Power battery monitoring and protection method and system based on deep learning and nitrogen protection

Abstract

The present invention relates to the technical field of power lithium batteries for new energy vehicles, and more specifically, to a monitoring and protection method and system for power batteries based on deep learning and nitrogen protection. The present invention applies nitrogen protection to the safety protection of power lithium batteries for new energy vehicles, prevents fires from occurring, and improves the safety performance of power lithium batteries for new energy vehicles during driving. The present invention detects the oxygen concentration, air pressure, temperature, and humidity in the battery box where the battery is located, and performs nitrogen filling operations in a timely manner, so that the battery is in an environment that does not have the conditions for combustion and explosion. Based on historical detection data, the present invention constructs a neural network for model training to predict the battery state, can prepare for nitrogen production and nitrogen storage in advance, ensure the normal operation of nitrogen filling, and further provide safety.

Description

Technical Field

[0001] The present invention relates to the technical field of power lithium batteries for new energy vehicles, and more specifically, to a monitoring and protection method and system for power batteries based on deep learning and nitrogen protection. Background Art

[0002] With the development of new technologies, more and more new energy vehicles are emerging in people's vision. The safety issue is the lifeline related to the development of new energy vehicles, and strengthening the safety prevention and control and management of new energy vehicles is the primary task at present. Among them, the situation of power lithium batteries catching fire in new energy vehicles also occurs, so the battery safety hazard has become one of the key factors determining whether potential buyers are willing to purchase new energy vehicles. Summary of the Invention

[0003] Based on this, in view of the problem that the existing power lithium batteries for new energy vehicles may catch fire and have safety hazards, it is necessary to provide a monitoring and protection method and system for power batteries based on deep learning and nitrogen protection to provide safety protection for the power lithium batteries of new energy vehicles.

[0004] The present invention is implemented by the following technical solutions:

[0005] In a first aspect, the present invention discloses a monitoring and protection method for power batteries based on deep learning and nitrogen protection, which is used for providing safety protection for power lithium batteries of new energy vehicles.

[0006] The monitoring and protection method for the power battery includes the following steps:

[0007] Step 1: Obtain the oxygen concentration O(t0), air pressure P(t0), temperature T(t0), and humidity W(t0) of the battery box where the battery is located at the current moment t0.

[0008] Step 2: Compare O(t0) with a preset oxygen concentration threshold O0; compare P(t0) with a first preset air pressure threshold P0 and a second preset air pressure threshold P0'; compare T(t0) with a preset temperature threshold T0; compare W(t0) with a preset humidity threshold W0.

[0009] If O(t0) > O0, produce nitrogen and fill nitrogen into the battery box; if O(t0) ≤ O0, stop filling nitrogen into the battery box and delay and then stop nitrogen production. If P0' < P(t0) < P0, reduce the nitrogen filling flow rate; if P(t0) ≥ P0, stop filling nitrogen into the battery box and synchronously stop nitrogen production, and issue an alarm. If T(t0) ≥ T0, issue an alarm. If W(t0) ≥ W0, issue an alarm.

[0010] Step 3: With Δt as the period, return to Step 1; where

[0011] Step 4: Obtain the oxygen concentration, air pressure, temperature, and humidity of historical data, and establish and train a prediction model using a deep learning neural network to predict the battery state.

[0012] Among them, the battery state prediction includes predicting the subsequent change trends of oxygen concentration, air pressure, temperature, and humidity.

[0013] If it is predicted that O(t1) > O0 at the subsequent moment t1, nitrogen production and nitrogen storage shall start at the moment t2 before t1 to make preparations in advance. If it is predicted that P(t3) ≥ P0 at the subsequent moment t3, a warning shall be issued. If it is predicted that T(t5) ≥ T0 at the subsequent moment t5, a warning shall be issued. If it is predicted that W(t7) ≥ W0 at the subsequent moment t7, a warning shall be issued.

[0014] The power battery monitoring and protection method based on deep learning and nitrogen protection implements the method or process according to the embodiments of the present disclosure.

[0015] In a second aspect, the present invention discloses a power battery monitoring and protection system based on deep learning and nitrogen protection, which uses the power battery monitoring and protection method described in the first aspect.

[0016] The power battery monitoring and protection system includes a sensor monitoring module, a nitrogen production module, a data processing module, a deep learning prediction module, an alarm and warning module, and a vehicle-mounted control module.

