Data acquisition and transmission method and system for energy storage power station

Abstract

The invention relates to the technical field of energy storage data transmission, in particular to an energy storage power station data acquisition and transmission method and system.The method comprises the steps that standardization and segmented text quantization are conducted on original logs of all electric energy storage units, and a digital vector sequence and a text vector sequence are obtained; performing first clustering analysis on the original log based on the text vector sequence to obtain a feature category of each electric energy storage unit; analyzing the distribution characteristics of the original logs in the characteristic categories to obtain the redundancy weight of each electric energy storage unit; analyzing the correlation and difference between the digital vector sequences, and combining the redundancy weight to obtain the abnormal representation of the original log; performing second clustering analysis on the original logs according to the abnormal expressivity to obtain a normal log set; according to the method, the sentence embedding model based on the keyword is obtained by training the normal log set, and the original log is transmitted based on the sentence embedding model, so that the load pressure of storage and transmission can be effectively reduced, and the transmission efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage data transmission, and in particular to a method and system for data collection and transmission of an energy storage power station. Background Art

[0002] As an important part of the modern power system, energy storage power stations are designed to solve the intermittent and unstable problems of renewable energy generation. In the operation of energy storage power stations, log data plays a key role. These log data record the real-time operating status of the power station, energy storage and release, equipment performance and other information, providing an important basis for operation and maintenance decisions. By analyzing the log data, the scheduling strategy of the energy storage system can be optimized, the charging and discharging efficiency can be improved, and the operating costs can be reduced. In addition, log data also provides data support for fault diagnosis, predictive maintenance and performance evaluation, helping to achieve intelligent management.

[0003] With the advancement of technology and the growth of market demand, the application scenarios of energy storage power stations are constantly expanding, including electricity market participation, demand response and grid regulation. However, to fully tap the potential of energy storage power stations, it is still necessary to increase investment in data collection, processing and analysis capabilities to improve their intelligence level and economic benefits.

[0004] Since the energy storage power station is composed of multiple groups of energy storage units, that is, each energy storage unit has multiple batteries, the status of the energy storage unit is mainly described in the form of monitoring logs during the monitoring process of the energy storage power station. However, in the actual monitoring process, there are a lot of identical texts in the monitoring logs, that is, the similarity of the log texts is extremely high, so a lot of redundant data will be generated during storage, which leads to load pressure on the storage and transmission of the collected monitoring logs. Summary of the invention

[0005] In order to solve the above technical problems, the purpose of the present invention is to provide a method and system for data collection and transmission of an energy storage power station.

[0006] According to a first aspect of an embodiment of the present invention, a method for data collection and transmission of an energy storage power station is provided, and the technical solution adopted is specifically as follows: Performing data monitoring on each electric energy storage unit of the energy storage power station to obtain an original log, wherein the original log includes digital data and text data; Performing standardization processing on the digital data to obtain a digital vector sequence, and performing segmented text quantization on the text data to obtain a text vector sequence; Based on the text vector sequence, performing a first cluster analysis on the original log to obtain a feature category of each electric energy storage unit; Analyzing the distribution characteristics of the original logs in the characteristic category to obtain the redundancy weight of each electric energy storage unit; Analyze the correlation and difference between the digital vector sequences, and combine the redundant weights to obtain the abnormal performance of the original log; According to the abnormal performance, a second cluster analysis is performed on the original log to obtain a normal log set and a suspected abnormal log set; The normal log set is used to train a keyword-based sentence embedding model, and data transmission is monitored based on the sentence embedding model.

[0007] In some embodiments of the present invention, segmenting the text data to obtain a text vector sequence includes: According to the stop word list, the stop words in the text data are removed to obtain a valid text; Using the Jieba word segmentation algorithm, the valid text is segmented to obtain keywords of the valid text; All the original logs are divided into equal parts to obtain divided segments; Based on the keywords, a dictionary is constructed for the text data in the segmented segments corresponding to all the original logs in turn to obtain a word vector dictionary; Based on the word vector dictionary, the text data is vectorized to obtain a text vector sequence.

[0008] In some embodiments of the present invention, based on the text vector sequence, a first cluster analysis is performed on the original log to obtain a feature category of each electric energy storage unit, including: Modulo-value the word vectors in the text vector sequence to obtain the text sequence of the original log and the modulus value of the word vector; Using the Mann-Kendall (MK) trend test algorithm, the text sequence is tested and analyzed to obtain a test statistic corresponding to the text vector sequence; Calculating the average value of the modulus values ​​of all word vectors in the text vector sequence to obtain the data distribution of the text vector sequence; Based on the test statistic and the data distribution, the ISODATA algorithm is used to perform a first cluster analysis on the original log to obtain the characteristic category of each electric energy storage unit.

