Power plant unit load distribution system and method

Abstract

The invention discloses a power plant unit load distribution system and method, and the method comprises the steps: monitoring and collecting a data set of key operation parameters of each unit in real time through a sensor monitoring system; a data processing and analysis algorithm based on artificial intelligence and deep learning is introduced at the rear end to carry out collaborative correlation analysis on the data sets of the key operation parameters of the different units, so that the operation state characteristics of the units are extracted; and carrying out priority ranking on the units based on the running state characteristics of the different units and the variable load direction of the thermal power plant. Therefore, the priority order of each unit can be dynamically adjusted based on the operation state and the variable load direction of each unit of the power plant, so that more flexible and intelligent power plant unit load distribution and scheduling are realized to cope with continuously changing power markets and user requirements; the rationality and reliability of load distribution can be ensured, and the operation efficiency and economy of a power plant are ensured.

Description

Technical Field

[0001] The present application relates to the field of intelligent distribution, and more specifically, to a system and method for distributing loads of power plant units. Background Art

[0002] With the growth of electricity demand and the improvement of environmental protection requirements, thermal power plants need to consider how to optimize operations to reduce emissions and reduce costs while meeting grid dispatch instructions. Reasonable load distribution of power plant units can not only ensure the safe and stable operation of the power grid, but also improve power generation efficiency, reduce unnecessary fuel consumption, and reduce environmental pollution. However, the traditional load distribution method of power plant units is usually averaged according to the capacity of the units, but this method ignores the differences in operating economy between units, especially when the equipment conditions are inconsistent, which may lead to energy waste and low operating efficiency. In addition, the traditional load distribution method may lead to unreasonable load distribution when facing complex and changeable operating conditions, thereby affecting the overall economy and safety of the power plant. For example, in the process of variable load, if the load of each unit cannot be adjusted reasonably, some units may be in a non-economic operating state, increasing operating costs.

[0003] In response to the above technical problems, Chinese patent CN110048468A provides a method and device for distributing loads of units in a thermal power plant, which can determine whether the power plant needs to increase the load, reduce the load, or keep the original load unchanged based on the target load and actual load of the power plant. Then, the units are prioritized according to the operating cost data of each unit in the power plant and the load change direction of the power plant. The operating cost data of the unit and its first-order derivative are considered during the sorting to determine which unit should adjust the load first when increasing or reducing the load. Furthermore, based on the load change direction and rate of the power plant, the target load and load change rate of each unit after priority sorting are determined.

[0004] In the above-mentioned load distribution method for thermal power plant units, the operating cost data of each unit of the power plant adopts the operating cost curve, and the accuracy of the operating cost curve is crucial to priority sorting. The operating cost curve needs to be updated regularly to reflect changes in the unit status. If the operating cost curve is not updated in time or the fitting is not accurate enough, the accuracy of the priority sorting will be affected. This means that when the equipment status changes (for example, after technical transformation), the cost curve needs to be recalibrated to ensure the validity of the sorting results. If this is ignored, it may lead to unreasonable load distribution, which in turn affects the operating efficiency and economy of the power plant.

[0005] Therefore, an optimized load distribution scheme for power plant units is desired. Summary of the invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a power plant unit load distribution system and method, which uses a sensor monitoring system to monitor and collect data sets of key operating parameters of each unit in real time, and introduces a data processing and analysis algorithm based on artificial intelligence and deep learning at the back end to perform collaborative correlation analysis on the data sets of key operating parameters of these different units, so as to extract the operating status characteristics of each unit, and prioritize the units based on the operating status characteristics of these different units and the variable load direction of the thermal power plant. In this way, the priority ranking of each unit can be dynamically adjusted based on the operating status and variable load direction of each unit in the power plant, thereby realizing a more flexible and intelligent load distribution and scheduling of power plant units to cope with the ever-changing power market and user needs. In this way, the rationality and reliability of load distribution can be ensured, and the operating efficiency and economy of the power plant can be guaranteed.

[0007] According to one aspect of the present application, a method for distributing loads of power plant units is provided, comprising:

[0008] Determining the load change direction of the thermal power plant according to the target load of the thermal power plant and the actual load of the thermal power plant, wherein the load change direction includes increasing the load, decreasing the load and maintaining the original load;

[0009] According to the load variation direction of the thermal power plant, the units of the thermal power plant are prioritized;

[0010] According to the load change direction and load change rate of the thermal power plant, the target load and load change rate of each unit after priority sorting are determined.

[0011] According to another aspect of the present application, a power plant unit load distribution system is provided, comprising:

[0012] A load change direction determination module is used to determine the load change direction of the thermal power plant according to the target load of the thermal power plant and the actual load of the thermal power plant, wherein the load change direction includes increasing the load, decreasing the load and maintaining the original load;

[0013] A priority sorting module is used to prioritize the various units of the thermal power plant according to the load variation direction of the thermal power plant;

[0014] The target load and load change rate determination module is used to determine the target load and load change rate of each unit after priority sorting according to the load change direction and load change rate of the thermal power plant.

[0015] Compared with the prior art, the present application provides a power plant unit load distribution system and method, which uses a sensor monitoring system to monitor and collect data sets of key operating parameters of each unit in real time, and introduces data processing and analysis algorithms based on artificial intelligence and deep learning at the back end to perform collaborative correlation analysis on the data sets of key operating parameters of these different units, so as to extract the operating status characteristics of each unit, and prioritize the units based on the operating status characteristics of these different units and the variable load direction of the thermal power plant. In this way, the priority ranking of each unit can be dynamically adjusted based on the operating status and variable load direction of each unit in the power plant, thereby realizing a more flexible and intelligent load distribution and scheduling of power plant units to cope with the ever-changing power market and user needs. In this way, the rationality and reliability of load distribution can be ensured, and the operating efficiency and economy of the power plant can be guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1 is a flow chart of a method for distributing loads of power plant units according to an embodiment of the present application;

[0018] Figure 2 A data flow diagram of a method for distributing loads of power plant units according to an embodiment of the present application;

[0019] Figure 3 is a flowchart of sub-step S2 of the method for distributing loads of power plant units according to an embodiment of the present application;

[0020] Figure 4 is a flowchart of sub-step S22 of the method for distributing loads of power plant units according to an embodiment of the present application;

[0021] Figure 5 It is a block diagram of a load distribution system for power plant units according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0023] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0024] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0025] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0026] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0027] In the technical solution of the present application, a method for distributing loads of power plant units is proposed. Figure 1 The present invention is a flowchart of a method for distributing loads of power plant units according to an embodiment of the present application. Figure 2 FIG. 1 is a data flow diagram of a method for distributing loads of power plant units according to an embodiment of the present application. Figure 1 and Figure 2 As shown, the load distribution method for power plant units according to an embodiment of the present application includes the steps of: S1, determining the load change direction of the thermal power plant according to the target load of the thermal power plant and the actual load of the thermal power plant, wherein the load change direction includes increasing the load, reducing the load and maintaining the original load; S2, prioritizing the units of the thermal power plant according to the load change direction of the thermal power plant; S3, determining the target load and load change rate of each unit after priority sorting according to the load change direction of the thermal power plant and the load change rate of the thermal power plant.

