Evaluation method for regulation response capability of polymorphic energy unit in park

Abstract

The invention relates to the technical field of park polymorphic energy, and discloses a park polymorphic energy unit regulation response capability evaluation method, which comprises the following steps: S1, constructing a specific energy unit regulation response data model; s2, analyzing the operation characteristics of the polymorphic energy unit; s3, establishing a basic data model of polymorphic energy; s4, using a multi-head feature capturing algorithm; s5, focus point determination and multi-head feature output planning; s6, designing a dimension reduction matrix of the attention head; according to the method, a related data model is established through the operation characteristics of the energy supply unit, then a data model of the multi-energy energy supply system is extracted through a big data intelligent algorithm, then corresponding internal characteristics are extracted through massive operation scenes, and finally quantitative evaluation of the response capability is obtained in combination with the elastic affine principle. And the response characteristics of the polymorphic energy unit can be accurately evaluated.

Description

Technical Field

[0001] The present invention relates to the field of park polymorphic energy technology, and in particular to a method for evaluating the control response capability of a park polymorphic energy unit. Background Art

[0002] With the advancement of energy conservation and low-carbon goals, higher optimization goals are proposed for the operation of specific parks. In response to the needs of different energy usage scenarios in a park, different energy supply units in the park will play different roles and thus jointly meet the energy needs. For different energy usage scenarios and different energy supply unit compositions, it is necessary to formulate optimized scheduling and control plans.

[0003] In different energy usage scenarios, in order to meet the energy load requirements, the polymorphic energy unit has a complex operation mode, so it is difficult to accurately evaluate the response characteristics of the polymorphic energy unit. To this end, the present invention proposes a method for evaluating the control response capability of a park polymorphic energy unit. Summary of the invention

[0004] The purpose of the present invention is to provide a method for evaluating the control response capability of a multi-state energy unit in a park. A relevant data model is established through the operating characteristics of the energy supply unit, and then the data model of the multi-energy supply system is extracted through a big data intelligent algorithm. Then, the corresponding intrinsic characteristics are extracted through massive operating scenarios. Finally, an elastic affine principle is combined to obtain a quantitative evaluation of the response capability to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a method for evaluating the control response capability of a multi-state energy unit in a park, comprising the following steps:

[0006] S1: Construct a specific energy unit control response data model, summarize the characteristic data of different energy units according to their operating characteristics and response forms, and determine and preliminarily classify the types of polymorphic energy units;

[0007] S2: Operational characteristic analysis of polymorphic energy units, conduct characteristic analysis on polymorphic energy units in actual operation, abstract their key features, extract data of relevant features and their data collection methods;

[0008] S3: Establish the basic data model of polymorphic energy, conduct preliminary screening of the data that are considered to be usable, prepare the corresponding testing equipment, determine the specific data item regulations, and thus form a relatively complete data model matrix;

[0009] S4: Use a multi-head feature capture algorithm to design the corresponding Q, K, V matrix according to the specificity and sensitivity required, summarize the feature vector set to form a multi-dimensional matrix, reduce its dimension to obtain the final output matrix that matches its input information volume, and construct a multi-head feature model of the response capability of different energy units;

[0010] S5: Focus point determination and multi-head feature output planning, determine the capture focus points of the multi-head feature model;

[0011] S6: Design of the dimension reduction matrix of the attention head. After obtaining the Q, K, and V matrices, a similarity comparison is performed, that is, solving the attention head matrix and performing corresponding dimension reduction processing on it;

[0012] S7: An evaluation algorithm based on the affine principle is proposed. According to the multi-head feature model of the polymorphic energy unit composition and energy usage scenarios of the park, an evaluation algorithm based on the affine principle is proposed.

[0013] Preferably, in S1, the types of polymorphic energy units are divided into non-electrical energy units and electrical energy units.

[0014] Preferably, in S2, the operating characteristic analysis of the multi-state energy unit includes at least: operating characteristic analysis of the electric energy unit and operating characteristic analysis of the non-electric energy unit.

[0015] Preferably, in S3, when the basic data model of polymorphic energy is established, the following data items are included for non-electric energy units: the daily working data of the energy unit is collected accordingly, the average value is taken, and the data is listed by time with the typical example to form an energy conversion efficiency matrix The energy unit release, feedback, and energy generation rates are quantified, and a variety of typical scenarios are selected for analysis. When a fault occurs or the load changes over a large range, the energy transfer rate matrix is ​​obtained by listing the data by time with typical examples. Take the more typical load change scenarios and fault scenarios in the park, detect the reaction rate of the corresponding energy unit, and thus obtain the reaction speed matrix τ of the energy unit to load changes; take the more typical load change scenarios and fault scenarios in the park, and monitor the fluctuation degree of energy transmission of the energy unit in the park on ordinary natural days to form the corresponding matrix σ; the communication power and power output data matrix P of the energy unit and the power grid l , W l .