[0017] The sensor monitoring module is used to detect the oxygen concentration O(t0), air pressure P(t0), temperature T(t0), and humidity W(t0) of the battery box where the battery is located at the current moment t0. The sensor monitoring module includes an oxygen concentration sensor, an air pressure sensor, a temperature sensor, and a humidity sensor. The sensor monitoring module is evenly arranged in the battery box where the battery is located. The nitrogen production module is used for nitrogen production, nitrogen filling, and nitrogen storage to keep the nitrogen in the battery box at a preset concentration. The data processing module is used to collect and statistically analyze the data collected by the sensor detection module to obtain processed data. The deep learning prediction module is used to preprocess the data processed by the data processing module, and then obtain a prediction result after training the prediction model. The alarm and warning module is used to provide an alarm to the user when the data collected by the sensor monitoring module exceeds a preset threshold, and at the same time provide a warning when a problem is about to occur according to the predicted battery state. The vehicle-mounted control module is used to adjust the nitrogen production module based on the data processed by the data processing module and the prediction result of the deep learning prediction module to realize the automatic control of nitrogen production, nitrogen filling, and nitrogen storage.

[0018] The power battery monitoring and protection system based on deep learning and nitrogen protection implements the method or process according to the embodiments of the present disclosure.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] The present invention applies nitrogen protection to the safety protection of power lithium batteries for new energy vehicles, preventing fires from occurring and improving the safety performance of power lithium batteries for new energy vehicles during driving. The present invention detects the oxygen concentration, air pressure, temperature, and humidity in the battery box where the battery is located, and performs nitrogen filling operations in a timely manner, so that the battery is in an environment where combustion and explosion conditions are not available. Based on historical detection data, the present invention constructs a neural network for model training to predict the battery state, can prepare for nitrogen production and storage in advance, ensure the normal operation of nitrogen filling, and further improve safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flowchart of the power battery monitoring and protection method based on deep learning and nitrogen protection in an embodiment of the present invention;

[0022] Figure 2 is Figure 1 The flowchart of step four in

[0023] Figure 3 It is a structural diagram of the power battery monitoring and protection system based on deep learning and nitrogen protection in an embodiment of the present invention;

[0024] Figure 4 is Figure 3 The structural diagram of the nitrogen production module in

[0025] Figure 5 is Figure 3 The control flow diagram of the power battery monitoring and protection system in

[0026] Figure 6 is Figure 3 The result diagram of the prediction model training performed by the deep learning prediction module in DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0028] It should be noted that when a component is referred to as "installed on" another component, it can be directly on the other component or there can also be an intermediate component. When a component is considered to be "set on" another component, it can be directly set on the other component or there may be an intermediate component at the same time. When a component is considered to be "fixed to" another component, it can be directly fixed to the other component or there may be an intermediate component at the same time.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this invention belongs. The terms used in the description of the present invention herein are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "or / and" used herein includes any and all combinations of one or more of the related listed items.

[0030] Embodiment 1

[0031] Please refer to Figure 1 , Figure 1 , which is a flowchart of the power battery monitoring and protection method based on deep learning and nitrogen protection in the present invention. It should be emphasized that this power battery monitoring and protection method is used for the safety protection of power lithium batteries of new energy vehicles. As is well known, power lithium batteries of new energy vehicles are installed in battery boxes.

[0032] The power battery monitoring and protection method includes the following steps:

[0033] Step 1: Obtain the oxygen concentration O(t0), air pressure P(t0), temperature T(t0), and humidity W(t0) of the battery box where the battery is located at the current moment t0.

[0034] Step 1 is to detect the parameters of the battery and its environment in real time. It should be noted that for the battery box, the minimum value of its nitrogen concentration should be ensured to be 97%.

[0035] The nitrogen concentration can be indirectly obtained by measuring the oxygen concentration O(t) in the battery box. This is because the gas components in the air are calculated by volume fraction: nitrogen accounts for about 78%, oxygen accounts for about 21%, and other gases and impurities account for about 1%. Ignoring the other gases and impurities accounting for 1% (which will not constitute factors for battery combustion and explosion), after determining the oxygen concentration O(t), the nitrogen concentration can be obtained by 1 - oxygen concentration O(t). In short, when the actually measured oxygen concentration O(t) is greater than 3% subsequently, nitrogen filling will be carried out.

[0036] Step 2: Compare the oxygen concentration O(t0), air pressure P(t0), temperature T(t0), and humidity W(t0) with preset thresholds, and perform control based on the comparison results.

[0037] (1) Compare O(t0) with the preset oxygen concentration threshold O0: If O(t0) > O0, produce nitrogen and charge nitrogen into the battery box at the nitrogen charging flow rate v; if O(t0) ≤ O0, stop charging nitrogen into the battery box and delay and then stop nitrogen production.

[0038] Specifically, according to the explanation in step one, in this embodiment, O0 = 3%.

[0039] (2) Compare P(t0) with the preset first air pressure threshold P0 and the preset second air pressure threshold P'0, where P'0 < P0:

[0040] If P(t0) ≤ P'0, charge nitrogen at the nitrogen charging flow rate v; if P'0 < P(t0) < P0, reduce the nitrogen charging flow rate v; if P(t0) ≥ P0, stop charging nitrogen into the battery box and simultaneously stop nitrogen production, and issue an alarm.