[0028] In particular, the S1 determines the load change direction of the thermal power plant according to the target load of the thermal power plant and the actual load of the thermal power plant, and the load change direction includes increasing the load, reducing the load and maintaining the original load. That is, in one example, when the actual load of the thermal power plant is lower than the target load, it is necessary to increase the output of the generator set so that the actual load approaches or reaches the target load; conversely, when the actual load of the thermal power plant is higher than the target load, it is necessary to reduce the output of the generator set so that the actual load approaches or reaches the target load. When the actual load of the thermal power plant is basically consistent with the target load, the current load level is maintained unchanged. In this way, the load change direction of the thermal power plant can be effectively determined, further ensuring the rationality and reliability of load distribution, and ensuring the operating efficiency and economy of the power plant.

[0029] In particular, the S2 prioritizes the units in the thermal power plant according to the load variation direction of the thermal power plant. Figure 3 As shown, the S2 includes: S21, collecting a data set of key operating parameters of each unit through a sensor monitoring system, wherein the key operating parameters include temperature value, pressure value, vibration amplitude value, load value, fuel consumption rate and thermal efficiency; S22, extracting unit operating status characteristics from the data set of key operating parameters of each unit to obtain a set of unit operating status characteristics; S23, prioritizing the units based on the set of unit operating status characteristics and the variable load direction of the thermal power plant.

[0030] Specifically, S21 and S22 collect data sets of key operating parameters of each unit through a sensor monitoring system, wherein the key operating parameters include temperature value, pressure value, vibration amplitude value, load value, fuel consumption rate and thermal efficiency; and extract unit operating state characteristics from the data sets of key operating parameters of each unit to obtain a set of unit operating state characteristics. In a specific example of the present application, Figure 4 As shown, the S22 includes: S221, after regularizing the data sets of the key operating parameters of each unit into a set of multi-source tracking representation matrices of the unit operating status, performing timing feature extraction based on the operating status on the set of multi-source tracking representation matrices of the unit operating status to obtain a set of multi-source timing tracking feature graphs of the unit operating status; S222, performing feature enhancement processing based on local energy on each multi-source timing tracking feature graph of the unit operating status in the set of multi-source timing tracking feature graphs of the unit operating status to obtain the set of the unit operating status features.

[0031] More specifically, in S221, after the data sets of the key operating parameters of each unit are regularized into a set of multi-source tracking representation matrices of the unit operating status, the set of the multi-source tracking representation matrices of the unit operating status is subjected to time series feature extraction based on the operating status to obtain a set of multi-source time series tracking feature graphs of the unit operating status. Specifically, the data sets of the key operating parameters of each unit are first arranged into a set of multi-source tracking representation matrices of the unit operating status according to the time dimension and the parameter sample dimension; by arranging the key operating parameters according to the time dimension and the parameter sample dimension, the operating status of the unit can be monitored in real time and abnormal conditions can be discovered in time. Furthermore, each multi-source tracking representation matrix of the unit operating status in the set of the multi-source tracking representation matrix of the unit operating status is respectively input into the operating status time series feature extractor based on the deep separable convolutional neural network model to obtain a set of multi-source time series tracking feature graphs of the unit operating status. Here, each unit operating state multi-source tracking representation matrix in the set of the unit operating state multi-source tracking representation matrix is ​​respectively input into the operating state time series feature extractor based on the deep separable convolutional neural network model for feature mining, so as to respectively extract the time series correlation feature information of the key operating parameters of each unit in the time dimension, and these feature information can help understand the working state of each unit, including but not limited to load, efficiency, etc., to represent the time series dynamic semantics of the operating state of each unit, thereby obtaining a set of unit operating state multi-source time series tracking feature maps. It is worth noting that the deep separable convolutional neural network is an optimized convolutional neural network structure. The traditional convolution layer performs convolution operations on the input data through a set of filters (also called kernels), and each filter considers the spatial dimension (width and height) and the channel dimension at the same time. Although the traditional convolution layer is effective, it will lead to a large amount of computational requirements when processing high-resolution or multi-channel inputs. In contrast, the deep separable convolution decomposes the standard convolution into deep convolution and point-by-point convolution, and the deep convolution applies a single convolution kernel independently to each channel of the input. This means that if the input has C channels, then there will be C convolution kernels, each of which acts only on the corresponding single channel. This process retains the spatial information of the input, but reduces the information exchange across channels. Point-wise convolution uses a 1x1 convolution kernel to convolve the result of the depthwise convolution in order to mix the features between different channels.

[0032] More specifically, in S222, each of the multi-source timing tracking feature graphs of the unit operating status in the set of the multi-source timing tracking feature graphs of the unit operating status is subjected to feature enhancement processing based on local energy to obtain a set of the unit operating status features. It should be understood that since each of the multi-source timing tracking feature graphs of the unit operating status in the set of the multi-source timing tracking feature graphs of the unit operating status respectively represents the timing semantic information of the operating status of each unit, this timing semantics is obtained by performing timing collaborative correlation analysis on the multi-source key operating parameters of the unit, and for the unit, the timing correlation of different key operating parameters has different importance and contribution to characterizing the operating status of the unit, and these correlation information may also be interfered by noise, thereby reducing the performance of the multi-source timing tracking feature graph of the unit operating status in the subsequent unit priority sorting task. Therefore, in order to strengthen the more important features that are considered to be related to the subsequent unit priority sorting tasks, while suppressing the expression of less relevant features, thereby improving the robustness and effectiveness of feature representation, in the technical solution of the present application, each unit operating state multi-source time series tracking feature graph in the set of the unit operating state multi-source time series tracking feature graph is further subjected to feature enhancement processing based on local energy to obtain the set of unit operating state features. The feature enhancement processing method based on local energy can use the grid to standardize the attention window and imitate the attention structure based on the energy significance in the grid, so as to improve the expression ability and model performance of the unit operating state time series semantic features in the unit operating state multi-source time series tracking feature graph.

[0033] Specifically, first, the multi-source timing tracking feature graph of the unit operating status is grid-divided to obtain a set of local feature graphs of the multi-source timing semantics of the unit operating status; that is, the multi-source timing tracking feature graph of the unit operating status is decomposed into multiple local feature graphs through grid division, and each of the local feature graphs represents a small segment in the original multi-source timing tracking feature graph of the unit operating status. This division enables the model to focus on specific areas in the feature graph, which helps to identify and process the local features of the multi-source timing semantics of the unit operating status, and provide more refined control for subsequent feature enhancement and analysis. Next, the set of multi-source temporal semantic local feature maps of the unit operating state is subjected to energy significant description and enhancement processing to obtain a set of enhanced multi-source temporal semantic local feature maps of the unit operating state; that is, the statistical features of each multi-source temporal semantic local feature map of the unit operating state in the set of multi-source temporal semantic local feature maps of the unit operating state are used to calculate the energy significant description factor of each multi-source temporal semantic local feature map of the unit operating state, and the energy significant description factor reflects the significance or importance of the multi-source temporal semantic local feature map of the unit operating state in the global feature map. It should be understood that the calculation of the energy significant description factor helps to identify the key areas in the original multi-source temporal semantic local feature map of the unit operating state. That is, to identify which parts are more important or more worthy of attention, and these areas may contain information that has an important impact on the multi-source temporal semantic understanding of the unit operating state and the unit priority sorting decision task. These significant description factors are actually a quantification of the importance of the local feature map in the global context, so that the model can allocate attention resources more intelligently. Then, the set of the local energy significance descriptive factors of the multi-source temporal semantics of the unit operation state is input into the local feature adaptive selector based on the gating function for processing, so as to use the gating function to dynamically adjust the weight according to the energy significance of each local feature map of the multi-source temporal semantics of the unit operation state. This means that the model can self-regulate during the processing, strengthen those more important features related to the subsequent unit state recognition and priority sorting tasks, and suppress those less relevant features, so as to enhance the model's responsiveness to key information in the feature map, thereby improving the robustness and effectiveness of feature representation. Then, each local significant modulation weight of the multi-source temporal semantics of the unit operation state in the set of the local significant modulation weights of the multi-source temporal semantics of the unit operation state is used as a weight, and each local feature map of the multi-source temporal semantics of the unit operation state in the set of the local feature map of the multi-source temporal semantics of the unit operation state is weighted respectively, so as to further highlight and strengthen those highly significant local semantic features of the multi-source temporal semantics of the unit operation state, while reducing the influence of noise or irrelevant features.It should be understood that the essence of the weighted enhancement step is attention application, which enables the model to more effectively utilize the key action semantic information in the feature map, improves the discrimination and expressiveness of the feature representation, and helps to improve the performance and generalization ability of the model in subsequent tasks.