[0009] In some embodiments of the present invention, based on the test statistic and the data distribution, the ISODATA algorithm is used to perform a first cluster analysis on the original log to obtain a feature category of each electric energy storage unit, including: The data distribution is the horizontal axis, and the test statistic is the vertical axis, constructing a two-dimensional analysis space; Based on the two-dimensional analysis space, the ISODATA algorithm is used to perform a first cluster analysis on the original logs to obtain several categories; The number of original logs of each electric energy storage unit in each of the categories is obtained, and the category containing the largest number of original logs of the electric energy storage unit is selected as the characteristic category of the electric energy storage unit.

[0010] In some embodiments of the present invention, analyzing the distribution characteristics of the original logs in the feature category to obtain the redundancy weight of each electric energy storage unit includes: Obtain the number of original logs of each electric energy storage unit in its corresponding feature category, and obtain the total number of original logs contained in the feature category corresponding to each electric energy storage unit; The ratio of the number of original logs of each electric energy storage unit in its corresponding feature category to the total number of original logs in the feature category corresponding to each electric energy storage unit is calculated to obtain the redundancy weight of each electric energy storage unit.

[0011] In some embodiments of the present invention, analyzing the correlation and difference between the digital vector sequences and combining the redundancy weights to obtain the abnormal performance of the original log includes: Counting the number of word vectors in all the digital vector sequences to obtain a standard digital vector sequence; Perform DTW matching on each of the digital vector sequences and the standard digital vector sequence to obtain a regular digital vector sequence corresponding to each of the digital vector sequences; Calculating the Pearson correlation coefficient between each of the regular digital vector sequences and the standard digital vector sequence to obtain the correlation corresponding to each of the regular digital vector sequences; Calculate the module value difference between each of the regular digital vector sequences and the corresponding vector in the standard digital vector sequence to obtain the difference corresponding to each of the regular digital vector sequences; According to the correlation and the difference, and in combination with the redundancy weight, the abnormal performance of the original log is obtained.

[0012] In some embodiments of the present invention, a second cluster analysis is performed on the original log according to the abnormal performance to obtain a normal log set and a suspected abnormal log set, including: Based on the abnormal performance of all original logs, the density clustering algorithm DBSCAN was used for the second clustering analysis, where the radius was 0.1 and the number of neighboring points within the radius was the number of energy storage units, and several clusters were obtained; The original logs that can form a cluster are regarded as ordinary logs, and the original logs that cannot form a cluster are regarded as differential logs. All common logs constitute a normal log set, and all difference logs constitute a suspected abnormal log set.

[0013] In some embodiments of the present invention, monitoring data transmission based on the sentence embedding model includes: The suspected abnormal log set is transmitted, and the keywords of the original logs in the normal log set are transmitted; At the analysis end, the original log in the normal log set is restored using the sentence embedding model and the keywords of the original log.

[0014] According to a second aspect of an embodiment of the present invention, a data acquisition and transmission system for an energy storage power station is provided, comprising: a memory and a processor, wherein: The memory is used to store program code; The processor is used to read the program code stored in the memory and execute the method described in the first aspect of the embodiment of the present invention.

[0015] In some embodiments of the present invention, the processor comprises: A log data monitoring module is used to monitor the data of each electric energy storage unit of the energy storage power station to obtain an original log, wherein the original log includes digital data and text data; A log data processing module, used for performing standardization processing on the digital data to obtain a digital vector sequence, and performing segmented text quantization on the text data to obtain a text vector sequence; A redundancy weight analysis module, configured to perform a first cluster analysis on the original log based on the text vector sequence to obtain a feature category of each electric energy storage unit; and to analyze the distribution characteristics of the original log within the feature category to obtain a redundancy weight of each electric energy storage unit; A log judgment module is used to analyze the correlation and difference between the digital vector sequences, and obtain the abnormal performance of the original log in combination with the redundant weight; and according to the abnormal performance, perform a second cluster analysis on the original log to obtain a normal log set and a suspected abnormal log set; The sentence embedding model training module is used to use the normal log set to train a keyword-based sentence embedding model, and monitor data transmission based on the sentence embedding model.

[0016] Compared with the prior art, the energy storage power station data collection and transmission method and system provided by the present invention has the following beneficial effects: First, by segmenting the text data in the original log and quantizing it to obtain a text vector sequence, we can avoid the situation where the word vectors with differences are ranked at the back of the dictionary and are far away from the normal vocabulary vectors, that is, the vector difference is too large, resulting in distortion of the meaning of the vector.

[0017] Second, by analyzing the differences in text vector sequences caused by the internal resistance of the same energy storage unit, it is used as a redundant weight for the digital vector sequence with higher sensitivity when analyzing the similarity of the energy storage units.