[0041] It is well-known that the atmospheric pressure is 101 KPa. In this embodiment, P0 = 120 KPa and P'0 = 110 KPa. Specifically, although the battery box can use pressure-resistant materials, if the air pressure inside the battery box exceeds P0, there are relatively large potential safety hazards during operation, and users should be reminded to pay attention.

[0042] The above nitrogen charging flow rate v adopts a preset value and can be modified as needed.

[0043] Of course, the nitrogen charging flow rate v can also be dynamically adjusted based on PID control. The nitrogen charging flow rate v is based on the following formula:

[0044]

[0045] where U(k) is the nitrogen charging flow rate, k represents time (k is the current time, k - 1 is the previous time), e() is the difference function, K p is the proportionality coefficient, K i is the integral coefficient, K d is the derivative coefficient.

[0046] e() represents the difference between the target nitrogen concentration and the existing nitrogen concentration. The first term in the above formula is the proportional term. Since the nitrogen charged is related to the difference, the first term is used to determine the nitrogen charging flow rate. However, in actual situations, there may be gas leakage, so there will be a steady-state error. The second term in the above formula is the integral term (accumulation), which is used to accumulate the errors of the previous several times and can well eliminate the steady-state error. The third term in the above formula is the derivative term (the difference of e()), so that when approaching the target nitrogen concentration, the nitrogen production speed is reduced to reduce the oscillation during the nitrogen charging process.

[0047] In this embodiment, when the oxygen concentration in the battery box is greater than 3%, it indicates that nitrogen needs to be filled. The nitrogen generation module operates to fill nitrogen into the battery box, and the oxygen concentration in the battery box is monitored in real time. It is recommended to set the nitrogen filling flow rate at the start to 2 Nm 3 / h. Take K p as 0.8, and K i as 1.2, and K d as 1.3. Subsequently, the nitrogen filling flow rate v is determined according to the value of U(k).

[0048] (3) Compare T(t0) with the preset temperature threshold T0: If T(t0) < T0, it indicates that the battery temperature is normal; if T(t0) ≥ T0, an alarm is issued, indicating that the battery is abnormally overheated. Specifically, the battery temperature is an important basis for reflecting its state. In this embodiment, it is recommended that T0 = 60°C.

[0049] (4) Compare W(t0) with the preset humidity threshold W0: If W(t0) < W0, it indicates that the humidity of the environment where the battery is located is normal; if W(t0) ≥ W0, an alarm is issued, indicating that the humidity of the environment where the battery is located is too high and the battery is prone to short - circuit during operation. In this embodiment, it is recommended that W(t0) = 75%.

[0050] Step two provides a specific control method to automatically fill nitrogen into the battery box, so that the nitrogen concentration in the battery box reaches above the safety value, thereby making the conditions for combustion and explosion not exist in the battery box, and truly realizing prevention before "ignition".

[0051] Step three, with Δt as the period, return to step one. Among them

[0052] It should be noted that the above - mentioned current time t0 represents the real - time time, and because after Δt, the current time t0 corresponding to "return to step one" is also updated in real time. In this embodiment, it is recommended to set Δt = 2 s.

[0053] And it can ensure that at least one more oxygen concentration acquisition and comparison are carried out within, and the control is carried out according to the above - mentioned control method. In this embodiment, it is recommended

[0054] By periodically acquiring the real - time parameters of the battery and its environment, adaptive nitrogen filling control can be carried out on the battery box, creating a safe environment for the battery and avoiding waste caused by continuous nitrogen filling.

[0055] Step four, obtain the historical data of oxygen concentration, air pressure, temperature, and humidity, and establish and train a prediction model using a deep - learning neural network to predict the battery state.

[0056] Refer to Figure 2, which is the flowchart of Step 4. It should be emphasized that before establishing and training a prediction model using a deep learning neural network, the oxygen concentration, air pressure, temperature, and humidity at historical moments need to be preprocessed:

[0057] Among them, the historical moment t 历史 is earlier than the current moment t0. t 历史 and t0 form a time interval [t 历史 , t0], which ends at the current moment t0. In this way, the oxygen concentration, air pressure, temperature, and humidity within the time interval [t 历史 , t0] are historical data, which also constitute the original data and are sorted in chronological order.

[0058] Analyze and process the original data, transform it into features that can be used by the model, better describe the potential laws to the prediction model, and thus improve the accuracy of the model for unseen data. Specifically, it includes:

[0059] 1. Select the EDA data analysis library pandas profiling for exploratory data analysis.

[0060] 2. Determine data outliers through the 3σ criterion, delete or replace them, and train a regression model to predict and fill in the missing values.

[0061] 3. Use the featuretools tool to generate features for the data, make up for the limited expression of basic features for sample information, increase the non-linear expression ability of features, and improve the model effect.

[0062] After completing the preprocessing of the original data,

[0063] Construct a neural network model and select the sigmoid activation function to ensure the learning efficiency of the neural network.

[0064] Initialize the model weights using the normal Gaussian distribution to accelerate the model convergence speed.

