Natural language interaction-based method for extracting parameter of energy storage management system

Abstract

Disclosed is a natural language interaction-based method for extracting a parameter of an energy storage management system, belonging to the technical field of energy storage of new energy power systems. The method comprises parameter type division, natural language interaction corpus collection, parameter labeling, construction of a word segmentation set, natural language interaction corpus encoding, computation of a prior probability and a conditional probability, persistence of the prior probability and the conditional probability, new corpus input, loading of the prior probability and the conditional probability, computation of a posterior probability on the basis of a naive Bayes formula, and extraction of a parameter corresponding to the maximum value of the posterior probability.

Description

A natural language interaction-based energy storage management system parameter extraction method TECHNICAL FIELD

[0001] The present application belongs to the technical field of energy storage of new energy power systems, and particularly relates to a natural language interaction-based energy storage management system parameter extraction method. BACKGROUND

[0002] With the continuous expansion of new energy power generation capacity, the proportion of new energy power generation in the power grid is becoming higher and higher. However, due to the small single-machine capacity, large number, scattered distribution, and significant intermittency, volatility, randomness and other characteristics of new energy power generation, high proportion of new energy grid connection will inevitably bring unprecedented challenges to power system supply and demand balance, safety and stability control, etc. Energy storage system is a key link in regulating the imbalance between new energy power generation and power system supply and demand, energy management and optimization. The energy storage system usually includes energy storage devices, energy converters, control system connection devices, auxiliary devices, etc. To ensure the safe, stable and efficient operation of the energy storage system, accurate management and control of each part of the energy storage system are required, which requires the energy storage management system user or technician to input the requirements including time, location and action parameters to achieve accurate management and control of the energy storage system.

[0003] The current traditional energy storage management system parameter extraction method has the following disadvantages: (1) complex operation: the current traditional energy storage management system parameter extraction method usually needs to select from a large number of parameters in the interactive interface, which has a use threshold; (2) poor universality: the current traditional energy storage management system parameter selection is usually isolated between different functions, and there is no correlation between different functions; (3) poor scalability: when there are new functions or new parameters of existing functions, the traditional energy storage management system parameter extraction method usually needs to redesign the parameter structure, which has no scalability; (4) poor robustness: the traditional energy storage management system parameter extraction method usually cannot handle input errors or out-of-limit, and cannot run normally when abnormal conditions occur. SUMMARY

[0004] In view of the parameter extraction problem in the energy storage management system, a natural language interaction-based energy storage management system parameter extraction method is invented, which reduces the operation complexity of the energy storage management system and reduces the use threshold through the natural language interaction-based parameter extraction method; the universality between management functions is improved by standard coding of different tasks, functions and type parameters; the accuracy and scalability of the parameter extraction method are improved through artificial parameter labeling; and the stability of system operation is improved through machine learning fuzzy processing technology.

[0005] A natural language interaction-based energy storage management system parameter extraction method, comprising:

[0006] S1: Parameter type classification and organization, parameter type set C = (C 1 C 2 ,...,C m ), m represents the number of types after partitioning; C i This represents the set of the i-th parameter type after partitioning, and the parameter set of the i-th parameter type set. in: For parameter set C i The j-th parameter, n i For parameter set C i The number of elements, i = 1, 2, ..., m, j = 1, 2, ..., n i ;

[0007] S2: Natural language interaction corpus collection, each piece of natural language interaction corpus corresponds to a label vector b = (b1, b2, ..., bm) containing m parameter values. m ), where b i The parameter value corresponding to the type of the i-th parameter in the current statement;

[0008] S3: Parameter labeling, labeling each parameter type;

[0009] S4: Construct a word segmentation set, word segmentation set W = (w1, w2, ..., w...). L ), where wl is the l-th element of the word segmentation set W, l = 1, 2, ..., L;

[0010] S5: Encode the natural language interaction corpus. For each piece of natural language interaction corpus, initialize the encoding vector v with a zero vector of length L. If w l In natural language interaction corpora, v is set l =1, otherwise set v l =0, where v l This is the l-th element of the encoding vector v;