[0018] Third, by analyzing all text vector sequences and digital vector sequences of the same energy storage unit separately, and using the redundant weights obtained by analyzing the text vector sequences as the weights when analyzing the digital vector sequences, the accuracy of judging the abnormal performance of the original logs is improved, thereby improving the reliability of normal log sets and suspected abnormal log sets.

[0019] Fourth, using normal log sets to train a keyword-based sentence embedding model can effectively reduce the storage and transmission load pressure and improve transmission efficiency while ensuring transmission quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. 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 paying creative work.

[0021] Figure 1 A basic flow chart of a method for data collection and transmission of an energy storage power station provided by an embodiment of the present invention; Figure 2 A schematic diagram of the basic composition of a data acquisition and transmission system for an energy storage power station provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the energy storage power station data acquisition and transmission method and system proposed by the present invention, its specific implementation method, structure, characteristics and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. Terms such as "comprises", "comprising" or any other variants thereof are intended to cover non-exclusive inclusion, so that a circuit structure, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such article or device. In the absence of further restrictions, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the article or device including the element.

[0024] The specific scheme of the data collection and transmission method of an energy storage power station provided by the present invention is described in detail below with reference to the accompanying drawings.

[0025] See also Figure 1 , which shows the basic process of a data collection and transmission method for an energy storage power station provided by an embodiment of the present invention.

[0026] like Figure 1 As shown, an embodiment of the present invention provides a method for data collection and transmission of an energy storage power station, which specifically includes: S100: Perform data monitoring on each electric energy storage unit of the energy storage power station to obtain an original log, which includes digital data and text data.

[0027] The monitoring log data of the energy storage unit usually includes the operating status data of each energy storage unit (battery, SOC status, voltage, current, temperature), performance indicators (charging and discharging efficiency, system power input and output, records of faults and abnormal events), as well as environmental data and maintenance records.

[0028] Therefore, firstly, sensors are installed at the corresponding positions that need to be monitored based on each energy storage unit, and the data of each sensor is collected and summarized by using the data acquisition system (SCADA). During the summary, the data transmission between sensors, equipment, and acquisition systems is realized through the industrial standard protocol (MODBUS), so as to obtain the original log of each energy storage unit in the data acquisition system at each time node. The log is generated every 15 minutes, that is, all energy storage units have a unique original log at each time node. Each original log contains recorded digital data and text data (text description log), and because the data acquisition system has the same rules when summarizing all data, the positions of digital data and text data of the same description object in the log are the same.

[0029] S200: performing standardization processing on digital data to obtain a digital vector sequence, and performing segmented text quantization on text data to obtain a text vector sequence.

[0030] At present, the main steps for extracting keywords from text are to remove stop words, model training, keyword extraction, and obtain a keyword set. However, when extracting keywords from original logs, the stop words belonging to digital data may be more important than those belonging to text data. Using the same rule to extract keywords from digital data and text data will lead to extraction deviations. Therefore, this embodiment performs standardization processing on digital data to obtain a digital vector sequence, and performs segmented text quantization on text data to obtain a text vector sequence. The specific operations are: First, for digital data, all characters (strings), identifiers, numbers, and stop words in each original log are obtained through the stop word list. All characters (strings) (digital text units) and numbers are arranged in the order of text content, and a word vector table is built based on these characters to obtain the digital vector sequence in each original log. Since the data acquisition system has the same rules for summarizing all data, the digital vector sequences of different original logs are only different in numbers, and the characters (strings) as digital units are the same.

[0031] At the same time, the text data is segmented and quantized to obtain a text vector sequence, including: First, according to the stop word list, the stop words in the text data are removed to obtain the valid text; the Jieba word segmentation algorithm is used to segment the valid text to obtain the keywords of the valid text, that is, all the keywords in each original log can be obtained.

[0032] Then, all the original logs are divided into equal proportions to obtain segmented parts; based on the keywords, the text data in the segmented parts corresponding to all the original logs are constructed into dictionaries in turn to obtain a word vector dictionary. Traditionally, when the text is vectorized, it is necessary to complete the construction of the word vector dictionary for all the keywords of each text before constructing the word vector dictionary for the next original log. Then, after the original log is vectorized based on the constructed word vector dictionary, the regularity in the text will no longer exist. In order to ensure that the text vector sequence obtained after the original log is vectorized using the constructed word vector dictionary still has the regularity described, in this embodiment, all the original logs are divided into equal proportions to obtain segmented parts, and the number of segmented parts is ten. Based on the keywords, the text data in the segmented parts corresponding to all the original logs are constructed into dictionaries in turn to obtain a word vector dictionary. Further elaboration is: construct a dictionary for all text data corresponding to the same copy after segmentation. When the first dictionary of all original logs is constructed, construct a dictionary for the second copy, so that the texts with similar word vector dictionaries have similar trend, that is, similar log information and content are located together in the text vector. Then, when the text vector of an original log is different from the text vector of other original logs in position, it means that the similarity between the text vector and other text vectors of the same category is different, which can highlight the difference between the original logs. If the traditional dictionary construction method is used, the vectors of different words will be ranked at the back of the dictionary, far away from the normal word vectors, that is, the vector difference is too large, resulting in the distortion of the meaning of the vector. Based on the word vector dictionary, the text data is vectorized to obtain a text vector sequence.