[0065] Perform batch normalization on the values. In this way, while retaining the input information, the distribution differences between layers can be eliminated, which has the effect of accelerating convergence and is similar to introducing noise regularization.

[0066] The batch normalization process is based on the formula:

[0067]

[0068] Among them, x * is the processed data, x is the data before processing, μ is the mean, and σ is the variance.

[0069] Use L2 parameter regularization to reduce the generalization error, suppress overfitting, and improve the generalization ability of the model.

[0070] Specifically, L2 parameter regularization makes the weights closer to the origin and the model simpler (smaller container) by adding a regularization term Ω(θ) to the objective function. From a Bayesian perspective, the constraint term of L2 can be regarded as introducing a prior Gaussian distribution to the model parameters, as shown in the formula:

[0071]

[0072] where X is the training sample, y is the corresponding label of the sample, ω is the weight coefficient vector, J() is the objective function, is the regularized objective function, and α is the adjustment parameter used to control the strength of regularization.

[0073] Update the weights of the model parameters for the L2 objective function:

[0074]

[0075] where ε is the learning rate, is the gradient direction.

[0076] Divide the preprocessed data into a training set and a test set, and then further divide the training set into a sub-training set and a sub-validation set. The sub-training set is used to run the learning algorithm and train the model. The sub-validation set is used to adjust the model hyperparameters, select features, etc. to select a suitable prediction model. The test set is only used to evaluate the performance of the selected prediction model, but will not change the learning algorithm or parameters accordingly.

[0077] It should be noted that the grid search method is used to adjust the hyperparameters to select suitable hyperparameters.

[0078] In this way, a prediction model is obtained based on the historical oxygen concentration O(t), air pressure P(t), temperature T(t), and humidity W(t), and the future oxygen concentration O(t), air pressure P(t), temperature T(t), and humidity W(t) can be predicted.

[0079] Specifically, if it is predicted that O(t1) > O0 at the subsequent time t1, then nitrogen production and storage start at time t2 before t1 to make preparations in advance. If it is predicted that P(t3) ≥ P0 at the subsequent time t3, a warning is issued. If it is predicted that T(t5) ≥ T0 at the subsequent time t5, a warning is issued. If it is predicted that W(t7) ≥ W0 at the subsequent time t7, a warning is issued.

[0080] where t2 - t1 = Δt1, and Δt1 can be adjusted according to needs.

[0081] Of course, for the warning issuance times of air pressure, temperature, and humidity, the same method as that for oxygen can be adopted, and the corresponding times t4 before t3, t6 before t5, and t8 before t7 are selected.

[0082] Among them, t4 - t3 = Δt2, t6 - t5 = Δt3, t8 - t7 = Δt4. Similarly, Δt2, Δt3, and Δt4 can also be adjusted according to needs. In this embodiment, it is recommended that Δt1 = Δt2 = Δt3 = Δt4 = 3 min.

[0083] It should be noted that as the number of operating cycle times increases, t0 is updated in real time, and the time interval where the historical moment is located is also updated, so that the historical data is also updated synchronously. In this way, the prediction model established based on the historical data is continuously iterated, and better hyperparameters are selected to make the prediction result more accurate.

[0084] The method for determining historical data in this embodiment is as follows:

[0085] Taking t0 as the starting moment, pushing back N data points to the historical moment for the first time, the moment corresponding to the Nth data point is t N ; calculating the variance σ of the N data points pushed back for the first time N ;

[0086] If σ N < 0.8, it indicates that the difference of these N data points is small, and the number of values taken from them should be less: arranging these N data points into A groups in chronological order, calculating one mean value for each group, a total of A mean values, as the first component of the historical data; if σ N ≥ 0.8, it indicates that the difference of these N data points is large, and the number of values taken from them should be more: arranging these N data points into B groups in chronological order, calculating one mean value for each group, a total of B mean values, as the first component of the historical data; A < B;

[0087] Taking t N as the starting moment, pushing back N data points to the historical moment for the second time, the moment corresponding to the 2Nth data point is t 2N ; calculating the variance σ of the N data points pushed back for the second time 2N ;

[0088] If σ 2N < 0.8, it indicates that the difference of these N data points is small, and the number of values taken from them should be less: arranging these N data points into A groups in chronological order, calculating one mean value for each group, a total of A mean values, as the second component of the historical data; if σ 2N≥0.8 indicates that the N data points have large differences, and more data points should be selected from them. Arrange the N data points in chronological order and divide them into B groups evenly. Calculate one average value for each group, with a total of B average values, which serve as the second component of the historical data.

