the voltage and current fluctuations of the battery can be reflected in the change of the battery temperature. Based on this, the present application proposes a concept of overheating risk points. Historical overheating risk points refer to single cells with overheating risks in the historical information of thermal management of battery clusters. The areas where these single cells are located are usually closely related to phenomena such as the emission of harmful gases by the battery and voltage and current fluctuations. A neural network-based overheating warning model is generated by utilizing the changing trend of the overheating risk points. Compared with the existing neural network prediction model, by collecting and training multi-dimensional temperature change data at the overheating risk points, the changes in battery temperature can be monitored with high precision, which can significantly reduce the types of sensors deployed in traditional fire warning systems, so that high-precision fire warnings can be achieved by retaining only the relevant equipment for temperature monitoring. This method greatly reduces the cost and difficulty of sensor deployment without sacrificing warning accuracy, while simplifying the maintenance and management of the system.

[0038] Other advantages, objectives, and features of the present invention will be described in the following description and will be apparent to those skilled in the art to some extent, or may be taught by those skilled in the art from the practice of the present invention. The purposes and other advantages of the present invention may be realized and obtained through the structures particularly pointed out in the written description and the accompanying drawings.

[0039] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0041] Figure 1 The present invention is a flowchart of a method for fire warning of energy storage equipment based on a neural network in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0043] like Figure 1 As shown, the present invention proposes a fire warning method for energy storage equipment based on a neural network, comprising:

[0044] S101. Obtain historical overheating risk points in each battery cluster in the energy storage device;

[0045] S102, collecting temperature change data of any historical overheating risk point within a collection period to construct a temperature change curve;

[0046] S103, analyzing the temperature change curve to extract slope characteristics, inflection point information, and statistically related characteristics to form a feature vector;

[0047] S104, inputting the feature vector into a pre-trained neural network-based overheating warning model, and outputting a warning information label;

[0048] S105, judging whether an alarm is required based on the warning information label;

[0049] The working principle of the above technical solution is as follows: In actual situations, energy storage equipment is usually composed of several battery clusters for realizing charging and discharging operations; in this application, in order to realize accurate fire warning of energy storage equipment, compared with the existing technology that monitors the battery cluster as the basic unit, it is chosen to directly use the risk single battery in the battery cluster as the basic monitoring unit; because the single batteries in the battery cluster of energy storage equipment are usually connected in series, based on the barrel effect, their maximum performance depends on the minimum performance of the single battery. Similarly, it can be known that the maximum probability of fire risk in a battery cluster depends on the maximum fire risk probability of the single battery therein. Based on this, this application proposes a concept of overheating risk point, that is, the single battery in each battery cluster that is most likely to cause fire risk is specially marked, which is used for subsequent accurate fire warning of energy storage equipment in combination with neural network;

[0050] When the method proposed in the present application is actually applied, it is first necessary to obtain the historical overheating risk points in each battery cluster in the energy storage device. It is worth noting that the historical overheating risk points in each battery cluster usually do not change in the absence of maintenance or replacement; then, among all the historical overheating risk points, taking any historical overheating risk point as an example, the temperature change data of any historical overheating risk point within the collection time period is collected to construct a temperature change curve, and the temperature change curve is subjected to feature analysis to extract the slope feature, inflection point information and statistically related features to form a feature vector, wherein the slope feature represents the rate of change of the overheating risk point. By increasing the collection frequency, the collection accuracy of the temperature change can be correspondingly improved, and the inflection point information The information characterizes the abnormal information of the overheating risk point. By collecting the inflection point information, it is determined whether there is a temperature mutation. The statistical related features are used to characterize the discrete degree and average state of the collected temperature of the overheating risk point. By integrating the above related features to form a feature vector, it is used to accurately characterize the temperature change of the overheating risk point, and then determine the fire risk change of the battery cluster. Finally, the feature vector collected and processed in each collection time period is input into the pre-trained neural network-based overheating warning model, and the corresponding warning information label is output. This warning information label is used to characterize whether the current battery cluster has a fire warning risk. Finally, all warning information labels are counted to determine whether the energy storage device needs to be alarmed.

