Intelligent operation and maintenance management platform for underground cable trench operation environment

Abstract

The invention relates to the technical field of cable trench operation environment management platforms, in particular to an intelligent operation and maintenance management platform for an underground cable trench operation environment. The platform comprises a data acquisition module used for acquiring various external environment data outside a cable trench and one kind of operation environment data of one monitoring point in the cable trench; the matching analysis module is used for acquiring different analysis intervals in the to-be-analyzed sequence and further acquiring the maximum matching degree and lag time of the analysis intervals; the lagging analysis module is used for acquiring the matching degree sequence and the lagging time sequence, and further performing smoothing to obtain the smoothed matching degree sequence and lagging time sequence; the training set acquisition module is used for acquiring training samples and forming a training set based on the to-be-analyzed sequence and the matching degree and the lag time of each moment of the various external environment data; and the prediction module is used for obtaining the operation environment data of the monitoring point at the to-be-predicted moment based on the trained GRU model. According to the invention, the operation environment of the cable trench can be effectively managed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cable trench operation environment management platform, and in particular to an intelligent operation and maintenance management platform for underground cable trench operation environment. BACKGROUND

[0002] With the continuous expansion of urban power grid scale, as the important channel for the centralized laying and operation of power cables, the safety and stability of the operation environment of underground cable trenches are directly related to the reliable power supply and overall operation efficiency of the power system. Due to the characteristics of space closure, limited ventilation, complex environment and high difficulty of artificial inspection, the internal operation environment of the cable trench is easily affected by various factors. For example, external climate change can cause significant fluctuations in temperature and humidity in the trench, heavy rain may cause water seepage or accumulation, harmful gases will gradually accumulate when the ventilation is poor, and the continuous operation of electrical equipment will also bring heating effect and insulation aging risk. If there is no effective monitoring and management of these environmental factors, it is easy to cause the decline of cable insulation performance, short circuit, fire and even large-area power failure, which seriously threatens the safe and stable operation of the power grid. Therefore, it is of great significance to establish an intelligent monitoring and operation and maintenance management platform for the operation environment of the cable trench to improve the safety, reliability and management level of the power system.

[0003] The existing underground cable trench operation environment monitoring and operation and maintenance method generally relies on fixed threshold and static rules for alarm, which is simple to implement, but has obvious shortcomings in complex and changeable environment, lacks adaptability and intelligent ability. For example, the temperature threshold is usually set to 60℃, in the summer high temperature weather, the overall temperature of the trench may be close to or even exceed the value for a long time, frequent invalid alarm is triggered; while in cold weather, the cable abnormal heating is not obvious, and may be missed. Similarly, in the rainy season, the overall humidity is high, if the fixed threshold is still used, a large number of redundant alarms will be generated, which will cover the real abnormal signal. At the same time, the existing method lacks comprehensive analysis and dynamic adjustment ability of multi-dimensional environmental data, and it is difficult to identify potential risks in time. SUMMARY

[0004] In order to solve the above technical problems, the purpose of the present application is to provide an intelligent operation and maintenance management platform for underground cable trench operation environment, and the technical scheme adopted is as follows: An embodiment of the present application provides an intelligent operation and maintenance management platform for underground cable trench operation environment, which comprises: A data acquisition module is used to acquire various external environmental data outside the cable trench and one kind of operation environment data of one monitoring point inside the cable trench, and one kind of operation environment data of one monitoring point forms one sequence to be analyzed. The matching analysis module is configured to obtain different analysis intervals in the sequence to be analyzed based on each extreme point of the sequence to be analyzed; and establish a sliding window based on a length of one analysis interval, and obtain a maximum matching degree and a lag time of the analysis interval based on the sliding window and one external environment data. The lag analysis module is configured to form a matching degree sequence and a lag time sequence based on the maximum matching degrees and the lag times of each analysis interval of the sequence to be analyzed obtained based on one external environment data, respectively; and obtain a smoothed matching degree sequence by adaptively smoothing each data in one matching degree sequence based on a difference between the data and data around the data, and obtain a smoothed lag time sequence in the same manner. The training set obtaining module is configured to obtain a matching degree and a lag time at each time of one external environment data based on the smoothed matching degree sequence and the lag time sequence corresponding to the external environment data; and obtain training samples based on the sequence to be analyzed and the matching degrees and the lag times at each time of various external environment data, and form a training set. The prediction module is configured to construct a loss function of a GRU model, and train the GRU model based on the loss function and the training set to obtain a trained GRU model; and obtain the external environment data of the monitoring point at the time to be predicted based on the trained GRU model.

