Method, device and equipment for predicting load of power system

Abstract

The invention provides a power system load prediction method, device and equipment, and the method comprises the steps: determining a multi-input-output table of policy text data based on a policy consistency index model; determining a first matrix by adopting a word frequency-inverse document frequency statistical method according to the multi-input-output table; determining a second matrix by adopting an analytic hierarchy process according to the multi-input-output table; inputting the first matrix and the second matrix into a policy data processing unit for feature extraction; inputting the geographic area data into a geographic area data processing unit; inputting the economic data into an economic data processing unit; and obtaining prediction data of the power system load. According to the method, the policy text data is subjected to quantitative analysis and qualitative analysis, so that the analysis matrix of the policy text data is obtained. And by combining geographic area data and economic data and adopting an advanced neural network architecture, stable and accurate future power system load trend prediction is provided, and the decision-making process of the power industry is assisted.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence, and in particular, to a method, apparatus, and device for predicting the load of a power system. Background Art

[0002] Under the guidance of the concept of green development and the "dual carbon" strategy, the installed capacity of new energy has been increasing year by year, and the traditional power system dominated by fossil energy has developed into a new energy power system dominated by new energy (wind energy, solar energy, etc.).

[0003] However, with the popularization of the new energy power system, relevant policies have great uncertainties. Under the background of the new energy power system facing multiple policy impacts, the law of load change has become more and more complex. The impact of policies is often difficult to quantify, and the specific effects of policy changes may become unclear due to various factors. For example, policy decisions usually involve extensive economic, social, and technological considerations, and the specific effects of these considerations on power demand may not be immediately apparent and may also interact with other influencing factors. In addition, there are uncertainties in the process of policy formulation and implementation. For example, policy changes, implementation efficiency, and the response of the audience will all increase the difficulty of predicting the load of the new energy power system. Therefore, accurately predicting the impact of policies on the load of the power system requires not only in-depth analysis and complex models but also a profound understanding of the motivation behind the policies and their interaction with the power system. Summary of the Invention

[0004] Embodiments of the present disclosure provide a method, apparatus, and device for predicting the load of a power system to solve the problem of difficult prediction of the load of the power system affected by existing policies.

[0005] Based on the above problems, in a first aspect, a method for predicting the load of a power system provided by an embodiment of the present disclosure includes:

[0006] Obtain policy text data, geographical area data, and economic data;

[0007] Perform word frequency statistics on the policy text data, and determine the multi-input-output table of the policy text data based on the policy consistency index model;

[0008] Determine a first matrix according to the multi-input-output table by using the term frequency-inverse document frequency statistical method;

[0009] Determine a second matrix according to the multi-input-output table by using the analytic hierarchy process;

[0010] Construct a prediction model for the load of the power system, where the prediction model for the load of the power system includes: a policy data processing unit, a geographical area data processing unit, an economic data processing unit, and a power system load prediction unit;

[0011] Input the first matrix and the second matrix into the policy data processing unit to extract first feature data; input the geographical area data into the geographical area data processing unit to obtain second feature data; input the economic data into the economic data processing unit to obtain third feature data;

[0012] Input the first feature data, the second feature data and the third feature data into the power system load forecasting unit to obtain the forecasting data of the power system load.

[0013] Combined with the first aspect, in a possible implementation manner, the word frequency statistics of the policy text data and the determination of the multi-input-output table of the policy text data based on the policy consistency index model include:

[0014] Perform word frequency statistics on the policy text data, and determine a preset number of first-level variables in the policy text data based on the policy consistency index model;

[0015] Classify according to the first-level variables and their keywords to determine the second-level variables after classifying the first-level variables;

[0016] Determine the multi-input-output table of the policy text data according to the first-level variables and the second-level variables.

[0017] Combined with the first aspect, in a possible implementation manner, the determination of the first matrix by using the term frequency-inverse document frequency statistical method according to the multi-input-output table includes:

[0018] Determine the keyword dictionary corresponding to each second-level variable in the multi-input-output table;

[0019] Determine the value of each second-level variable by using the term frequency-inverse document frequency statistical method according to the keyword dictionary and the policy text data;

[0020] Determine the first matrix according to the multi-input-output table and the value of each second-level variable.

[0021] Combined with the first aspect, in a possible implementation manner, the determination of the second matrix by using the analytic hierarchy process according to the multi-input-output table includes:

[0022] Determine the first pairwise comparison matrix by using the analytic hierarchy process according to the multi-input-output table;

[0023] Determine the weights of the first-level variables and the weights of the second-level variables according to the first pairwise comparison matrix;

[0024] Determine the probability value of each second-level variable;

[0025] Determine a second matrix according to the weights of the first-level variables, the weights of the second-level variables, and the probability values of the second-level variables;

[0026] Among them, the probability values of the second-level variables are determined by using a latent Dirichlet allocation model or by using a term frequency-inverse document frequency statistical method.

[0027] Combined with the first aspect, in a possible implementation manner, the policy data processing unit includes: a first input sub-layer, a residual network sub-layer, and a first max pooling layer;

[0028] The step of inputting the first matrix and the second matrix into the policy data processing unit for feature extraction to obtain first feature data includes:

[0029] Determine the input data of the first input sub-layer according to the first matrix and the second matrix;

[0030] Perform feature extraction on the input data of the first input sub-layer in sequence by using the residual network sub-layer and the first max pooling layer to obtain first feature data.

[0031] Combined with the first aspect, in a possible implementation manner, the geographical area data processing unit includes: a second input sub-layer, a convolutional layer, and a second max pooling layer;

[0032] The step of inputting the geographical area data into the geographical area data processing unit to obtain second feature data includes:

[0033] Determine the input data of the second input sub-layer according to the geographical area data;

[0034] Perform feature extraction on the input data of the second input sub-layer in sequence by using the convolutional layer and the second max pooling layer to obtain second feature data;

[0035] Among them, the geographical area data includes at least one of the following: climate data, environmental data, and industrial data.