[0016] Preferably, in S3, when the basic data model of polymorphic energy is established, the following data items are included for the electric energy unit: the fixed storage capacity E of the energy unit si ; Energy transfer power P of energy unit si ; Power fluctuation during monitoring period under stable conditions e st; The reaction rate of the energy unit to load changes and regulation requirements v i ; Transmission efficiency of energy unit and its quantified loss ε s .

[0017] Preferably, in S3, the formation of the data matrix includes: standardization and formation of the energy unit attribute matrix; granularity alignment and parameter matching; typical data capture and selection.

[0018] Preferably: in S5, the focus includes: regulating the response ramp rate; regulating the economic efficiency of the response; regulating the environmental protection index of the response; the correlation degree of each energy unit; and the integration performance of each energy unit.

[0019] Preferably, S6: designing the dimension reduction matrix of the attention head includes the following steps:

[0020] Step 1: Comparison of the similarity between Q and K: The similarity can be compared by matrix dot product multiplication, finding the vector cosine value, and introducing the MLP network. This comparison can obtain a matrix containing attention score information;

[0021] SCORE i =Q i ·K i T #(2.4)

[0022] Step 2: Weight V assignment and normalization matrix design: According to different evaluation mechanisms, the weight matrix V is obtained. This weight matrix reflects the degree of concern for the initial capture of different features:

[0023]

[0024] Step 3: Feature combination and dimensionality reduction: Note that the data of the attention heads will obtain stronger feature communication performance after splicing and processing. In addition, the splicing of the attention heads makes the output appear much larger than the number of mismatches before and after the input data. Therefore, the dimensionality reduction operation is performed on it. Therefore, the design of the dimensionality reduction matrix should have the ability of feature integration and joint evaluation:

[0025] Multihead=Concatenate(Attentionhead i )·L#(2.6).

[0026] Preferably, an evaluation algorithm based on the affine principle is proposed in S7, including: deriving the matrix to be projected; establishing the evaluation space; introducing the mathematical method of projection and the elastic affine method.

[0027] Preferably, the derivation of the matrix to be projected uses row vectors to represent various features of a certain energy unit captured by multiple features of the protection point 2, and column vectors are used to represent the scores of different energy units under the same evaluation mechanism, giving a matrix X to be projected i ; Establishment of evaluation space: After the matrix is ​​generated, an evaluation space needs to be established according to the evaluation needs. The evaluation space is composed of n linearly independent evaluation bases;

[0028] Mathematical method of projection: transform the matrix X i All row vectors β in i Projected to B i Each evaluation basis b in the space i On the surface, the shadow vector p is formed i ; After clarifying the above tasks, the following methods are proposed:

[0029] First, the row vector β i Need to be projected to the evaluation base b i On, there are:

[0030]

[0031] On this basis, this calculation method is extended to matrix operations, as follows:

[0032] P=B i (B i B i T ) -1 B i T X i #(3.6)

[0033] At this point, the shadow matrix P contains all the evaluation information of the polymorphic energy unit under different evaluation dimensions, and the modulus of each column vector is the specific score information;

[0034] Introduction of elastic affine method: If the influence of a certain data in the integrated big data model is too large or too small, it is necessary to introduce an elastic coefficient to reduce its proportion or increase its proportion, so that the final big data model can fully and fairly reflect the contribution of each element to the evaluation index.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] The present invention proposes a method for evaluating the control response capability of polymorphic energy units in a park. A relevant data model is established through the operating characteristics of the energy supply unit, and then the data model of the multi-energy supply system is extracted through a big data intelligent algorithm. Then, the corresponding intrinsic characteristics are extracted through massive operating scenarios. Finally, an elastic affine principle is combined to obtain a quantitative evaluation of the response capability. This method can accurately evaluate the response characteristics of the polymorphic energy unit, and is convenient for formulating optimized scheduling and control schemes for different energy usage scenarios and different energy supply unit compositions.

[0037] The parts not involved in the device are the same as those in the prior art or can be implemented by using the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of a method for evaluating the control response capability of a multi-state energy unit in a park according to the present invention; DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0040] See also Figure 1 The present invention provides a technical solution: a method for evaluating the control response capability of a multi-state energy unit in a park, comprising the following steps:

[0041] S1: Construct a specific energy unit control response data model, summarize the characteristic data of different energy units according to their operating characteristics and response forms, and determine and preliminarily classify the types of polymorphic energy units;

[0042] Since there are many types of energy supply units in the park with different characteristics, the same energy supply units have different models and parameters and specifications; in addition, different energy usage scenarios in the park also put forward different energy requirements, which makes the park's polymorphic energy units present differentiated operating characteristics to the outside world. In order to further study the operating characteristics and control response of the park's polymorphic energy, a relatively unified specific energy unit control response data model was constructed.