[0034] Among them, the process of performing energy significant description and enhancement processing on the set of multi-source temporal semantic local feature maps of the unit operating status to obtain a set of enhanced multi-source temporal semantic local feature maps of the unit operating status includes: calculating the energy significant description factor of each multi-source temporal semantic local feature map of the unit operating status in the set of multi-source temporal semantic local feature maps of the unit operating status to obtain a set of multi-source temporal semantic local energy significant description factors of the unit operating status; inputting the set of multi-source temporal semantic local energy significant description factors of the unit operating status into a local feature adaptive selector based on a gating function to obtain a set of multi-source temporal semantic local significant modulation weights of the unit operating status; using each multi-source temporal semantic local significant modulation weight of the unit operating status in the set of multi-source temporal semantic local significant modulation weights of the unit operating status as a weight, and weighting each multi-source temporal semantic local feature map of the unit operating status in the set of multi-source temporal semantic local feature maps of the unit operating status to obtain the set of enhanced multi-source temporal semantic local feature maps of the unit operating status. More specifically, the energy significance description factor of each unit operation state multi-source temporal semantic local feature graph in the set of the unit operation state multi-source temporal semantic local feature graph is calculated to obtain a set of unit operation state multi-source temporal semantic local energy significance description factors, including: extracting the maximum value and the minimum value of the unit operation state multi-source temporal semantic local feature graph respectively to obtain the maximum value of the unit operation state multi-source temporal semantic local feature and the minimum value of the unit operation state multi-source temporal semantic local feature; calculating the difference between the maximum value of the unit operation state multi-source temporal semantic local feature and the minimum value of the unit operation state multi-source temporal semantic local feature to obtain the difference value of the unit operation state multi-source temporal semantic local feature; calculating the unit operation state multi-source temporal semantic local feature difference value respectively. The global mean and variance of the multi-source temporal semantic local feature map of the operating status are used to obtain the mean of the multi-source temporal semantic local features of the unit operating status and the variance of the multi-source temporal semantic local features of the unit operating status; the value obtained by multiplying the variance of the multi-source temporal semantic local features of the unit operating status by a constant of two and the mean of the multi-source temporal semantic local features of the unit operating status are added to the hyperparameter to obtain the statistical modulation value of the multi-source temporal semantic local features of the unit operating status; the difference value of the multi-source temporal semantic local features of the unit operating status is divided by the statistical modulation value of the multi-source temporal semantic local features of the unit operating status to obtain the multi-source temporal semantic local energy significance description factor of the unit operating status corresponding to the multi-source temporal semantic local feature map of the unit operating status.More specifically, the set of the unit operation state multi-source temporal semantic local energy significance description factors is input into the local feature adaptive selector based on the gate function to obtain the set of the unit operation state multi-source temporal semantic local significance modulation weights, including: calculating an exponential function with the natural constant e as the base and the negative of each unit operation state multi-source temporal semantic local energy significance description factor in the set of the unit operation state multi-source temporal semantic local energy significance description factors as the exponent to obtain the set of unit operation state multi-source temporal semantic local energy significance description index factors; calculating the reciprocal of the sum of each unit operation state multi-source temporal semantic local energy significance description index factor in the set of the unit operation state multi-source temporal semantic local energy significance description index factors and the constant one to obtain the set of normalized unit operation state multi-source temporal semantic local energy significance description factors; and adding the normalized unit operation state multi-source temporal semantic local energy significance description factors to the set of the normalized unit operation state multi-source temporal semantic local energy significance description factors. Each normalized unit operating state multi-source temporal semantic local energy significance description factor is input into a gating function for masking processing to obtain a set of multi-source temporal semantic local significant modulation weights of the unit operating state; wherein, each normalized unit operating state multi-source temporal semantic local energy significance description factor in the set of normalized unit operating state multi-source temporal semantic local energy significance description factors is input into a gating function for masking processing to obtain a sequence of multi-source temporal semantic local significant modulation weights of the unit operating state, including: in response to a normalized unit operating state multi-source temporal semantic local energy significance description factor in the set of normalized unit operating state multi-source temporal semantic local energy significance description factors being greater than a predetermined threshold, taking the original value of the normalized unit operating state multi-source temporal semantic local energy significance description factor greater than the predetermined threshold, otherwise setting it to 0 to obtain a set of multi-source temporal semantic local significant modulation weights of the unit operating state.

[0035] More specifically, the process of performing feature aggregation and uniform dispersion processing on the set of the enhanced multi-source temporal semantic local feature graphs of the unit operation status to obtain the enhanced multi-source temporal tracking feature graph of the unit operation status as the unit operation status feature includes: first, performing feature aggregation on the set of the enhanced multi-source temporal semantic local feature graphs of the unit operation status according to the grid division method to obtain the multi-source temporal semantic local significant guidance enhanced feature graph of the unit operation status; performing feature aggregation on the set of the enhanced multi-source temporal semantic local feature graphs of the unit operation status according to the grid division method to form the multi-source temporal semantic local significant guidance enhanced feature graph of the unit operation status. In this way, it is ensured that the enhanced multi-source temporal semantic local feature graph of the unit operation status can be integrated back into a complete representation form, retaining the global information while enhancing the local details. Furthermore, the multi-source temporal semantic local significant guidance enhanced feature graph of the unit operation status is input into the feature uniform dispersion module based on the hole convolution layer to obtain the enhanced multi-source temporal tracking feature graph of the unit operation status. Here, considering that the attention granularity of this processing method is the grid granularity, there will be feature transition gaps at the grid boundary. In order to avoid the abnormal feature expression transition caused by grid division, the hole convolution is used to disperse the features, that is, the multi-source time series semantic local significant guidance enhanced feature map of the unit operation status is input into the feature dispersion module based on the hole convolution layer to obtain the enhanced multi-source time series tracking feature map of the unit operation status. The hole convolution can capture more contextual information without increasing the receptive field, thereby smoothing the feature expression differences between different grids and ensuring the consistency and coherence of the entire feature map.