[0036] Combined with the first aspect, in a possible implementation manner, the economic data processing unit includes: a third input sub-layer, a long short-term memory network model sub-layer, and a third max pooling layer;

[0037] The step of inputting the economic data into the economic data processing unit to obtain third feature data includes:

[0038] Determine the input data of the third input sub-layer according to the economic data;

[0039] Feature extraction is performed on the input data of the third input sublayer in sequence using the long short-term memory network model sublayer and the third max pooling layer to obtain third feature data;

[0040] Wherein, the economic data at least includes one of the following: economic trend data and historical load data of the power system.

[0041] Combined with the first aspect, in a possible implementation manner, the power system load prediction unit includes: a splicing layer, a fully connected layer, and an output layer;

[0042] Inputting the first feature data, the second feature data, and the third feature data into the power system load prediction unit to obtain prediction data of the power system load includes:

[0043] Inputting the first feature data, the second feature data, and the third feature data into the splicing layer for splicing to obtain fourth feature data;

[0044] Inputting the fourth feature data into the fully connected layer and the output layer in sequence to obtain prediction data of the power system load.

[0045] In a second aspect, a power system load prediction device is provided, including:

[0046] A data acquisition module, configured to acquire policy text data, geographical area data, and economic data;

[0047] A policy text data processing module, configured to perform word frequency statistics on the policy text data, determine a multi-input-output table of the policy text data based on a policy consistency index model; determine a first matrix according to the multi-input-output table using the term frequency-inverse document frequency statistical method; determine a second matrix according to the multi-input-output table using the analytic hierarchy process;

[0048] A power system load prediction module, configured to construct a prediction model for the power system load, where the prediction model for the power system load includes: a policy data processing unit, a geographical area data processing unit, an economic data processing unit, and a power system load prediction unit; input the first matrix and the second matrix into the policy data processing unit for feature extraction to obtain first feature data; input the geographical area data into the geographical area data processing unit to obtain second feature data; input the economic data into the economic data processing unit to obtain third feature data; input the first feature data, the second feature data, and the third feature data into the power system load prediction unit to obtain prediction data of the power system load.

[0049] In a third aspect, a prediction device for power system load is provided, including: the prediction device for power system load as described in the second aspect.

[0050] The beneficial effects of the embodiments of the present disclosure include:

[0051] A prediction method, device and equipment for power system load provided by the present disclosure include: obtaining policy text data, geographical area data and economic data; performing word frequency statistics on the policy text data, and determining the multi-input-output table of the policy text data based on the policy consistency index model; determining the first matrix according to the multi-input-output table by using the term frequency-inverse document frequency statistical method; determining the second matrix according to the multi-input-output table by using the analytic hierarchy process; constructing a prediction model for power system load, where the prediction model for power system load includes: a policy data processing unit, a geographical area data processing unit, an economic data processing unit and a power system load prediction unit; inputting the first matrix and the second matrix into the policy data processing unit for feature extraction to obtain first feature data; inputting the geographical area data into the geographical area data processing unit to obtain second feature data; inputting the economic data into the economic data processing unit to obtain third feature data; inputting the first feature data, the second feature data and the third feature data into the power system load prediction unit to obtain the prediction data of the power system load. The prediction method for power system load provided by the embodiments of the present disclosure quantitatively analyzes the policy text data by using the policy consistency index model, and qualitatively analyzes the policy text data by using the analytic hierarchy process. Thus, an analysis matrix of the policy text data is obtained, and in combination with the geographical area data and the economic data, by using an advanced neural network architecture, a robust and accurate prediction of the future power system load trend is provided, which helps the decision-making process of the power industry. Description of the Drawings

[0052] Figure 1 is a flowchart of the prediction method for power system load provided by the embodiments of the present disclosure;

[0053] Figure 2 is a schematic structural diagram of the prediction model for power system load provided by the embodiments of the present disclosure;

[0054] Figure 3 is a structural diagram of the prediction device for power system load provided by the embodiments of the present disclosure. Detailed Embodiments

[0055] The embodiments of the present disclosure provide a prediction method, device and equipment for power system load. The following describes the preferred embodiments of the present disclosure with reference to the accompanying drawings of the specification. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present disclosure, and are not used to limit the present disclosure. And without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0056] An embodiment of the present disclosure provides a method for predicting the load of a power system, as Figure 1 shown, including the following steps:

[0057] S101. Obtain policy text data, geographical area data, and economic data;

[0058] S102. Perform word frequency statistics on the policy text data, and determine the multi-input-output table of the policy text data based on the policy consistency index model;

[0059] S103. Determine the first matrix according to the multi-input-output table by using the term frequency-inverse document frequency statistical method;

[0060] S104. Determine the second matrix according to the multi-input-output table by using the analytic hierarchy process;

[0061] S105. Construct a prediction model for the load of the power system. The prediction model for the load of the power system includes: a policy data processing unit, a geographical area data processing unit, an economic data processing unit, and a power system load prediction unit;

[0062] S106. Input the first matrix and the second matrix into the policy data processing unit for feature extraction to obtain first feature data; input the geographical area data into the geographical area data processing unit to obtain second feature data; input the economic data into the economic data processing unit to obtain third feature data;

[0063] S107. Input the first feature data, the second feature data, and the third feature data into the power system load prediction unit to obtain the prediction data of the load of the power system.