[0043] In order to establish a reasonable data model, it is necessary to summarize the characteristic data of different energy units according to their operating characteristics and response forms, and classify them on this basis, first classified into the following parts:

[0044] In S1, the types of polymorphic energy units are divided into non-electrical energy units and electrical energy units.

[0045] Non-electric energy units and related restrictions: The following provisions are made for non-electric energy units. Non-electric energy units should be energy aggregates and their carrier concepts that exist, function and participate in energy exchange in the park in the form of non-electric energy, that is, heat energy, gas energy, etc. other than electric energy. After investigating the target park, it is learned that its general forms are high-temperature heating, environmental heating, natural gas, etc.;

[0046] Electric energy units and related restrictions: The following provisions are made for electric energy units. Electric energy units should be a conceptual collection of energy combinations and their carriers that exist, function and supply electric energy to the park. After investigating the target park, it is learned that their general forms are power supply from the grid, distributed power sources, energy storage systems and charging piles, etc.

[0047] Note: There are still other classification methods for the above concepts in use, such as movable electric energy units and fixed energy units, which will directly affect the subsequent evaluation of dispatch response and operating characteristics, and will be introduced in more detail in the characteristic research.

[0048] S2: Operational characteristic analysis of polymorphic energy units, conduct characteristic analysis on polymorphic energy units in actual operation, abstract their key features, extract data of relevant features and their data collection methods;

[0049] After clarifying the classification method of polymorphic energy units, we need to conduct characteristic analysis research on the polymorphic energy units in actual operation, abstract their key features, extract the data of relevant features and their data collection methods. Here we will list typical but limited energy units for research: In S2, the operating characteristic analysis of polymorphic energy units at least includes: operating characteristic analysis of electric energy units and operating characteristic analysis of non-electric energy units.

[0050] Analysis and research on the operating characteristics of power units: The research on the operating characteristics of power units mainly focuses on the following power units, and studies and explores their operating characteristics, summarizing the following characteristics that may affect the corresponding regulation:

[0051] (1) Grid power supply: The largest energy unit in the park, with the following characteristics: energy can be transmitted in both directions, has a great impact on economic indicators, relatively stable power transmission, strong interference resistance, high power quality, and strong dispatchability.

[0052] (2) Distributed power source: In the park, it mainly refers to photovoltaic power generation units. Wind power generation units are the main forms of power supply units. They have the following characteristics: energy cannot be transmitted in both directions, which has a certain impact on economic indicators, poor power transmission stability, insufficient anti-interference ability, poor power quality, and greater difficulty in scheduling.

[0053] (3) Energy storage system: In the park, it usually refers to related equipment for storing electrical energy. It has the following characteristics: energy can be transmitted in both directions, it has certain anti-interference and crisis backup functions, and it can also be divided into mobile energy storage equipment and fixed mobile energy storage equipment. Energy storage equipment with different flexibility has different response capabilities to regulation.

[0054] (4) Charging pile: namely, V2G charging pile, which has the following characteristics: it can transmit energy in both directions, has high randomness of energy transmission, and has high scheduling flexibility.

[0055] Study on the operation characteristics of non-electrical energy units:

[0056] (1) High-temperature heating: Energy units that supply high temperatures and large amounts of heat in the park have a slow response to regulation (it takes a long time to achieve the corresponding state switch), but can provide high-temperature steam for power generation, thereby playing a role in communicating with the park's power grid.

[0057] (2) Environmental heating: It is assumed that environmental heating is basically supplied by electrical energy, that is, electrical energy is converted into thermal energy so that the ambient temperature is maintained within the allowable range. This usually occurs in a load state in the park.

[0058] (3) P2G equipment: It can realize the bidirectional conversion of gas energy and electric energy and can be regarded as a special energy storage device.

[0059] S3: Establish the basic data model of polymorphic energy, conduct preliminary screening of the data that are considered to be usable, prepare the corresponding testing equipment, determine the specific data item regulations, and thus form a relatively complete data model matrix;

[0060] After clarifying the types of polymorphic energy units and the relevant operating characteristics of their different energy units, the basic data model of polymorphic energy units is established. First, it is necessary to conduct a preliminary screening of the data that is considered to be usable, and then prepare the corresponding testing equipment. Finally, the specific data item regulations are determined, so as to form a more complete data model matrix.