[0005] Preferably, obtaining different analysis intervals in the sequence to be analyzed based on each extreme point of the sequence to be analyzed comprises: obtaining extreme points in the sequence to be analyzed, taking time points included by four adjacent extreme points at a front end of the sequence to be analyzed as a first analysis interval, moving a last extreme point of the first analysis interval to a next extreme point to obtain a second analysis interval, moving a last extreme point of the second analysis interval to a next extreme point to obtain a third analysis interval, and obtaining all analysis intervals in the sequence to be analyzed in the same manner.

[0006] Preferably, the sliding window is established based on a length of one analysis interval, and the maximum matching degree and the lag time of the analysis interval are obtained based on the sliding window and one external environment data, comprising: For one kind of external environment data, a data group of preset time length is intercepted forwardly in the kind of external environment data from the last time of one analysis interval of a sequence to be analyzed, to form a first sequence; a sliding window is constructed with the length of the analysis interval, and the sliding window is slid on the first sequence according to a set step length, the mean square error of the frequency spectrum data of the kind of external environment data and the frequency spectrum data of the running environment data in the analysis interval in one sliding window in the sliding process is obtained, and the matching degree corresponding to the sliding window is obtained by using the exponential function with the natural constant as the base to negatively correlate the mean square error, and the maximum value of the matching degrees corresponding to all sliding windows is taken as the maximum matching degree of the analysis interval; the time difference of the last time in the analysis interval and the last time in the sliding window corresponding to the maximum matching degree is taken as the lag time of the analysis interval.

[0007] Preferably, the adaptive smoothing is performed on each data in a matching degree sequence according to the difference between each data and the data around each data to obtain a smoothed matching degree sequence, including: The product of the spectral entropy of a matching degree sequence and the number of data points in the matching degree sequence is obtained, and the product is rounded to obtain the selected data amount of each data in the matching degree sequence; and the data range of one data in the matching degree sequence is determined according to the selected data amount; Each data in the data range of one data is recorded as a neighborhood data, and the time interval weight corresponding to the neighborhood data is obtained by using the exponential function with the natural constant as the base to negatively correlate the time interval between one neighborhood data and the data in the data range of one data in the matching degree sequence; the weighted average difference is obtained by using the time interval weight corresponding to each neighborhood data in the data range of one data to weight and average the absolute value of the difference between the data and each neighborhood data in the data range of the data, and the smoothing degree value of the data is obtained by normalizing the weighted average difference; The smoothing degree value of one data in the matching degree sequence is added to the first preset value, and then multiplied by the initial kernel bandwidth to obtain the kernel bandwidth parameter value corresponding to the data; and the smoothing is performed on each data in the matching degree sequence by using the kernel bandwidth parameter value corresponding to each data in the matching degree sequence to obtain a smoothed matching degree sequence.

[0008] Preferably, the matching degree and the lag time of each time of one kind of external environment data are obtained based on the smoothed matching degree sequence and the lag time sequence corresponding to the kind of external environment data, including: Fitting data in the smoothed matching degree sequence corresponding to one kind of external environment data to obtain a matching degree fitting curve, and obtaining the matching degree of each time of the kind of external environment data on the matching degree fitting curve; fitting data in the smoothed lag time sequence corresponding to one kind of external environment data to obtain a lag time fitting curve, and obtaining the lag time of each time of the kind of external environment data on the lag time fitting curve.

[0009] Preferably, the training samples are obtained based on the to-be-analyzed sequence and the matching degree and lag time of each time of various external environment data, and the training set is composed, including: Determining the time corresponding to one training sample as a prediction time; grouping one kind of external environment data into a sequence, denoted as an external environment sequence; in the external environment sequence, based on the lag time of each time before the prediction time, obtaining the running environment data corresponding to each time before the prediction time in the to-be-analyzed sequence to form a second sequence; the external environment data and the matching degree of each time of the second sequence form a third sequence and a fourth sequence respectively, thereby obtaining the second, third and fourth sequences corresponding to the kind of external environment data, and the second, third and fourth sequences of other kinds of external environment data are obtained in the same way, and the second, third and fourth sequences of each kind of external environment data are aligned in the shortest sequence of each kind of external environment data, to obtain the second, third and fourth sequences of each kind of external environment data after data alignment, which form one training sample, and the label of the training sample is the actual value of the running environment data at the prediction time; all training samples form a training set.

[0010] Preferably, the loss function is specifically: , Wherein, represents the loss function; represents the time index value of the t time; A represents the total number of data in the to-be-analyzed sequence; and respectively represent the predicted value and the actual value of one kind of running environment data at the t time.

