Intelligent management method, detection system and management front end of electric bicycle shed

Abstract

The invention discloses an intelligent management method for an electric bicycle shed, and the method comprises the steps: enabling an intelligent sensing device to be connected into the electric bicycle shed, obtaining the data of an electric bicycle from the electric bicycle shed through the intelligent sensing device, dividing the data of the electric bicycle through the combination of an improved RBF neural network, and obtaining the divided data of the electric bicycle; an abnormal identification and control algorithm is obtained through continuous training and testing, so that electric bicycle battery faults are rapidly identified, fire hazards of an electric bicycle shed are reduced, 1s-15min high-frequency acquisition, high-frequency ammeter simulation and digital twinborn simulation of power consumption data of a user can be realized through a data acquisition device, the charging load of the electric bicycle is detected, and the charging load of the electric bicycle is detected. The fault of the single battery of the electric bicycle is diagnosed through an RBF neural network, and abnormal modes are further diagnosed and classified; and the alarm module is used for sending a real-time fault alarm and notification to a manager or a user so as to ensure that the manager or the user is timely notified to take corresponding measures.

Description

Technical Field

[0001] The present invention relates to the technical field of electric bicycle sheds, and in particular to an intelligent management method, a detection system and a management front end for an electric bicycle shed. Background Art

[0002] In recent years, as the concept of green, energy-saving and environmentally friendly life has been deeply rooted in people's hearts, electric bicycles have gradually become the main means of transportation for short-distance travel for urban residents due to their convenience, speed, economy and environmental protection. Faced with the explosive growth in the number of electric bicycles, the supporting facilities in some old residential communities are updated and improved slowly, and the residents of the communities have a weak sense of safety. They privately pull wires to charge in the ground floor corridors, hallways and even safety passages, posing a major safety hazard.

[0003] In order to effectively prevent and control electric bicycle fires, protect personal and property safety and maintain community public safety, it is strictly forbidden to park and charge electric bicycles in public areas such as the ground floor lobby, shared corridors, stairwells, corridors, and indoor areas of personal rooms and crowded places of buildings; property management, village committees and neighborhood committees are required to organize regular special inspections on parking and charging of electric vehicles to eliminate hidden dangers in a timely manner; however, on the one hand, regulatory entities such as property management do not have law enforcement power and compulsory constraints, and on the other hand, electric bicycles usually move during the day and are parked and charged at night, and the charging behavior is random and hidden. Therefore, in the actual supervision process, the efficiency and effectiveness of human inspection work are low, and it is impossible to effectively prevent electric bicycle fire accidents; therefore, an intelligent management method, detection system and management front end of electric bicycle sheds are proposed to solve the above problems. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: an intelligent management method for electric bicycle sheds, comprising:

[0006] An intelligent sensing device is connected to an electric bicycle shed, the intelligent sensing device obtains electric bicycle data from the electric bicycle shed, divides the electric bicycle data into a training set and a test set, trains and tests an improved RBF neural network, and finally forms an abnormality recognition algorithm.

[0007] As a preferred solution of the electric bicycle shed intelligent management method described in the present invention, the intelligent sensing device is composed of a high-frequency data acquisition device, a refined electricity consumption data acquisition and application service platform, and an advanced application development service.

[0008] As a preferred solution of the electric bicycle shed intelligent management method described in the present invention, a state transition removal algorithm is used to remove load events in the total active power P and total reactive power Q with a window length of 6 hours.

[0009] As a preferred solution of the electric bicycle shed intelligent management method described in the present invention, the sampling frequency is reduced to 1 / 30 Hz, and Savitzky-Golay filtering is used to reduce the noise in the power signal after the frequency reduction; and the difference of the power signal is calculated.

[0010] As a preferred solution of the electric bicycle shed intelligent management method described in the present invention, the continuous negative subsequences in the active power differential signal are matched with the continuous positive subsequences in the reactive power differential signal, and the distance between the matched continuous subsequences and the electric bicycle charging load template is calculated.

[0011] In order to solve the above technical problems, the present invention also provides the following technical solutions: an electric bicycle shed battery detection system, comprising:

[0012] The training data set is input into the optimized radial basis function (RBF) neural network for learning and training to identify the fault characteristics of electric bicycle batteries.

[0013] As a preferred solution of the electric bicycle shed battery detection system of the present invention, after the training is completed, the test data set is used to perform fault diagnosis classification test on the trained RBF neural network to verify the diagnostic accuracy of the model.

[0014] As a preferred solution of the electric bicycle shed battery detection system of the present invention, the diagnosis result of the neural network is interpreted by the battery management system to achieve accurate diagnosis of electric bicycle battery failure.

[0015] To solve the above technical problems, the present invention also provides the following technical solutions: an intelligent management front end for an electric bicycle shed, comprising six interfaces, namely, a device status module, an alarm information module, an alarm ratio module, a total power module, a total power module, and a charging information module.