[0061] In S3, when the basic data model of polymorphic energy is established, the following data items are included for non-electric energy units: the daily working data of the energy unit is collected accordingly, the average value is taken, and the data is listed by time with the typical example to form an energy conversion efficiency matrix The energy unit release, feedback, and energy generation rates are quantified, and a variety of typical scenarios are selected for analysis. When a fault occurs or the load changes over a large range, the energy transfer rate matrix is ​​obtained by listing the data by time with typical examples. Take the more typical load change scenarios and fault scenarios in the park, detect the reaction rate of the corresponding energy unit, and thus obtain the reaction speed matrix τ of the energy unit to load changes; take the more typical load change scenarios and fault scenarios in the park, and monitor the fluctuation degree of energy transmission of the energy unit in the park on ordinary natural days to form the corresponding matrix σ; the communication power and power output data matrix P of the energy unit and the power grid l , W l .

[0062] In S3, when the basic data model of polymorphic energy is established, the following data items are included for the electric energy unit: the fixed storage capacity E of the energy unit si ; Energy transfer power P of energy unit si ; Power fluctuation during monitoring period under stable conditions e st ; The reaction rate of the energy unit to load changes and regulation requirements v i ; Transmission efficiency of energy unit and its quantified loss ε s .

[0063] In S3, the formation of the data matrix includes: standardization and formation of the energy unit attribute matrix; granularity alignment and parameter matching; typical data capture and selection.

[0064] Specifically: For the above non-electrical energy units (including high-temperature heating, environmental heating, and P2G equipment), the following items should be examined:

[0065] (1) Collect the daily working data of the energy unit (including the typical scenario of the example), take the average value, and list the data by time with the typical example to form an energy conversion efficiency matrix (The time corresponds to other items in the example, and will not be repeated below)

[0066]

[0067] (2) Quantify the release, feedback, and energy generation rates of energy units, and select a variety of typical scenarios for analysis, such as failures and large-scale load changes. Compare typical examples and list data by time to derive an energy transfer rate matrix.

[0068]

[0069] (3) Take the more typical load change scenarios and fault scenarios in the park, detect the reaction rate of the corresponding energy unit, and thus obtain the reaction speed matrix τ of the energy unit to the load change (corresponding to the time of other items in the example, the same below will not be repeated).

[0070] (4) Take the more typical load change scenarios and fault scenarios in the park, and monitor the fluctuation degree of energy transmitted by the energy units in the park on ordinary natural days to form the corresponding matrix σ.

[0071] (5) The data matrix P of the communication power and power output between the energy unit and the power grid l , W l

[0072] Data item regulations and processing of electric energy units:

[0073] (1) Fixed storage capacity E of energy unit si ,The corresponding energy units mentioned above, such as distributed ,power sources and charging piles, all have certain storage units for storing ,electrical energy, so the data should include the energy storage level of the ,energy unit.

[0074] (2) Energy transfer power P of energy unit si , the elements can be listed according to the time period of the example to form a corresponding matrix, and the unit-grid power AC status W in the corresponding time period can be found at the same time si , which reflects the communication status between the energy unit and the power grid.

[0075] (3) Power fluctuation during the monitoring period (under stable conditions) st , list the elements according to the time period of the example to form a data matrix.

[0076] (4) The reaction rate v of the energy unit to load changes and regulation requirements i ,The monitoring area mainly focuses on multiple time periods when external indicators actively change, and ,the calculation cases are arranged to form a corresponding matrix.

[0077] (5) Transmission efficiency of energy unit and its quantified loss ε s .

[0078] Integration of data models and formation of data matrix:

[0079] (1) Standardization and formation of energy unit attribute matrix: For any attribute or data item of energy unit, the data is arranged by time to form a vector α with a dimension of t, which is the number of time periods in a calculation example. i :

[0080]

[0081] In particular, for attributes or data items that do not change over time, all elements can be listed as unchanging constants.

[0082] (2) Granularity alignment and parameter matching: For different individuals or types of energy units, the highest principle should be to use the different attributes of the energy units to determine their contribution to the park power grid control response. Select the ones with the highest similarity for one-to-one correspondence and matching. If the data size is very different, weighted integration or normalization by the softmax function can be used before loading.

[0083] (3) Typical data capture and selection: It is particularly important to note that the presentation of routine operation data has a time axis, so the correspondence of similar data requires the use of the same example to ensure the content consistency of the matrix data. The selection of examples should also be differentiated. For example, the example of collecting response speed should capture the response speed of the energy unit when the external active adjustment or load actively changes. The example should not be the same as the example selected for the stability capacity of the energy unit during routine operation.