[0036] In summary, in the above embodiment, each of the multi-source time series tracking feature graphs of the unit operating state in the set of the multi-source time series tracking feature graphs of the unit operating state is subjected to feature enhancement processing based on local energy to obtain the set of the unit operating state features, including: each of the multi-source time series tracking feature graphs of the unit operating state in the set of the multi-source time series tracking feature graphs of the unit operating state is subjected to feature enhancement processing based on local energy to obtain the set of the unit operating state features according to the following significant gating enhancement formula; wherein the significant gating enhancement formula is:

[0037]

[0038] F z =DilatedConv(F t )

[0039] Among them, F init is the multi-source time series tracking characteristic diagram of the unit operation status, Partition(·) is the grid partitioning operation, are the first, second, i-th, and m-th multi-source temporal semantic local feature maps of the unit operation status in the set of multi-source temporal semantic local feature maps of the unit operation status, respectively. max(·) and min(·) extract the maximum and minimum values ​​in the feature maps, respectively. and They are The mean and variance of , λ is a hyperparameter, ES i represents the local energy significance description factor of the multi-source temporal semantics of the operating state of the i-th unit corresponding to the multi-source temporal semantics local feature map, mask(·) is the masking process, w si is the local significant modulation weight of the multi-source temporal semantics of the operating state of the i-th unit corresponding to the multi-source temporal semantics local feature map of the operating state of the unit, θ is the predetermined threshold, Aggregate(·,...,·) is the feature aggregation operation for each weighted feature map, F t is the multi-source temporal semantic local significant guided enhanced feature map of the unit operation status, DilatedConv(·) is the dilated convolution operation, and F z It is the multi-source timing tracking characteristic diagram of the enhanced unit operation status.

[0040] Specifically, the S23, based on the set of the unit operating state characteristics and the variable load direction of the thermal power plant, prioritizes each unit. In the process of unit priority sorting, since each enhanced unit operating state multi-source time series tracking feature diagram in the set of the enhanced unit operating state multi-source time series tracking feature diagram respectively represents the multi-source time series characteristics of the unit operating state after the characteristics are significantly enhanced, in order to be able to perform a difference calculation based on the anchor reference center for the operating state time series semantics of each unit, so as to highlight the difference between the operating state and the operating state center characteristics of these different units, thereby realizing the current priority sorting of each unit. Specifically, first, a feature aggregation analysis of cluster contribution calculation is performed on the set of the enhanced unit operating state multi-source time series tracking feature diagram to obtain a pseudo-anchor reference feature diagram of the unit operating state; that is, a feature aggregation analysis of cluster contribution calculation is performed on the set of the enhanced unit operating state multi-source time series tracking feature diagram to obtain a pseudo-anchor reference feature diagram of the unit operating state. The feature aggregation analysis of cluster contribution calculation aims to use the simulated field analysis theory to perform explicit field-based modeling on the cluster contribution of each enhanced unit operating state multi-source timing tracking feature diagram to improve the accuracy of the feature aggregation analysis, thereby obtaining a more accurate unit operating state anchor reference center. Then, the cosine distance between each enhanced unit operating state multi-source timing tracking feature diagram in the set of enhanced unit operating state multi-source timing tracking feature diagrams and the unit operating state pseudo-anchor reference feature diagram is calculated to obtain a set of unit operating state difference coefficients. It should be understood that the unit operating state pseudo-anchor reference feature diagram, as a benchmark reference, can be regarded as the characteristic performance of each unit operating state under an ideal state. In order to measure the difference of each unit relative to the ideal state or reference state, so as to more accurately perceive and identify the operating state of each unit, so as to prioritize, in the technical solution of the present application, the cosine distance between each enhanced unit operating state multi-source timing tracking feature diagram in the set of the enhanced unit operating state multi-source timing tracking feature diagram and the unit operating state pseudo-anchor reference feature diagram is further calculated to obtain a set of unit operating state difference coefficients. In this way, the priority of each unit can be sorted based on the operating state characteristics of these different units and the load change direction of the thermal power plant. Finally, if the load change direction is to increase the load, the priority of each unit is sorted based on the forward sorting result of the set of the unit operating state difference coefficients; if the load change direction is to reduce the load, the priority of each unit is sorted based on the reverse sorting result of the set of the unit operating state difference coefficients; if the load change direction is to maintain the original load, the current priority of each unit is maintained.In this way, the priority ranking of each unit can be dynamically adjusted based on the operating status and load change direction of each unit in the power plant, thereby achieving more flexible and intelligent load distribution and scheduling of power plant units to cope with the ever-changing electricity market and user needs. In this way, the rationality and reliability of load distribution can be ensured, and the operating efficiency and economy of the power plant can be guaranteed.

[0041] Among them, the process of performing feature aggregation analysis of cluster contribution calculation on the set of enhanced unit operating status multi-source timing tracking feature graphs to obtain the unit operating status pseudo-anchor reference feature graph includes: first, performing global mean pooling on each feature matrix of each enhanced unit operating status multi-source timing tracking feature graph along the channel dimension in the set of enhanced unit operating status multi-source timing tracking feature graphs to obtain a set of enhanced unit operating status multi-source timing tracking global pooling feature vectors; then, performing cluster analysis on the set of enhanced unit operating status multi-source timing tracking global pooling feature vectors to obtain enhanced unit operating status multi-source timing correlation feature self-supervised clustering feature vectors; here, an unsupervised learning method is used to perform cluster analysis on the set of enhanced unit operating status multi-source timing tracking global pooling feature vectors to automatically discover the patterns and structures in the multi-source timing features of each unit operating status to generate enhanced unit operating status multi-source timing correlation feature self-supervised clustering feature vectors. It should be understood that self-supervised clustering can reveal the intrinsic structure of the data, help the model learn meaningful feature representations without explicit labels, and enhance the model's understanding of the distribution of data features. Then, based on the self-supervised clustering feature vector of the enhanced unit operating state multi-source time series correlation feature, the set of the enhanced unit operating state multi-source time series tracking feature graphs is subjected to feature clustering field modulation aggregation to obtain the unit operating state multi-source time series significant aggregation representation vector; in the technical solution of the present application, by calculating the implicit clustering contribution factor of each enhanced unit operating state multi-source time series tracking global pooling feature vector relative to the enhanced unit operating state multi-source time series correlation feature self-supervised clustering feature vector, the implicit clustering contribution factor reflects the contribution of each enhanced unit operating state multi-source time series tracking global pooling feature vector to the cluster to which it belongs. Accordingly, the calculation of the implicit clustering contribution factor helps to identify and quantify the role of each unit operating state multi-source time series feature in the clustering process, and provides a basis for subsequent feature selection and weight allocation, so that the model can pay more attention to the unit operating state features that contribute significantly to the clustering. Next, the sequence of implicit cluster contribution factors of the multi-source time series correlation characteristics of the enhanced unit operation status calculated is converted into a cluster contribution field distribution vector, that is, the sequence of implicit cluster contribution factors is integrated into the vector field by data integration to construct the cluster contribution field. Then, the self-attention mechanism is used to explicitly model the distribution vector of the cluster contribution field of the multi-source time series correlation characteristics of the enhanced unit operation status to generate the cluster contribution field modulation weight vector of the multi-source time series correlation characteristics of the enhanced unit operation status, that is, the self-attention mechanism is used to explicitly model the cluster contribution of each fine-grained factor in the cluster contribution field based on the field autocorrelation global analysis, so that the attention to the multi-source time series correlation characteristics of the unit operation status with large cluster contribution can be strengthened, while suppressing unimportant features, improving the quality of feature representation and the prediction accuracy of the model.Then, the set of the enhanced unit operating state multi-source time series tracking global pooling feature vectors is weighted fused using the enhanced unit operating state multi-source time series correlation feature clustering contribution field modulation weight vector to generate the unit operating state multi-source time series significant aggregation representation vector. This optimized feature aggregation representation not only retains the time series feature information of the original unit operating state, but also incorporates the knowledge learned in the clustering process, so that the model can more accurately capture the key features in the data, thereby improving the performance and generalization ability of the model. Finally, the multi-source time series significant aggregation representation vector of the unit operating state is dimensionally reconstructed to obtain the pseudo-anchor reference feature map of the unit operating state. In this way, the clustering contribution analysis of multiple units can be used to obtain the pseudo-anchor reference feature map of the unit operating state with the meaning of the anchor reference center, so that the differences between the operating state features of each unit and the anchor reference center features can be used in the subsequent processing process to more accurately perceive and identify the operating state of each unit, providing a basis for the priority sorting task.