[0064] In the embodiments of the present disclosure, the prediction of the power system load may refer to the process of estimating and predicting the load demand of the power system or the energy system over a period of time in the future. Its main objective is to analyze the variation law of the load, predict the power load level in a certain period in the future, so as to serve the energy management of power companies and grid operators, reasonably arrange the power generation plan, dispatch the power generation equipment and optimize the energy distribution, which helps to avoid the problems of power system overload or deficiency, and improve the stability and reliability of the power system. Under the guidance of the concept of green development and the "dual carbon" strategy, the installed capacity of new energy has been increasing year by year, and the power system has developed from a traditional power system dominated by fossil energy to a new energy power system dominated by new energy (wind energy, solar energy, etc.). However, affected by climate and environmental factors, new energy has intermittency, randomness and volatility. In addition, under the guidance of policies, a large number of distributed power sources and energy storage devices are connected to the grid, and the factors affecting the load increase sharply, and the load characteristics tend to be complicated, resulting in more difficult prediction of the load change of the power system, increasing the difficulty and error of load prediction. Therefore, accurate load prediction is more necessary in the new energy power system. In addition, the large-scale consumption of new energy requires the power system to have sufficient flexible regulation ability. However, the energy structure and operation mode of the new energy power system are different from those of the traditional power system. With the replacement of traditional power sources, the controllability and regulation ability of the grid load have decreased to a certain extent, and the challenges of peak supply and valley consumption are more significant. Policies have great uncertainties, including many factors, and the relationships between these factors and their impact mechanisms on power loads are complex. At the same time, in the context of power loads being affected by multiple policies, the law of load change has become more and more complex. The impact of the same policy on the relevant loads in different industries, regions and times is also different. For example, the time-of-use electricity price policy can reduce the peak-valley difference rate of the load for most industries. However, the proportion of electricity cost, electricity consumption characteristics and profit margin in different industries determine the implementation effect of the time-of-use electricity price policy. Among them, industries such as wine, beverage and refined tea manufacturing (the peak-valley difference rate is reduced by 3.27%) and automobile manufacturing (the peak-valley difference rate is reduced by 2.82%) have more significant effects, while industries such as software and information technology services (the peak-valley difference rate is reduced by -4.21%) and wind energy prime mover manufacturing (the peak-valley difference rate is reduced by -2.80%) do not have a significant effect on reducing the peak-valley difference rate of the load.

[0065] Predicting the impact of policies on the load of new energy power systems is an extremely complex task. The impact of policies is often difficult to quantify, and the specific effects of policy changes may become unclear due to various factors. Policy decisions usually involve extensive economic, social, and technical considerations, and the specific impacts of these considerations on electricity demand may not be immediately apparent and may also interact with other influencing factors. In addition, uncertainties in the policy formulation and implementation processes, such as policy changes, implementation efficiency, and the responses of the audience, will increase the difficulty of prediction. Therefore, accurately predicting the impact of policies on electricity load requires not only in-depth analysis and complex models but also a profound understanding of the motivations behind the policies and their interactions with the power system.

[0066] In the embodiments of the present disclosure, first, preprocess the policy text, including removing blank characters and punctuation marks, which can eliminate the noise in the text and thus improve the accuracy of word segmentation and word frequency analysis. Since Chinese texts lack clear word boundaries, word segmentation processing is required, and high-frequency stop words that do not contribute to semantic analysis are removed to obtain the preprocessed policy text data. Then, a word frequency analysis tool can be used to perform word frequency statistics on the policy text data. The Policy Modeling Consistency Index Model (PMC) is a quantitative method for policy evaluation and analysis. To perform quantitative analysis on the policy text data based on the PMC index model, first, according to the extraction results of high-frequency words in the word frequency statistics, determine the multi-input-output table of the policy text data. For example, as shown in Table 1, the relevant variables of the multi-input-output table are listed.

[0067]

[0068]

[0069] Table 1

[0070] The Term Frequency-Inverse Document Frequency (TF-IDF) statistical method can be used to evaluate the importance of a word for a document set or a single document in a corpus. According to the multi-input-output table, a PMC index matrix can be quantified, and the Term Frequency-Inverse Document Frequency statistical method can be used to assign values to the variables of the PMC index matrix. And determine the PMC index matrix as the first matrix, thereby obtaining the quantitative analysis of the policy text data.

[0071] To conduct qualitative analysis on policy text data and ensure objectivity and accuracy, the selection of evaluation indicators follows four key principles: comprehensiveness, operability, importance, and relevance. To scientifically evaluate the impact of policies on the load of the new energy power system, it is necessary to comprehensively consider the policy target group, ultimate goal, policy tools, and the degree of relevance to the power system. Based on the analysis of policy text data, a multi-level theme framework is constructed according to the multi-input-output table, as shown in Table 2, to more accurately evaluate the degree of policy impact on the load and cover a wider evaluation scope. The criteria and sub-criteria in Table 2 can serve as the first-level variables and second-level variables in the multi-input-output table, or the multi-input-output table can be adjusted according to the policy text data and then constructed.

[0072]

[0073] Table 2

[0074] The Analytic Hierarchy Process (AHP) is a method for multi-criteria decision analysis. By decomposing complex decision problems into multiple levels and factors and using pairwise comparison methods to determine the weights of each factor, it helps decision-makers make scientific decisions. Using the AHP to analyze the criteria and sub-criteria in Table 2 above and determine their weight values can achieve qualitative analysis of policy text data. The analysis results are determined as the second matrix.

[0075] Furthermore, a prediction model for the power system load is constructed, which includes: a policy data processing unit, a geographical area data processing unit, an economic data processing unit, and a power system load prediction unit. The first matrix and the second matrix are input into the policy data processing unit for predicting the power system load demand. Combining geographical area data and economic data, they are respectively input into the geographical area data processing unit and the economic data processing unit for feature extraction. The obtained first feature extraction data, second feature extraction data, and third feature extraction data are input into the power system load prediction unit to obtain the predicted data of the power system load, thus ensuring the integrity of the prediction task.

[0076] In the embodiments of this application, the policy consistency index model and the Analytic Hierarchy Process are respectively used to conduct quantitative analysis and qualitative analysis on policy text data, thus providing a comprehensive quantitative analysis result for the impact of policies on the power system load. The obtained analysis matrix is combined with geographical area data and economic data and jointly input into the prediction model of the power system load, thus providing a robust and accurate prediction of future load trends to assist the decision-making process in the power industry.

[0077] In another embodiment of the present disclosure, in the above step S102, for the policy text data, word frequency statistics are performed, and based on the policy consistency index model, a multi-input-output table of the policy text data is determined, including the following steps:

[0078] Step 1: Perform word frequency statistics on the policy text data, and based on the policy consistency index model, determine a preset number of first-level variables in the policy text data;

[0079] Step 2: Classify according to the first-level variables and their keywords, and determine the second-level variables after classifying the first-level variables;

[0080] Step 3: Determine the multi-input-output table of the policy text data according to the first-level variables and the second-level variables.