[0033] Finally, based on the word vector dictionary, the text data is vectorized to obtain a text vector sequence, that is, the text vector sequence corresponding to each original log can be obtained.

[0034] S300: Based on the text vector sequence, a first clustering analysis is performed on the original log to obtain a feature category of each electric energy storage unit.

[0035] After obtaining the digital vector sequence and text vector sequence of all the original logs, since the information described by the digital vector sequence and the text vector sequence is different, that is, the digital vector sequence describes the sensor detection information of the electric energy storage unit, and the text vector sequence describes the operating status of the electric energy storage unit, it is necessary to analyze all the text vector sequences and digital vector sequences of the same electric energy storage unit separately, so as to distinguish the normal original logs from the abnormal original logs.

[0036] For different energy storage units, the internal resistance of their battery packs (the internal resistance of a battery refers to the internal resistance of the battery, which is a scalar, usually expressed in ohms. The smaller the internal resistance, the better the stability of the battery output voltage and the greater the output power) is different. This internal resistance will affect the current input, current output and power. Therefore, it is necessary to first analyze the differences in text vector sequences caused by the internal resistance influence coefficient of the same energy storage unit, which will be used as a redundant weight for a more sensitive digital vector sequence when analyzing the similarity of energy storage units.

[0037] The information reflected by the text vector is more ambiguous than that of the digital vector. Since the influence of the internal resistance is mainly on the overall influence of the electric energy storage unit, and the corresponding text vector is also a description of the state of the electric energy storage power supply, this embodiment analyzes all text vectors of the original log based on the text vector sequence, and then analyzes whether the same electric energy storage unit meets the uniqueness, and then obtains the redundant weight.

[0038] Based on the above analysis, this embodiment first performs a first cluster analysis on the original log based on the text vector sequence to obtain the characteristic category of each electric energy storage unit. Further, it includes: modularizing the word vectors in the text vector sequence to obtain the text sequence and the modular value of the word vector of the original log; using the Mann-Kendall (MK) trend test algorithm to perform a test analysis on the text sequence to obtain the test statistic corresponding to the text vector sequence; calculating the average value of the modular value of all word vectors in the text vector sequence to obtain the data distribution of the text vector sequence; based on the test statistic and data distribution, using the ISODATA algorithm to perform a first cluster analysis on the original log to obtain the characteristic category of each electric energy storage unit. Furthermore, based on the test statistic and data distribution, the ISODATA algorithm is used to perform the first cluster analysis on the original logs to obtain the characteristic categories of each electric energy storage unit, including: the test statistic is used as the horizontal axis, the data distribution is used as the vertical axis, and a two-dimensional analysis space is constructed; based on the two-dimensional analysis space, the ISODATA algorithm is used to perform the first cluster analysis on the original logs to obtain several categories; the number of original logs of each electric energy storage unit in each category is obtained, and the category with the largest number of original logs containing electric energy storage units is selected as the characteristic category of the electric energy storage unit.

[0039] First Taking the original log as an example, the further explanation is as follows: For All word vectors in the text vector sequence of the original log are modulized, and each word vector corresponds to a modulus value. Then the text vector sequence is converted into a sequence composed of multiple modulus values, called a text sequence, that is, the first The text sequence of the original log and the modulus of all word vectors. Then the Mann-Kendall (MK) trend analysis algorithm is used to analyze the The text sequence of the original log is tested and analyzed to obtain the The test statistic of the text sequence corresponding to the original log is mainly used to describe the The trend of the text vector sequence of the original log; affected by the word vector dictionary construction method of the text vector, the trend of similar text vector sequences is the same, and the text vector sequence that is offset from the normal text vector sequence will have different test statistic results due to the difference in the modulus of the word vector and other text vector sequences. When the test statistic is greater than 0, it indicates an upward trend, when it is less than 0, it indicates a downward trend, and when it is equal to 0, it indicates an unchanged trend. Therefore, the test statistic can measure the overall situation described by the text data in different original logs.

[0040] Since the test statistic can only describe the trend relationship between different text vector sequences but cannot describe the similarity or difference relationship between vector values, this embodiment calculates the average value of all word vector modulus values ​​in the text vector sequence to obtain the data distribution of the text vector sequence as the first The data distribution corresponding to the original logs. Then the value distribution of similar original logs and the trend under the value distribution are similar.