[0042] Among them, the process of performing cluster analysis on the set of global pooling feature vectors of multi-source timing tracking of enhanced unit operating status to obtain self-supervised clustering feature vectors of multi-source timing correlation characteristics of enhanced unit operating status includes: calculating the mean vector of the set of global pooling feature vectors of multi-source timing tracking of enhanced unit operating status to obtain self-supervised clustering feature vectors of multi-source timing correlation characteristics of enhanced unit operating status.

[0043] More specifically, based on the self-supervised clustering feature vector of the enhanced unit operation state multi-source time series correlation feature, the process of performing feature clustering field modulation aggregation on the set of the enhanced unit operation state multi-source time series tracking feature graphs to obtain the unit operation state multi-source time series significant aggregation representation vector includes: calculating the implicit clustering contribution factor of each enhanced unit operation state multi-source time series tracking global pooling feature vector in the set of the enhanced unit operation state multi-source time series tracking global pooling feature vector relative to the enhanced unit operation state multi-source time series correlation feature self-supervised clustering feature vector to obtain a sequence of implicit clustering contribution factors of the enhanced unit operation state multi-source time series correlation feature; The sequence of implicit clustering contribution factors of sequence-related features is vectorized to obtain the distribution vector of the clustering contribution field of the multi-source time series correlation features of the enhanced unit operation status; the distribution vector of the clustering contribution field of the multi-source time series correlation features of the enhanced unit operation status is input into the clustering contribution field explicit modeling module based on the self-attention mechanism to obtain the modulation weight vector of the clustering contribution field of the multi-source time series correlation features of the enhanced unit operation status; each eigenvalue in the modulation weight vector of the clustering contribution field of the multi-source time series correlation features of the enhanced unit operation status is used as a weight to calculate the position-weighted sum of the set of global pooling feature vectors of the multi-source time series tracking of the enhanced unit operation status to obtain the multi-source time series significant aggregation representation vector of the unit operation status. More specifically, the implicit clustering contribution factor of each enhanced unit operating state multi-source time series tracking global pooling feature vector in the set of the enhanced unit operating state multi-source time series tracking global pooling feature vector is calculated relative to the enhanced unit operating state multi-source time series correlation feature self-supervised clustering feature vector to obtain a sequence of implicit clustering contribution factors of the enhanced unit operating state multi-source time series correlation feature, including: calculating the absolute value of the positional division between the enhanced unit operating state multi-source time series tracking global pooling feature vector and the enhanced unit operating state multi-source time series correlation feature self-supervised clustering feature vector to obtain the enhanced unit operating state multi-source time series self-supervised clustering interactive correlation feature vector; calculating the enhanced unit operating state with base 2 as the base The logarithmic function value of each position feature value in the state multi-source time series self-supervised clustering interactive correlation feature vector is used to obtain the enhanced unit operation state multi-source time series self-supervised clustering interactive logarithmic representation vector; using each position feature value in the enhanced unit operation state multi-source time series self-supervised clustering interactive logarithmic representation vector as the weighting coefficient, the weighted sum of each position feature value in the enhanced unit operation state multi-source time series tracking global pooling feature vector is calculated to obtain the enhanced unit operation state multi-source time series correlation feature clustering contribution representation factor; calculate the exponential function value of the enhanced unit operation state multi-source time series correlation feature clustering contribution representation factor with the natural constant e as the base to obtain the enhanced unit operation state multi-source time series correlation feature implicit clustering contribution factor.More specifically, the enhanced unit operation state multi-source time series correlation feature cluster contribution field distribution vector is input into the cluster contribution field explicit modeling module based on the self-attention mechanism to obtain the enhanced unit operation state multi-source time series correlation feature cluster contribution field domain modulation weight vector, including: calculating the matrix multiplication between the enhanced unit operation state multi-source time series correlation feature cluster contribution field distribution vector and the query weight matrix, the key weight matrix and the value weight matrix respectively to obtain the unit operation state multi-source time series cluster contribution field distribution query vector, the unit operation state multi-source time series cluster contribution field distribution key vector and the unit operation state multi-source time series cluster contribution field distribution value vector; calculating the unit operation state multi-source time series cluster contribution field distribution query vector and the unit operation state multi-source time series cluster contribution field distribution key vector. After multiplying the transpose vectors of the vectors, the obtained matrix is ​​divided by the square root of the length of the key vector of the multi-source time series clustering contribution field distribution of the unit operation status by position to obtain the query-key unit operation status multi-source time series clustering contribution field semantic interaction representation matrix; the query-key unit operation status multi-source time series clustering contribution field semantic interaction representation matrix is ​​normalized by using the Softmax function to obtain the normalized query-key unit operation status multi-source time series clustering contribution field semantic interaction representation matrix; the multiplication between the distribution value vector of the multi-source time series clustering contribution field of the unit operation status and the query-key unit operation status multi-source time series clustering contribution field semantic interaction representation matrix is ​​calculated to obtain the enhanced unit operation status multi-source time series correlation feature clustering contribution field domain modulation weight vector.

[0044] In summary, in the above embodiment, the set of the enhanced multi-source time series tracking feature graphs of the unit operation state is subjected to feature aggregation analysis of cluster contribution calculation to obtain the pseudo-anchor reference feature graph of the unit operation state, including: performing cluster analysis and feature clustering field modulation aggregation on the set of global pooling feature vectors of the multi-source time series tracking of the unit operation state using the following feature aggregation formula to obtain the multi-source time series significant aggregation representation vector of the unit operation state; wherein the feature aggregation formula is:

[0045] X={x1,x2,...,x k ,...,x n}

[0046]

[0047] v d ={d1; d2; ...; d j ;...;d n}

[0048]

[0049] v q =W q vd

[0050] v k =W k v d

[0051] v v =W b v d

[0052]