[0089] Generally speaking, starting from time t (i-1)N as the starting moment, push back N data points to the historical moment for the i-th time. The moment corresponding to the iN-th data point is t iN ; calculate the variance σ of the N data points pushed back for the i-th time iN ; i ≥ 1;

[0090] Among them, when i = 1, t (i-1)N take t0, t iN is t N , σ iN take σ N ;

[0091] If σ iN < 0.8, it indicates that the N data points have small differences, and fewer data points should be selected from them. Arrange the N data points in chronological order and divide them into A groups evenly. Calculate one average value for each group, with a total of A average values, which serve as the i-th component of the historical data. If σ iN ≥0.8, it indicates that the N data points have large differences, and more data points should be selected from them. Arrange the N data points in chronological order and divide them into B groups evenly. Calculate one average value for each group, with a total of B average values, which serve as the i-th component of the historical data.

[0092] Obtain the first component to the i-th component of the historical data and count the total number of the historical data until the total number of the historical data reaches M.

[0093] In this embodiment, it is recommended that A be 2, B be 5, N be 100, and M be 300. Of course, the above parameters can also be adjusted.

[0094] It should be emphasized that in this embodiment, the size of the time interval [t 历史 , t0] is not fixed, and t 历史 is not fixed either. The method for determining the historical data in this embodiment is more representative and accurate compared to using a fixed-size time interval; compared to fixing t 历史 , it can reduce the learning pressure of the neural network.

[0095] In addition, since step four requires operations such as data processing, if there are reasons such as insufficient processing performance, it cannot be completed within a short time such as Δt = 2s. Therefore, considering costs and practical applications, in this embodiment, the prediction in step four is set to be once per minute, that is, the battery state is predicted once every minute.

[0096] This embodiment also discloses a power battery monitoring and protection system based on deep learning and nitrogen protection, which uses the above-mentioned power battery monitoring and protection method.

[0097] Specifically, refer to Figure 3 which is the structure diagram of the power battery monitoring and protection system. The power battery monitoring and protection system includes: a sensor monitoring module, a nitrogen generation module, a data processing module, a deep learning prediction module, an alarm and early warning module, and a vehicle-mounted control module.

[0098] The sensor monitoring module is used to detect the oxygen concentration O(t0), air pressure P(t0), temperature T(t0), and humidity W(t0) of the battery box where the battery is located at the current moment t0. The sensor monitoring module includes an oxygen concentration sensor, an air pressure sensor, a temperature sensor, and a humidity sensor. The sensor monitoring module is uniformly arranged in the battery box where the battery is located. Specifically, in this embodiment, the oxygen concentration sensor can adopt the S40XV module, which is uniformly arranged in the battery box to detect the oxygen concentration and then determine the nitrogen concentration. The air pressure sensor can adopt the HX710B, which is uniformly arranged in the battery box and is linked with the nitrogen generation module to control the nitrogen filling flow according to the pressure. The temperature sensor adopts the DHT11 temperature and humidity module and is uniformly arranged on the battery to detect the battery temperature and the humidity state around the battery.

[0099] It should be noted that multiple oxygen concentration sensors, air pressure sensors, temperature sensors, and humidity sensors are provided. The oxygen concentration O(t0) is the average value of the collected values of multiple oxygen concentration sensors. The same applies to the air pressure P(t0), temperature T(t0), and humidity W(t0), and all of them adopt the average value.

[0100] The nitrogen generation module is used for nitrogen generation, nitrogen filling, and nitrogen storage to keep the nitrogen in the battery box at a preset concentration.

[0101] Refer to Figure 4 and the nitrogen generation module includes an air compressor, a compressed air tank, an air filter, a membrane separator, and a nitrogen buffer tank.

[0102] The air compressor is used to compress the outside air. One end of the compressed air tank is connected to the air compressor and is used to store the outside air compressed by the air compressor. The inlet of the air filter is connected to the other end of the air tank and is used to adsorb, filter, and purify the compressed outside air. The inlet of the membrane separator is connected to the outlet of the air filter and is used for oxygen-nitrogen adsorption separation. One end of the nitrogen buffer tank is connected to the outlet of the membrane separator and is used to store the separated nitrogen. The other end of the nitrogen buffer tank is connected to the battery box through a ventilation pipe, and the ventilation pipe is equipped with an electronically controlled flow valve.

[0103] The air compressor compresses the outside air and transports it into the compressed air tank. Then, it is adsorbed, filtered, and purified through an air filter, and then passed into a membrane separator. The molecular membrane is used for oxygen-nitrogen adsorption separation. Then, nitrogen gas flows out from the outlet end of the membrane separator and enters the nitrogen buffer tank for storage. In this way, by controlling the electronic control flow valve, nitrogen gas can be filled into the battery box and the nitrogen filling flow rate can be controlled.

[0104] The data processing module is used to collect and count the data collected by the sensor detection module to obtain the processed data. It should be noted that the data processing module classifies the collected data according to the data type and time period. For example, the oxygen concentration sensor corresponds to the oxygen concentration at a certain time. The data processing module also sends the data to the deep learning prediction module in the order of time.

[0105] The deep learning prediction module is used to preprocess the data processed by the data processing module, and then obtain the prediction result after training the prediction model. The deep learning prediction module receives the data sent by the data processing module, which is the original data sorted in the order of time. The training of the prediction model is carried out according to the fourth step of the above method, which will not be elaborated here.