[0011] S6: Construct the dataset;

[0012] S7: Calculate the prior probability matrix PB for all parameters, where PB is an m-row matrix. A two-dimensional matrix of zeros;

[0013] S8: Calculate the conditional probability matrix PBV of each word in the natural language interactive corpus. PBV is... A three-dimensional matrix;

[0014] S9: Persistently store the prior probability matrix PB and the conditional probability matrix PBV;

[0015] S10: Input a new natural language interaction corpus and obtain the encoding vector v′ of the natural language interaction corpus to be predicted;

[0016] S11: load the prior probability matrix PB and the conditional probability matrix PBV;

[0017] S12: calculate the parameter vector

[0018] S13: output the parameter vector b'.

[0019] Further, the specific steps of constructing the data set in S6 include:

[0020] S6.1: initialize the encoding matrix V and the label matrix B of the natural language interaction corpus, wherein the encoding matrix V is a 0 matrix with K rows and L columns, each row representing the encoding of a piece of natural language interaction corpus; the label matrix B is a 0 matrix with K rows and m columns, each row representing a parameter value label vector of a piece of natural language interaction corpus; wherein k = 1, 2, …, K, K is the number of natural language interaction corpora;

[0021] S6.2: encode the kth corpus using the encoding method of the natural language interaction corpus in S5, and assign it to the kth row of the encoding matrix V; use the parameter labeling method in S3 to label the kth natural language interaction corpus, and assign it to the kth row of the matrix B.

[0022] Further, the process of calculating the prior probability matrix PB of all parameters in S7 is as follows:

[0023] The prior probability of the jth parameter in the parameter type i is calculated according to the Laplace correction method, and the calculation formula is:

[0024] where PB ij is the element in the ith row and jth column of the vector matrix PB, represents the probability that all rows in the matrix B contain the element , ni,j is the number of elements in the ith column of the matrix B equal to , K is the number of natural language interaction corpora, and m is the number of columns of the label matrix B.

[0025] Further, the calculation formula of the conditional probability matrix PBV of the occurrence of each word in the natural language interaction corpus in S8 is as follows:

[0026] where PBV ijl is the element in the (i, j, l) position of the matrix PBV, represents the conditional probability of the occurrence of the word w in the natural language interaction corpus labeled with the parameter l , N(V lwherein L represents the number of columns of the encoding matrix V, and represents the number of elements in the lth column of the matrix V which are equal to 1.

[0027] Further, the parameter vector is calculated in S12 wherein The calculation formula of is as follows:

[0028] Further, according to the naive Bayes formula, The derivation process of the calculation formula of is as follows:

[0029] wherein: represents the probability of the parameter selection of the parameter type i when the encoding vector of the natural language interaction corpus is v', and p(v') represents the probability of the encoding vector of the natural language interaction corpus being v'.

[0030] Further, in the parameter labeling of S3, the artificial parameter value labeling is performed on each parameter type.

[0031] Further, in the construction of the segmentation set in S4, the segmentation of all natural language interaction corpora is performed using a segmenter, and the segmentation set W is obtained by removing the duplicates after the segmentation.

[0032] A computer device, comprising: a memory, a processor to store a computer program on the memory and executable on the processor, and the processor executes the computer program to realize the natural language interaction-based parameter extraction method of the energy storage management system.

[0033] A computer readable storage medium, having a computer program stored thereon, the computer program being executed by a processor to realize the natural language interaction-based parameter extraction method of the energy storage management system.