[0081] In the embodiment of the present disclosure, quantitative analysis is performed on the policy text data based on the policy consistency model. For the above step 1, a word frequency statistics tool can be used to perform word frequency statistics on the policy text data, and in combination with the extraction results of high-frequency words, a preset number of first-level variables are determined. Exemplarily, as shown in Table 1 above, nine first-level variables are defined. For the above step 2, each first-level variable is further refined into specific second-level variables, and a quantifiable standard is assigned to each variable. For the above step 3, a multi-input-output table is constructed according to the first-level variables and the second-level variables. As shown in Table 1, the multi-input-output table lists the variables based on the policy consistency index model. The multi-input-output table can display the set of variables for evaluating policies related to the power industry. Based on the characteristics of the power industry and in combination with the results of the thematic analysis of policies related to the power industry, several first-level variables including policy type, policy direction, policy scope, etc. are selected, and each first-level variable is further refined into specific multiple second-level variables. A quantification standard is designed for each second-level variable. For example, the implementation scope of the policy includes national, provincial, municipal, county-level, etc. The classification of each first-level variable is refined into specific classifications of second-level variables, and each classification of second-level variables has corresponding keywords, aiming to comprehensively cover all aspects involved in power policy modeling.

[0082] In another embodiment of the present disclosure, in the above step S103, according to the multi-input-output table, the word frequency-inverse document frequency statistical method is used to determine the first matrix, including the following steps:

[0083] Step 1: Determine the keyword dictionary corresponding to each second-level variable in the multi-input-output table;

[0084] Step 2: According to the keyword dictionary and the policy text data, use the word frequency-inverse document frequency statistical method to determine the value of each second-level variable;

[0085] Step 3: Determine the first matrix according to the multi-input-output table and the value of each second-level variable.

[0086] In the embodiments of the present disclosure, the term frequency-inverse document frequency statistical method can ensure the accuracy of assigning values to each variable. The term frequency-inverse document frequency statistical method (TF-IDF, Term Frequency-Inverse Document Frequency) can be used to evaluate the importance of a word for a document set or a single document in a corpus. The TF-IDF value increases in direct proportion to the frequency of the word in the document, but at the same time decreases in inverse proportion to the frequency of the word in the corpus. This means that TF-IDF tends to filter out common words and retain important words. The term frequency (TF, Term Frequency) can refer to the number of times a word appears in a document. If a word appears more times in a document, then its term frequency is higher. The inverse document frequency (IDF, Inverse Document Frequency) can refer to the reciprocal of the frequency of a word in the corpus. If a word appears more frequently in the corpus, then its IDF is lower. As shown in Table 1 above, the policy text data can be quantified as the PMC index matrix M, where each element M ij represents the assignment of the j-th secondary variable X i under the i-th primary variable X ij . For step 1 above, for each secondary variable V j , first compile a dictionary of relevant keywords {k j1 , k j2 , …, k jn}. These keywords reflect specific topics or elements related to V j .

[0087] For step 2 above, for a policy text data d, calculate the term frequency-inverse document frequency value of each keyword. The calculation formula for the TF-IDF score of keyword k in the policy text data d is as follows:

[0088] TF-IDF(k, d) = TF(k, d) × IDF(k)

[0089] where TF(k, d) represents the relative term frequency of keyword k in the policy text data d, and its calculation formula is the number of times f t，d that keyword k appears in the policy text data d divided by the total number of all terms in the policy text data d. The calculation formula is as follows:

[0090]

[0091] IDF(k) represents the inverse document frequency of keyword k, and its calculation formula is as follows:

[0092]

[0093] Where N represents the total number of policy text data in corpus D, and |d ∈ D: k ∈ d| is the number of policy text data including keyword k. For the value related to secondary variable V in the given policy text data d, it is determined by the average TF-IDF value of all keywords related to V j related, and is determined by the average TF-IDF value of all keywords related to V j related, and the calculation formula is:

[0094]

[0095] Where n is the total number of keywords related to V j related, and TF-IDF(k ji , d) is the TF-IDF value of keyword k ji in the policy text data d. For step 3 above, according to the multi-input-output table, it is quantified into the PMC index matrix M and the value of each determined secondary variable, and the PMC index matrix M is determined as the first matrix. The value of each element in the first matrix is calculated through the comprehensive analysis of the policy text data, so as to realize the quantitative analysis of the policy text data.

[0096] In another embodiment of the present disclosure, in the above step S104, according to the multi-input-output table, the analytic hierarchy process is used to determine the second matrix, including the following steps:

[0097] Step 1: Determine the first pairwise comparison matrix according to the multi-input-output table using the analytic hierarchy process;

[0098] Step 2: Determine the weights of the primary variables and the weights of the secondary variables according to the first pairwise comparison matrix;

[0099] Step 3: Determine the probability value of each secondary variable;

[0100] Step 4: Determine the second matrix according to the weights of the primary variables, the weights of the secondary variables, and the probability values of the secondary variables;

[0101] Among them, the probability value of the secondary variable is determined by using the latent Dirichlet allocation model or the term frequency-inverse document frequency statistical method.