[0106] The alarm and early warning module is used to provide an alarm to the user when the data collected by the sensor monitoring module exceeds the preset threshold, and at the same time, according to the predicted battery state, provide an early warning when a problem is about to occur.

[0107] Specifically, the vehicle-mounted control module receives the battery state prediction of the deep learning prediction module. If the predicted value indicates that the battery state is about to reach the critical value (i.e., the preset threshold), it will immediately send the early warning or alarm information to the alarm module. The alarm and early warning module receives the early warning or alarm information sent by the vehicle-mounted control module and promptly notifies the driver through the vehicle-mounted large screen and voice prompts to ensure the safety of the vehicle during driving.

[0108] Of course, the alarm and early warning module can also provide functions such as viewing and deleting historical alarm information for management operations.

[0109] The vehicle-mounted control module is used to adjust the nitrogen generation module based on the data processed by the data processing module and the prediction result of the deep learning prediction module to achieve automatic control of nitrogen generation, nitrogen filling, and nitrogen storage.

[0110] The vehicle-mounted control module adjusts the nitrogen generation module according to the above power battery monitoring and protection method, which specifically includes two aspects: one is real-time monitoring, and the other is prediction protection:

[0111] (1) Real-time monitoring:

[0112] Compare O(t0) with the preset oxygen concentration threshold O0: If O(t0) > O0, start the nitrogen generation module to generate nitrogen, open the electronically controlled flow valve, and adjust the opening amplitude of the electronically controlled flow valve to fill nitrogen into the battery box at a nitrogen filling flow rate v; if O(t0) ≤ O0, close the electronically controlled flow valve and delay and then shut down the nitrogen generation module to stop filling nitrogen into the battery box.

[0113] It should be noted that this ensures that oxygen concentration acquisition and comparison are performed at least once more within [], and control is carried out according to the above control method, which can avoid frequent start and stop of the nitrogen generation module.

[0114] Compare P(t0) with the preset pressure threshold P0 and the preset pressure threshold P'0, where P'0 < P0: If P(t0) ≤ P'0, adjust the opening amplitude of the electronically controlled flow valve to fill nitrogen at a nitrogen filling flow rate v; if P'0 < P(t0) < P0, reduce the opening amplitude of the electronically controlled flow valve to decrease the nitrogen filling flow rate v; if P(t0) ≥ P0, close the electronically controlled flow valve and simultaneously shut down the nitrogen generation module to stop filling nitrogen into the battery box, and send an alarm through the vehicle-mounted control module.

[0115] For the control of the nitrogen filling flow rate v, refer to the above method and will not be elaborated here.

[0116] It should be noted that although the electronically controlled flow valve and the nitrogen generation module are controlled based on the oxygen concentration O(t0) and the pressure P(t0), for safety considerations, the control priority of the pressure P(t0) is higher than that of the oxygen concentration O(t0). For example, if it is detected that P(t0) ≥ P0, even if O(t0) > O0, it should be executed according to P(t0) ≥ P0, that is, close the electronically controlled flow valve, simultaneously shut down the nitrogen generation module, stop filling nitrogen into the battery box, and send an alarm through the vehicle-mounted control module.

[0117] Compare T(t0) with the preset temperature threshold T0: If T(t0) ≤ T0, operate normally; if T(t0) > T0, send an alarm through the vehicle-mounted control module.

[0118] Compare W(t0) with the preset humidity threshold W0: If W(t0) ≤ W0, operate normally; if W(t0) > W0, send an alarm through the vehicle-mounted control module.

[0119] (2) Prediction protection:

[0120] If it is predicted that O(t1) > O0 at the subsequent time t1, then at the time t2 before t1, keep the electronically controlled flow valve closed, start the nitrogen generation module to generate and store nitrogen in advance for preparation.

[0121] This is because it takes a certain amount of time for the nitrogen generation module to start working and output nitrogen. If the nitrogen generation module is started to generate nitrogen only at t1, it will cause a lag.

[0122] If it is predicted that P(t3) ≥ P0 at a subsequent time t3, then at time t4 before t3, a warning is issued through the vehicle-mounted control module.

[0123] If it is predicted that T(t5) ≥ T0 at a subsequent time t5, then at time t6 before t5, a warning is issued through the vehicle-mounted control module.

[0124] If it is predicted that W(t7) ≥ W0 at a subsequent time t7, then at time t8 before t7, a warning is issued through the vehicle-mounted control module.

[0125] In addition, the vehicle-mounted control module can also be remotely controlled. Specifically, the vehicle-mounted control module uploads the data collected by the sensor monitoring module to the cloud in real time, reads the data through remote access to the cloud for monitoring, and can also issue commands through the cloud for remote management.

[0126] Considering the interaction experience, the power battery monitoring and protection system can also include a data display module. The data display module receives the data processed by the data processing module and displays it in a graphical form by category and time period on the vehicle-mounted large screen and the page, so that it is convenient for users to intuitively understand the battery usage situation.