[0011] The embodiment of the present application has at least the following beneficial effects: the present application collects various operating environment data and various external environment data of each monitoring point in the cable trench, then forms an operating environment data of a monitoring point into a to-be-analyzed sequence, obtains extreme points for determining different analysis intervals in the to-be-analyzed sequence, analyzes the influence of various external environment data on the to-be-analyzed sequence, then determines the maximum matching degree and the lag time corresponding to each analysis interval to form a matching degree sequence and a lag time sequence, and then performs adaptive smoothing on each data in the matching degree sequence and the lag time sequence to obtain a smoothed matching degree sequence and a lag time sequence; based on the smoothed matching degree sequence and the lag time sequence corresponding to a kind of external environment data, the matching degree and the lag time of each time of the kind of external environment data are obtained; based on the to-be-analyzed sequence and the matching degree and the lag time of each time of various external environment data, a training sample is obtained and a training set is formed, a loss function of the GRU model is constructed, and the GRU model is trained based on the loss function and the training set to obtain a trained GRU model, so that the prediction model is more accurate; based on the trained GRU model, the kind of operating environment data of the monitoring point at the to-be-predicted time is obtained, and then the purpose of effectively managing the operating environment of the cable trench is achieved according to each kind of operating environment data of each monitoring position obtained by prediction. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, and the advantages thereof, a brief introduction will be given to the drawings needed in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0013] Figure 1 The system block diagram of the intelligent operation and maintenance management platform for the operating environment of the underground cable trench provided by the embodiment of the present application. DETAILED DESCRIPTION

[0014] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the specific implementation, structure, features and effects of the intelligent operation and maintenance management platform for the operating environment of the underground cable trench according to the present application are described in detail as follows by combining with the drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0015] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0016] The following description, in conjunction with the accompanying drawings, details the specific solution of the intelligent operation and maintenance management platform for the underground cable trench operating environment provided by this invention.

[0017] Example: The main application scenario of this invention is to analyze and predict the operating environment of cable trenches, and to conduct subsequent analysis based on the predicted environmental data.

[0018] Please see Figure 1 The diagram illustrates a block diagram of an intelligent operation and maintenance management platform for an underground cable trench operating environment provided by an embodiment of the present invention. The platform includes the following modules: The data acquisition module is used to collect various external environmental data outside the cable trench and one type of operating environment data of a monitoring point inside the cable trench. The operating environment data of a monitoring point constitutes a sequence to be analyzed.

[0019] In the implementation of this application, firstly, multiple types of sensors are deployed at various monitoring points in the underground cable trench to collect multi-dimensional data about the operating environment of the cable trench, such as core parameters related to the operating environment, such as temperature, cable current, humidity, gas concentration, and air velocity at each monitoring point inside the cable trench. Each type of data is recorded as a type of operating environment data. In this way, various operating environment data of each monitoring point inside the cable trench are collected. This application uses one type of operating environment data of a monitoring point as an example for illustration. Thus, one type of operating environment data of a monitoring point is formed into a sequence to be analyzed for subsequent analysis. Preferably, this application uses the temperature of a monitoring point inside the cable trench as an example for illustration.

[0020] In addition, it is necessary to collect various external environmental data of the cable trench, such as external temperature and humidity. It should be noted that when collecting external environmental data, the external environment of the area where the cable trench is located can be used as the data source to control the stability of the data source. Furthermore, for the convenience of subsequent analysis, all operating environment data and external environmental data should be standardized.

[0021] By jointly collecting data from both internal and external environments, the impact of external environmental fluctuations on the operating environment within the cable trench can be effectively captured, avoiding biases caused by monitoring a single parameter. The collected data is then transmitted to a data processing platform to provide comprehensive data support for building more accurate predictive models, thereby improving the accuracy and robustness of environmental monitoring within the cable trench and laying the foundation for early warning of potential risks.

[0022] The matching analysis module is used for obtaining different analysis intervals in the sequence to be analyzed based on each extreme point of the sequence to be analyzed; a sliding window is established based on the length of an analysis interval, and the maximum matching degree and the lag time of the analysis interval are obtained based on the sliding window and one kind of external environment data.

[0023] In the operation environment monitoring of underground cable trench, external factors (such as temperature and humidity) may have a certain influence on the temperature and humidity data in the cable trench, which may make it difficult to accurately identify the abnormality by using the fixed threshold monitoring method. For example, high temperature in summer may cause a large number of invalid over-temperature alarms, and low temperature in winter may cover up the abnormal heating of the cable, thereby causing misjudgment or omission. It is difficult to distinguish between normal environmental fluctuations and potential abnormalities by relying only on static thresholds, which reduces the effectiveness of monitoring.

[0024] For a sequence to be analyzed composed of one kind of operation environment data of a monitoring point, the periodicity needs to be analyzed, and then the periodicity of the fluctuation is captured. Specifically, the extreme points in the sequence to be analyzed are obtained, the time points included in the four adjacent extreme points at the front end of the sequence to be analyzed (i.e., one period is one interval, and in the actual process, more than or equal to four extreme points can be selected as an analysis interval, but considering the calculation efficiency, the present application selects at least four extreme points as one period of the data for analysis) are taken as the first analysis interval, the last extreme point of the first analysis interval is moved to the next extreme point to obtain the second analysis interval, the last extreme point of the second analysis interval is moved to the next extreme point to obtain the third analysis interval, and so on, to obtain all the analysis intervals in the sequence to be analyzed. That is, four extreme points form an analysis interval, and then the last extreme point in the analysis interval is moved to the first extreme point after the analysis interval to obtain the next analysis interval every time, and the step length of the movement is the distance between every two extreme points.