[0102] In the embodiments of the present disclosure, the analytic hierarchy process is used to qualitatively analyze policy text data. To ensure objectivity and accuracy, the selection of evaluation indicators follows four key principles: comprehensiveness, operability, importance, and relevance. To scientifically evaluate the impact of policies on the load of the new energy power system, it is necessary to comprehensively consider the policy target group, the ultimate goal, policy tools, and the degree of relevance to the power system. For the above step 1, a multi-level theme framework is constructed based on the multi-input-output table, and a corresponding hierarchical structure is established, thereby constructing a policy evaluation index system table. As shown in Table 2, the criteria and sub-criteria in this policy evaluation index system table can be the first-level variables and second-level variables in the multi-input-output table, or can be constructed after adaptively adjusting the multi-input-output table according to the policy text data, where the first-level variables correspond to the criteria in Table 2, and the second-level variables correspond to the sub-criteria in Table 2. The analytic hierarchy process (AHP, Analytic Hierarchy Process) can be used to qualitatively analyze this policy evaluation index system table. The analytic hierarchy process can construct a pairwise comparison matrix for each layer of the hierarchical structure to evaluate the effectiveness of the policy in reducing the power load. Exemplarily, let the indicators in Table 2 be denoted as D(D i can represent Figure 2 each criterion B i or each sub-criterion C corresponding to the criterion i ), and the pairwise comparison matrix is A. Then, in the pairwise comparison matrix A, a ij represents the relative importance of D i relative to D j . The pairwise comparison matrix A is shown as follows:

[0103]

[0104] Among them, the diagonal element a ii is equal to 1, indicating that any indicator has the same importance when compared with itself. The non-diagonal element a ij (where i≠j) value is based on the judgment results of experts in the relevant field and can be marked using the Saaty 1-9 scale method. In the analytic hierarchy process, the Saaty 1-9 scale method is used to construct the pairwise comparison matrix. By comparing the importance of each indicator pairwise, the weights of each indicator are calculated. This method can effectively reduce the difficulty of comparison between different indicators and improve the accuracy and consistency of decision-making. The pairwise comparison matrix A satisfies the reciprocal property, that is, a ji = 1 / a ij, to ensure the consistency of the comparison. The constructed pairwise comparison matrix A can be applied to the main criteria for evaluating the main criteria that affect policy effectiveness, or to the sub-criteria under each main criterion for evaluating the further refined sub-criteria, thereby enhancing the granularity of the evaluation. The pairwise comparison matrix constructed for the main criteria and sub-criteria is determined as the first pairwise comparison matrix. For step 2 above, according to the first pairwise comparison matrix, the weight vector W can be calculated using the geometric mean method, and the formula is expressed as follows:

[0105]

[0106] To ensure the reliability of the judgments in the pairwise comparison matrix, a consistency test is required. The purpose of this test is to verify whether the comparisons made are logically consistent, thereby verifying the validity of the results obtained through AHP. For the maximum eigenvalue λ of the pairwise comparison matrix A max The calculation formula is as follows:

[0107]

[0108] The calculation formula for the consistency ratio (CR) is as follows:

[0109]

[0110] Among them, the calculation formula for the consistency index (CI) is as follows:

[0111]

[0112] When the CR value is less than or equal to 0.1, it indicates that the pairwise comparison is consistent. When the CR value is greater than 0.1, it indicates that the pairwise comparison is not consistent, and the judgment needs to be modified to improve the consistency. The weight of the determined first-level variables, that is, the main criterion weight, can be expressed as W Criterion . The weight of the determined second-level variables, that is, the sub-criterion weight, can be expressed as W Sub-Criterion . For step 3 above, determine the probability value of each second-level variable. The probability value of each second-level variable can represent the correlation weight of any policy text data on each sub-criterion (Sub-Criterion), and the obtained probability value of the second-level variable can be expressed as W PolicyThe probability value of each secondary variable can be determined using the Latent Dirichlet Allocation (LDA) model. Latent Dirichlet Allocation (LDA) is a generative topic model based on Bayesian statistics. First, the number of sub-criteria is corresponded to the number of topics in the LDA model. For example, as shown in Table 2, if the number of sub-criteria is 13 (C11, C12, C13, C21, C22, etc.), when using the LDA model, the number of topics is determined to be 13 (or equivalent to the number of sub-criteria), so that each sub-criterion approximately corresponds to a topic or has a close semantics. For each policy text data, the LDA model can output a list of topic distributions, which shows the proportion of each topic in the policy text data, and the sum of these proportions is 1, thus obtaining the relevance weight W of the policy text data for this sub-criterion. Policy The probability value of each secondary variable can also be determined using the Term Frequency-Inverse Document Frequency (TF-IDF) statistical method. First, a set of discriminative keywords is set for each sub-criterion to represent the core meaning of the sub-criterion. For example, the keywords for enterprises (C11) can include: enterprise, legal person, industrial sector, industry, etc., and the keywords for residents (C12) can include: resident, family, individual, community, etc. For any sub-criterion C i the set of keywords can be represented as {k 1 ,k 2 ,…,k m}. The sum or weighted average of the TF-IDF scores of the keywords in the policy text data is calculated to obtain the total score of the sub-criterion. The calculation formula is:

[0113]

[0114] where TF-IDF(t, d) = TF(t, d) × IDF(t), TF(t, d) represents the number of times (or normalized frequency) that the word t appears in the policy text data d, and IDF(t) is related to the occurrence of the word t in the corpus composed of multiple policy text data d. On this basis, normalization is performed so that the total scores of all sub-criteria are 1, and the relevance weight W of each policy text data on different sub-criteria is obtained. Policy The calculation formula is:

[0115] For step 4 above, according to the weight W Criterion of the primary variable, the weight W Sub-Criterion of the secondary variable, and the probability value W Policy of the secondary variable, the second matrix is determined. The calculation formula is as follows:

[0116] E = Σ(W Criterion × W Sub-Criterion × W Policy )

[0117] By adopting the analytic hierarchy process, qualitative analysis of policy text data can be realized. The policy effectiveness evaluation indicators and their weights can provide a robust evaluation framework for evaluating the effectiveness of policies on different topics.

[0118] In another embodiment of the present disclosure, the policy data processing unit includes: a first input sublayer, a residual network sublayer, and a first max pooling layer;

[0119] In the above step S106, inputting the first matrix and the second matrix into the policy data processing unit for feature extraction to obtain the first feature data includes the following steps:

[0120] Step 1: Determine the input data of the first input sublayer according to the first matrix and the second matrix;

[0121] Step 2: Sequentially perform feature extraction on the input data of the first input sublayer by using the residual network sublayer and the first max pooling layer to obtain the first feature data.