[0084] Auxiliary data model establishment based on historical operation data:

[0085] This data item is set for reference. Select operating data that has not undergone preliminary feature screening, select a generally ordinary time period to establish a corresponding auxiliary model, and only select one of the above-mentioned extracted features as the object of investigation. List the corresponding auxiliary model. This model is also included in the data combination as a backup reference.

[0086] S4: Use a multi-head feature capture algorithm to design the corresponding Q, K, V matrix according to the specificity and sensitivity required, summarize the feature vector set to form a multi-dimensional matrix, reduce its dimension to obtain the final output matrix that matches its input information volume, and construct a multi-head feature model of the response capability of different energy units;

[0087] S5: Focus point determination and multi-head feature output planning, determine the capture focus points of the multi-head feature model;

[0088] In S5, the focus includes: regulating the response ramp rate; regulating the economic efficiency of the response; regulating the environmental protection indicators of the response; the degree of correlation between each energy unit; and the integration performance of each energy unit.

[0089] Specific:

[0090] The capture of multi-head feature models needs to include a certain purpose, that is, the capture results should include the following information: the reflection of the operating characteristics and adjustment response capabilities of the park's target multi-state energy units; the interconnection relationship and integrated performance between the park's multi-state energy units in terms of adjustment response. Specifically, it is reflected in the following aspects:

[0091] (1) Regulation response ramp rate: It is the most important indicator reflecting the regulation response capability of the energy unit, which indicates the energy unit's ability to respond to external environmental disturbances under a unified measurement standard.

[0092] (2) Economic efficiency of regulation response: Certain losses will occur in energy transmission and energy conversion during the regulation process. The regulation action will bring certain economic benefits or the passive regulation caused by load change will increase the economic cost. The integration of this information should be reflected in the economic efficiency of the regulation response.

[0093] (3) Environmental indicators of regulation response: Energy will produce certain losses or carbon emissions during the transmission process. These indicators will be used as one of the reference items in the regulation response evaluation indicators.

[0094] (4) The degree of correlation among energy units: The various indicators among energy units may have certain connections during the adjustment response process. For example, the same indicators of different energy units may change in the same way or in the opposite way during the adjustment process. The above scenario actually reflects the degree of correlation among different energy units.

[0095] (5) Integration performance of each energy unit: Different polymorphic energy units are classified and integrated to form corresponding energy unit clusters, and the regulatory response capability of the entire cluster is evaluated.

[0096] These goals and evaluation parameters are the basis for designing the weight matrix and deformation matrix below. That is, through result-oriented matrix design, the parameters of the intermediate tools are continuously adjusted to observe whether the output target can contain the information required by the target.

[0097] Formation of Q, K, and V matrices: In the previous step, we have obtained the corresponding data set, that is, the matrix set that has been feature summarized. After inputting it into the multi-head feature capture algorithm, we need to design different weight matrices to multiply the main matrix to obtain the corresponding Q, K, and V matrices.

[0098] (1) Generation of Q matrix: The Q matrix needs to be designed based on the relevant content of the expected query, so the corresponding designed weight matrix is It is necessary to include the feature information that the above evaluation mechanism is subjectively concerned about. On this basis, it is necessary to design multiple sets of weight matrices to multiply the corresponding set of original data matrices.

[0099] (2) Generation of K matrix: The K matrix needs to be designed based on the autonomous features of the objects to be compared that have been summarized in step 1. Therefore, the corresponding designed weight matrix The output matrix K should be i It contains the objective characteristic information of the above energy units. On this basis, it is necessary to design multiple sets of weight matrices to multiply the corresponding set of original data matrices.

[0100] (3) Generation of V matrix: The V matrix needs to be designed according to the characteristic output size in the regulation response, so the corresponding designed weight matrix It is necessary to reflect the degree to which the weights care about different features. On this basis, it is necessary to design multiple sets of weight matrices to multiply the corresponding set of original data matrices.

[0101]

[0102] S6: Design of the dimension reduction matrix of the attention head. After obtaining the Q, K, and V matrices, a similarity comparison is performed, that is, solving the attention head matrix and performing corresponding dimension reduction processing on it;

[0103] S6: The design of the dimensionality reduction matrix of the attention head includes the following steps:

[0104] Step 1: Comparison of the similarity between Q and K: The similarity can be compared by matrix dot product multiplication, finding the vector cosine value, and introducing the MLP network. This comparison can obtain a matrix containing attention score information;

[0105] SCORE i =Q i ·K i T #(2.4)

[0106] Step 2: Weight V assignment and normalization matrix design: According to different evaluation mechanisms, the weight matrix V is obtained. This weight matrix reflects the degree of concern for the initial capture of different features:

[0107]

[0108] Step 3: Feature combination and dimensionality reduction: Note that the data of the attention heads will obtain stronger feature communication performance after splicing and processing. In addition, the splicing of the attention heads makes the output appear much larger than the number of mismatches before and after the input data. Therefore, the dimensionality reduction operation is performed on it. Therefore, the design of the dimensionality reduction matrix should have the ability of feature integration and joint evaluation:

[0109] Multihead=Concatenate(Attentionhead i )·L#(2.6).

