A Method and System for Predicting Hydropower Generation Capacity of Hydropower Stations Based on Hydrological Information

Abstract

This invention provides a method and system for predicting the power generation of hydropower stations based on hydrological information, belonging to the field of hydropower generation prediction technology. First, a set of dynamic hydrological interaction information of the hydropower station is obtained. Then, a hydrological interaction transmission network is constructed based on this set, with state change records as nodes and mutual triggering relationships as network edges. The edge attributes are determined by the temporal continuity and influence range of the triggering relationships. Next, key transmission nodes related to power generation and their corresponding network edge attributes are extracted from the hydrological interaction transmission network to generate a power response correlation factor sequence. Based on the power response correlation factor sequence and the historical power generation time-series records of the hydropower station, a power response correlation rule base is constructed. Finally, the currently generated power response correlation factor sequence is matched with the rule base to generate the power generation prediction result of the hydropower station for the current period, significantly improving the accuracy and reliability of hydropower generation prediction.

Description

Technical Field

[0001] This invention relates to the field of hydropower generation prediction technology, and more specifically, to a method and system for predicting hydropower generation based on hydrological information. Background Technology

[0002] In the field of hydropower station operation and management, accurate prediction of power generation is crucial for optimizing power resource allocation, ensuring stable grid operation, and improving the economic benefits of hydropower stations. Traditional methods for predicting hydropower generation mostly focus on considering single hydrological factors, such as making simple predictions based solely on historical data of upstream water flow or estimating power generation solely based on changes in reservoir water levels.

[0003] However, the hydrological system of a hydropower station is a complex and dynamically interacting whole. The timing information of upstream water inflow not only affects the inflow into the reservoir but also has a cascading effect on the reservoir's water level; fluctuations in the reservoir's water level alter the water flow velocity, thus affecting downstream water demand; and the response of downstream water demand, in turn, influences the reservoir's water storage and release strategies, ultimately impacting power generation. Existing forecasting methods, lacking a comprehensive and detailed analysis of the dynamic interactions of hydrological information, fail to fully capture the mutual triggering relationships and transmission mechanisms between various hydrological information segments, resulting in significant deviations between forecast results and actual power generation, making it difficult to meet the needs of refined management and scientific decision-making in hydropower stations. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, the present invention provides a method for predicting the power generation capacity of a hydropower station based on hydrological information, the method comprising:

[0005] A set of dynamic hydrological interaction information for a hydropower station is obtained. The set of dynamic hydrological interaction information includes upstream water inflow time sequence information, reservoir water level fluctuation information, water flow velocity change information, and downstream water demand response information. Each piece of information in the set of dynamic hydrological interaction information is marked with a continuous time series, and each piece of dynamic hydrological interaction information contains state change records for multiple time periods.

[0006] A hydrological interaction transmission network is constructed based on the aforementioned set of dynamic hydrological information. The hydrological interaction transmission network uses the state change records of each hydrological information as network nodes and the mutual triggering relationships between the state changes of each hydrological information as network edges. The attributes of the network edges are determined by the temporal continuity and influence range of the triggering relationships.

[0007] From the hydrological interaction transmission network, key transmission nodes and corresponding network edge attributes related to power generation are extracted to generate a power response correlation factor sequence. Each power response correlation factor in the power response correlation factor sequence contains the state change record of the key transmission node and the attribute record of the corresponding network edge.

[0008] Based on the power response correlation factor sequence and the historical power generation time series record of the hydropower station, a power response correlation rule base is constructed. The power response correlation rule base contains the correspondence between different power response correlation factor combinations and power generation change patterns.

[0009] The power response correlation factor sequence generated by the currently acquired dynamic hydrological information set is matched with the power response correlation rule base to generate the power generation prediction result of the hydropower station for the current period.

[0010] In another aspect, the present invention also provides a hydropower station power generation prediction system based on hydrological information, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the machine-readable storage medium to implement the above-mentioned method.