[0051] The beneficial effects of the above technical solution are as follows: through the above technical solution, a fire warning method for energy storage equipment based on a neural network is proposed. Compared with the traditional method that usually relies on setting a fixed temperature threshold or other fixed parameters to determine whether there is an abnormality, the present invention uses the learning, recognition and prediction capabilities of the neural network to construct an overheating warning model for fire warning; the neural network can identify complex nonlinear relationships by learning a large amount of historical data, and make high-precision predictions on future temperature change trends; this data-driven dynamic warning method can effectively improve the accuracy of fire warnings and reduce the risk of false alarms and missed alarms; at the same time, since the battery emits harmful gases and the battery voltage and current fluctuations are related to the battery temperature, that is, the battery emits harmful gases and the battery voltage and current fluctuations can be reflected in the battery temperature changes. In terms of chemistry, based on this, the above technical solution proposes a concept of overheating risk points. Historical overheating risk points refer to single cells with overheating risks in the historical information of thermal management of battery clusters. The areas where these single cells are located are usually closely related to phenomena such as harmful gas emission from batteries and voltage and current fluctuations. The changing trend of overheating risk points is used to generate an overheating warning model based on a neural network. Compared with the existing neural network prediction model, by collecting and training multi-dimensional temperature change data at overheating risk points, high-precision monitoring of battery temperature changes can significantly reduce the types of sensors deployed in traditional fire warning systems, so that high-precision fire warnings can be achieved by retaining only temperature monitoring-related equipment. This method greatly reduces the cost and difficulty of sensor deployment without sacrificing warning accuracy, while simplifying system maintenance and management.

[0052] In one embodiment, obtaining historical overheating risk points in each battery cluster in the energy storage device includes:

[0053] Obtain the battery cluster thermal management history information from the thermal management system database corresponding to the energy storage device;

[0054] Single cells with overheating risks in the battery cluster thermal management history information are extracted and recorded as historical overheating risk points; wherein, the overheating risk can be determined by a corresponding thermal management system.

[0055] In one embodiment, collecting temperature change data of any historical overheating risk point within a preset time period to construct a temperature change curve includes:

[0056] Determine the collection time period based on user-set parameters, and identify any historical overheating risk point as a point to be collected;

[0057] When the energy storage device is in the startup state, the temperature change data of the points to be collected are periodically collected based on the collection time period to construct a temperature change curve that changes over time;

[0058] The working principle and beneficial effects of the above technical solution are as follows: guiding the user to set the collection parameters in advance. If no settings are set, the collection time period is determined based on the default collection parameters, usually every 15 minutes, and all historical overheating risk points are defined as points to be collected. Taking any historical overheating risk point as an example, when the energy storage device is in the startup state, based on the collection time period, the temperature change data of the point to be collected is periodically collected to construct a temperature change curve that changes over time, thereby realizing periodic monitoring of the battery cluster and improving monitoring reliability.

[0059] In one embodiment, the temperature change curve is analyzed to extract slope characteristics, inflection point information, and statistically relevant characteristics to form a feature vector, including:

[0060] Obtain all temperature change curves and perform feature analysis on any temperature change curve based on a preset feature extraction method. The slope characteristics, inflection point information, and statistically relevant features of the temperature change curve are extracted and recorded as features to be fused. Statistically relevant features include the temperature mean and temperature variance.

[0061] Based on the feature fusion method, the features to be fused are fused and normalized to obtain the feature vector;

[0062] The working principle and beneficial effects of the above technical solution are as follows: all temperature change curves are obtained, and based on a preset feature extraction method, feature analysis is performed on any temperature change curve, wherein the preset feature extraction method can adopt feature extraction technologies such as convolutional neural networks or autoencoders to extract the slope characteristics, inflection point information and statistically related features of the temperature change curve, which are recorded as features to be fused; wherein the statistically related features include temperature mean and temperature variance, etc.; finally, based on the feature fusion method, the features to be fused are fused and normalized to obtain feature vectors; the feature vectors generated by using multi-dimensional temperature change data can improve the monitoring accuracy of the temperature changes of single cells, thereby improving the early warning accuracy of the fire risk of the energy storage equipment corresponding to the battery cluster.

[0063] In one embodiment, the training process of the neural network-based overheating warning model includes:

[0064] Obtain temperature change curves of several historical overheating risk points, and use a preset feature extraction method to extract the resulting feature vectors as first training data; wherein the first training data corresponds to a warning information label indicating safety;

[0065] Acquire experimental data from multiple overheating risk points with fire hazards to construct a temperature change curve, and use a preset feature extraction method to extract the resulting feature vector as second training data; wherein the second training data corresponds to a warning information label indicating danger;

[0066] Acquire multiple groups of feature vectors from the first training data and the second training data as test data;

[0067] An initial neural network prediction model is constructed, and the first training data and the second training data are input into the initial neural network prediction model for training. The trained initial neural network prediction model is verified using test data, and an overheating warning model is generated when the training conditions are met. The result output by the overheating warning model is a warning information label.