[0016] As a preferred solution of the electric bicycle shed intelligent management front end of the present invention, wherein: the device status module realizes remote monitoring of the device status;

[0017] The alarm information module pushes the alarm information of all intelligent sensing devices in real time;

[0018] The alarm proportion module calculates the proportion of different types of alarm information based on all historical alarm information;

[0019] The total power module displays the cumulative power of all intelligent sensing devices from 0 to 24 hours of the day in the form of a bar graph;

[0020] The total power module displays the accumulated power of all intelligent sensing devices from 0 to 24 hours of the day in the form of a curve;

[0021] The charging information module is mainly used to display basic information such as the number of charges today, the amount of electricity charged today, the charging time today, and the number of alarms today.

[0022] The beneficial effects of the present invention are as follows: the data acquisition device can realize high-frequency acquisition of the user's electricity consumption data of 1s to 15min, high-frequency electric meter simulation, and digital twin simulation, so as to detect the charging load of the electric bicycle, diagnose the battery monomer failure of the electric bicycle through the RBF neural network, and further diagnose and classify the abnormal mode; the alarm module is used to send real-time fault alarms and notifications to managers or users, ensuring that managers or users are notified in time to take corresponding measures.

[0023] At the same time, the development of an electric bicycle safe charging management platform has been completed. It is deeply integrated with intelligent sensing devices and coordinated with software and hardware to achieve remote monitoring, unified management, and real-time early warning of the charging status of electric bicycles. It can promptly identify and cut off abnormal charging loads to maintain the safety of the park. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0025] Figure 1 It is a schematic diagram of the overall structure of the present invention.

[0026] Figure 2 It is a functional schematic diagram of the data acquisition device in the present invention.

[0027] Figure 3 This is an example of data collection by the intelligent data collection device in the present invention and an example diagram of data collection by an electric bicycle shed.

[0028] Figure 4 It is a charging load power curve diagram of an abnormal electric bicycle battery in the present invention.

[0029] Figure 5 Schematic diagram of bilateral filtering in the present invention.

[0030] Figure 6 It is a schematic diagram of state transition removal in the present invention.

[0031] Figure 7 This is a schematic diagram of SG filtering in the present invention.

[0032] Figure 8 Schematic diagram of PLR in the present invention.

[0033] Fig. 9 This is a flow chart of the online rapid detection method of the electric bicycle charging load based on local features in the present invention.

[0034] Fig.10 This is a schematic diagram of the constant voltage charging stage model of an electric bicycle in the present invention.

[0035] Fig.11 This is a model diagram of the charging loads of 10 electric bicycles in the present invention.

[0036] Fig.12 This is a schematic diagram of SG filtering after load event detection and state transition removal for user No. 1 on a certain day in the present invention.

[0037] Fig.13 Schematic diagram of the power difference signal and the subsequences determined to be consecutively positive or negative in the present invention.

[0038] Fig.14 This is a schematic diagram of the misdetection result of the charging load of an electric bicycle of user No. 2 in the present invention.

[0039] Fig.15 This is a schematic diagram of the online rapid detection result of the charging load of an electric bicycle of user No. 1 in the present invention.

[0040] Fig.16 This is a diagram showing the duration of the constant voltage phase when the charging load of the electric bicycle is detected in the present invention.

[0041] Fig.17 This is a flow chart of fault diagnosis of battery cells of an electric bicycle in the present invention.

[0042] Fig.18 This is the improved RBF neural network diagram in the present invention. DETAILED DESCRIPTION

[0043] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0044] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0045] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0046] Example 1, reference Figures 1 to 9 This embodiment provides an intelligent management method for electric bicycle sheds, which solves the problem of a flowchart of an online rapid detection method for electric bicycle charging load based on local features, and includes:

[0047] Charging data preprocessing:

[0048] Bilateral filtering algorithm

[0049] Bilateral filtering can effectively retain the edge information of data when smoothing redundant values ​​and removing noise, and is widely used in the field of image data processing. In the household electricity consumption data with a sampling frequency of 1 / 60Hz, the transition between the states of electrical load startup and shutdown, that is, load events, is basically manifested as steps. When the small amplitude fluctuations or noise generated during its operation coincide with the constant voltage charging stage of electric bicycles, it will interfere with its detection. Commonly used filtering algorithms, such as Gaussian filtering, smooth the step edges in the signal while smoothing the signal noise, resulting in unnecessary transition processes in the filtered step edges. Therefore, this topic uses bilateral filtering to remove measurement noise and retain step edges. In the bilateral filtering algorithm, the weight of each sample in the neighborhood is first determined according to the similarity between the target sample and the neighborhood sample, and then the weighted strength of the neighborhood sample is calculated to determine the strength of the target sample. The difference from Gaussian filtering is that the size of the bilateral filtering weight is not only related to the physical distance between samples, but also to factors such as the intensity distance between samples. Its kernel function is the product of the spatial smoothing function and the distance smoothing function. This topic uses Gaussian bilateral filtering, that is, both smoothing functions are Gaussian functions. Formula (1) is the calculation formula for bilateral filtering weights:

[0050]

[0051] Among them, i represents the spatial position index of the target sample; j represents the spatial position index of the neighborhood sample; f i and f j Represent the sample values ​​of the target sample and the neighborhood sample respectively; σ d and σ r Represent the smoothing factors of the spatial domain and the distance domain respectively. From formula (2), it can be seen that the closer the spatial position and sample value of the neighborhood sample is to the target sample, the greater the weight of the neighborhood sample, otherwise the smaller the weight. The output of the bilateral filtering algorithm is the weighted value of the neighborhood sample, as shown in formula (2):

[0052]

[0053] Figure 5 This is an example of a bilaterally filtered smoothed signal, where the signal Z is generated by the active power signal and the reactive power signal. It can be observed that the difference between the filtered signal Z' and the signal before filtering is small, and the bilateral filtering retains the trend and integrity of the signal. In the enlarged diagram, it is further found that the bilateral filtering retains the step change caused by the load state transition in the signal, the edge information is retained, the small fluctuations in the signal are smoothed, the noise is effectively removed, and the gentle slope characteristics corresponding to the constant voltage stage of charging of the electric bicycle are retained for further identification.

[0054] 2. State transition removal

[0055] Since there may be some sharp state transitions, that is, step changes in the power signal, this topic proposes a method to remove this state transition, the purpose is to restore the gentle slope of the constant voltage charging stage of the electric bicycle that is divided by the redundant step state transition. First, it is necessary to detect the state transition that occurs in the signal, that is, the amplitude of the signal change ΔZ' τ Compare with the threshold c set by heuristic method. If the signal change amplitude is greater than or equal to the threshold, the load operation state has changed. If the signal change amplitude is less than the threshold, the load operation state has not changed. Then, each detected state transition is removed from the original signal by formula (3):

[0056]

[0057] Where τ represents the time index of the analyzed sample, Z' represents the signal after bilateral filtering, ΔZ' represents the differential signal of Z', and L represents the time series signal after the state transition is removed. Figure 6 Shown is a schematic diagram of state transition removal.

[0058] Figure 6 It can be clearly seen that the proposed method removes the edge caused by the load state transition in the signal and retains the small fluctuations and trend characteristics of the electric bicycle charging load.

[0059] 3. Savitzky-Golay Filter

[0060] Since the signal still contains some useless fluctuations after the state transition is removed, in order to further reduce this fluctuation and smooth the gentle slope characteristics corresponding to the constant voltage charging stage of electric bicycles, this project uses Savitzky-Golay filtering (SG filtering) to smooth the signal. SG filtering is a filtering method based on local least squares polynomial fitting using a sliding window. This filtering method retains the peak and width of the original signal while eliminating noise of different frequencies. It is widely used in signal denoising with non-Gaussian noise. Compared with filtering methods such as mean filtering and Kalman filtering, SG filtering has better signal shape preservation and denoising performance without losing resolution.

[0061] Given a local symmetric data window n=[l -m ,l -m+1 ,...,l0,...,l m-1 ,l m ], then the signal after SG filtering is:

[0062]

[0063] Among them, p<2m, represents the order of the least squares polynomial, a k represents the coefficients of the polynomial. Therefore, the goal of SG filtering is to minimize the error of equation (5):

[0064]

[0065] Figure 7 The figure shows the smoothing effect of the signal L after the state transition is removed by the SG filtering algorithm using a sliding window and local least squares polynomial fitting.

[0066] from Figure 7 It can be seen that the SG filtering algorithm smoothes the noise in the signal L after the state transition is removed, retains the signal trend, width and other features, and makes the gentle slope trend of the constant voltage charging stage of the electric bicycle more obvious, which is conducive to the detection of this feature.

[0067] Electric bicycle charging load detection and battery health status assessment:

[0068] Constant voltage charging stage detection

[0069] The piecewise linear representation (PLR) divides the time series signal into several subsequences according to the rising or falling trend of the signal. The first and last sample points of each subsequence are taken from the points in the original sequence, and the other sample points are linearized, which further refines the trend characteristics of the time series and compresses the data. Since the SG filter smoothes the noise in the signal L, making the rising and falling trend characteristics in the signal curve more obvious, the PLR ​​algorithm can segment the signal L' into several subsequences according to these trend characteristics, so as to further determine whether the time series contains the constant voltage charging stage of the electric bicycle.