[0110] S7: An evaluation algorithm based on the affine principle is proposed. According to the multi-head feature model of the polymorphic energy unit composition and energy usage scenarios of the park, an evaluation algorithm based on the affine principle is proposed.

[0111] S7 proposes an evaluation algorithm based on the affine principle, including: the derivation of the matrix to be projected; the establishment of the evaluation space; the introduction of the mathematical method of projection and the elastic affine method.

[0112] Derivation of the matrix to be projected: After integrating the above steps, we can always get the following data matrix, using row vectors to represent the various features of a certain energy unit captured by the multi-head feature of the protection point 2, and column vectors to represent the scores of different energy units under the same evaluation mechanism, giving a matrix X to be projected i ; As an example:

[0113]

[0114] This matrix X i It contains m items of data information of n energy units (represented by n m-dimensional vectors). Of course, the matrix output by the protection point 2 may be different, so the data matrix here is not unique.

[0115] Establishment of evaluation space: After the matrix is ​​generated, it is necessary to establish an evaluation space according to the needs of the evaluation. The evaluation space is composed of n linearly independent evaluation bases. It is particularly important to note here that the various characteristic evaluation items obtained by multiple captures at the protection point are numerous and need to be summarized. Therefore, the number of evaluation bases obtained is much smaller than the evaluation characteristic items. Therefore, determining fewer evaluation items, namely the evaluation bases, plays a role in summarizing and summarizing.

[0116] Defining the target base of projection: i That is, the evaluation basis is the target basis of the projection. i A row vector β i (used to characterize the performance of energy units under various characteristics (column vector evaluation)) is projected onto each evaluation basis b i , and obtain a new projection vector p i , this projection vector actually reflects the performance of the projected energy unit containing multiple features on this evaluation benchmark.

[0117] The evaluation basis here is b i Follow these design principles:

[0118] (1) Evaluation base b i Evaluation item: Evaluation basis b i It represents an evaluation dimension, which can be evaluated as follows: the comprehensive response capability of the energy unit to regulation, the robustness of the energy unit in the face of large load changes or system failures, the impact of the energy unit on the overall stability of the system during the regulation-response action, etc. (the number of evaluation items is less than the number of features of the basis to be evaluated).

[0119] (2) Evaluation base b i Parameter determination: Evaluation base b iis a column vector whose elements are used to assign different weights to different features of the vector, in order to determine the degree to which the vector to be projected is biased towards the evaluation basis. The weight coefficient C ij The size is determined by the contribution of the object feature to the evaluation item:

[0120]

[0121] (3) Evaluation base b i Difference: It is important to note that under the goal orientation, the evaluation bases must be different, because it is unnecessary to set evaluation bases with high similarity or even proportionality, which naturally makes the evaluation base b i There is no linear independence between them, and this high-dimensional space is indeed spanned by the above-mentioned basis.

[0122] Mathematical method of projection: transform the matrix X i All row vectors β in i Projected to B i Each evaluation basis b in the space i On the surface, the shadow vector p is formed i ; After clarifying the above tasks, the following methods are proposed:

[0123] First, the row vector β i Need to be projected to the evaluation base b i On, there are:

[0124]

[0125] On this basis, this calculation method is extended to matrix operations, as follows:

[0126] P=B i (B i B i T ) -1 B i T X i #(3.6)

[0127] At this point, the shadow matrix P contains all the evaluation information of the polymorphic energy unit under different evaluation dimensions, and the modulus of each column vector is the specific score information;

[0128] Introduction of elastic affine method: In the above evaluation method, we introduce an adjustment method based on the elastic affine principle to improve the flexibility of the evaluation mechanism, so that the evaluation model can have better adaptability and can be applied to the evaluation requirements in different scenarios. In some other scenarios, the vector has a proportional distortion phenomenon, which also requires the introduction of scaling changes to restore its true proportion. The elastic coefficient can be introduced in the following links to scale the vector and space:

[0129] Principles for establishing elastic coefficients:

[0130] (1) If the influence of a certain data item in the integrated big data model is too large or too small, it is necessary to introduce an elasticity coefficient to reduce its proportion or increase its proportion, so that the final big data model can fully and fairly reflect the contribution of each element to the evaluation index. Its necessity is reflected in the following: if directly modifying the data is a tampering with the authenticity, if stretching the entire dimension will alleviate the adverse effects of this situation (stretching action can be remembered and reflected by changes in other elements). Therefore, the determination of the stretching coefficient follows the following principles:

[0131] Parameter k r That is, there is a weight imbalance between energy units. Assuming that the unbalanced data line is the i-th row, the threshold of the imbalance is:

[0132]

[0133]

[0134] When the threshold is exceeded, k is introduced r , so that the corrected element k r x ij :

[0135]

[0136] The imbalance of column elements here (i.e., the abnormal proportion of a certain indicator of the energy unit) also follows the above principles. You can determine k based on the actual situation by analogy with the above ideas. c The specific value of

[0137] (2) It is also necessary to introduce the expansion coefficient of the evaluation space. In the real-time adjustment process, it is often the case that the correlation between a certain feature of the evaluation mechanism and the evaluation result does not meet the established requirements. The established requirements may change immediately, so the corresponding k can be commanded. i As it changes, the corresponding time-varying function is designed:

[0138] k i =k i (t)

[0139] At the same time, the result-oriented relevance provisions for each evaluation mechanism can also be modified by a weakened form of the above solution, with a certain projection vector p i For example, we found:

[0140]

[0141] So we introduce k i Modify it so that it can meet the required threshold.

[0142] Use of elastic modulus:

[0143] (1) Scaling transformation of the original data matrix: There may be some problems with the matrix input into the projection transformation. For example, the row vector representing a certain energy unit in the input data matrix is ​​too large compared to the row vectors of other energy units, or the influence weight of a certain feature item or element in the data matrix is ​​too large. Therefore, it is beneficial to appropriately elastically scale the target vector and matrix by left-multiplying or right-multiplying the target matrix:

[0144]

[0145]

[0146] (2) Scaling and transformation of the evaluation space: The evaluation space needs to be adjusted according to different evaluation mechanisms. The adjustment here should design transformations in multiple dimensions. Of course, in order to ensure the displacement of the evaluation space, the linear independence between vectors needs to be verified during the process:

[0147]

[0148] The present invention proposes a method for evaluating the control response capability of polymorphic energy units in a park. A relevant data model is established through the operating characteristics of the energy supply unit, and then the data model of the multi-energy supply system is extracted through a big data intelligent algorithm. Then, the corresponding intrinsic characteristics are extracted through massive operating scenarios. Finally, an elastic affine principle is combined to obtain a quantitative evaluation of the response capability. This method can accurately evaluate the response characteristics of the polymorphic energy unit, and is convenient for formulating optimized scheduling and control schemes for different energy usage scenarios and different energy supply unit compositions.

[0149] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making any creative work shall fall within the scope of protection of the present invention.

Claims

1. A method for evaluating the control response capability of a multi-state energy unit in a park, characterized by: The following steps are involved: S1: Construct a specific energy unit control response data model, summarize the characteristic data of different energy units according to their operating characteristics and response forms, and determine and preliminarily classify the types of polymorphic energy units; S2: Operational characteristic analysis of polymorphic energy units, conduct characteristic analysis on polymorphic energy units in actual operation, abstract their key features, extract data of relevant features and their data collection methods; S3: Establish the basic data model of polymorphic energy, conduct preliminary screening of the data that are considered to be usable, prepare the corresponding testing equipment, determine the specific data item regulations, and thus form a relatively complete data model matrix; S4: Use a multi-head feature capture algorithm to design the corresponding Q, K, V matrix according to the specificity and sensitivity required, summarize the feature vector set to form a multi-dimensional matrix, reduce its dimension to obtain the final output matrix that matches its input information volume, and construct a multi-head feature model of the response capability of different energy units; S5: Focus point determination and multi-head feature output planning, determine the capture focus points of the multi-head feature model; S6: Design of the dimension reduction matrix of the attention head. After obtaining the Q, K, and V matrices, a similarity comparison is performed, that is, solving the attention head matrix and performing corresponding dimension reduction processing on it; S7: An evaluation algorithm based on the affine principle is proposed. According to the multi-head feature model of the polymorphic energy unit composition and energy usage scenarios of the park, an evaluation algorithm based on the affine principle is proposed.

2. A method for evaluating the control response capability of a park multi-state energy unit according to claim 1, characterized in that: In S1, the types of polymorphic energy units are divided into non-electrical energy units and electrical energy units.

3. A method for evaluating the control response capability of a multi-state energy unit in a park according to claim 2, characterized in that: In S2, the operation characteristic analysis of the multi-state energy unit at least includes: the operation characteristic analysis of the electric energy unit and the operation characteristic analysis of the non-electric energy unit.