[0068] Furthermore, when an unrecognizable feature vector is encountered, an empty label is input, and a second alarm signal is sent back to the user based on the empty label. After the user defines the feature vector, the feature vector and its defined warning information label are used to update the overheating warning model, which is beneficial to improving the warning accuracy of the overheating warning model.

[0069] In one embodiment, a neural network-based energy storage device fire warning method further includes: obtaining the locations of all energy storage devices and corresponding warning information tags within a target area, and generating a monitoring image in real time; wherein the monitoring image includes a plurality of warning information tags and their corresponding locations within the target area; the monitoring image for user viewing preferably uses a white background, and the image feature vectors generated by the warning information tags are other colors, the number of superimposed features is determined by color depth, and the corresponding location coordinates can be displayed at the same time;

[0070] A preset time interval, preferably a collection time period, is obtained, and image frames are extracted from the surveillance image to obtain several key image frames corresponding to different time nodes. Similarity calculations are performed on each pair of key image frames according to the extraction order. For example, four key image frames A, B, C, and D are collected in sequence, and similarity calculations are performed on AB, BC, and CD respectively. The image feature vectors corresponding to different warning information labels in the key image frames are different. Warning information labels include safety labels and danger labels. Furthermore, the labels can be divided more finely according to actual conditions.

[0071] Based on the similarity calculation results and the preset emergency ladder diagram, the warning information corresponding to the warning level is generated for warning, and the real-time monitoring image is output accordingly;

[0072] It is worth noting that if the number of danger labels in the initial key image frame is greater than the preset danger threshold, a large-scale fire warning information is directly issued;

[0073] Among them, the above-mentioned similarity calculation results are combined with the preset emergency level ladder diagram to generate warning information corresponding to the warning level, including:

[0074] Obtaining a preset emergency ladder diagram; wherein the preset emergency ladder diagram includes multiple warning levels, each warning level corresponding to a similarity sudden change range or a similarity change rate range; the similarity sudden change range and the similarity change rate range are preferably determined by user customization or default settings, and the default setting can be determined by the sudden change in the number of accident points of energy storage equipment in large-scale fire accidents determined from experiments or real events, and the rate of occurrence of batch accident points before the accident;

[0075] According to the key frame extraction order, the nth and n+1th key image frames are sequentially obtained for similarity calculation to obtain the nth similarity calculation result; wherein n is an integer with an initial value of 1;

[0076] Determine whether the similarity difference between the nth similarity calculation result and the n+1th similarity calculation result falls within the similarity sudden change range. If so, generate warning information corresponding to the warning level according to the preset emergency level ladder diagram; if not, obtain the similarity difference and store it; when the number of similarity difference values ​​obtained is greater than the preset judgment number threshold, construct a similarity change trend curve diagram based on all similarity differences, and calculate the similarity slope of the similarity change trend curve diagram;

[0077] Determine whether the similarity slope falls within the similarity change rate range. If so, generate warning information corresponding to the warning level according to the preset urgency ladder diagram; if not, clear the currently stored similarity difference and re-obtain a new similarity difference for storage;

[0078] It is worth noting that when performing similarity calculations, if multiple dangerous labels appear at the same energy storage device location, their corresponding image feature vectors are superimposed to form a new image feature vector. If multiple safe labels appear at the same energy storage device location, their corresponding image feature vectors are deduplicated, and only one image feature vector is retained.

[0079] The beneficial effects of the above technical solution are as follows: in the existing technology, all energy storage devices are usually monitored independently, and it is impossible to predict large-scale fire anomalies in advance based on the status of multiple energy storage devices. Through the above technical solution, joint fire monitoring is carried out on the energy storage devices in the entire target area, and sensitive data (i.e., image feature vectors corresponding to hazard labels) are established. The hidden feature information of the sensitive data is amplified from a global perspective, and then by monitoring the changing trend of the hidden feature information, the accurate fire warning level of the energy storage devices in the entire area is determined. Compared with the existing technology, it is beneficial to improve the response speed of the fire warning system, and through the feedback of monitoring images, it can help users quickly formulate reasonable fire fighting measures to minimize fire losses.

[0080] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.