[0070] This topic adopts the top-down PLR algorithm, which realizes the piecewise linearization of time series by iteratively updating the segmentation points. For a time series L' with a length of N, define the first sample point l'1 and the last sample l' N is the initial segmentation point, and then a linear discrete signal I of length N is generated, where the first sample point i1 is equal to l'1 and the last sample point i N Equal to l' N , and the remaining sample points are assigned linearly. Then, subtract the signal L' and the signal I point by point to obtain the difference between the two signals, and determine whether there is a point in the differential signal with an amplitude greater than the threshold h set by the heuristic method. If not, the signal L' can be directly represented by the linear representation of the signal I. If so, the new segmentation point of the signal L' is determined according to the index of the point with the maximum amplitude, and the signal I is updated as described above. Finally, repeat the above steps for the two segmented signals, iterate to obtain new segmentation points, until the amplitude of all points in the differential signal is less than the threshold h, at which time the signal L' can be represented by the piecewise linear signal I. Figure 8 An example of applying PLR is shown.

[0071] After segmenting the time series signal through PLR, several subsequences with different trend characteristics are formed. Based on experience, the detection rule of the subsequence containing the gentle slope of the constant voltage charging stage of electric bicycles is defined as follows:

[0072] (1) The difference between the end point and the starting point of the subsequence is less than A;

[0073] (2) The time span between the end point and the start point of the subsequence is [t1, t2].

[0074] When an electric bicycle is in the constant-voltage charging stage, the decreasing slope of active power and the increasing slope of reactive power are significantly different from those of other typical household appliance loads. Taking into account the characteristics of the small amplitude change and long duration of the ramp power, the above detection rules are set to distinguish those subsequences that are not related to the constant-voltage charging stage, and roughly detect all subsequences that include the constant-voltage charging stage of an electric bicycle.

[0075] The charging power of different electric bicycles is inconsistent. The charging power of most lead-acid batteries is greater than that of lithium-ion batteries. In the PLR ​​step, setting a smaller segmentation threshold may cause the time series containing the constant-voltage charging stage of electric bicycles for some users to be split into different subsequences, and the duration of each subsequence is short, which is not conducive to detection. Setting a larger segmentation threshold may cause the time series containing the constant-voltage charging stage of electric bicycles for some users to only detect the two segmentation points of the starting point and the end point, and completely miss the detection. This project sets the same but smaller segmentation threshold to avoid missed detection of the constant-voltage charging stage of electric bicycles, but there are still problems that the time series containing the charging load of electric bicycles is divided into different subsequences, and the identified subsequences contain a certain length of non-constant-voltage stage charging loads. Therefore, it is necessary to merge adjacent subsequences containing the gentle slope of the constant-voltage charging stage of electric bicycles, and finely locate the running time of the constant-voltage charging stage.

[0076] Although the initial amplitude of reactive power in the constant voltage charging stage of lead-acid batteries and lithium-ion batteries is small, they both have the characteristics of gradually decreasing and eventually dropping to zero. This topic will further refine the positioning of the operation time of the constant voltage charging stage of electric bicycles based on this characteristic. First, the reactive power Q sequence corresponding in time is determined according to the electric bicycle constant voltage charging subsequence (after merging), and then the signal H is obtained by bilateral filtering and state transition removal, and further by SG filtering to obtain the signal H', and the noise in the differential signal of the signal H' is smoothed by median filtering to obtain the signal H", and finally, according to the continuity of positive samples in the filtered differential signal, it is divided into several subsequences. Due to the load characteristics of the reactive power amplitude gradually decreasing and the decreasing speed gradually slowing down, its differential signal will appear as a positive sample and gradually decrease to zero. Therefore, the subsequences of continuous positive samples in the filtered sequence may contain the constant voltage charging stage of electric bicycles. The identified charging subsequences are refined using the following heuristically defined rules:

[0077] (1) The maximum value of the subsequence is greater than A';

[0078] (2) The time length of the subsequence is greater than t1.

[0079] Online rapid detection method for charging load of electric bicycle

[0080] Based on the electric bicycle charging load detection method, in order to meet the requirements of fast online detection, the local characteristics of the constant voltage stage, that is, the load characteristics of the gradually increasing difference between the active power and reactive power amplitudes, are used to detect the electric bicycle charging load.

[0081] Firstly, the state transition removal algorithm is used to remove the load events in the total active power P and total reactive power Q with a window length of 6 hours; secondly, the sampling frequency is reduced to 1 / 30Hz, and the Savitzky-Golay filter is used to reduce the noise in the power signal after the frequency reduction; then, the difference of the power signal is calculated, the continuous negative subsequences in the active power difference signal are matched with the continuous positive subsequences in the reactive power difference signal, and the distance between the matched continuous subsequences and the charging load template of the electric bicycle is calculated; finally, if the distance is less than the set threshold, the electric bicycle battery is in the constant voltage charging stage and the user is charging illegally, otherwise the electric bicycle battery is not in the constant voltage charging stage or there is no charging load, the sliding window moves back 10 minutes, and the total active power P and total reactive power Q with a window length of 6 hours are continuously analyzed.