[0122] In the embodiment of the present disclosure, an advanced neural network model is used to perform feature extraction on the first matrix and the second matrix. As Figure 2 shown, for the above step 1, the first matrix and the second matrix can form the structured policy data 201 and be input into the first input sublayer 202. The structured policy data 201 comes from the analysis of policy text data and integrates the quantification results of various policy impacts. For the above step 2, the input data of the first input sublayer 202 is sequentially input into the residual network sublayer 203 and the first max pooling layer 204. The residual network sublayer 203 can provide deep feature extraction capabilities and residual connection techniques to solve the problem of gradient disappearance, and then learn the complex relationships in policy impacts, such as the interactions between regulatory frameworks, economic measures, and environmental initiatives. This ensures that the complex and abstract features of policy text data can be effectively captured and retained for downstream tasks in the prediction model of power system load. The residual network sublayer 203 can include a preset number of residual network models to sequentially optimize the features extracted in the previous residual network model to ensure that important information can be retained and enhanced. The first max pooling layer 204 can achieve dimension reduction and reduction of computational complexity. This layer reduces the size of the feature map, thereby improving the processing efficiency and retaining the most important features. By performing feature extraction on the first matrix and the second matrix through an advanced neural network model, the important features of the analysis of policy text data are retained.

[0123] In another embodiment of the present disclosure, the geographical area data processing unit includes: a second input sublayer, a convolutional layer, and a second max pooling layer;

[0124] In the above step S106, inputting the geographical area data into the geographical area data processing unit to obtain the second feature data includes the following steps:

[0125] Step 1: Determine the input data of the second input sublayer according to the geographical area data;

[0126] Step 2: Perform feature extraction on the input data of the second input sublayer in sequence using the convolutional layer and the second max pooling layer to obtain the second feature data;

[0127] Among them, the geographical area data includes at least one of the following: climate data, environmental data, and industrial data.

[0128] In the embodiments of the present disclosure, an advanced neural network model is used to extract features from geographical area data. Geographical area data is one of the key factors affecting the load of the power system. Among them, the geographical area data includes at least one of the following: climate data, environmental data, and industrial data. As Figure 2 shown, for the above step 1, the geographical area data 205 is input into the second input sublayer 206. For the above step 2, the input data of the second input sublayer 206 is sequentially input into the convolutional layer 207 and the second max pooling layer 208. Through the convolutional layer 207, the features of the geographical area data 205 can be extracted to obtain spatial and local patterns. The convolutional layer 207 can efficiently capture the short-term dependencies and interaction relationships among multiple variables, such as temperature, humidity, and industrial data, and provides the ability to generalize from local features, which can be used to model the short-term impact of external factors on power load. The convolutional layer 207 may include a preset number of sub-convolutional layers. For example, the convolutional layer 207 includes a first convolutional layer and a second convolutional layer. The first convolutional layer can capture basic patterns and features from the geographical area data 205. These features are further optimized by the second convolutional layer, thereby enhancing the ability of the prediction model of the power system load to detect complex patterns and relationships in the data. The second max pooling layer 208 can achieve dimensionality reduction and reduction of computational complexity. This layer reduces the size of the feature map, thereby improving the processing efficiency and retaining the most important features. By using an advanced neural network model to extract features from geographical area data, complex patterns and relationships in the geographical area data can be detected.

[0129] In another embodiment of the present disclosure, the economic data processing unit includes: a third input sublayer, a long short-term memory network model sublayer, and a third max pooling layer;

[0130] In the above step S106, inputting the economic data into the economic data processing unit to obtain the third feature data includes the following steps:

[0131] Step 1: Determine the input data of the third input sub-layer according to economic data;

[0132] Step 2: Sequentially use the long short-term memory network model sub-layer and the third max pooling layer to extract features from the input data of the third input sub-layer to obtain third feature data;

[0133] Among them, the economic data includes at least one of the following: economic trend data and historical load data of the power system.

[0134] In the embodiments of the present disclosure, an advanced neural network model is used to extract features from economic data. Economic data is one of the key factors affecting the load of the power system. Among them, the economic data includes at least one of the following: economic trend data and historical load data of the power system, providing background information for the current and future load trends. As Figure 2 shown, for the above-mentioned Step 1, the economic data 209 is input into the third input sub-layer 210. For the above-mentioned Step 2, the input data of the third input sub-layer 210 is sequentially input into the long short-term memory network model sub-layer 211 and the third max pooling layer 212. Through the long short-term memory network model sub-layer 211, the long-term dependencies and time dynamics in the economic data 209 can be captured. Through the gating mechanism, the long short-term memory network model sub-layer 211 can retain relevant information across time steps, so as to understand seasonal trends, daily changes, and other periodic patterns, ensuring that historical load patterns are effectively integrated into the prediction process and improving the overall prediction accuracy. The long short-term memory network model sub-layer 211 may include a preset number of long short-term memory network models. For example, the long short-term memory network model sub-layer 211 includes a first long short-term memory network model and a second long short-term memory network model. The first long short-term memory network model can process the economic data 209 and learn the time dynamics of load demand. The second long short-term memory network model further optimizes these time features, improving the ability to make accurate predictions based on historical trends. The third max pooling layer 212 can achieve dimensionality reduction and reduction of computational complexity. This layer reduces the size of the feature map, thereby improving the processing efficiency and retaining the most important features. Extracting features from economic data through an advanced neural network model can improve the ability to make accurate predictions based on historical trends.

[0135] In another embodiment of the present disclosure, the power system load prediction unit includes: a splicing layer, a fully connected layer, and an output layer;

[0136] In the above-mentioned step S107, the first feature data, the second data, and the third feature data are input into the power system load prediction unit to obtain the predicted data of the power system load, the following steps:

[0137] Step 1: Input the first feature data, the second feature data, and the third feature data into a splicing layer for splicing to obtain the fourth feature data;

[0138] Step 2: Input the fourth feature data into a fully connected layer and an output layer in sequence to obtain the predicted data of the power system load.