[0011] Based on the above, a comprehensive set of dynamic hydrological information is acquired, including upstream water inflow time series information, reservoir water level fluctuation information, water flow velocity change information, and downstream water demand response information. Each piece of information is marked with a continuous time series and records of state changes over multiple time periods. A hydrological interaction transmission network is constructed based on this set of dynamic hydrological information. Using the state change records of each piece of hydrological information as network nodes and the mutual triggering relationships as network edges, and clearly defining the attributes of the network edges, the dynamic transmission mechanism between hydrological information can be accurately presented. Key transmission nodes related to power generation and their corresponding network edge attributes are extracted from the network, generating a power response correlation factor sequence. This effectively extracts the core factors that directly affect power generation. A power response correlation rule base is constructed, containing the correspondence between different combinations of power response correlation factors and power generation change patterns. Finally, the currently generated power response correlation factor sequence is matched with the rule base to generate power generation prediction results. This significantly improves the accuracy and reliability of hydropower generation prediction, contributing to the refined management and efficient operation of hydropower stations. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the execution flow of the hydropower generation prediction method based on hydrological information provided in an embodiment of the present invention.

[0013] Figure 2This is a schematic diagram of exemplary hardware and software components of a hydropower station power generation prediction system based on hydrological information provided in an embodiment of the present invention. Detailed Implementation

[0014] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a hydropower generation power prediction method based on hydrological information, provided in one embodiment of the present invention. The following is a detailed description of this hydropower generation power prediction method based on hydrological information.

[0015] Step S110: Obtain the hydrological dynamic interaction information set of the hydropower station. The hydrological dynamic interaction information set includes upstream water inflow time sequence information, reservoir water level fluctuation information, water flow velocity change information, and downstream water demand response information. Each piece of information in the hydrological dynamic interaction information set is marked with a continuous time series, and each piece of hydrological dynamic interaction information contains state change records for multiple time periods.

[0016] In this embodiment, the power generation scheduling of the hydropower station needs to comprehensively consider multiple factors such as upstream water conditions, reservoir water storage, water flow characteristics, and downstream water use. To achieve accurate prediction of the power generation of the hydropower station, it is first necessary to systematically collect various dynamic interactive information on water conditions closely related to the power generation process.

[0017] Upstream water inflow time-series information is collected through hydrological monitoring stations deployed at different cross-sections upstream of the hydropower station's intake. These stations are equipped with devices capable of continuously recording water flow. The collected data forms a continuous time series at fixed time intervals, with each data point accompanied by a precise collection time stamp. This information reflects the changes in water inflow caused by factors such as precipitation and snowmelt in the upstream basin, including inflow patterns under different seasons and weather conditions. Reservoir water level fluctuation information is obtained from level gauges installed at specific locations on the reservoir dam. These gauges monitor real-time changes in reservoir water surface elevation and record the data chronologically, forming a time series of water level fluctuations. This series not only includes the real-time water level height but also reflects the water level change characteristics under different operating conditions such as water storage, flood discharge, and power generation. Water flow velocity change information is collected through velocity measurement devices deployed in key water flow channels such as the hydropower station's intake channels and pressure pipelines. These devices can sense changes in water flow velocity at different locations and under different operating conditions and generate corresponding time-series data. Changes in water flow velocity directly affect the output of the turbine and are a crucial intermediate parameter linking hydrological conditions and power generation. Downstream water demand response information is obtained through a data sharing mechanism with downstream water management departments. This information covers the planned and real-time water consumption data of various water users, including those engaged in agricultural irrigation, industrial production, and urban life. This data is also organized in time series, reflecting the dynamic changes in downstream water demand over time and the response to hydropower station water release scheduling.

[0018] During the collection and recording of all the aforementioned hydrological information, continuous time-series markers were ensured to facilitate subsequent correlation analysis over time. Furthermore, each hydrological dynamic interaction was divided into multiple consecutive time periods. The state change record within each time period includes the initial state value at the beginning of the period, the final state value at the end of the period, and a detailed description of the state's change from the initial to the final value throughout the entire period, such as whether the state exhibits continuous stable change, fluctuating change, or a step-like change.