[0025] Further, the matching degree between the external environment data and the operation environment data in the analysis interval is obtained by using the sliding matching method for the analysis interval in the corresponding external environment data, and then the time lag influence of the environment on the operation environment data is identified.

[0026] A sliding window is established based on the length of an analysis interval, and the maximum matching degree and the lag time of the analysis interval are obtained based on the sliding window and one kind of external environment data.

[0027] Specifically, for one kind of external environment data, a first sequence is formed by taking data of a preset time length forward from the last time of one analysis interval in the to-be-analyzed sequence in the kind of external environment data; a sliding window is constructed with the length of the analysis interval, the sliding window is slid on the first sequence according to a set step, the mean square error of the frequency spectrum data of the kind of external environment data and the frequency spectrum data of the running environment data in the analysis interval in one sliding window in the sliding process is obtained, and the matching degree corresponding to the sliding window is obtained by using an exponential function with a natural constant as a base to negatively correlate the mean square error, and the maximum value of the matching degrees corresponding to all sliding windows is taken as the maximum matching degree of the analysis interval; the time difference between the last time in the analysis interval and the last time in the sliding window corresponding to the maximum matching degree is taken as the lag time of the analysis interval.

[0028] It should be noted that the preset time length reference value is 5 hours (because the present application believes that the delay lag of external influencing factors on the data cannot exceed this hour, so this time is the upper limit of the lag of the factor data in the sliding matching, for different factor data, the lag constraint can be flexibly adjusted, for example, when monitoring the temperature in the trench, the external environment takes a relatively long time to transfer the temperature inside the cable trench, and in this case, a longer lag time range can be set, and for the thermal effect of cable current, the cable current factor data is located inside the cable trench, so the influence is transmitted faster, and a shorter lag time range can be set, and when selecting, the lag time range of the factor data can be reasonably selected through artificial experience and actual application environment characteristics, which will not be described here), which can be adjusted according to actual conditions. Here, the acquisition of the first sequence is to map the last time in one analysis interval in the to-be-analyzed sequence to one kind of external environment data, and then take 5 hours of data forward from the time to form a first sequence in the kind of external environment data; one data is taken as the set step when the sliding window slides; the frequency spectrum data is obtained by using short-time Fourier transform (STFT), and then the similarity of the two frequency spectrum data is corresponded, and then the dependence of the change of the running environment data in one analysis interval in the to-be-analyzed sequence on the environmental factor corresponding to the kind of external environment data is evaluated, and the accuracy of subsequent analysis is improved by analyzing the interference of the external environment.

[0029] The calculation model of the matching degree is specifically: , Among them, represents the matching degree of the running environment data in any one analysis interval in the to-be-analyzed sequence and the s-th sliding window in the sliding process of the sliding window corresponding to the analysis interval on the first sequence of one kind of external environment data obtained based on the analysis interval, and exp represents an exponential function with a natural constant as a base, which is used for negative correlation mapping; spectrum data representing the running environment data in the analysis interval at this time; spectrum data representing the external environment data in the s-th sliding window at this time, and MSE represents a mean square error function, a mean square error representing the two spectrum data, the smaller the value is, the larger the value is, the closer the similarity of the changes of the two at the position is, and the greater the matching degree is likely to be.

[0030] Thus, for a to-be-analyzed sequence and an external environment data, the maximum matching degree and the lag time of each analysis interval in the to-be-analyzed sequence for the external environment data can be obtained, and the lag time refers to the hysteresis of the influence of the external environment data on the running environment data in the cable trench.

[0031] a lag analysis module, configured to form a matching degree sequence and a lag time sequence respectively by the maximum matching degrees and the lag times of each analysis interval of the to-be-analyzed sequence based on one external environment data; and obtain a smoothed matching degree sequence by adaptively smoothing each data in a matching degree sequence according to the difference between each data and the data around each data, and obtain a smoothed lag time sequence in the same way.

[0032] The maximum matching degrees and the lag times of each analysis interval of the to-be-analyzed sequence based on each external environment data are obtained, and thus the maximum matching degrees and the lag times of each analysis interval of the to-be-analyzed sequence based on one external environment data are formed into a matching degree sequence and a lag time sequence respectively, so that for the to-be-analyzed sequence, the matching degree sequence and the lag time sequence corresponding to each external environment data can be obtained.