[0053] Where X is the set of global pooling feature vectors for multi-source time series tracking of the enhanced unit operation status, x1, x2, x k ,x n are respectively the first, second, kth and nth enhanced unit operation state multi-source time series tracking global pooling feature vectors in the set of the enhanced unit operation state multi-source time series tracking global pooling feature vectors, x i is the ith enhanced unit operation state multi-source time series tracking global pooling feature vector in the set of enhanced unit operation state multi-source time series tracking global pooling feature vectors, N is the number of feature graphs in the set of enhanced unit operation state multi-source time series tracking global pooling feature vectors, x c In order to enhance the self-supervised clustering feature vector of multi-source time series correlation features of unit operation status, is the feature value of the kth position of the jth enhanced unit operation state multi-source time series tracking global pooling feature vector in the set of the enhanced unit operation state multi-source time series tracking global pooling feature vectors, is the eigenvalue of the kth position in the self-supervised clustering feature vector of the multi-source time series correlation feature of the enhanced unit operation status, exp represents the value of the natural exponential function, d1, d2, d j ,d n are the implicit clustering contribution factors of the multi-source time series correlation features of the enhanced unit operating status corresponding to the first, second, jth, and nth enhanced unit operating status multi-source time series tracking global pooling feature vectors, v d is the distribution vector of the clustering contribution field of the multi-source time series correlation characteristics of the operation status of the enhanced units obtained by quantizing the implicit clustering contribution factor vectors of the multi-source time series correlation characteristics of the operation status of the enhanced units, W q , W k and W b are the query weight matrix, key weight matrix and value weight matrix respectively, v q 、v k and v vare the query vector of the multi-source time series clustering contribution field distribution of the unit operation status, the key vector of the multi-source time series clustering contribution field distribution of the unit operation status, and the value vector of the multi-source time series clustering contribution field distribution of the unit operation status, L is the length of the key vector of the multi-source time series clustering contribution field distribution of the unit operation status, Softmax(·) is the Softmax function, is vector multiplication, v w To enhance the clustering of multi-source temporal correlation features of unit operation status, the contribution field modulation weight vector, x f is a multi-source time series significant aggregation representation vector of the unit operation status.

[0054] In the technical solution of the present application, each enhanced unit operating state multi-source time series tracking feature graph in the set of enhanced unit operating state multi-source time series tracking feature graphs respectively represents the local implicit enhanced correlation features of the key operating parameters of each unit across sample dimensions and time dimensions. In this way, when the set of enhanced unit operating state multi-source time series tracking feature graphs is input into the feature significant aggregation network based on cluster contribution field analysis, the difference in cluster contribution of each enhanced unit operating state multi-source time series tracking feature graph will make the unit operating state pseudo-anchor reference feature graph have a complex spatial structure based on different spatial significance distribution structures. Furthermore, this will cause each enhanced unit operating state multi-source time series tracking feature graph in the set of enhanced unit operating state multi-source time series tracking feature graphs to have a spatial structure distribution difference with the unit operating state pseudo-anchor reference feature graph, thereby affecting the calculation accuracy of the unit operating state difference coefficient.

[0055] Based on this, in a preferred example, before calculating the cosine distance between each enhanced unit operating state multi-source timing tracking feature graph in the set of enhanced unit operating state multi-source timing tracking feature graphs and the unit operating state pseudo-anchor reference feature graph to obtain a set of unit operating state difference coefficients, it also includes: performing spatial distribution structure optimization on the unit operating state pseudo-anchor reference feature graph to obtain an optimized unit operating state pseudo-anchor reference feature graph.

[0056] Specifically, the spatial distribution structure of the pseudo-anchor reference feature map of the unit operation status is optimized, including:

[0057] Determine the mean and variance of the feature set composed of the feature values ​​of the pseudo-anchor reference feature graph of the unit operation status, and divide the mean by the variance to obtain the pseudo-anchor reference probability weight value of the unit operation status:

[0058]

[0059] Among them, μ and σ 2represent the mean and variance of the feature set composed of the feature values ​​of the pseudo-anchor reference feature graph of the unit operation status respectively, and θ represents the pseudo-anchor reference probability weight value of the unit operation status;

[0060] The pseudo-anchor reference feature vector of the unit operating state obtained by expanding the pseudo-anchor reference feature map of the unit operating state is multiplied by the inverse of the maximum eigenvalue of the pseudo-anchor reference feature map of the unit operating state to obtain the pseudo-anchor reference probability constraint vector of the unit operating state:

[0061] V1=V⊙f max -1

[0062] Among them, f max -1 represents the inverse of the maximum eigenvalue of the pseudo-anchor reference feature map of the unit operation state, V represents the pseudo-anchor reference feature vector of the unit operation state, ⊙ represents the point multiplication by position, and V1 represents the pseudo-anchor reference probability constraint vector of the unit operation state;

[0063] The pseudo-anchor reference probability weight value of the unit operating state is added to the pseudo-anchor reference probability constraint vector of the unit operating state, and the logarithmic value with base 2 is calculated to obtain the pseudo-anchor reference information interaction vector of the unit operating state:

[0064] V2=log2(V1⊕θ)

[0065] Among them, V1 represents the pseudo-anchor reference probability constraint vector of the unit operation status, θ represents the pseudo-anchor reference probability weight value of the unit operation status, ⊕ represents adding by position point, and V2 represents the pseudo-anchor reference information interaction vector of the unit operation status;

[0066] After subtracting the pseudo-anchor reference probability constraint vector of the unit operating state from the pseudo-anchor reference probability weight value of the unit operating state, the inverse of each eigenvalue is calculated to obtain the pseudo-anchor reference sequence constraint vector of the unit operating state:

[0067]

[0068] Among them, V1 represents the pseudo-anchor reference probability constraint vector of the unit operation status, θ represents the pseudo-anchor reference probability weight value of the unit operation status, represents the subtraction according to the position point, V3 represents the pseudo-anchor reference sequence constraint vector of the unit operation status;

[0069] Finally, the unit operating state pseudo-anchor reference information interaction vector point is added to the unit operating state pseudo-anchor reference sequence constraint vector to obtain an optimized unit operating state pseudo-anchor reference feature vector, and the optimized unit operating state pseudo-anchor reference feature vector is restored to an optimized unit operating state pseudo-anchor reference feature graph.

[0070] Accordingly, in the preferred example, the probability statistical characteristics of the pseudo-anchor reference feature map of the unit operating state are used to simulate the mesoscale interaction structure under probability constraints between the eigenvalue scale and the eigenvector scale of the pseudo-anchor reference feature vector of the unit operating state, so as to construct a bidirectional hidden variable motif based on the mesoscale short sequence relative to the probability statistical value to perform mesoscale bidirectional migration, and perform posterior recovery based on the sequence constraint of the interactive information to ensure the spatial transformation (translation, scaling and rotation) invariance of the pseudo-anchor reference feature map of the unit operating state under the interaction of the feature space, thereby improving the spatial structure distribution homogeneity between the pseudo-anchor reference feature map of the unit operating state and each enhanced unit operating state multi-source time series tracking feature map in the set of the enhanced unit operating state multi-source time series tracking feature map, thereby improving the calculation accuracy of the unit operating state difference coefficient. In this way, the priority ranking of each unit can be dynamically adjusted based on the operating state and variable load direction of each unit in the power plant, so as to achieve more flexible and intelligent load distribution and scheduling of power plant units to cope with the ever-changing power market and user needs.