[0019] Step S120: Construct a hydrological interaction transmission network based on the hydrological dynamic interaction information set. The hydrological interaction transmission network uses the state change records of each hydrological information as network nodes and the mutual triggering relationship between the state changes of each hydrological information as network edges. The attributes of the network edges are determined by the temporal continuity and influence range of the triggering relationship.

[0020] Step S121: The state change records of the upstream water inflow time sequence information, reservoir water level fluctuation information, water flow velocity change information and downstream water demand response information of the hydropower station in the hydrological dynamic interaction information set are divided into multiple continuous state segments. Each state segment corresponds to the same time span, and each state segment contains the state start value, state end value and state change process description within the corresponding time period.

[0021] First, a fixed time span should be uniformly set for all types of hydrological information. The selection of this time span should take into account both the rate of change of hydrological information and the actual response time of hydropower station power generation regulation, so as to ensure that meaningful state change processes can be captured.

[0022] For upstream water inflow time-series information, the continuous time series records are divided into multiple consecutive, non-overlapping state segments according to a set time span. Each state segment corresponds to a specific time period, including the inflow flow value at the beginning of the time period as the initial value, the inflow flow value at the end of the time period as the final value, and a detailed description of the entire change process of the inflow flow from the initial value to the final value within that time period, such as whether the flow shows a gradually increasing trend, a gradually decreasing trend, an initial increasing trend followed by a decreasing trend, or a stable fluctuation within a certain range. Similarly, reservoir water level fluctuation information is also divided into multiple state segments according to the same time span. Each segment includes the initial water level height, the final water level height, and a description of the water level fluctuation process within that time period, such as whether the water level rises slowly and continuously, falls rapidly, or oscillates within a certain range.

[0023] The state segmentation method for water flow velocity change information is similar to that described above. Each segment includes the initial water flow velocity, the final water flow velocity, and a description of the characteristics of the water flow velocity change within that time period, such as whether the velocity change is uniform or whether there are abrupt changes. Downstream water demand response information is also segmented into state segments corresponding to the time span, including the initial water demand, the final water demand, and a description of the change process of demand within the time period, such as whether the demand is steadily increasing, remaining constant, or fluctuating in stages.

[0024] Through the above decomposition process, the continuous state changes of various types of hydrological information are transformed into a series of state fragments with clear time boundaries and internal characteristics.

[0025] Step S122: Select any two different types of hydrological information state segments as the trigger node and the triggered node, respectively. Analyze the temporal relationship between the state change process description of the trigger node and the state change process description of the triggered node to determine whether the trigger node changes state before the triggered node.

[0026] After decomposing all hydrological information state segments, the analysis begins on the potential triggering relationships between different types of hydrological information state segments. First, two different types of hydrological information state segments are randomly selected from all the state segments; for example, one segment is selected from the upstream inflow time series information state segment, and another segment is selected from the reservoir water level fluctuation information state segment. One of the selected segments is designated as a potential trigger node, and the other as a potential triggered node. The state change process descriptions of the trigger node and the triggered node are compared to determine the start time of the state change. By comparing the positional relationship of these two start times on the time axis, it is determined whether the state change of the trigger node occurs earlier than that of the triggered node. For example, if the start time of the state change of a certain state segment of the upstream inflow time series information is earlier than the start time of the state change of a certain state segment of the reservoir water level fluctuation information, then the trigger node (upstream inflow state segment) is considered to have changed state before the triggered node (reservoir water level state segment). The determination of the chronological order mentioned above is the primary condition for identifying triggering relationships, because only when one event occurs before another in time can the former be a triggering factor for the latter.

[0027] Step S123: If the triggering node changes state before the triggered node, further analyze the correlation trend between the state change amplitude of the triggering node and the state change amplitude of the triggered node, and determine whether there is a unidirectional or inverse correlation between the triggering node and the triggered node.