[0033] In the monitoring of the running environment of an underground cable trench, the matching degree sequence and the lag time sequence corresponding to one external environment data are often disturbed by environmental fluctuations or other environmental factors, and if directly used for prediction and analysis, the results may frequently fluctuate or abnormally jump, and it is difficult to truly reflect the action law of external factors on the internal temperature.

[0034] Therefore, by smoothing the matching degree sequence and the lag time sequence corresponding to one external environment data, the influence of occasional abnormal points and short-term fluctuations can be effectively weakened, the action trend of the influencing factors is more stable and reliable, and thus more robust input basis is provided for subsequent temperature prediction and abnormality detection, and the accuracy and robustness of the model in a complex environment are improved.

[0035] Since the fluctuation characteristics of the matching degree sequence and the lag time sequence corresponding to different types of external environment data are different, the smoothing parameters should be adaptively adjusted, including the size of the neighborhood range when analyzing data and the strength when smoothing, so as to improve the accuracy of the smoothing processing and provide more reliable data basis for modeling analysis.

[0036] Therefore, the smooth time should be adaptively selected according to the specific characteristics to select the appropriate neighborhood range and the smoothing degree parameter. Thus, the accuracy of the data smoothing processing is improved.

[0037] Therefore, first, the selected data point range of the matching degree sequence corresponding to one kind of external environment data and the lag time sequence smoothing is obtained according to the change characteristics of the matching degree sequence and the lag time sequence corresponding to the external environment data.

[0038] Thus, taking the matching degree sequence corresponding to one kind of external environment data as an example for analysis, specifically, the product of the spectral entropy of a matching degree sequence and the number of data points in the matching degree sequence is obtained, and the selected data amount of each data in the matching degree sequence is obtained by rounding the product. Taking one data in the matching degree sequence as the center, the data range of the data is determined according to the selected data amount.

[0039] The calculation model of the selected data amount is N=Z×H, Wherein, N represents the product of the matching degree sequence corresponding to one kind of external environment data and the number of data points in the matching degree sequence, and the selected data amount of each data in the matching degree sequence (taking even integer, for example, 2.3, 2.4, 2.7 are all 2, 3.1, 3.2, 3.7 are all 4) can be obtained by rounding. That is, the number of selected data points when smoothing. Z represents the total number of data points of the matching degree sequence; H represents the spectral entropy of the matching degree sequence (the spectral data needs to be obtained first, and then the spectral entropy is obtained, which is a known technology, and will not be described here). Its value range is between 0 and 1. The greater the value, the more dispersed the spectral distribution of the matching degree sequence, the more random and complex the fluctuation, and the weaker the regularity. Therefore, a larger data range needs to be selected when smoothing to ensure the stability of the analysis result. On the contrary, if H is smaller, it means that the sequence is more stable, the periodicity is stronger, and the regularity is more obvious. At this time, a smaller data range can be used for smoothing. When determining the data range of one data, the data is taken forward and backward from the center of the matching degree sequence. The number of data taken forward and backward is the same, which is half of the selected data amount. For example, if the selected data amount is 4, then two data are taken forward and two data are taken backward from the center of one data, and the data taken form the data range of the data.

[0040] After determining the data range required for smoothing, the fluctuation characteristics of different data in the matching degree sequence also need to be considered. Since the fluctuation degree of each data in the matching degree sequence may differ, the required smoothing strength when smoothing also differs: data with greater fluctuations requires stronger smoothing to weaken the influence of abnormal fluctuations; while data with smaller fluctuations and stable trends can use weaker smoothing to avoid information loss caused by excessive smoothing. This enables more targeted adaptive smoothing, allowing the characteristics of the matching degree sequence to be more truly preserved.

[0041] Thus, each data in the data range of a data is recorded as a neighborhood data, the time interval weight corresponding to a neighborhood data in the data range of a data in the matching degree sequence is obtained by using an exponential function with a natural constant as the base to negatively correlate the time interval between the neighborhood data and the data; the weighted average difference is obtained by using the time interval weight corresponding to each neighborhood data in the data range of a data to weight and average the absolute value of the difference between the data and each neighborhood data in its data range, and the smoothing degree value of the data is obtained by normalizing the weighted average difference.

[0042] The calculation model of the smoothing degree value of a data in the matching degree sequence is specifically: , wherein, represents the smoothing degree value of the i-th data in the matching degree sequence, and norm represents a normalization function; NZ represents the number of neighborhood data in the data range of the i-th data, excluding the i-th data itself; 、 respectively represent the value of the i-th data in the matching degree sequence and the value of the s-th neighborhood data in the data range of the i-th data; e represents a natural constant; represents the time interval between the s-th neighborhood data in the data range of the i-th data and the i-th data, the smaller this value, the greater the weight when analyzing the data, represents the time interval weight corresponding to the s-th neighborhood data in the data range of the i-th data; represents the absolute value of the difference between the i-th data and the s-th neighborhood data in the data range of the i-th data, the greater this value, the greater the difference between the two data; represents the weighted difference average of the i-th data and the neighborhood data in the data range of the i-th data in the matching degree sequence, the greater this value, the greater the relative difference between the data and the neighborhood data, and the greater the degree when smoothing.