[0082] Example 2, reference Figures 1 to 18 , which is the second embodiment of the present invention, is different from the previous embodiment in that this embodiment provides an electric bicycle shed battery detection system, which solves the problem of detecting specific faults of electric bicycle batteries, and includes:

[0083] Construction of electric bicycle charging load template

[0084] As mentioned above, the constant voltage charging stage of electric bicycles presents the load characteristics of active power decreasing slowly and reactive power increasing slowly. It is worth noting that the difference of the amplitude of the slow slope also has the characteristic of gradually increasing. This section will establish the charging load template of electric bicycles based on this characteristic to realize the online rapid detection of the constant voltage charging stage. In the total power consumption data with a sampling frequency of 1 / 60Hz, the characteristic of the gradual increase of the difference of the slow slope amplitude is not obvious enough. In the total power consumption data with a sampling frequency of 1Hz, although this characteristic is retained to the greatest extent, there may also be other interferences in the constant voltage charging process, and its operation fluctuations and noise interfere with the detection of the charging load of electric bicycles. Therefore, when detecting the charging load of electric bicycles, it is considered to reduce the sampling frequency to 1 / 30Hz. The charging load template in this section is also constructed at this sampling frequency. Although the electric bicycle constant voltage charging stage maintains a downward trend as a whole, it is actually a oscillating decline. Therefore, this section first calculates the mean value every 30 sampling points as a new sampling point to achieve filtering and frequency reduction, and then uses SG filtering to further smooth the signal. Finally, the difference is calculated to extract the load template matching of the constant voltage charging stage.

[0085] like Fig.10 As shown in (a) and (b), SG filtering makes the transition process from constant current charging to constant voltage charging smoother, and the small fluctuations in the constant voltage charging stage are also smoothed. Fig.10(c) and (d) show the differential signals of active power and reactive power after filtering. It can be seen that the power differential signal in the constant voltage charging stage can be fitted into a line segment, as shown by the red dotted line in the figure. This project will establish the active power differential signal template of the electric bicycle charging load based on this characteristic. and reactive power differential signal template Further from Fig.10 It can be seen that the maximum point of the line segment does not correspond to the moment when constant current charging is converted to constant voltage charging. This is because the SG filter only retains the shape characteristics of the signal and smoothes the two-stage conversion process.

[0086] In this section, the charging load templates for 10 electric bicycle batteries are established based on the maximum value point and slope of the line segment fitted by the power differential signal during the constant voltage charging stage of the electric bicycle. Since the unsupervised method is used, it is assumed that the electric bicycle batteries of the users are all healthy batteries, so Figure 4 The three electric bicycles equipped with unhealthy batteries are modeled, and the charging load model is shown in Table 6.

[0087] Table 610 Models of charging loads for electric bicycles

[0088]

[0089]

[0090] According to the data in Table 6, the maximum points and slope diagrams of 10 electric bicycle charging loads are drawn, as shown in Fig.11 As shown in the figure, it can be seen that the model of the No. 5 lead-acid battery is obviously deviated from the other 9 models, so it is eliminated, and the remaining 9 models are averaged to obtain the charging load template of the electric bicycle, that is, Lowest point: -1.2836; Slope: 0.0100; Highest point: 0.5548; Slope: -0.0054.

[0091] Event Detection

[0092] Since in the electricity consumption data with a sampling frequency of 1Hz, the state transition no longer appears in the form of a step in the data with a sampling frequency of 1 / 60Hz, it is necessary to use a load event detection algorithm to locate the state transition in the power signal in order to further eliminate its impact on the detection of the charging load of electric bicycles. This project adopts an adaptive two-stage event detection method, which adaptively adjusts the event detection threshold according to the different degrees of load fluctuation in the total electricity consumption data, and for step-like events and long transient events with different waveform characteristics, an improved edge detection method and a window-based detection method combining moving average and sliding T test are used to achieve accurate detection and positioning of load events in the total electricity consumption data. Since the constant-voltage charging stage of electric bicycles lasts for a long time, the transitions between other load states such as on, running, and off will destroy or even submerge the charging ramp. Therefore, after locating the event in the total data with a sampling frequency of 1Hz through the event detection algorithm, the load event in the total data is removed with the help of the state transition removal algorithm. Fig.12 Schematic diagram of load event detection and state transition removal for user No. 1 on a certain day.

[0093] from Fig.12 (a) It can be seen that the load event detection algorithm accurately detects the events in the total power consumption data, determines the load events in the reactive power according to the events in the active power, and then uses the state transition removal algorithm to remove the detected events. After removing the events, there are still small load fluctuations and noise in the power signal. Further combined with SG filtering to eliminate the noise while retaining the trend of the power signal, as shown in Fig.12 (b).