[0028] Once it is determined that the state change of the triggering node occurs before that of the triggered node, it is necessary to further analyze the correlation trend between the magnitudes of the state changes of the two nodes. The magnitude of the state change is obtained by calculating the difference between the state termination value and the state start value of the state segment. The sign of this difference reflects the direction of the state change, and the absolute value of the difference reflects the degree of change.

[0029] Compare the signs of the state change amplitudes of the triggering node and the triggered node. If the state change amplitude of the triggering node is positive (i.e., the state value is increasing) and the state change amplitude of the triggered node is also positive; or if the state change amplitude of the triggering node is negative (i.e., the state value is decreasing) and the state change amplitude of the triggered node is also negative, it indicates that the state changes of the triggering node and the triggered node are in the same direction, and there is a correlation of change in the same direction.

[0030] Conversely, if the state change amplitude of the triggering node is positive and the state change amplitude of the triggered node is negative, or if the state change amplitude of the triggering node is negative and the state change amplitude of the triggered node is positive, it indicates that the state changes of the two nodes are in opposite directions and there is an inverse relationship between them.

[0031] The aforementioned correlations between changes in the same or opposite directions indicate that the changes in the two state segments are not isolated, accidental events, but rather that there may be some inherent causal relationship or mutual influence mechanism.

[0032] Step S124: If there is a change association in the same or opposite direction between the triggering node and the triggered node, a network edge is established between the triggering node and the triggered node. The time interval from the triggering node to the triggered node is used as the temporal continuity attribute of the network edge, and the length of the coverage period of the change association is used as the influence range attribute of the network edge.

[0033] Once a unidirectional or inverse relationship between the triggering node and the triggered node is confirmed, a directed network edge is established between these two state fragment nodes. The direction of the network edge points from the triggering node to the triggered node to clearly indicate the direction of the triggering effect. Simultaneously, two key attributes are defined for this network edge: temporal continuity and influence range. The temporal continuity attribute is the time interval between the start of the state change of the triggering node and the start of the state change of the triggered node. This interval reflects the time delay from the occurrence of the triggering effect to the start of the response from the triggered node. This delay may vary between different types of hydrological information and is influenced by various factors such as water flow propagation speed and reservoir storage capacity.

[0034] The influence range attribute is the duration for which the state change of the triggering node affects the state change of the triggered node. Specifically, it is the time elapsed from the moment the triggered node begins its state change until the triggered node's state change is no longer affected by the triggering node. This attribute reflects the sustained influence of the triggering action and is related to factors such as the intensity of the triggering node's state change and the inertia of the triggered node itself. By assigning the above attributes to the network edges, the hydrological interaction transmission network can not only represent whether a triggering relationship exists between nodes but also quantitatively describe the characteristics of this relationship in the time dimension.

[0035] Step S125: Repeat the steps of state fragment splitting, time sequence analysis, change correlation judgment and network edge establishment, traverse all state fragment combinations of different types of hydrological information, and establish multiple sets of network edges.

[0036] To comprehensively capture all possible triggering relationships between various types of hydrological information, a systematic traversal analysis of state fragment combinations of all different types of hydrological information is required. This includes state fragment combinations of upstream inflow time series information with reservoir water level fluctuation information, water flow velocity change information, and downstream water demand response information; state fragment combinations of reservoir water level fluctuation information with water flow velocity change information and downstream water demand response information; and state fragment combinations of water flow velocity change information with downstream water demand response information.

[0037] For each type of combination, all possible state fragment pairs need to be extracted, and their temporal relationships analyzed one by one (as described in step S122) to determine whether the triggering node changes state before the triggered node. For state fragment pairs that satisfy the temporal relationship, further analysis is conducted to determine whether there is a unidirectional or inverse correlation between their state change amplitudes (as described in step S123). Once a pair of state fragments is found to satisfy the temporal relationship and have a unidirectional or inverse correlation, a network edge is established between them according to the method in step S124, and the temporal continuity attribute and influence range attribute of the network edge are determined. By traversing all possible combinations as described above, the mutual triggering relationships between various types of hydrological information state changes can be found as comprehensively as possible, thereby establishing multiple sets of network edges between numerous state fragment nodes and initially constructing the connection structure of the hydrological information interaction transmission network.