[0127] The data display module also shows the prediction results of the deep learning prediction module, so that it is convenient for users to understand the subsequent change trend of the battery state, know in advance when an abnormality is about to occur, make preparations for protection or replacement, and prevent accidents. See Figure 5 , which is the control flow chart of the above power battery monitoring and protection system.

[0128] Embodiment 2

[0129] This embodiment provides an example of the application of Embodiment 1, and its object of action is a certain model of new energy vehicle, and the battery specification is a 310Ah lithium iron phosphate battery.

[0130] The new energy vehicle was actually measured for one month. Based on the test data of the first 24 days, a prediction model was established to obtain the prediction data for the next 7 days.

[0131] First, when controlling the nitrogen of the power lithium battery of the new energy vehicle, the nitrogen generation module does not always start working, nor does it start and stop frequently, and the power consumption is also saved.

[0132] Secondly, see Figure 6 , which is the result graph of the prediction model training by the deep learning prediction module. Figure 6The horizontal axis represents the number of deep learning training times, and the loss on the vertical axis is MSELoss, that is, the mean square loss function (the MSE of the difference function between the predicted value and the true value), and EMA_loss is the EMA (exponential moving average) of loss.

[0133] It can be seen that as the number of training times increases, the curve shows a downward trend, indicating that the difference between the predicted value and the true value (experimental data) is smaller, that is, the error of the model is getting smaller and smaller.

[0134] Compare the predicted data for the next 7 days with the experimental data, as shown in the following table:

[0135] Table 1 Comparison of predicted data and experimental data for the next 7 days

[0136]

[0137] It can be seen that after multiple trainings, the error rate gradually decreases and finally stabilizes at 0.59%, and the accuracy meets the requirements.

[0138] It should be noted that the error rate is obtained by calculating MSE based on the data in Table 1:

[0139] Among them, y is the true value and y' is the predicted value.

[0140] Embodiment 3

[0141] This embodiment also discloses a readable storage medium. When computer program instructions stored in the readable storage medium are read and run by a processor, they execute the power battery monitoring and protection method based on deep learning and nitrogen protection in Embodiment 1.

[0142] When the method in Embodiment 1 is applied, it can be applied in the form of software. For example, it can be designed as a program that can run independently on a computer-readable storage medium. The computer-readable storage medium can be a USB flash drive, designed as a USB key, and designed as a program that starts the entire method through external triggering via the USB flash drive.

[0143] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0144] The above-described embodiments merely represent several implementation manners of the present invention. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. A power battery monitoring and protection method based on deep learning and nitrogen protection, which is used for safety protection of power lithium batteries in new energy vehicles, and is characterized in that: The power battery monitoring and protection method includes the following steps: Step 1: Obtain the oxygen concentration O(t0), air pressure P(t0), temperature T(t0), and humidity W(t0) of the battery box where the battery is located at the current moment t0. Step 2: Compare O(t0) with the preset oxygen concentration threshold O0; compare P(t0) with the first preset air pressure threshold P0 and the second preset air pressure threshold P0'; compare T(t0) with the preset temperature threshold T0; compare W(t0) with the preset humidity threshold W0. If O(t0) > O0, nitrogen production is carried out and nitrogen is filled into the battery box; if O(t0) ≤ O0, nitrogen filling into the battery box is stopped and nitrogen production is stopped after a time delay ; if P0' < P(t0) < P0, the nitrogen filling flow rate is reduced; if P(t0) ≥ P0, nitrogen filling into the battery box is stopped and nitrogen production is stopped synchronously, and an alarm is issued; if T(t0) ≥ T0, an alarm is issued; if W(t0) ≥ W0, an alarm is issued; Step 3, taking Δt as the period, return to Step 1; where, Step 4: Obtain the oxygen concentration, air pressure, temperature, and humidity of historical data, and use a deep learning neural network to establish and train a prediction model for battery state prediction. Among them, the battery state prediction includes predicting the subsequent change trends of oxygen concentration, air pressure, temperature, and humidity; if it is predicted that O(t1) > O0 at the subsequent moment t1, nitrogen production and nitrogen storage will start at the moment t2 before t1 to make preparations in advance. If it is predicted that P(t3) ≥ P0 at the subsequent moment t3, a warning will be issued; if it is predicted that T(t5) ≥ T0 at the subsequent moment t5, a warning will be issued; if it is predicted that W(t7) ≥ W0 at the subsequent moment t7, a warning will be issued.

2. The method for monitoring and protecting a power battery based on deep learning and nitrogen protection according to claim 1, characterized in that, In Step 2, when O(t0) > O0, the nitrogen filling flow rate for nitrogen filling is based on PID control and is calculated according to the following formula: Among them, U(k) is the nitrogen filling flow rate, k represents time, e() is the difference function, and K p is the proportionality coefficient, and K i is the integral coefficient, and K d is the differential coefficient.