[0043] The smoothness value of each data in the matching degree sequence is obtained through the above steps, and the kernel bandwidth corresponding to each data during smoothing is further adaptively adjusted based on the smoothness value. Specifically, the kernel bandwidth parameter value corresponding to the data is obtained by adding the smoothness value of the data in the matching degree sequence to a first preset value and multiplying the initial kernel bandwidth. The first preset value is 1, and the greater the smoothness value, the greater the kernel bandwidth parameter value of the data during smoothing, the greater the smoothness of the data, and vice versa.

[0044] Thus, for the matching degree sequence corresponding to various external environment data of a to-be-analyzed sequence, the kernel bandwidth parameter value of each data can be obtained, and then the kernel bandwidth parameter value is used to smooth each data in the matching degree sequence to obtain a smoothed matching degree sequence. Similarly, the same is true for the smoothing of the lag time sequence.

[0045] For the to-be-analyzed sequence, the smoothed matching degree sequence and the lag time sequence corresponding to each kind of external environment data can be obtained.

[0046] The training set acquisition module is configured to acquire the matching degree and the lag time of each time instant of a kind of external environment data based on the smoothed matching degree sequence and the lag time sequence corresponding to the kind of external environment data, and acquire training samples based on the to-be-analyzed sequence and the matching degree and the lag time of each time instant of each kind of external environment data, and compose a training set.

[0047] The above obtains the smoothed matching degree sequence and the lag time sequence corresponding to each kind of external environment data, and further, the data in the smoothed matching degree sequence corresponding to a kind of external environment data can be fitted to obtain a matching degree fitting curve, and the matching degree of each time instant of the kind of external environment data is obtained on the matching degree fitting curve; the data in the smoothed lag time sequence corresponding to a kind of external environment data is fitted to obtain a lag time fitting curve, and the lag time of each time instant of the kind of external environment data is obtained on the lag time fitting curve.

[0048] Further, the time corresponding to a training sample is determined as the prediction time; a sequence of external environment data is formed, denoted as an external environment sequence; in the external environment sequence, the running environment data corresponding to each time before the prediction time is obtained in the to-be-analyzed sequence based on the lag time of each time before the prediction time, to form a second sequence; the external environment data of each time of the second sequence and the matching degree are respectively formed into a third sequence and a fourth sequence, thereby obtaining the second, third and fourth sequences corresponding to the external environment data of this kind, and the second, third and fourth sequences of other kinds of external environment data are obtained in the same way, and the shortest sequence of the second, third and fourth sequences of each kind of external environment data is aligned with the second, third and fourth sequences of each kind of external environment data, to obtain the second, third and fourth sequences of each kind of external environment data after data alignment, which form a training sample, and the label of the training sample is the actual value of the running environment data of the prediction time. Thus, based on the running environment data (to-be-analyzed sequence) of one monitoring point and the smoothed matching degree sequence and lag time sequence corresponding to various external environment data, different training samples can be obtained, and then all the training samples form a training set.

[0049] It should be noted that for a prediction time, if 20 times are collected during data collection, the prediction time is the 14th time, and the lag time corresponding to the 13th time of one kind of external environment data before the prediction time is 9, then the data of the 13th time of the external environment data corresponds to the data of the 22nd time in the to-be-analyzed sequence, but at this time, the data of the to-be-analyzed sequence is only up to the 20th time, and therefore the second sequence only contains the times that can find corresponding data in the to-be-analyzed sequence; the lengths of the second, third and fourth sequences corresponding to one kind of external environment data are the same, but since the lag times of each time of each kind of external environment data may be different, the lengths of the second, third and fourth sequences corresponding to each kind of external environment data may be different, and at this time, the shortest sequence needs to be aligned with the data, so that the lengths of the second, third and fourth sequences corresponding to each kind of external environment data after alignment are the same, that is, the intersection in time sequence is taken.

[0050] The prediction module is configured to construct a loss function of the GRU model, train the GRU model based on the loss function and the training set, and obtain the running environment data of the monitoring point at the to-be-predicted time based on the trained GRU model.

[0051] The training set for training is obtained above, and the GRU model is trained using the training set. In the initial stage of model training, the prediction result may have a certain deviation due to the limited number of available historical samples. However, as the monitoring data gradually accumulates, the model can adaptively capture the action law of more complete time sequence characteristics and environmental factors, thereby continuously improving the prediction accuracy and stability, finally realizing high-precision prediction and dynamic optimization of the cable trench temperature, and embodying the continuous optimization ability and convergence of the model in a complex environment.