[0041] Based on the above logic, a two-dimensional analysis space is constructed. The horizontal axis of the two-dimensional analysis space is the data distribution, and the vertical axis is the test statistic. The ISODATA algorithm is used to cluster the original logs to obtain several categories, where the input initial value of the ISODATA algorithm is the same as the number of electric energy storage units.

[0042] Among the obtained categories, if the same category only contains the original logs of the same energy storage unit, and does not contain or contains fewer original logs of other energy storage units, it means that the energy storage unit is special compared with other energy storage units, which means that it may be affected by the internal resistance of the battery, resulting in a deviation in its data. Therefore, the number of original logs of each energy storage unit in each category is obtained, and the category with the largest number of original logs of the energy storage unit is selected as the characteristic category of the energy storage unit.

[0043] S400: Analyze the distribution characteristics of the original logs in the characteristic category to obtain the redundancy weight of each electric energy storage unit.

[0044] Analyze the distribution characteristics of the original logs in the feature category to obtain the redundant weight of each energy storage unit. Specifically, by obtaining the number of original logs of each energy storage unit in its corresponding feature category and the total number of original logs contained in the feature category corresponding to each energy storage unit; then calculate the ratio of the number of original logs of each energy storage unit in its corresponding feature category to the total number of original logs contained in the feature category corresponding to each energy storage unit to obtain the redundant weight of each energy storage unit. For example, the redundancy weight of an energy storage unit is The calculation method is: Where: Indicates The redundancy weight of the energy storage unit; Indicates The number of raw logs of an energy storage unit in its corresponding feature category; Indicates The feature category corresponding to the energy storage unit contains the total number of original logs.

[0045] Indicates The number of raw logs of the energy storage unit in its corresponding feature category is The ratio of the total number of original logs contained in the feature category corresponding to the energy storage unit. The larger the ratio, the The larger the proportion of an energy storage unit in its feature category, the less the category contains original logs of other energy storage units, the more special the energy storage unit is compared with other energy storage units, and the more likely the energy storage unit is affected by internal resistance, resulting in overall problems. The corresponding redundancy weight of the energy storage unit is larger.

[0046] Similarly, the redundant weights of all energy storage units are obtained.

[0047] S500: Analyze the correlation and difference between the digital vector sequences, and combine the redundant weights to obtain the abnormal performance of the original log.

[0048] The digital vector sequence mainly contains numbers and digital units. Since the order of the data acquisition system is the same when it is summarized, the units are matched. Therefore, before analyzing the correlation and difference between the digital vector sequences, the different original logs must be regularized first. In this embodiment, the number of word vectors in all digital vector sequences is counted to obtain a standard digital vector sequence; each digital vector sequence is matched with the standard digital vector sequence by DTW to obtain a regular digital vector sequence corresponding to each digital vector sequence. The specific operations are: First, the number of word vectors in all digital vector sequences is counted, and the digital vector sequence containing the most word vectors is recorded as the standard digital vector sequence; and the number of times each word vector appears in the standard digital vector sequence is counted, and then the Otsu threshold is used for segmentation, and the larger part of all word vectors is recorded as unit word vectors (characters such as units representing numbers).

[0049] Then, DTW matching is performed on the digital vector sequence corresponding to each original log and the standard digital vector sequence. Since the unit word vector is unchanged, the standard digital vector sequence is segmented based on every two unit word vectors of the standard digital vector sequence; in the digital vector sequence, the word vector with the smallest DTW distance to each unit word vector of the standard digital vector sequence is used as the unit word vector in the digital vector sequence of each original log. It should be noted that the DTW here is not a generalized DTW. The principle of DTW is to traverse and subtract two digital vector sequences to obtain a difference, and then select the one with the smallest difference in each sequence as a matching element (possibly one-to-many or many-to-one), and then sum up all the differences to obtain the DTW distance. In this embodiment, the word vector with the smallest DTW distance is to select the word vector with the smallest difference in one-to-many or many-to-one; then, based on the DTW matching result, equal-length interpolation is performed to make the length of the digital vector sequence of each original log the same as that of the standard digital vector sequence, that is, since the DTW matching is random, there may be one-to-many, which is equivalent to finding the left and right anchor points here. To make the number of vectors in the middle the same, equal-length interpolation is performed to obtain a regular digital vector sequence of each original log.

[0050] After obtaining the regular digital vector sequence corresponding to each digital vector sequence, the Pearson correlation coefficient between each regular digital vector sequence and the standard digital vector sequence is calculated to obtain the correlation corresponding to each regular digital vector sequence, that is, the correlation corresponding to each original log can be obtained. At the same time, the modulus difference between the corresponding numerical vectors in each regular digital vector sequence and the standard digital vector sequence is calculated to obtain the difference corresponding to each regular digital vector sequence, where the numerical vector represents the vector in the regular digital vector sequence and the standard digital vector sequence except the unit word vector. Based on the correlation and difference, combined with the redundant weight, the abnormal performance of the original log is obtained.