3. The power battery monitoring and protection method based on deep learning and nitrogen protection according to claim 1, characterized in that The method for obtaining historical data is as follows: Starting from time t (i-1)N backward N data points from the historical time for the i-th time, the time corresponding to the iN-th data point is t iN ; Calculate the variance σ of the N data points for the i-th backtracking iN ; i ≥ 1; Among them, when i = 1, t (i-1)N Take t0, t iN as t N , σ iN Take σ N ; If σ iN < 0.8, the N data points are evenly divided into A groups in chronological order, and one mean value is calculated for each group, with a total of A mean values, which are used as the i-th component of the historical data; if σ iN ≥ 0.8, the N data points are evenly divided into B groups in chronological order, and one mean value is calculated for each group, with a total of B mean values, which are used as the i-th component of the historical data; A < B; Obtain the first composition to the i-th composition of historical data, and count the total number of historical data until the total number of historical data reaches M.

4. The power battery monitoring and protection method based on deep learning and nitrogen protection according to claim 1, characterized in that In Step 4, before using a deep learning neural network to establish and train a prediction model, preprocess the oxygen concentration, air pressure, temperature, and humidity of the obtained historical data. The preprocessing includes: performing exploratory data analysis, outlier and missing value processing, and feature generation processing.

5. The power battery monitoring and protection method based on deep learning and nitrogen protection according to claim 4, wherein, In Step 4, the method for establishing the prediction model is as follows: Construct a neural network model structure and select the sigmoid activation function. Initialize the model weights, perform batch normalization on the values, and regularize the L2 parameters. Divide the sorted data into a training set and a test set. The training set is further divided into a sub-training set and a sub-validation set, and the model is trained until convergence or reaches the preset number of iterations. Among them, the sub-training set is used to run the learning algorithm and train the model; the sub-validation set is used to adjust the model hyperparameters and select features to select a suitable prediction model; the hyperparameter debugging uses the grid search method; the test set is used to evaluate the performance of the selected prediction model.

6. A power battery monitoring and protection system based on deep learning and nitrogen protection, characterized in that, It uses the power battery monitoring and protection method described in any one of claims 1-5. The power battery monitoring and protection system includes: A sensor monitoring module, which is used to detect the oxygen concentration O(t0), air pressure P(t0), temperature T(t0), and humidity W(t0) of the battery box where the battery is located at the current moment t0; the sensor monitoring module includes an oxygen concentration sensor, an air pressure sensor, a temperature sensor, and a humidity sensor; the sensor monitoring module is uniformly arranged in the battery box where the battery is located. A nitrogen production module, which is used for nitrogen production, nitrogen filling, and nitrogen storage to keep the nitrogen in the battery box at a preset concentration. A data processing module, which is used to collect and statistically analyze the data collected by the sensor monitoring module to obtain processed data; A deep learning prediction module, which is used to preprocess the data processed by the data processing module, and then obtain a prediction result after training a prediction model; An alarm and early warning module, which is used to provide an alarm to the user when the data collected by the sensor monitoring module exceeds a preset threshold, and at the same time provide an early warning when a problem is about to occur according to the predicted battery state; and A vehicle-mounted control module, which is used to adjust the nitrogen generation module based on the data processed by the data processing module and the prediction result of the deep learning prediction module to realize the automatic control of nitrogen generation, nitrogen filling, and nitrogen storage.

7. The power battery monitoring and protection system based on deep learning and nitrogen protection according to claim 6, characterized in that, The nitrogen generation module includes: An air compressor, which is used to compress external air; A compressed air tank, one end of which is connected to the air compressor and is used to store the external air compressed by the air compressor; An air filter, the inlet of which is connected to the other end of the air tank and is used to adsorb, filter, and purify the compressed external air; A membrane separator, the inlet of which is connected to the outlet of the air filter and is used for oxygen-nitrogen adsorption and separation; and A nitrogen buffer tank, one end of which is connected to the outlet of the membrane separator and is used to store the separated nitrogen; the other end of the nitrogen buffer tank is connected to the battery box through a ventilation pipe, and an electronically controlled flow valve is provided on the ventilation pipe.

8. The power battery monitoring and protection system based on deep learning and nitrogen protection according to claim 6, wherein The data processing module classifies the data collected by the sensor monitoring module according to data types and time periods, and sends it to the deep learning prediction module in chronological order.

9. The power battery monitoring and protection system based on deep learning and nitrogen protection according to claim 6, characterized in that The vehicle-mounted control module also uploads the data collected by the sensor monitoring module in real time to realize remote management and monitoring.

10. The power battery monitoring and protection system based on deep learning and nitrogen protection according to claim 6, characterized in that, The power battery monitoring and protection system further includes: A data display module, which is used to receive the data processed by the data processing module and display it in a graphical form on the vehicle-mounted large screen and the page by category and time period; Or / and, the data display module also displays the prediction result of the deep learning prediction module.

Images