[0034] Compared with the prior art, the present application has at least the following beneficial effects:

[0035] The present application uses the natural language interaction mode to extract parameters, reduces the use threshold of the traditional energy storage management system, solves the problem of poor generality of the parameter extraction method, improves the accuracy and expandability of the traditional energy storage management system parameter extraction method through artificial parameter labeling, and improves the robustness of the traditional energy storage management system parameter extraction method. BRIEF DESCRIPTION OF DRAWINGS

[0036] Fig. 1 is a flow structure diagram of the present application;

[0037] Fig. 2 is a conditional probability heat map of each segmentation about each parameter in a parameter type set in the present application;

[0038] ​Fig. 3 is a schematic diagram of the accuracy of each type of parameter extraction in the present application. DETAILED DESCRIPTION

[0039] The principles and features of the present application are described below in conjunction with all the accompanying drawings, and the examples are only used to explain the present application and not to limit the scope of the present application.

[0040] Referring to Figs. 1-3, the embodiment of the present application discloses a parameter extraction method of energy storage management system based on natural language interaction, which specifically comprises:

[0041] S1: parameter type division and arrangement.

[0042] Specifically, the parameter types are divided according to the use and working logic of the energy storage management system. The parameter types are abstracted from the actual application scenarios. Each type of parameter is enumerated according to the functional requirements. The parameter type set C=(C 1 ,C 2 ,…,C m ), wherein Ci represents the i-th parameter type set after division, i=1, 2, …, m, and m represents the number of types after division; the parameter set of the i-th parameter type set is wherein: is the j-th parameter of the parameter set C i , n i is the number of elements of the parameter set C i , j=1, 2, …, n i .

[0043] S2: natural language interaction corpus collection. Each piece of natural language interaction corpus corresponds to a label vector b containing m parameter values, b=(b1, b2, …, b m ), wherein b i is the parameter value corresponding to the i-th parameter type of the current sentence.

[0044] Specifically, the natural language description of the demand parameters of the user or the operation and maintenance engineer using the energy storage management system is collected, i.e. the natural language interaction corpus. Each piece of natural language interaction corpus corresponds to an accurate, complete and implementable use demand containing several parameters.

[0045] S3: parameter labeling.

[0046] Specifically, the values in the parameter types divided in S1 are used to label the natural language interaction corpus, i.e. manual parameter value labeling is performed on each parameter type. The process of converting into a label vector is actually the parameter labeling of S3.

[0047] For example, the parameter type division may be: project name, container number, cluster number, task name, etc.

[0048] Given corpus: Please analyze the battery module temperature consistency of A project 10th container 4th cluster.

[0049] Annotated result: (A, 10, 4, module temperature consistency, …).

[0050] S4: Construct a word segmentation set.

[0051] Specifically, use a word segmenter to segment all natural language interaction corpora, and obtain a word segmentation set W by de-duplication after segmentation, W = (w1, w2, …, wL), where: l = 1, 2, …, L, wL is the lth element of the word segmentation set W. In this embodiment, the word segmenter can be a Chinese word segmenter, such as Jieba segmentation, or a custom word segmenter. L l

[0052] S5: Encode the natural language interaction corpus.

[0053] Specifically, for each piece of natural language interaction corpus, initialize the encoding vector v with a 0 vector of length L, and if w l In the natural language interaction corpus, set v l = 1, otherwise set v l = 0, where v l is the lth element of the encoding vector v of the natural language interaction corpus.

[0054] S6: Construct a data set.

[0055] Specifically, S6 specifically includes the following steps:

[0056] S6.1: Initialize the encoding matrix V and the label matrix B of the natural language interaction corpus, where the encoding matrix V is a 0 matrix of K rows and L columns, each row representing the encoding of a piece of natural language interaction corpus; the label matrix B is a 0 matrix of K rows and m columns, each row representing a parameter value label vector of a piece of natural language interaction corpus, k = 1, 2, …, K.

[0057] S6.2: Encode the kth corpus using the encoding method of the natural language interaction corpus in S5, and assign it to the kth row of the encoding matrix V; use the parameter annotation method in S3 to annotate the kth natural language interaction corpus, and assign it to the kth row of the label matrix B, where K is the number of natural language interaction corpora, which can also be understood as the number of rows of the encoding matrix V or the label matrix B.

[0058] S7: Calculate the prior probability matrix PB of all parameters, PB is a two-dimensional 0 matrix of m rows and n columns.