[0051] Build The energy storage unit Abnormal performance of raw logs The calculation formula is: Where: Indicates The energy storage unit The abnormal performance of the original log; Indicates The redundancy weight of the energy storage unit; Indicates The energy storage unit The Pearson correlation coefficient between the regularized digital vector sequence of the original log and the standard digital vector sequence, that is, The correlation corresponding to the regularized digital vector sequence of the original logs, ; Indicates The energy storage unit In the regular digital vector sequence of the original log, except for the unit word vector The modulus of the word vector; Indicates the first The modulus of the word vector; Indicates The energy storage unit The total number of word vectors excluding unit word vectors in the original log; Represents the sigmoid normalization function.

[0052] Redundant Weight The larger the value, the more likely the abnormality of the energy storage unit is caused by the internal resistance, and the smaller the abnormal performance of the corresponding energy storage unit; the absolute value of the Pearson correlation coefficient The larger the value, the more likely the energy storage unit is normal, and the smaller the abnormal performance of the corresponding energy storage unit; Indicates The energy storage unit In the regular digital vector sequence and standard digital vector sequence of the original log, except for the unit word vector The modulus difference of the word vectors, that is, The difference between the regular digital vector sequence of the original log, the larger the value, the greater the difference between the regular digital vector sequence of the energy storage unit and the standard digital vector sequence, and the greater the abnormal performance of the corresponding energy storage unit. The situation, that is, The energy storage unit There is no correlation between the regular digital vector sequence of the original log and the standard digital vector sequence. At this time, it can be directly judged that the original log is a difference log and is included in the suspected abnormal log set.

[0053] Here, only the word vectors other than the unit word vectors in the regular digital vector sequence of the original log are used to analyze the abnormal expressiveness in order to avoid the problem that the value of the abnormal expressiveness is weakened due to the same unit word vectors.

[0054] S600: Perform a second cluster analysis on the original logs according to the abnormal performance to obtain a normal log set and a suspected abnormal log set.

[0055] According to the abnormal performance, the original logs are clustered for the second time to obtain the normal log set and the suspected abnormal log set. Specifically, based on the abnormal performance of all the original logs of all the energy storage units, the density clustering algorithm DBSCAN is used for the second clustering analysis, where the radius is 0.1, and the number of neighboring points within the radius is the number of energy storage units, to obtain several clusters; the original logs that can form a cluster are regarded as ordinary logs, and the original logs that cannot form a cluster are regarded as difference logs; all ordinary logs constitute the normal log set, and all difference logs constitute the suspected abnormal log set.

[0056] S700: A keyword-based sentence embedding model is obtained by training a normal log set, and monitoring data transmission is performed based on the sentence embedding model.

[0057] The information in the normal log set has extremely high similarity. This embodiment trains a keyword-based sentence embedding model based on the normal log set, so that the keywords are used as the storage information of each ordinary log in the normal log set, and other non-keyword words are discarded. Then, when the ordinary log in the normal log set needs to be called in the end, it is only necessary to input the keywords into the keyword-based sentence embedding model to obtain the complete log information.

[0058] The training process of the keyword-based sentence embedding model is: The BERT model is a revolutionary pre-training model proposed by Google in 2018. It uses the Transformer architecture to significantly improve the performance of natural language processing tasks through bidirectional context understanding. In the BERT-KPE project, BERT is trained as a keyword locator that can identify which words or phrases are essential to the topic of the original text. Traditional methods usually rely on statistics and rules, while BERT-KPE introduces deep learning, which adaptively grasps the rules of keyword selection by learning a large amount of unlabeled text.

[0059] Therefore, this embodiment uses the normal log set as the training set and the validation set, where the ratio of the training set to the validation set is 3:7, for model training. The training architecture is: Transformer architecture; The loss function is: the mean square error loss function of the triplet; The training steps are: Input layer: Use the Input class to define an input layer that accepts a string array (i.e., a sentence) of shape (1,) and is named "sentence"; The backbone network processes the encoded input: the encoded input is passed to the backbone network backbone; Global average pooling: Use the GlobalAveragePooling1D layer to perform global average pooling on the output of the backbone network; Create a model: Use the Model class to combine the input layer, preprocessor, backbone network, and global average pooling layer into a complete model roberta_encoder. The input of the model is a sentence, and the output is the embedded representation of the sentence.

[0060] Based on the above training steps, a keyword-based sentence embedding model is obtained.