[0139] In the embodiments of the present disclosure, by integrating multiple data sources and an advanced neural network architecture, the prediction model of the power system load can capture various characteristics of the load demand. As Figure 2 shown, for the above Step 1, the first feature data, the second feature data, and the third feature data are input into the splicing layer 213 for splicing to obtain the fourth feature data. The splicing layer 213 can integrate information from different layers, combine the features extracted from the residual network sublayer 203, the convolutional layer 207, and the long short-term memory network model sublayer 211, generate a unified data representation, capture spatial and temporal patterns, and then obtain the fourth feature data. For the above Step 2, the fourth feature data is input into the fully connected layer 214 and the output layer 215 in sequence to obtain the predicted data of the power system load. The fully connected layer 214 may include a preset number of sub-fully connected layers. Through these sub-fully connected layers, high-level abstraction is performed to learn and extract complex relationships between features, so as to achieve more accurate load prediction. The output layer 215 can be a fully connected layer with a linear activation function, enabling the prediction model of the power system load to generate a continuous scalar value representing the predicted load. By mapping the learned high-level feature representation to a single output, the output layer 215 can effectively model the relationship between the input variables and the target value, thereby obtaining the predicted data of the power system load. By combining policy text data, geographical area data, and economic data, the prediction model of the power system load can provide a robust and accurate prediction of future load trends, assisting the decision-making process in the power industry.

[0140] Based on the same inventive concept, the embodiments of the present disclosure also provide a prediction device and equipment for the power system load. Since the principles of the problems solved by these devices and equipment are similar to those of the aforementioned prediction method for the power system load, the implementation of these devices and equipment can refer to the implementation of the aforementioned method, and the repeated parts will not be elaborated.

[0141] The embodiments of the present disclosure provide a prediction device for the power system load, as Figure 3 shown, including:

[0142] A data acquisition module 301, configured to acquire policy text data, geographical area data, and economic data;

[0143] The policy text data processing module 302 is used to perform word frequency statistics on the policy text data, determine the multi-input-output table of the policy text data based on the policy consistency index model; use the term frequency-inverse document frequency statistical method according to the multi-input-output table to determine the first matrix; use the analytic hierarchy process according to the multi-input-output table to determine the second matrix.

[0144] The power system load forecasting module 303 is used to construct a forecasting model for the power system load. The forecasting model for the power system load includes: a policy data processing unit, a geographical area data processing unit, an economic data processing unit, and a power system load forecasting unit; input the first matrix and the second matrix into the policy data processing unit for feature extraction to obtain first feature data; input the geographical area data into the geographical area data processing unit to obtain second feature data; input the economic data into the economic data processing unit to obtain third feature data; input the first feature data, the second feature data, and the third feature data into the power system load forecasting unit to obtain the forecasting data of the power system load.

[0145] In another embodiment of the present disclosure, the policy text data processing module 302 is used to perform word frequency statistics on the policy text data and determine a preset number of first-level variables in the policy text data based on the policy consistency index model.

[0146] Classify according to the first-level variables and their keywords to determine the second-level variables after classifying the first-level variables.

[0147] Determine the multi-input-output table of the policy text data according to the first-level variables and the second-level variables.

[0148] In another embodiment of the present disclosure, the policy text data processing module 302 is used to determine the keyword dictionary corresponding to each second-level variable in the multi-input-output table.

[0149] Use the term frequency-inverse document frequency statistical method according to the keyword dictionary and the policy text data to determine the value of each second-level variable.

[0150] Determine the first matrix according to the multi-input-output table and the value of each second-level variable.

[0151] In another embodiment of the present disclosure, the policy text data processing module 302 is used to determine the first pairwise comparison matrix according to the multi-input-output table by using the analytic hierarchy process.

[0152] Determine the weights of the first-level variables and the weights of the second-level variables according to the first pairwise comparison matrix.

[0153] Determine the probability value of each of the secondary variables;

[0154] Determine the second matrix according to the weight of the primary variable, the weight of the secondary variable, and the probability value of the secondary variable;

[0155] Among them, the probability value of the secondary variable is determined by using a latent Dirichlet allocation model or by using a term frequency-inverse document frequency statistical method.

[0156] In another embodiment of the present disclosure, the policy data processing unit includes: a first input sublayer, a residual network sublayer, and a first max pooling layer;

[0157] The power system load forecasting module 303 is configured to determine the input data of the first input sublayer according to the first matrix and the second matrix;

[0158] Feature extraction is sequentially performed on the input data of the first input sublayer by using the residual network sublayer and the first max pooling layer to obtain first feature data.

[0159] In another embodiment of the present disclosure, the geographical area data processing unit includes: a second input sublayer, a convolutional layer, and a second max pooling layer;

[0160] The power system load forecasting module 303 is configured to determine the input data of the second input sublayer according to the geographical area data;

[0161] Feature extraction is sequentially performed on the input data of the second input sublayer by using the convolutional layer and the second max pooling layer to obtain second feature data;

[0162] Among them, the geographical area data includes at least one of the following: climate data, environmental data, and industrial data.

[0163] In another embodiment of the present disclosure, the economic data processing unit includes: a third input sublayer, a long short-term memory network model sublayer, and a third max pooling layer;

[0164] The power system load forecasting module 303 is configured to determine the input data of the third input sublayer according to the economic data;

[0165] Feature extraction is sequentially performed on the input data of the third input sublayer by using the long short-term memory network model sublayer and the third max pooling layer to obtain third feature data;

[0166] Among them, the economic data includes at least one of the following: economic trend data and historical load data of the power system.

[0167] In another embodiment of the present disclosure, the power system load prediction unit includes: a splicing layer, a fully connected layer, and an output layer;

[0168] The power system load prediction module 303 is configured to input the first feature data, the second feature data, and the third feature data into the splicing layer for splicing to obtain fourth feature data;

[0169] The fourth feature data is sequentially input into the fully connected layer and the output layer to obtain prediction data of the power system load.

[0170] An embodiment of the present disclosure provides a prediction device for the power system load, including: the prediction device for the power system load as described in any of the above embodiments.

[0171] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments of the present disclosure can be implemented by hardware or by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solutions of the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present disclosure.

[0172] Those skilled in the art can understand that the drawings are only schematic diagrams of a preferred embodiment, and the modules or processes in the drawings are not necessarily essential for implementing the present disclosure.