[0059] Specifically, the prior probability vector of each parameter in the parameter type i is calculated according to the Laplace correction method. ​​​The calculation formula of the conditional probability matrix PBV of the words in the natural language interaction corpus is:

[0060] wherein, is the jth element of the vector PB, j = 1, 2, …, n i , represents the probability that the element is contained in all rows of the matrix B, is the number of elements equal to in the ith column of the matrix B, K is the number of natural language interaction corpus, and m is the column number of the label matrix B.

[0061] S8: Calculate the conditional probability matrix PBV of the words in the natural language interaction corpus, wherein PBV is a three-dimensional matrix.

[0062] Specifically, the calculation formula of the conditional probability matrix of the words in the natural language interaction corpus is:

[0063] wherein: PBV ijl is the element of the matrix PBV at the (i, j, l) position, represents the conditional probability of the word w l in the natural language interaction corpus with the annotation parameter , N(V l = 1) represents the number of elements of the lth column of the matrix V equal to 1, L encodes the column number of the matrix V, i = 1, 2, …, m, j = 1, 2, …, n i , and l = 1, 2, …, L.

[0064] S9: Persistently store the prior probability matrix PB and the conditional probability matrix PBV.

[0065] S10: Input a new natural language interaction corpus, encode the natural language interaction corpus using S5, and obtain the encoding vector v' of the to-be-predicted natural language interaction corpus.

[0066] S11: Load the prior probability matrix PB and the conditional probability matrix PBV.

[0067] S12: Calculate the parameter vector

[0068] Specifically, the calculation formula of the posterior probability is:

[0069] According to the Naive Bayes formula to calculate the posterior probability, that is, the derivation process of the calculation formula of

[0070] ​Wherein: represents the parameter selection of the parameter type i when the encoding vector of the natural language interaction corpus is v' , p(v') represents the probability that the encoding vector of the natural language interaction corpus is v'.

[0071] S13: output the parameter vector b'.

[0072] Specifically, the parameter corresponding to the maximum posterior probability in each parameter type is extracted, that is, the parameter vector b' is output.

[0073] In the embodiment, S1-S9 are executed by a training module, and S10-S13 are executed by an extraction module, as shown in FIG. 1.

[0074] The application uses the natural language interaction to extract parameters, reduces the use threshold of the traditional energy storage management system, solves the problem of poor generality of the parameter extraction method, improves the accuracy and expansibility of the parameter extraction method of the traditional energy storage management system through artificial parameter labeling, and improves the robustness of the parameter extraction method of the traditional energy storage management system.

[0075] The embodiment of the application further discloses a computer device.

[0076] The computer device comprises a memory and a processor to store a computer program on the memory and run the computer program on the processor, and the processor executes the computer program to realize the natural language interaction-based energy storage management system parameter extraction method.

[0077] The embodiment of the application further discloses a computer readable storage medium.

[0078] The computer readable storage medium stores a computer program, and the computer program is executed by a processor to realize the natural language interaction-based energy storage management system parameter extraction method.

[0079] The above only describes the preferred embodiments of the application, and is not used to limit the application, and any modification, equivalent replacement, improvement, etc. within the spirit and principle of the application should be included in the protection scope of the application.