Claims

1. A fire warning method for energy storage equipment based on neural network, characterized in that: include: Obtain historical overheating risk points within each battery cluster in the energy storage device; Collect temperature change data of any historical overheating risk point within the collection period to construct a temperature change curve; Analyze the temperature change curve to extract slope characteristics, inflection point information and statistically related features to form a feature vector; Input the feature vector to a pre-trained neural network-based overheating warning model, and output the warning information label; Determine whether an alarm is needed based on the warning information label; Obtain the locations of all energy storage devices and corresponding warning information tags within the target area, and generate a monitoring image in real time; wherein the monitoring image includes several warning information tags and their corresponding locations within the target area; Obtain a preset time interval, extract image frames from the surveillance image, obtain several key image frames corresponding to different time nodes, and calculate the similarity between each pair of key image frames based on the extraction order; where the image feature vectors corresponding to different warning information labels in the key image frames are different, and the warning information labels include safety labels and danger labels; Based on the similarity calculation results and the preset emergency ladder diagram, the warning information corresponding to the warning level is generated for warning, and the real-time monitoring image is output accordingly; Among them, when performing similarity calculation; if multiple danger labels appear at the same energy storage device location, their corresponding image feature vectors are superimposed to form a new image feature vector. If multiple safety labels appear at the same energy storage device location, their corresponding image feature vectors are deduplicated and only one image feature vector is retained.

2. A fire alarm method for energy storage equipment based on neural network according to claim 1, characterized in that: Obtain historical overheating risk points within each battery cluster in the energy storage device, including: Obtain the battery cluster thermal management history information from the thermal management system database corresponding to the energy storage device; The single cells with overheating risk in the battery cluster thermal management history information are extracted and recorded as historical overheating risk points.

3. A fire alarm method for energy storage equipment based on neural network according to claim 1, characterized in that: Collect temperature change data of any historical overheating risk point within a preset time period to construct a temperature change curve, including: Determine the collection time period based on user-set parameters, and identify any historical overheating risk point as a point to be collected; When the energy storage device is in the startup state, the temperature change data of the points to be collected are periodically collected based on the collection time period to construct a temperature change curve that changes over time.

4. A fire alarm method for energy storage equipment based on neural network according to claim 1, characterized in that: Analyze the temperature change curve to extract slope characteristics, inflection point information, and statistically related features to form a feature vector, including: Obtain all temperature change curves and perform feature analysis on any temperature change curve based on a preset feature extraction method. The slope characteristics, inflection point information, and statistically relevant features of the temperature change curve are extracted and recorded as features to be fused. Statistically relevant features include the temperature mean and temperature variance. Based on the feature fusion method, the features to be fused are fused and normalized to obtain a feature vector.

5. The fire warning method for energy storage equipment based on neural network according to claim 1, characterized in that: The training process of the neural network-based overheating warning model includes: Obtain temperature change curves of several historical overheating risk points, and use a preset feature extraction method to extract the resulting feature vectors as first training data; wherein the first training data corresponds to a warning information label indicating safety; Acquire experimental data from multiple overheating risk points with fire hazards to construct a temperature change curve, and use a preset feature extraction method to extract the resulting feature vector as second training data; wherein the second training data corresponds to a warning information label indicating danger; Acquire multiple groups of feature vectors from the first training data and the second training data as test data; An initial neural network prediction model is constructed, and the first training data and the second training data are input into the initial neural network prediction model for training. The trained initial neural network prediction model is verified using test data, and an overheating warning model is generated after the training conditions are met. The result output by the overheating warning model is a warning information label.

6. The fire alarm method for energy storage equipment based on neural network according to claim 1, characterized in that: Also includes: If the number of danger labels in the initial key image frame is greater than the preset danger threshold, a large-scale fire warning information is issued.

7. The method for fire alarm of energy storage equipment based on neural network according to claim 1, characterized in that: Based on the similarity calculation results and the preset emergency ladder diagram, warning information corresponding to the warning level is generated, including: Obtaining a preset emergency ladder diagram; wherein the preset emergency ladder diagram includes multiple warning levels, each warning level corresponding to a similarity sudden change range or a similarity change rate range; According to the key frame extraction order, the nth and n+1th key image frames are sequentially obtained for similarity calculation to obtain the nth similarity calculation result; wherein n is an integer with an initial value of 1; Determine whether the similarity difference between the nth similarity calculation result and the n+1th similarity calculation result falls within the similarity sudden change range. If so, generate warning information corresponding to the warning level according to the preset emergency level ladder diagram; if not, obtain the similarity difference and store it; when the number of similarity difference values ​​obtained is greater than the preset judgment number threshold, construct a similarity change trend curve diagram based on all similarity differences, and calculate the similarity slope of the similarity change trend curve diagram; Determine whether the similarity slope falls within the similarity change rate range. If so, generate warning information corresponding to the warning level according to the preset urgency ladder diagram; if not, clear the currently stored similarity difference and re-acquire a new similarity difference for storage.

Images