[0052] Therefore, when training it, a weighted loss function is constructed for use as its stopping condition, and the formula is as follows: , wherein, represents the loss function, represents the time index value at the t-th moment, that is, when the training sample at the t-th moment is used for training, the time index value of the training sample, that is, the moment corresponding to the training sample, for example, the corresponding moment is the 10th moment, then is 10, and A represents the total number of collected data (total number of data in the sequence to be analyzed), and respectively represent the predicted value and the actual value of one kind of operating environment data at the t-th moment; respectively represent the absolute value of the difference between the predicted value and the actual value of one kind of operating environment data at the t-th moment, that is, the prediction error at this moment; represents the ratio of the time index value at the t-th moment to the total number of collected data, that is, the weight of the moment, the earlier the value, the greater the error due to the limited number of input historical data, and therefore the smaller the error value determines the accuracy evaluation weight of the model. represents the weighted difference mean of the predicted data and the actual data, the greater the value, the greater the error of the current model prediction.

[0053] Therefore, by training the prediction model, then is the minimum, the training parameters in the model are retained, and the trained GRU model is obtained, which is then used for subsequent prediction of the operating environment data in the cable trench.

[0054] Specifically, for the to-be-predicted moment, the input data corresponding to the to-be-predicted moment is obtained using the above method, that is, the aligned second, third and fourth sequences of each external ring environment data are used as input data, which are input into the trained GRU model to obtain the operating environment data of the monitoring point at the to-be-predicted moment.

[0055] When the running environment data is the temperature of a monitoring point in the cable trench, the predicted value of the temperature in the cable trench can be obtained through the above prediction model, and compared with the actual detection value. When the actual detection value is greater than the predicted value, and the residual error (i.e. the absolute difference between the predicted value and the actual value) between the two exceeds 0.1 times the historical weighted average error L(T) (which can be determined according to actual monitoring requirements, and is not limited here), it indicates that the temperature at this position may be abnormal, and timely warning prompt is triggered. The application can effectively utilize the ability of the prediction model to depict the normal operation trend, combined with the residual error dynamic threshold determination mechanism, to reduce the false positives and false negatives caused by the fixed threshold, so as to realize more accurate abnormal detection and warning. (For the remaining monitoring data in the cable trench, the prediction model can also be used for monitoring and warning through the same reason as above, which will not be described here again). It should be noted that if a smaller running environment data is more abnormal, when the actual detection value is less than the predicted value, and the residual error (i.e. the absolute difference between the predicted value and the actual value) between the two exceeds 0.1 times the historical weighted average error L(T), it indicates that an abnormality may occur.

[0056] It should be noted that the above-mentioned sequence of the embodiments of the application is only for description, and does not represent the advantages and disadvantages of the embodiments. The above describes a specific embodiment of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or may be advantageous.

[0057] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment focuses on the differences from other embodiments.

[0058] The above is only the preferred embodiment of the application, and is not intended to limit the application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application shall be included in the protection scope of the application.

Claims

1. An intelligent operation and maintenance management platform for the operating environment of underground cable trenches, characterized in that, The platform includes: The data acquisition module is used to collect various external environmental data outside the cable trench and one type of operating environment data of a monitoring point inside the cable trench. The operating environment data of a monitoring point constitutes a sequence to be analyzed. The matching analysis module is used to obtain different analysis intervals in the sequence to be analyzed based on each extreme point of the sequence to be analyzed; to establish a sliding window with the length of an analysis interval, and to obtain the maximum matching degree and lag time of the analysis interval based on the sliding window and a kind of external environment data; The lag analysis module is used to form a matching degree sequence and a lag time sequence respectively by taking the maximum matching degree and lag time of each analysis interval of the sequence to be analyzed based on external environmental data; and to obtain a smoothed matching degree sequence by adaptively smoothing each data according to the difference between each data in a matching degree sequence and the data around each data. Similarly, a smoothed lag time sequence is obtained. The training set acquisition module is used to acquire the matching degree and lag time of the external environment data at each time step based on the smoothed matching degree sequence and lag time sequence corresponding to the external environment data; and to acquire training samples and form a training set based on the sequence to be analyzed and the matching degree and lag time of various external environment data at each time step. The prediction module is used to construct the loss function of the GRU model, and to train the GRU model based on the loss function and the training set to obtain the trained GRU model; based on the trained GRU model, the operating environment data of the monitoring point at the time to be predicted is obtained.