[0094] Template matching of charging load for electric bicycles

[0095] After event detection, state transition removal, downsampling frequency, SG filtering and differentiation, the total power consumption data of different users is further divided into continuous negative subsequences, and the reactive power difference signal is divided into continuous positive subsequences. In this project, the active power difference is set to be negative when it is less than -0.1, and the reactive power difference is set to be positive when it is greater than 0.06. It is also stipulated that 8 or more consecutive sampling points in the same state can be divided into continuous subsequences. Fig.13 Shows the subsequences of the active power differential signal that are judged as continuous negative and the subsequences of the reactive power differential signal that are judged as continuous positive

[0096] Since the active power ramp down and reactive power ramp up of the electric bicycle constant voltage charging stage are synchronized in time, it is necessary to match the continuous negative active power differential signal subsequence and the continuous positive reactive power differential signal subsequence, that is, the subsequences are considered to be matched if they overlap in time. Fig.13 In the figure, the two subsequences located in the middle of the horizontal axis, that is, near 00:00, match.

[0097] For the subsequences of active power differential signals that are continuously negative and reactive power differential signals that are continuously positive and synchronized in time (i.e., successfully matched), their distances from the electric bicycle charging load template are calculated, i.e., formula (21):

[0098]

[0099] Among them, x is the template signal of the electric bicycle charging load, y is the successfully matched power difference continuous subsequence, x1 is the maximum point of the template, y1 is the maximum point of the continuous subsequence, and n is the number of data points from the maximum point in the continuous subsequence of the differential signal to the end point of the subsequence. It should be noted that in order to avoid the situation where the maximum point in the continuous subsequence is at the last bit and it is close to the highest point of the template, it is necessary to set n>10. The average of the active signal subsequence distance and the reactive signal subsequence distance is taken as the distance between the matching subsequence and the template. If it is less than the threshold η set by the heuristic method, it is considered that the subsequence matches the template and there is a constant voltage charging stage for the electric bicycle; if it is greater than the threshold, it is considered that the subsequence does not match the template and there is no constant voltage charging stage for the electric bicycle.

[0100] Experimental verification

[0101] The data used in the experiment is the same as that in Section 3.3.3, but the frequency does not need to be reduced to 1 / 60Hz. In order to realize online rapid detection of electric bicycle charging load, a sliding window with a window length of 6 hours and a step length of 10 minutes is used to analyze the total power consumption data of users. Table 7 shows the experimental results when the distance threshold η is set to 0.25.

[0102] Table 7 Online rapid detection results of electric bicycle charging load

[0103]

[0104] It can be seen from Table 7 that although the overall F1 score of the online rapid detection results of the electric bicycle charging load is only 0.86, which is lower than the F1 score of the detection results at a sampling frequency of 1 / 60Hz, the recall rate of the online rapid detection is higher, reaching 0.98, that is, almost every time the electric bicycle is charged, it is detected. Since the purpose of the electric bicycle charging load detection is to issue a warning message once an illegal charging behavior of the electric bicycle is found, the situation of undetected illegal charging should be avoided as much as possible. Therefore, a higher recall rate and a relatively high precision rate are relatively important for fire safety. In the table, each illegal charging behavior of users 1, 2, 3, 5, and 6 was accurately detected, and users 3, 4, 5, 6, and 7 all achieved an F1 score higher than 0.9. Among them, the batteries of users 5, 6, and 7 were abnormal batteries, but they still achieved a high F1 score, which proved the versatility of the electric bicycle charging load template. The detection result of user 2 was poor, with more false detections. Fig.14 The result of a false detection of user 2 is shown.

[0105] Since the experiment uses sliding window fitting for real-time data analysis, Fig.14 The horizontal axis is the 6-hour window length, from Fig.14 (b) and (c) show that the last matching continuous subsequence in the sliding window is mistakenly detected as the constant voltage charging stage of the electric bicycle. In the total amount data corresponding to Figure (a), it is observed that the active power curve has a gentle slope trend that lasts longer than the constant voltage charging stage of the electric bicycle and continues to decline, resulting in the active power differential signal after filtering being judged as a continuous negative subsequence, while the reactive power curve shows a small step that continues to rise, but the event detection algorithm does not detect the occurrence of the event. The reactive power shows a continuous rising curve after filtering, resulting in the differential signal being judged as a continuous positive subsequence, and the reactive power maximum value point in the charging load template is small, and it is ultimately mistakenly detected as the charging load of the electric bicycle, but it is not ruled out that the data is mixed with the charging of electrical loads such as laptops and mobile phones. Since User No. 2 has an electrical load that makes the active power curve "convex" almost every day, the filtered active power differential signal is easily judged as continuously negative. Once there is a continuous positive subsequence in the reactive power differential signal, it is easily misdetected. Therefore, User No. 2 has more misdetections, but each illegal charging behavior is accurately detected.