[0061] It should be noted that the normal log set used to train the sentence embedding model uses historical data or data generated in the early stage, that is, the sentence embedding model is obtained first; the original logs of each energy storage unit of the energy storage power station are generated in real time (a log is generated every 15 minutes), and the pre-trained sentence embedding model is used to restore the real-time original log data.

[0062] The data transmission is monitored based on the sentence embedding model. Specifically, when the original log data is transmitted, the suspected abnormal log set and the normal log set in the original log are identified, the suspected abnormal log set is transmitted in its entirety, the keywords of the original log in the normal log set are transmitted, and the original log in the normal log set is restored using the pre-trained sentence embedding model at the analysis end.

[0063] Based on the same inventive concept as the above method, this embodiment also provides a data collection and transmission system for an energy storage power station.

[0064] See also Figure 2 , which shows the basic composition of a data acquisition and transmission system for an energy storage power station provided by an embodiment of the present invention.

[0065] like Figure 2 As shown, a data collection and transmission system for an energy storage power station includes: a memory 10 and a processor 20, wherein: A memory 10, used for storing program codes; The processor 20 is used to read the program code stored in the memory 10, and execute data monitoring on each electric energy storage unit of the energy storage power station to obtain an original log, which includes digital data and text data; standardize the digital data to obtain a digital vector sequence, and segment the text data to obtain a text vector sequence; based on the text vector sequence, perform a first cluster analysis on the original log to obtain a feature category of each electric energy storage unit; analyze the distribution characteristics of the original log within the feature category to obtain the redundant weight of each electric energy storage unit; analyze the correlation and difference between the digital vector sequences, and combine the redundant weight to obtain the abnormal expression of the original log; according to the abnormal expression, perform a second cluster analysis on the original log to obtain a normal log set and a suspected abnormal log set; use the normal log set to train a keyword-based sentence embedding model, and monitor data transmission based on the sentence embedding model.

[0066] Furthermore, the processor 20 includes a log data monitoring module 21, a log data processing module 22, a redundant weight analysis module 23, a log judgment module 24, and a sentence embedding model training module 25. Among them: The log data monitoring module 21 is used to monitor the data of each electric energy storage unit of the energy storage power station to obtain the original log, which includes digital data and text data; The log data processing module 22 is used to perform standardization processing on digital data to obtain a digital vector sequence, and to perform segmented text quantization on text data to obtain a text vector sequence; The redundancy weight analysis module 23 is used to perform a first cluster analysis on the original log based on the text vector sequence to obtain the feature category of each electric energy storage unit; and analyze the distribution characteristics of the original log within the feature category to obtain the redundancy weight of each electric energy storage unit; The log judgment module 24 is used to analyze the correlation and difference between the digital vector sequences, and combine the redundant weights to obtain the abnormal performance of the original log; and according to the abnormal performance, perform a second cluster analysis on the original log to obtain a normal log set and a suspected abnormal log set; The sentence embedding model training module 25 is used to obtain a keyword-based sentence embedding model by using a normal log set for training, and to monitor data transmission based on the sentence embedding model.

[0067] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0068] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for data collection and transmission of energy storage power station, characterized in that: The method comprises: Performing data monitoring on each electric energy storage unit of the energy storage power station to obtain an original log, wherein the original log includes digital data and text data; Performing standardization processing on the digital data to obtain a digital vector sequence, and performing segmented text quantization on the text data to obtain a text vector sequence; Based on the text vector sequence, performing a first cluster analysis on the original log to obtain a feature category of each electric energy storage unit; Analyzing the distribution characteristics of the original logs in the characteristic category to obtain the redundancy weight of each electric energy storage unit; Analyze the correlation and difference between the digital vector sequences, and combine the redundant weights to obtain the abnormal performance of the original log; According to the abnormal performance, a second cluster analysis is performed on the original log to obtain a normal log set and a suspected abnormal log set; The normal log set is used to train a keyword-based sentence embedding model, and data transmission is monitored based on the sentence embedding model.

2. The energy storage power station data collection and transmission method according to claim 1 is characterized in that: The text data is segmented and quantized to obtain a text vector sequence, including: According to the stop word list, the stop words in the text data are removed to obtain a valid text; Using the Jieba word segmentation algorithm, the valid text is segmented to obtain keywords of the valid text; All the original logs are divided into equal parts to obtain divided segments; Based on the keywords, a dictionary is constructed for the text data in the segmented segments corresponding to all the original logs in turn to obtain a word vector dictionary; Based on the word vector dictionary, the text data is vectorized to obtain a text vector sequence.