2. The intelligent operation and maintenance management platform for the operating environment of underground cable trenches according to claim 1, characterized in that, The process of obtaining different analysis intervals in the sequence to be analyzed based on each extreme point includes: The extreme points in the sequence to be analyzed are obtained. The time span of the four adjacent extreme points at the beginning of the sequence is taken as the first analysis interval. The last extreme point of the first analysis interval is moved to the next extreme point to obtain the second analysis interval. The last extreme point of the second analysis interval is moved to the next extreme point to obtain the third analysis interval, and so on, to obtain all the analysis intervals in the sequence to be analyzed.

3. The intelligent operation and maintenance management platform for the operating environment of underground cable trenches according to claim 1, characterized in that, The step of establishing a sliding window with the length of an analysis interval, and obtaining the maximum matching degree and lag time of the analysis interval based on the sliding window and external environmental data, includes: For a given set of external environment data, starting from the last moment of an analysis interval of the sequence to be analyzed, data of a predetermined time length is extracted from the external environment data to form a first sequence. A sliding window is constructed based on the length of the analysis interval. The sliding window is moved along the first sequence at a predetermined step size. During the sliding process, the mean square error of the spectral data of the external environment data and the spectral data of the operating environment data within the analysis interval is obtained. The mean square error is negatively correlated with the mean square error using an exponential function with the natural constant as the base to obtain the matching degree corresponding to the sliding window. The maximum matching degree corresponding to all sliding windows is taken as the maximum matching degree of the analysis interval. The time difference between the last moment of the analysis interval and the last moment of the sliding window corresponding to the maximum matching degree is taken as the lag time of the analysis interval.

4. The intelligent operation and maintenance management platform for the operating environment of underground cable trenches according to claim 1, characterized in that, The step of adaptively smoothing each data point based on the difference between each data point in a matching degree sequence and the data surrounding each data point to obtain a smoothed matching degree sequence includes: Obtain the product of the spectral entropy of a matching degree sequence and the number of data points in the matching degree sequence, and round the product to obtain the selected data quantity for each data point in the matching degree sequence; determine the data range of a data point based on the selected data quantity, with a data point in the matching degree sequence as the center; Each data point within a data range is denoted as neighborhood data. An exponential function with a base of the natural constant is used to perform a negative correlation mapping between the neighborhood data within the data range of a data point and the time interval of the data point, thus obtaining the time interval weight corresponding to the neighborhood data. The absolute values ​​of the differences between the data point and each of the neighborhood data points within the data range are weighted and averaged to obtain the weighted average difference. The weighted average difference is then normalized to obtain the smoothness value of the data point. The smoothness value of a data point in the matching degree sequence is added to a first preset value and then multiplied by the initial kernel bandwidth to obtain the kernel bandwidth parameter value corresponding to that data point. The kernel bandwidth parameter values ​​corresponding to each data point in the matching degree sequence are then used to smooth each data point in the matching degree sequence to obtain a smoothed matching degree sequence.

5. The intelligent operation and maintenance management platform for the operating environment of underground cable trenches according to claim 1, characterized in that, The step of obtaining the matching degree and lag time of the external environment data at each moment based on the smoothed matching degree sequence and lag time sequence corresponding to the external environment data includes: A matching degree fitting curve is obtained by fitting the data in the smoothed matching degree sequence corresponding to a certain type of external environment data, and the matching degree of the external environment data at each time point is obtained from the matching degree fitting curve; a lag time fitting curve is obtained by fitting the data in the smoothed lag time sequence corresponding to a certain type of external environment data, and the lag time of the external environment data at each time point is obtained from the lag time fitting curve.

6. The intelligent operation and maintenance management platform for the operating environment of underground cable trenches according to claim 1, characterized in that, The process of obtaining training samples and forming a training set based on the matching degree and lag time of the sequence to be analyzed and various external environmental data at each time step includes: A time corresponding to a training sample is determined as the prediction time. External environment data of one type is grouped into a sequence, denoted as the external environment sequence. Based on the lag time of each time point before the prediction time, the corresponding operating environment data for each time point before the prediction time is obtained from the sequence to be analyzed, forming a second sequence. The external environment data and matching degree of each time point in the second sequence are used to form a third and fourth sequence, respectively, thus obtaining the second, third, and fourth sequences corresponding to this type of external environment data. Similarly, the second, third, and fourth sequences of other types of external environment data are obtained. The shortest sequence among the second, third, and fourth sequences of each type of external environment data is used to align the second, third, and fourth sequences of each type of external environment data, resulting in a training sample composed of the aligned second, third, and fourth sequences of each type of external environment data. The label of the training sample is the actual value of the operating environment data at the prediction time. All training samples are then combined into a training set.

7. The intelligent operation and maintenance management platform for the operating environment of underground cable trenches according to claim 1, characterized in that, The loss function is specifically as follows: ， in, Represents the loss function; Represents the time index value at time t; A represents the total number of data points in the sequence to be analyzed; and Let represent the predicted and actual values ​​of a certain operating environment data at time t, respectively.

Images