[0106] Table 7 shows that the detection effect of user No. 1 is also poor. Fig.15 The accurate detection result of the charging load of an electric bicycle of user No. 1 is shown. As can be seen from Figure (a), when the electric bicycle is in the constant voltage charging stage, there is a high-power electrical load running, and the active power of the load also has a downward trend. As a result, when the constant voltage stage just appears in the sliding window, the distance between the subsequence of the active power differential signal with continuous negative and the charging load template is large, and it cannot be detected immediately. The constant voltage stage lasts for a long time. When the operating power of the high-power electrical load tends to be stable, the constant voltage charging stage has not stopped. The difference between the differential signal entering the sliding window and the template is small. Since the distance is averaged, as time goes by, the distance between the subsequence of the active power differential signal with continuous negative and the charging load template gradually decreases, and finally drops below 0.25, realizing the detection of the constant voltage charging stage of the electric bicycle. Furthermore, since the continuous subsequence of the active power differential signal of the high-power load may be close to the charging load template, and the amplitude of the reactive power maximum point of the charging load template is small, if the maximum point of the constant-voltage charging stage of the electric bicycle in the filtered active power differential signal is far away from the maximum point of the high-power load in time, the subsequence in the reactive power differential signal that is close to the charging load template will be detected as the constant-voltage charging stage of the electric bicycle. This is the main reason for the 9 false detections. Fig.15The time difference between the start of the constant voltage stage and the end of the sliding window is 45 minutes, that is, the charging load of the electric bicycle is detected 45 minutes after the start of the constant voltage charging stage of the electric bicycle. Fig.16 The distribution of the duration of the constant voltage phase when the e-bike charging load was accurately detected for 150 times is shown.

[0107] from Fig.16 It can be seen that among the 150 correct online rapid detections of the electric bicycle charging load, 75 of them had a constant voltage charging stage lasting 25 to 35 minutes. Fig.10 In the figure, the time difference from the start of the constant voltage stage to the maximum point of the power differential signal is approximately 10 minutes. The time interval of the 7 users' electric bicycle batteries may be longer. To determine whether the continuous subsequence of the differential signal contains the constant voltage charging stage of the electric bicycle, 10 consecutive data points (i.e. 5 minutes) after the maximum point are required. Therefore, there is one correct detection within 20 minutes of the constant voltage stage, and the rest are mainly concentrated in 20 minutes to 45 minutes. To a certain extent, online rapid detection of the electric bicycle charging load is achieved.

[0108] Fault diagnosis of battery cells in electric bicycles

[0109] In addition to rapid detection of the charging load of electric bicycles, when an electric bicycle battery fails, it is also necessary to further identify the type of fault in order to determine the safety hazards of the electric vehicle battery.

[0110] Collecting the original normal data and fault data of the electric bicycle battery is the first step in building the diagnostic system. Subsequently, the collected raw data is finely processed to eliminate the noise in the data, fill in the missing values, and perform standardization to ensure the accuracy and reliability of the data. The processed data is further divided into a training data set and a test data set. The training data set is used for model learning and training, while the test data set is used to evaluate the diagnostic performance of the model.

[0111] The training data set is input into the optimized radial basis function (RBF) neural network for deep learning training to identify the fault characteristics of electric bicycle batteries. After the training is completed, the test data set is used to perform fault diagnosis classification tests on the trained RBF neural network to verify the diagnostic accuracy of the model. Finally, the diagnosis results of the neural network are interpreted by the battery management system to achieve accurate diagnosis of electric bicycle battery faults. The various fault conditions mainly include nine fault conditions: battery overcharge, over discharge, battery aging, overvoltage, undervoltage, increased internal resistance, reduced battery capacity, high temperature and low temperature.

[0112] Example 3, reference Figures 1 to 18, which is the third embodiment of the present invention. Different from the previous embodiment, this embodiment provides an electric bicycle shed intelligent management front end, which solves the problem of remote unified management of electric bicycle sheds, and includes:

[0113] The front-end interface of the smart management of electric bicycle sheds mainly includes six interfaces, namely, equipment status module, alarm information module, alarm ratio module, total power module, total power module and charging information module.

[0114] The device status module realizes remote monitoring of device status and displays the total number of devices, idle status, fault status, and usage status.

[0115] The alarm information module pushes the alarm information of all intelligent sensing devices in real time. The pushed information includes alarm time, alarm device, and alarm type.

[0116] The alarm proportion module calculates the proportion of different types of alarm information based on all historical alarm information and displays it in the form of a pie chart for analysis.

[0117] The total power module displays the cumulative power of all intelligent sensing devices from 0 to 24 hours of the day in the form of a bar graph. Click the time button on the right to view the power status of a specified date.