[0038] Step S126: Treat all state fragments as network nodes, and connect all established network edges according to the network node association relationship to form an initial hydrological interaction and transmission network.

[0039] After completing the traversal analysis of all different types of hydrological information state fragment combinations and establishing multiple sets of network edges, each of the decomposed hydrological information state fragments was treated as an independent network node. Each node contains complete information such as the corresponding hydrological information type, the time period it belongs to, the state start value, the state end value, and a description of the state change process.

[0040] Then, based on the previously established sets of network edges, the various network nodes are connected according to their triggering relationships. Specifically, each network edge connects a triggering node and a triggered node, and according to the directionality of the network edges, the path points from the triggering node to the triggered node, thus forming a series of directed connection paths throughout the network.

[0041] Through the above connection method, all state fragment nodes and network edges together constitute an initial hydrological information interaction and transmission network. In this initial network, nodes represent the state changes of different types of hydrological information at specific time periods, and edges represent the triggering relationships between nodes and their time attributes. The entire network initially demonstrates the complex interaction and transmission patterns between hydrological information.

[0042] Step S127: Perform time-series validity verification on the network edges in the initial hydrological information interaction and transmission network. Retain network edges that can maintain stable time-series continuity and influence range attributes over multiple consecutive time spans, and delete network edges that exist only in a single or two time spans to obtain the hydrological information interaction and transmission network.

[0043] The initial hydrological communication network may contain some unstable network edges. These edges may appear to have a triggering relationship only during certain time periods due to accidental factors, rather than being truly stable connections. Therefore, it is necessary to verify the temporal validity of network edges in the initial network to improve its reliability. The core idea of ​​the verification is to check whether the temporal continuity and influence range attributes of each network edge remain stable over multiple consecutive time spans. If a network edge appears only in a single time span or a few time spans, and its attribute values ​​fluctuate significantly, then the network edge is considered to have low validity and may be a spurious triggering relationship. Conversely, if a network edge exists stably over multiple consecutive time spans, and its temporal continuity and influence range attributes remain within a relatively stable range without drastic or irregular fluctuations, then the network edge is considered to have high temporal validity and is a genuine and stable triggering relationship. Through the above verification process, network edges with stable temporal attributes are retained, while unstable network edges that only exist in a few time periods are deleted. This optimizes the initial hydrological information interaction and transmission network, resulting in a more stable and reliable hydrological information interaction and transmission network.

[0044] Step S1271: Extract the temporal continuity attribute records and influence range attribute records of each network edge in the initial hydrological interaction transmission network over all time spans to form the attribute temporal sequence corresponding to each network edge.

[0045] To verify the time series validity, it is first necessary to collect attribute data for each network edge across all time spans in its existence. For each network edge in the initial hydrological interaction and transmission network, the temporal continuity attribute value and influence range attribute value for that edge across each time span are extracted from the network records. These attribute values ​​are then arranged sequentially according to the time spans to form the temporal continuity attribute time series and influence range attribute time series corresponding to that network edge. For example, a network edge might have a certain temporal continuity attribute value and influence range attribute value in the first time span; a different value in the second time span; and so on. Organizing these values ​​in chronological order yields the two attribute time series sequences for that network edge.

[0046] Step S1272: Perform continuous stability analysis on the attribute time series of each network edge to check whether there are attribute records with multiple consecutive time spans, wherein the change amplitude of the time series continuity attribute does not exceed the preset continuity change threshold, and the change amplitude of the influence range attribute does not exceed the preset range change threshold.