[0173] Those skilled in the art can understand that the modules in the device in the embodiment can be distributed in the device in the embodiment according to the description of the embodiment, or can be correspondingly changed to be located in one or more devices different from this embodiment. The modules of the above embodiments can be combined into one module, or further split into multiple sub-modules.

[0174] The serial numbers of the above embodiments of the present disclosure are only for description and do not represent the advantages and disadvantages of the embodiments.

[0175] Obviously, those skilled in the art can make various changes and modifications to the present disclosure without departing from the spirit and scope of the present disclosure. Thus, if these modifications and variations of the present disclosure fall within the scope of the claims of the present disclosure and their equivalent technologies, the present disclosure is also intended to include these changes and variations.

Claims

1. A method for predicting power system load, characterized in that: include: Obtain policy text data, geographic area data, and economic data; Perform word frequency statistics on the policy text data, and determine a multi-input-output table of the policy text data based on a policy consistency index model; Determine a first matrix using a word frequency-inverse document frequency statistical method according to the multi-input-output table; Determine the second matrix by using the analytic hierarchy process according to the multi-input-output table; Constructing a prediction model for power system load, the prediction model for power system load comprising: a policy data processing unit, a geographic area data processing unit, an economic data processing unit and a power system load prediction unit; Input the first matrix and the second matrix into the policy data processing unit for feature extraction to obtain first feature data; input the geographic area data into the geographic area data processing unit to obtain second feature data; input the economic data into the economic data processing unit to obtain third feature data; The first characteristic data, the second characteristic data and the third characteristic data are input into the power system load prediction unit to obtain prediction data of the power system load.

2. The method according to claim 1, characterized in that The performing word frequency statistics on the policy text data and determining a multi-input-output table of the policy text data based on a policy consistency index model includes: Performing word frequency statistics on the policy text data, and determining a preset number of primary variables in the policy text data based on a policy consistency index model; Classify the primary variables and their keywords to determine the secondary variables after classifying the primary variables; A multiple input-output table of the policy text data is determined according to the primary variables and the secondary variables.

3. The method according to claim 2, characterized in that The step of determining the first matrix by using a word frequency-inverse document frequency statistical method according to the multi-input-output table includes: Determine a keyword dictionary corresponding to each of the secondary variables in the multi-input-output table; Determine the value of each secondary variable using a word frequency-inverse document frequency statistical method based on the keyword dictionary and the policy text data; A first matrix is ​​determined according to the multiple input-output table and the value of each secondary variable.

4. The method according to claim 2, characterized in that The method of using the hierarchical analysis method according to the multi-input-output table to determine the second matrix includes: Determine a first pairwise comparison matrix using the analytic hierarchy process according to the multiple input-output table; Determining the weight of the primary variable and the weight of the secondary variable according to the first pairwise comparison matrix; determining a probability value for each of the secondary variables; Determining a second matrix according to the weight of the primary variable, the weight of the secondary variable and the probability value of the secondary variable; The probability value of the secondary variable is determined by using a latent Dirichlet allocation model or a word frequency-inverse document frequency statistical method.

5. The method according to claim 1, characterized in that The policy data processing unit includes: a first input sublayer, a residual network sublayer and a first maximum pooling layer; The step of inputting the first matrix and the second matrix into the policy data processing unit for feature extraction to obtain first feature data includes: Determine input data of the first input sublayer according to the first matrix and the second matrix; According to the input data of the first input sub-layer, the residual network sub-layer and the first maximum pooling layer are used in sequence to perform feature extraction to obtain first feature data.

6. The method according to claim 1, characterized in that The geographic area data processing unit includes: a second input sublayer, a convolutional layer, and a second maximum pooling layer; The step of inputting the geographic area data into the geographic area data processing unit to obtain second feature data comprises: Determining input data of the second input sublayer according to the geographic area data; According to the input data of the second input sublayer, the convolution layer and the second maximum pooling layer are used in sequence to perform feature extraction to obtain second feature data; The geographical area data includes at least one of the following: climate data, environmental data and industrial data.

7. The method according to claim 1, characterized in that The economic data processing unit comprises: a third input sublayer, a long short-term memory network model sublayer and a third maximum pooling layer; The step of inputting the economic data into the economic data processing unit to obtain the third characteristic data comprises: Determining input data of the third input sublayer according to the economic data; According to the input data of the third input sublayer, the long short-term memory network model sublayer and the third maximum pooling layer are used in sequence to perform feature extraction to obtain third feature data; The economic data includes at least one of the following: economic trend data and power system historical load data.

8. The method according to claim 1, characterized in that The power system load forecasting unit comprises: a splicing layer, a fully connected layer and an output layer; The step of inputting the first characteristic data, the second characteristic data and the third characteristic data into the power system load prediction unit to obtain prediction data of the power system load includes: Inputting the first feature data, the second feature data and the third feature data into the splicing layer for splicing to obtain fourth feature data; The fourth characteristic data is sequentially input into the fully connected layer and the output layer to obtain the predicted data of the power system load.

9. A device for predicting power system load, characterized in that: include: Data acquisition module, used to acquire policy text data, geographic area data and economic data; A policy text data processing module is used to perform word frequency statistics on the policy text data, determine a multi-input-output table of the policy text data based on a policy consistency index model; and determine a first matrix based on the multi-input-output table using a word frequency-inverse document frequency statistics method; Determine the second matrix by using the analytic hierarchy process according to the multi-input-output table; The power system load forecasting module is used to construct a forecasting model for the power system load, and the forecasting model for the power system load includes: a policy data processing unit, a geographic area data processing unit, an economic data processing unit and a power system load forecasting unit; the first matrix and the second matrix are input into the policy data processing unit for feature extraction to obtain first feature data; the geographic area data is input into the geographic area data processing unit to obtain second feature data; the economic data is input into the economic data processing unit to obtain third feature data; the first feature data, the second feature data and the third feature data are input into the power system load forecasting unit to obtain forecasting data for the power system load.

10. A power system load prediction device, characterized in that: It comprises a power system load prediction device as described in claim 9.

Images