[0071] In particular, the S3 determines the target load and load change rate of each unit after priority sorting according to the load change direction and load change rate of the thermal power plant. It should be understood that after the units are prioritized according to the load change direction and load change rate, the target load and load change rate of each unit are determined, which helps to ensure the stability and economy of load adjustment and the stable operation of the power grid. In one example, the units with high thermal efficiency, fast response speed and short start-up time are set to the load increase direction; according to the priority order, the load of each unit is gradually increased until the total load demand is met, and a target load value is set for each unit to ensure a smooth load increase process. On the contrary, the units with low thermal efficiency, fast response speed and high fuel consumption rate are set to the load reduction direction; according to the priority order, the load of each unit is gradually reduced until the total load demand is met, and a target load value is set for each unit to ensure a smooth load reduction process. Further, according to the power grid demand and the current state of the unit, a reasonable load change rate is set. Ensure that the load change rate does not exceed the maximum load change rate of the unit to avoid damage to the equipment. Dynamically adjust the load change rate according to actual operating conditions and changes in grid demand.

[0072] In summary, the load distribution method of the power plant units according to the embodiment of the present application is explained, which uses a sensor monitoring system to monitor and collect data sets of key operating parameters of each unit in real time, and introduces data processing and analysis algorithms based on artificial intelligence and deep learning at the back end to perform collaborative correlation analysis on the data sets of key operating parameters of these different units, so as to extract the operating status characteristics of each unit, and prioritize the units based on the operating status characteristics of these different units and the variable load direction of the thermal power plant. In this way, the priority ranking of each unit can be dynamically adjusted based on the operating status and variable load direction of each unit in the power plant, thereby realizing a more flexible and intelligent load distribution and scheduling of power plant units to cope with the ever-changing power market and user needs. In this way, the rationality and reliability of load distribution can be ensured, and the operating efficiency and economy of the power plant can be guaranteed.

[0073] Furthermore, a load distribution system for power plant units is also provided.

[0074] Figure 5 FIG. 1 is a block diagram of a load distribution system for a power plant unit according to an embodiment of the present application. Figure 5 As shown, according to the power plant unit load distribution system 300 of the embodiment of the present application, it includes: a variable load direction determination module 310, which is used to determine the variable load direction of the thermal power plant according to the target load of the thermal power plant and the actual load of the thermal power plant, and the variable load direction includes increasing load, reducing load and maintaining original load; a priority sorting module 320, which is used to prioritize the units of the thermal power plant according to the variable load direction of the thermal power plant; a target load and variable load rate determination module 330, which is used to determine the target load and variable load rate of each unit after priority sorting according to the variable load direction of the thermal power plant and the variable load rate of the thermal power plant.

[0075] As described above, the power plant unit load distribution system 300 according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with a power plant unit load distribution algorithm. In a possible implementation, the power plant unit load distribution system 300 according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the power plant unit load distribution system 300 can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the power plant unit load distribution system 300 can also be one of the many hardware modules of the wireless terminal.

[0076] Alternatively, in another example, the power plant unit load distribution system 300 and the wireless terminal may also be separate devices, and the power plant unit load distribution system 300 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0077] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for distributing loads of power plant units, characterized in that: include: Determining the load change direction of the thermal power plant according to the target load of the thermal power plant and the actual load of the thermal power plant, wherein the load change direction includes increasing the load, decreasing the load and maintaining the original load; According to the load variation direction of the thermal power plant, the units of the thermal power plant are prioritized; According to the load change direction and load change rate of the thermal power plant, the target load and load change rate of each unit after priority sorting are determined; Among them, according to the load change direction of the thermal power plant, the units of the thermal power plant are prioritized, including: Collecting data sets of key operating parameters of each unit through a sensor monitoring system, wherein the key operating parameters include temperature value, pressure value, vibration amplitude value, load value, fuel consumption rate and thermal efficiency; Extracting unit operating status characteristics from the data sets of key operating parameters of each unit to obtain a set of unit operating status characteristics; Based on the set of operating status characteristics of the units and the load variation direction of the thermal power plant, the units are prioritized.

2. The method for distributing loads of power plant units according to claim 1, characterized in that: Extracting unit operation status characteristics from the data set of key operation parameters of each unit to obtain a set of unit operation status characteristics, including: After the data sets of the key operating parameters of each unit are regularized into a set of multi-source tracking representation matrices of the unit operating status, the set of the multi-source tracking representation matrices of the unit operating status is subjected to time series feature extraction based on the operating status to obtain a set of multi-source time series tracking feature graphs of the unit operating status; Each of the multi-source time series tracking feature graphs of the unit operating status in the set of the multi-source time series tracking feature graphs of the unit operating status is subjected to feature enhancement processing based on local energy to obtain a set of the unit operating status features, including: gridding the multi-source time series tracking feature graphs of the unit operating status to obtain a set of multi-source time series semantic local feature graphs of the unit operating status; performing energy significant description and enhancement processing on the set of multi-source time series semantic local feature graphs of the unit operating status to obtain a set of enhanced multi-source time series semantic local feature graphs of the unit operating status; performing feature aggregation and uniform dispersion processing on the set of enhanced multi-source time series semantic local feature graphs of the unit operating status to obtain an enhanced multi-source time series tracking feature graph of the unit operating status as the unit operating status features.

3. The method for distributing loads of power plant units according to claim 2, characterized in that: After the data sets of the key operating parameters of each unit are regularized into a set of multi-source tracking representation matrices of the unit operating status, the set of the multi-source tracking representation matrices of the unit operating status is subjected to time series feature extraction based on the operating status to obtain a set of multi-source time series tracking feature graphs of the unit operating status, including: Arranging the data sets of the key operating parameters of each unit according to the time dimension and the parameter sample dimension into a set of multi-source tracking representation matrices of the unit operating status; Each unit operating status multi-source tracking representation matrix in the set of the unit operating status multi-source tracking representation matrix is ​​respectively input into the operating status time series feature extractor based on the deep separable convolutional neural network model to obtain the set of the unit operating status multi-source time series tracking feature graphs.

4. The method for distributing loads of power plant units according to claim 3, characterized in that: The set of multi-source temporal semantic local feature graphs of the unit operation status is subjected to energy significant description and enhancement processing to obtain a set of enhanced multi-source temporal semantic local feature graphs of the unit operation status, including: Calculating the energy significance description factor of each unit operation state multi-source temporal semantic local feature graph in the set of unit operation state multi-source temporal semantic local feature graphs to obtain a set of unit operation state multi-source temporal semantic local energy significance description factors; Inputting the set of local energy significance description factors of multi-source temporal semantics of the unit operation status into a local feature adaptive selector based on a gating function to obtain a set of local significance modulation weights of multi-source temporal semantics of the unit operation status; Taking each unit operating state multi-source temporal semantic local significant modulation weight in the set of the unit operating state multi-source temporal semantic local significant modulation weights as the weight, each unit operating state multi-source temporal semantic local feature map in the set of the unit operating state multi-source temporal semantic local feature map is weighted respectively to obtain the set of enhanced unit operating state multi-source temporal semantic local feature maps.