[0047] For each network edge, both the time series sequence of the continuity attribute and the time series sequence of the influence range attribute are subjected to continuous stability analysis. First, two thresholds need to be preset: a continuity change threshold and a range change threshold. These two thresholds are set based on statistical analysis of historical hydrological data and hydropower station operation experience, and are used to determine whether the changes in attribute values ​​are within an acceptable stability range. For the time series sequence of the continuity attribute, the attribute values ​​of multiple consecutive time spans in the sequence are examined sequentially, and the difference (variation amplitude) between attribute values ​​of two adjacent time spans is calculated. It is then determined whether the absolute values ​​of these differences do not exceed the preset continuity change threshold. Simultaneously, for the time series sequence of the influence range attribute, the same method is used to check whether the variation amplitude of attribute values ​​of multiple consecutive time spans does not exceed the preset range change threshold.

[0048] If, in the time series of two attributes of a network edge, there are consecutive time spans (e.g., three or more), such that within these time spans, the change magnitude of the time series continuity attribute and the change magnitude of the influence range attribute do not exceed their respective thresholds, then the network edge is considered to have exhibited attribute stability within these consecutive time spans.

[0049] Step S1273: If there are attribute records with multiple consecutive time spans that satisfy the stability conditions of the change amplitude of the temporal continuity attribute and the change amplitude of the influence range attribute, then mark the network edge as a candidate valid network edge and record the length of the longest consecutive stable time span.

[0050] After performing continuous stability analysis on the attribute time series of a network edge, if multiple consecutive time spans of attribute records satisfy the aforementioned stability conditions (i.e., the changes in both the time series continuity and the range of influence attributes are within their respective thresholds), the network edge is marked as a candidate valid network edge. This indicates that the network edge has the potential to become a stable and valid triggering relationship. Simultaneously, the longest continuous stable time span segment needs to be identified in the attribute time series of the network edge—that is, the segment with the largest number of time spans continuously satisfying the stability conditions—and its length (the number of time spans it contains) needs to be recorded. The length of this longest continuous stable time span segment is an important indicator of the stability of the candidate valid network edge; the longer the length, the better the stability of the network edge.

[0051] Step S1274: Perform correlation verification on the longest continuous stable time span of the candidate valid network edges, and analyze whether the state change process descriptions of the corresponding triggering node and the triggered node within the corresponding time period always maintain the same change correlation type.

[0052] Even if the properties of the candidate valid network edges show stability over the longest continuous stable time span, it is still necessary to further verify whether the change association type between the triggering node and the triggered node remains consistent throughout this period. The change association type refers to the same-direction change association or opposite-direction change association determined in the previous steps.

[0053] Specifically, the process involves extracting the specific time periods corresponding to the longest continuous stable time span of the candidate valid network edge. For each time period, the descriptions of the state change processes of the trigger node state segments and the triggered node state segments connected by the network edge are examined to reconfirm the type of change association between them (same direction or opposite direction).

[0054] Check whether the change association type remains consistent across all these time periods; that is, all are either unidirectional or unidirectional associations. If, within the longest continuous stable time span, the change association type differs between some time periods and other time periods, it indicates that the triggering relationship of the network edge has unstable factors in terms of association type.

[0055] Step S1275: If the state change process descriptions of the triggering node and the triggered node always maintain the same change association type, then the candidate valid network edge is confirmed as a valid network edge; if the state change process descriptions of the triggering node and the triggered node change the change association type within a continuous stable time span, then the valid time period of the network edge is extracted into a sub-time period with the same change association type, and the sub-time period is re-checked to see if it meets the requirement of multiple consecutive time spans.

[0056] After correlation verification, if the state change process descriptions of the triggering node and the triggered node remain consistent with the same correlation type (all in the same direction or all in opposite directions) within the longest continuous stable time span of the candidate valid network edge, and there is no change in the correlation type, then the candidate valid network edge can be confirmed as a valid network edge, and its stability and correlation are guaranteed.