4. A method for evaluating the control response capability of a park multi-state energy unit according to claim 2, characterized in that: In S3, when the basic data model of polymorphic energy is established, the following data items are included for non-electric energy units: the daily working data of the energy unit is collected accordingly, the average value is taken, and the data is listed by time with the typical example to form an energy conversion efficiency matrix The energy unit release, feedback, and energy generation rates are quantified, and a variety of typical scenarios are selected for analysis. When a fault occurs or the load changes over a large range, the energy transfer rate matrix is ​​obtained by listing the data by time with typical examples. Take the more typical load change scenarios and fault scenarios in the park, detect the reaction rate of the corresponding energy unit, and thus obtain the reaction speed matrix τ of the energy unit to load changes; take the more typical load change scenarios and fault scenarios in the park, and monitor the fluctuation degree of energy transmission of the energy unit in the park on ordinary natural days to form the corresponding matrix σ; the communication power and power output data matrix P of the energy unit and the power grid l , W l .

5. A method for evaluating the control response capability of a park multi-state energy unit according to claim 4, characterized in that: In S3, when the basic data model of polymorphic energy is established, the following data items are included for the electric energy unit: the fixed storage capacity E of the energy unit si ; Energy transfer power P of energy unit si ; Power fluctuation during monitoring period under stable conditions e st ; The reaction rate of the energy unit to load changes and regulation requirements v i ; Transmission efficiency of energy unit and its quantified loss ε s .

6. A method for evaluating the control response capability of a park multi-state energy unit according to claim 5, characterized in that: In S3, the formation of the data matrix includes: standardization and formation of the energy unit attribute matrix; granularity alignment and parameter matching; typical data capture and selection.

7. A method for evaluating the control response capability of a park multi-state energy unit according to claim 1, characterized in that: In S5, the focus includes: regulating the response ramp rate; regulating the economic efficiency of the response; regulating the environmental protection indicators of the response; the degree of correlation between each energy unit; and the integration performance of each energy unit.

8. A method for evaluating the control response capability of a park multi-state energy unit according to claim 1, characterized in that: S6: The design of the dimensionality reduction matrix of the attention head includes the following steps: Step 1: Comparison of the similarity between Q and K: The similarity can be compared by matrix dot product multiplication, finding the vector cosine value, and introducing the MLP network. This comparison can obtain a matrix containing attention score information; SCORE i =Q i ·K i T #(2.4) Step 2: Weight V assignment and normalization matrix design: According to different evaluation mechanisms, the weight matrix V is obtained. This weight matrix reflects the degree of concern for the initial capture of different features: Step 3: Feature combination and dimensionality reduction: Note that the data of the attention heads will obtain stronger feature communication performance after splicing and processing. In addition, the splicing of the attention heads makes the output appear much larger than the number of mismatches before and after the input data. Therefore, the dimensionality reduction operation is performed on it. Therefore, the design of the dimensionality reduction matrix should have the ability of feature integration and joint evaluation: Multihead=Concatenate(Attenionhead i )·L#(2.6)。 9. A method for evaluating the control response capability of a multi-state energy unit in a park according to claim 1, characterized in that: S7 proposes an evaluation algorithm based on the affine principle, including: the derivation of the matrix to be projected; the establishment of the evaluation space; the introduction of the mathematical method of projection and the elastic affine method.

10. A method for evaluating the control response capability of a park multi-state energy unit according to claim 9, characterized in that: The derivation of the matrix to be projected uses row vectors to represent various features of a certain energy unit captured by the multi-head feature of the protection point 2, and column vectors are used to represent the scores of different energy units under the same evaluation mechanism. A matrix X to be projected is given i ; Establishment of evaluation space: After the matrix is ​​generated, an evaluation space needs to be established according to the evaluation needs. The evaluation space is composed of n linearly independent evaluation bases; Mathematical method of projection: transform the matrix X i All row vectors β in i Projected to B i Each evaluation basis b in the space i On the surface, the shadow vector p is formed i ; After clarifying the above tasks, the following methods are proposed: First, the row vector β i Need to be projected to the evaluation base b i On, there are: On this basis, this calculation method is extended to matrix operations, as follows: P=B i (B i B i T ) -1 B i T X i #(3.6) At this point, the shadow matrix P contains all the evaluation information of the polymorphic energy unit under different evaluation dimensions, and the modulus of each column vector is the specific score information; Introduction of elastic affine method: If the influence of a certain data in the integrated big data model is too large or too small, it is necessary to introduce an elastic coefficient to reduce its proportion or increase its proportion, so that the final big data model can fully and fairly reflect the contribution of each element to the evaluation index.

Images