5. The method for distributing loads of power plant units according to claim 4, characterized in that: Calculating the energy significance description factor of each unit operation state multi-source time series semantic local feature graph in the set of the unit operation state multi-source time series semantic local feature graph to obtain a set of unit operation state multi-source time series semantic local energy significance description factors, including: Respectively extracting the maximum value and the minimum value of the multi-source temporal semantic local feature map of the unit operation status to obtain the maximum value of the multi-source temporal semantic local feature of the unit operation status and the minimum value of the multi-source temporal semantic local feature of the unit operation status; Calculate the difference between the maximum value of the multi-source temporal semantic local feature of the unit operation state and the minimum value of the multi-source temporal semantic local feature of the unit operation state to obtain the multi-source temporal semantic local feature difference value of the unit operation state; Respectively calculating the global mean and variance of the multi-source temporal semantic local feature map of the unit operation status to obtain the multi-source temporal semantic local feature mean and the multi-source temporal semantic local feature variance of the unit operation status; The value obtained by multiplying the variance of the multi-source temporal semantic local feature of the unit operation status by a constant of two and the mean value of the multi-source temporal semantic local feature of the unit operation status and the hyperparameter are added to obtain the statistical modulation value of the multi-source temporal semantic local feature of the unit operation status; The difference value of the multi-source temporal semantic local feature of the unit operation status is divided by the statistical modulation value of the multi-source temporal semantic local feature of the unit operation status to obtain the multi-source temporal semantic local energy significance description factor of the unit operation status corresponding to the multi-source temporal semantic local feature map of the unit operation status.

6. The method for distributing loads of power plant units according to claim 5, characterized in that: The set of the enhanced unit operation state multi-source time series semantic local feature graphs is subjected to feature aggregation and uniform dispersion processing to obtain an enhanced unit operation state multi-source time series tracking feature graph as the unit operation state feature, including: Performing feature aggregation on the set of the enhanced unit operation status multi-source temporal semantic local feature maps according to the grid division method to obtain the unit operation status multi-source temporal semantic local significant guidance enhanced feature map; The multi-source time series semantic local significant guided enhanced feature map of the unit operation status is input into a feature dispersion module based on a hole convolution layer to obtain the enhanced multi-source time series tracking feature map of the unit operation status.

7. The method for distributing loads of power plant units according to claim 6, characterized in that: Based on the set of the operating state characteristics of the units and the load variation direction of the thermal power plant, the units are prioritized, including: Performing feature aggregation analysis of cluster contribution calculation on the set of enhanced multi-source time series tracking feature graphs of unit operation status to obtain a pseudo-anchor reference feature graph of unit operation status; Calculating the cosine distance between each enhanced unit operating state multi-source time series tracking feature graph in the set of enhanced unit operating state multi-source time series tracking feature graphs and the unit operating state pseudo-anchor reference feature graph to obtain a set of unit operating state difference coefficients; If the load change direction is to increase the load, the priority of each unit is sorted based on the forward sorting result of the set of the unit operation status difference coefficients; if the load change direction is to reduce the load, the priority of each unit is sorted based on the reverse sorting result of the set of the unit operation status difference coefficients; if the load change direction is to maintain the original load, the current priority of each unit is maintained; The feature aggregation analysis of cluster contribution calculation is performed on the set of the enhanced multi-source time series tracking feature graphs of the unit operation status to obtain the pseudo-anchor reference feature graph of the unit operation status, including: Performing global mean pooling on each feature matrix along the channel dimension of each enhanced unit operation state multi-source time series tracking feature graph in the set of enhanced unit operation state multi-source time series tracking feature graphs to obtain a set of enhanced unit operation state multi-source time series tracking global pooling feature vectors; Performing cluster analysis on the set of global pooling feature vectors of the enhanced unit operation status multi-source time series tracking to obtain a self-supervised clustering feature vector of the enhanced unit operation status multi-source time series correlation feature; Based on the self-supervised clustering feature vector of the multi-source time series correlation feature of the enhanced unit operation state, a set of the multi-source time series tracking feature graphs of the enhanced unit operation state is subjected to feature clustering field modulation aggregation to obtain a multi-source time series significant aggregation representation vector of the unit operation state; The multi-source time series significant aggregation representation vector of the unit operation status is dimensionally reconstructed to obtain a pseudo-anchor reference feature map of the unit operation status.

8. The method for distributing loads of power plant units according to claim 7, characterized in that: Based on the self-supervised clustering feature vector of the multi-source time series correlation feature of the enhanced unit operation state, a set of the multi-source time series tracking feature graphs of the enhanced unit operation state is subjected to feature clustering field modulation aggregation to obtain a multi-source time series significant aggregation representation vector of the unit operation state, including: Calculate the implicit clustering contribution factor of each enhanced unit operation state multi-source time series tracking global pooling feature vector in the set of the enhanced unit operation state multi-source time series tracking global pooling feature vector relative to the enhanced unit operation state multi-source time series correlation feature self-supervised clustering feature vector to obtain a sequence of enhanced unit operation state multi-source time series correlation feature implicit clustering contribution factors; Vectorizing and arranging the sequence of implicit clustering contribution factors of the multi-source time series correlation characteristics of the enhanced unit operation status to obtain a clustering contribution field distribution vector of the multi-source time series correlation characteristics of the enhanced unit operation status; Inputting the distribution vector of the cluster contribution field of the multi-source temporal correlation feature of the enhanced unit operation state into the cluster contribution field explicit modeling module based on the self-attention mechanism to obtain the modulation weight vector of the cluster contribution field of the multi-source temporal correlation feature of the enhanced unit operation state; Taking each eigenvalue in the field modulation weight vector of the clustering contribution of the multi-source time series correlation feature of the enhanced unit operating status as the weight, the position-weighted sum of the set of global pooling feature vectors of the enhanced multi-source time series tracking of the unit operating status is calculated to obtain the multi-source time series significant aggregation representation vector of the unit operating status.

9. The method for distributing loads of power plant units according to claim 8, characterized in that: Calculating the implicit clustering contribution factor of each enhanced unit operating state multi-source time series tracking global pooling feature vector in the set of the enhanced unit operating state multi-source time series tracking global pooling feature vector relative to the enhanced unit operating state multi-source time series correlation feature self-supervised clustering feature vector to obtain a sequence of enhanced unit operating state multi-source time series correlation feature implicit clustering contribution factors, including: Calculating the absolute value of the positional division between the enhanced unit operation state multi-source time series tracking global pooling feature vector and the enhanced unit operation state multi-source time series correlation feature self-supervised clustering feature vector to obtain the enhanced unit operation state multi-source time series self-supervised clustering interactive correlation feature vector; Calculate the logarithmic function value of each position feature value in the enhanced unit operation state multi-source time series self-supervised clustering interactive correlation feature vector with base 2 to obtain the enhanced unit operation state multi-source time series self-supervised clustering interactive logarithmic representation vector; Taking each position feature value in the enhanced unit operation state multi-source time series self-supervised clustering interactive logarithm representation vector as a weighting coefficient, calculating the weighted sum of each position feature value in the enhanced unit operation state multi-source time series tracking global pooling feature vector to obtain the enhanced unit operation state multi-source time series correlation feature clustering contribution representation factor; The exponential function value of the clustering contribution degree representation factor of the multi-source time series correlation feature of the enhanced unit operation status is calculated with the natural constant e as the base to obtain the implicit clustering contribution factor of the multi-source time series correlation feature of the enhanced unit operation status.

10. A load distribution system for power plant units, characterized in that: include: A load change direction determination module is used to determine the load change direction of the thermal power plant according to the target load of the thermal power plant and the actual load of the thermal power plant, wherein the load change direction includes increasing the load, decreasing the load and maintaining the original load; A priority sorting module is used to prioritize the various units of the thermal power plant according to the load variation direction of the thermal power plant; The target load and load change rate determination module is used to determine the target load and load change rate of each unit after priority sorting according to the load change direction and load change rate of the thermal power plant.

Images