Claims

1. A natural language interaction based energy storage management system parameter extraction method, characterized in that, Comprise: S1: parameter type classification and arrangement, according to the energy storage management system usage and working logic to classify the parameter type, the parameter type is abstracted from the actual application scene, according to the functional requirement to enumerate each type parameter, the parameter type classification and arrangement is: project name, container number, cluster number, task name; parameter type set C=(C 1 ,C 2 ,…,C m ), m represents the number of classified types; C i represents the i-th parameter type set after classification, the parameter set of the i-th parameter type set wherein: For the jth parameter of the parameter set C i n i is the number of elements of the parameter set C i i = 1, 2, …, m, j = 1, 2, …, n i ; S2: natural language interaction corpus collection, each natural language interaction corpus corresponds to a label vector b = (b1, b2, …, b m m) containing m parameter values, where b i is the parameter value corresponding to the i-th parameter type of the current sentence; S3: parameter labeling, labeling each parameter type; S4: Construct a word segmentation set, word segmentation set W = (w1, w2, ..., w...). L ), where w l Let l be the l-th element of the word segmentation set W, where l = 1, 2, ..., L; S5: encode the natural language interaction corpus, for each piece of natural language interaction corpus, initialize the encoding vector v with a 0 vector of length L, if w l In the natural language interaction corpus, set v l = 1, otherwise set v l = 0, where v l is the lth element of the encoding vector v; S6: constructing a data set; S7: Compute the prior probability matrix PB of all parameters, PB is m rows A two-dimensional 0 matrix of columns; S8: calculate the conditional probability matrix PBV of the occurrence of each word in the natural language interaction corpus, PBV is A three-dimensional matrix of S9: persistently storing the prior probability matrix PB and the conditional probability matrix PBV; S10: inputting a new natural language interaction corpus to obtain an encoded vector v' of the to-be-predicted natural language interaction corpus; S11: loading the prior probability matrix PB and the conditional probability matrix PBV; S12: Calculate parameter vector S13: outputting a parameter vector b'.

2. The natural language interaction based energy storage management system parameter extraction method of claim 1, wherein: The specific steps of constructing a data set in S6 include: S6.1: initializing an encoded matrix V of natural language interaction corpora and a label matrix B, wherein the encoded matrix V is a 0 matrix of K rows and L columns, each row representing the encoding of a natural language interaction corpus; the label matrix B is a 0 matrix of K rows and m columns, each row representing a parameter value label vector of a natural language interaction corpus; wherein k = 1, 2, …, K, K is the number of natural language interaction corpora; S6.2: encode the kth corpus using the encoding method of the natural language interaction corpus in S5, assign it to the kth row of the encoded matrix V, and use the parameter labeling method in S3 to label the kth natural language interaction corpus and assign it to the kth row of the matrix B.

3. The natural language interaction based energy storage management system parameter extraction method of claim 1, wherein: The process of calculating the prior probability matrix PB of all parameters in S7 is as follows: The prior probability of the jth parameter in the parameter type i is calculated according to the Laplace correction method, and the calculation formula is: where: PB ij is the i-th row, j-th column element of the vector matrix PB, represents the elements contained in all rows of matrix B The probability, for the i-th column of matrix B is equal to The number of elements of 4. The natural language interaction based energy storage management system parameter extraction method of claim 3, wherein: The calculation formula of the conditional probability matrix PBV of the occurrence of each word in the natural language interaction corpus in S8 is specifically as follows: where: PBV ijl is the element of the matrix PBV at position (i,j,l), indicates in the labeling parameter occurrence of a word w in a natural language interaction corpus l the conditional probability of a word w l N(V l L is the number of columns of the encoding matrix V.

5. The natural language interaction based energy storage management system parameter extraction method of claim 4, wherein: In S12, the parameter vector is calculated i = 1, 2,..., m, where The calculation formula of is:

6. The natural language interaction based energy storage management system parameter extraction method of claim 5, wherein: According to the Naive Bayes formula, The derivation process of the calculation formula of wherein: Parameter selection of parameter type i when the encoding vector representing the natural language interaction corpus is v' The probability of 7. The natural language interaction based energy storage management system parameter extraction method of claim 1, wherein: When labeling parameters in S3, manually label the parameter values for each parameter type.

8. The natural language interaction based energy storage management system parameter extraction method of claim 1, wherein: When constructing the word segmentation set in S4, use a word segmenter to segment all natural language interaction corpora, and obtain the word segmentation set W after de-duplication.

9. A computer apparatus comprising: A memory, a processor, a computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the natural language interaction-based energy storage management system parameter extraction method of any one of claims 1-8.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the natural language interaction-based energy storage management system parameter extraction method of any one of claims 1-8.

Images