3. The energy storage power station data collection and transmission method according to claim 1, characterized in that: Based on the text vector sequence, a first cluster analysis is performed on the original log to obtain feature categories of each electric energy storage unit, including: Modulo-value the word vectors in the text vector sequence to obtain the text sequence of the original log and the modulus value of the word vector; Using the Mann-Kendall (MK) trend test algorithm, the text sequence is tested and analyzed to obtain a test statistic corresponding to the text vector sequence; Calculating the average value of the modulus values ​​of all word vectors in the text vector sequence to obtain the data distribution of the text vector sequence; Based on the test statistic and the data distribution, the ISODATA algorithm is used to perform a first cluster analysis on the original log to obtain the characteristic category of each electric energy storage unit.

4. The energy storage power station data collection and transmission method according to claim 3 is characterized in that: Based on the test statistic and the data distribution, the ISODATA algorithm is used to perform a first cluster analysis on the original log to obtain the characteristic categories of each electric energy storage unit, including: The data distribution is the horizontal axis, and the test statistic is the vertical axis, constructing a two-dimensional analysis space; Based on the two-dimensional analysis space, the ISODATA algorithm is used to perform a first cluster analysis on the original logs to obtain several categories; The number of original logs of each electric energy storage unit in each of the categories is obtained, and the category containing the largest number of original logs of the electric energy storage unit is selected as the characteristic category of the electric energy storage unit.

5. The energy storage power station data collection and transmission method according to claim 1 or 4, characterized in that: Analyze the distribution characteristics of the original logs in the characteristic category to obtain the redundancy weight of each electric energy storage unit, including: Obtain the number of original logs of each electric energy storage unit in its corresponding feature category, and obtain the total number of original logs contained in the feature category corresponding to each electric energy storage unit; The ratio of the number of original logs of each electric energy storage unit in its corresponding feature category to the total number of original logs in the feature category corresponding to each electric energy storage unit is calculated to obtain the redundancy weight of each electric energy storage unit.

6. The energy storage power station data collection and transmission method according to claim 1, characterized in that: Analyze the correlation and difference between the digital vector sequences, and combine the redundant weights to obtain the abnormal performance of the original log, including: Counting the number of word vectors in all the digital vector sequences to obtain a standard digital vector sequence; Perform DTW matching on each of the digital vector sequences and the standard digital vector sequence to obtain a regular digital vector sequence corresponding to each of the digital vector sequences; Calculating the Pearson correlation coefficient between each of the regular digital vector sequences and the standard digital vector sequence to obtain the correlation corresponding to each of the regular digital vector sequences; Calculate the module value difference between each of the regular digital vector sequences and the corresponding vector in the standard digital vector sequence to obtain the difference corresponding to each of the regular digital vector sequences; According to the correlation and the difference, and in combination with the redundancy weight, the abnormal performance of the original log is obtained.

7. The energy storage power station data collection and transmission method according to claim 1 or 6, characterized in that: According to the abnormal performance, a second cluster analysis is performed on the original log to obtain a normal log set and a suspected abnormal log set, including: Based on the abnormal performance of all original logs, the density clustering algorithm DBSCAN was used for the second clustering analysis, where the radius was 0.1 and the number of neighboring points within the radius was the number of energy storage units, and several clusters were obtained; The original logs that can form a cluster are regarded as ordinary logs, and the original logs that cannot form a cluster are regarded as differential logs. All common logs constitute a normal log set, and all difference logs constitute a suspected abnormal log set.

8. The energy storage power station data collection and transmission method according to claim 1 is characterized in that: Monitoring data transmission based on the sentence embedding model includes: The suspected abnormal log set is transmitted, and the keywords of the original logs in the normal log set are transmitted; At the analysis end, the original log in the normal log set is restored using the sentence embedding model and the keywords of the original log.

9. A data collection and transmission system for an energy storage power station, characterized in that: The system comprises: a memory and a processor, wherein: The memory is used to store program codes; The processor is configured to read the program code stored in the memory and execute the method according to any one of claims 1 to 8.

10. The energy storage power station data acquisition and transmission system according to claim 9, characterized in that: The processor comprises: A log data monitoring module is used to monitor the data of each electric energy storage unit of the energy storage power station to obtain an original log, wherein the original log includes digital data and text data; A log data processing module, used for performing standardization processing on the digital data to obtain a digital vector sequence, and performing segmented text quantization on the text data to obtain a text vector sequence; A redundancy weight analysis module, configured to perform a first cluster analysis on the original log based on the text vector sequence to obtain a feature category of each electric energy storage unit; and to analyze the distribution characteristics of the original log within the feature category to obtain a redundancy weight of each electric energy storage unit; A log judgment module is used to analyze the correlation and difference between the digital vector sequences, and obtain the abnormal performance of the original log in combination with the redundant weight; and according to the abnormal performance, perform a second cluster analysis on the original log to obtain a normal log set and a suspected abnormal log set; The sentence embedding model training module is used to use the normal log set to train a keyword-based sentence embedding model, and monitor data transmission based on the sentence embedding model.

Images