[0118] The total power module displays the accumulated power of all intelligent sensing devices from 0 to 24 hours of the day in the form of a curve. Click the time button on the right to view the power status of a specified date.

[0119] The charging information module is mainly used to display basic information such as the number of charges today, the amount of electricity charged today, the charging time today, and the number of alarms today.

[0120] Importantly, it should be noted that the construction and arrangement of the present application shown in a plurality of different exemplary embodiments are only exemplary. Although only a few embodiments are described in detail in this disclosure, it should be readily understood by those who refer to this disclosure that many modifications are possible (e.g., the size, scale, structure, shape and proportion of various elements, and parameter values ​​(e.g., temperature, pressure, etc.), installation arrangement, use of materials, color, directional changes, etc.) without substantially departing from the novel teachings and advantages of the subject matter described in the application. For example, the element shown as integrally formed can be composed of multiple parts or elements, the position of the element can be inverted or otherwise changed, and the nature or number or position of the discrete element can be changed or changed. Therefore, all such modifications are intended to be included in the scope of the present invention. The order or sequence of any process or method steps can be changed or reordered according to alternative embodiments. In the claims, any "device plus function" clause is intended to cover the structure of performing the function described herein, and is not only structurally equivalent but also equivalent structure. Without departing from the scope of the present invention, other replacements, modifications, changes and omissions can be made in the design, operating conditions and arrangement of the exemplary embodiments. Therefore, the invention is not limited to a specific embodiment, but extends to several modifications still falling within the scope of the appended claims.

[0121] Additionally, in order to provide a concise description of exemplary embodiments, all features of an actual embodiment (ie, those features that are not relevant to the best mode presently contemplated for carrying out the invention or those that are not relevant to implementing the invention) may not be described.

[0122] It will be appreciated that in the development of any actual implementation, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort may be complex and time-consuming, but will be a routine task of design, fabrication, and production for those of ordinary skill having the benefit of this disclosure without undue experimentation.

[0123] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A smart management method for electric bicycle sheds, characterized in that: include, An intelligent sensing device is connected to an electric bicycle shed, the intelligent sensing device obtains electric bicycle data from the electric bicycle shed, divides the electric bicycle data into a training set and a test set, trains and tests an improved RBF neural network, and finally forms an abnormality recognition algorithm.

2. The electric bicycle shed intelligent management method according to claim 1, characterized in that: The intelligent sensing device is composed of a high-frequency data acquisition device, a refined electricity consumption data acquisition and application service platform, and an advanced application development service.

3. The electric bicycle shed intelligent management method according to claim 1, characterized in that: The state transition removal algorithm is used to remove load events in the total active power P and total reactive power Q with a window length of 6 hours.

4. The electric bicycle shed intelligent management method according to claim 3, characterized in that: The sampling frequency is reduced to 1 / 30 Hz, and Savitzky-Golay filtering is used to reduce the noise in the power signal after the frequency reduction; the difference of the power signal is calculated.

5. The electric bicycle shed intelligent management method according to claim 4, characterized in that: The continuous negative subsequences in the active power differential signal are matched with the continuous positive subsequences in the reactive power differential signal, and the distance between the matched continuous subsequences and the electric bicycle charging load template is calculated.

6. An electric bicycle shed battery detection system, characterized in that: include, The training data set is input into the optimized radial basis function (RBF) neural network for learning and training to identify the fault characteristics of electric bicycle batteries.

7. The electric bicycle shed battery detection system according to claim 6, characterized in that: After the training is completed, the test data set is used to perform fault diagnosis classification tests on the trained RBF neural network to verify the diagnostic accuracy of the model.

8. The electric bicycle shed battery detection system according to claim 7, characterized in that: The diagnostic results of the neural network are interpreted by the battery management system to achieve accurate diagnosis of electric bicycle battery failures.

9. An intelligent management front end for an electric bicycle shed, characterized in that: The electric bicycle shed battery detection system comprises any one of claims 6-8, and includes six interfaces, namely, a device status module, an alarm information module, an alarm ratio module, a total power module, a total power module, and a charging information module.

10. The electric bicycle shed intelligent management front end as claimed in claim 9, characterized in that: include, The equipment status module realizes remote monitoring of equipment status; The alarm information module pushes the alarm information of all intelligent sensing devices in real time; The alarm proportion module calculates the proportion of different types of alarm information based on all historical alarm information; The total power module displays the cumulative power of all intelligent sensing devices from 0 to 24 hours of the day in the form of a bar graph; The total power module displays the accumulated power of all intelligent sensing devices from 0 to 24 hours of the day in the form of a curve; The charging information module is mainly used to display basic information such as the number of charges today, the amount of electricity charged today, the charging time today, and the number of alarms today.

Images