[0057] If the state change association type between the triggering node and the triggered node changes within the longest continuous stable time span—for example, the first half is a unidirectional association, the second half becomes a reversible association, or there is no association in the middle—then the longest continuous stable time span needs to be segmented. It should be divided into several sub-segments, with the change association type remaining consistent within each sub-segment. After segmentation, for each sub-segment, it should be re-checked to ensure it meets the requirement of multiple consecutive time spans (i.e., whether the number of time spans contained in the sub-segment meets the previously set standard for multiple consecutive segments, such as three or more segments).

[0058] Step S1276: If the truncated sub-time period still meets the requirement of multiple consecutive time spans, then the network edge is confirmed as a valid network edge, and the valid time period of the network edge is marked; if the truncated sub-time period does not meet the requirement of multiple consecutive time spans, then the network edge is marked as an invalid network edge.

[0059] For the sub-time periods obtained after truncation, if the number of time spans contained in a certain sub-time period still meets the requirement of multiple consecutive time spans (such as no less than three segments), it indicates that the network edge in the sub-time period still has a certain stability and correlation. Therefore, the network edge is confirmed as a valid network edge in the sub-time period, and the valid time period range of the network edge (i.e., the specific time span corresponding to the sub-time period) is clearly marked.

[0060] If none of the extracted sub-periods meet the requirement of multiple consecutive time spans, i.e., each sub-period contains fewer than three time spans, it indicates that even within a period of stable attributes, the change association type of the network edge is unstable, and the duration of the stable association type is short. Therefore, the network edge is marked as an invalid network edge and removed from the network.

[0061] Step S1277: Collect all confirmed valid network edges and the network nodes corresponding to the valid network edges, and reconstruct the network structure according to the relationship between network nodes and network edges.

[0062] After verifying and filtering all candidate valid network edges, all network edges confirmed as valid are collected. Simultaneously, all network nodes connected to these valid network edges are identified, as only nodes associated with valid network edges are meaningful for network propagation.

[0063] Then, based on the relationships between these effective network edges and their corresponding network nodes, the structure of the hydrological information interaction transmission network is reconstructed. The new network structure contains only effective network edges and the nodes they connect to, with nodes connected via these effective network edges according to the direction of the triggering relationship.

[0064] Step S1278: Perform a network node connectivity check on the reconstructed network structure. Find at least one propagation path for all network nodes through valid network edges. If there are network nodes that are not connected by any valid network edges, delete the network nodes that are not connected by any valid network edges.

[0065] The reconstructed network structure may contain some isolated network nodes that are not connected to any other nodes through effective network edges. That is, they are neither triggering nodes nor triggered nodes of any effective network edge. These isolated nodes cannot participate in the interactive transmission of hydrological information and therefore do not contribute to subsequent power generation prediction and analysis.

[0066] Therefore, a network node connectivity check is required for the reconstructed network structure. Specifically, for each node in the network, an attempt is made to find at least one path connecting it to other nodes through existing valid network edges. This path can originate from that node and lead to other nodes, or vice versa. If a node can establish at least one such path with other nodes in the network through valid network edges, then that node is considered connected. If the check finds that a node has no valid network edges connecting it to any other node (i.e., no valid network edges originating from or ending at that node), then that node is an isolated node. Isolated nodes are then removed from the network structure.

[0067] Step S12781: Assign a unique network node identifier to each network node in the reconstructed network structure and establish a list of network node identifiers.

[0068] To facilitate node connectivity checks and management, a unique identifier is first assigned to each network node in the reconstructed network structure. This identifier is generated by encoding information such as the node's hydrological information type and the time span it belongs to, ensuring that each node has a unique identifier. Then, all the unique identifiers of the network nodes are compiled into a network node identifier list, which contains the identification information of all nodes in the current network structure.

[0069] Step S12782: Select the first network node in the network node identifier list as the starting network node. Through the valid network edges connected to the starting network node, find other network nodes directly connected to the starting network node and mark the found network nodes as connected network nodes.

[0070] Select the first network node from the list of network node identifiers as the starting point for connectivity checks. Examine all valid network edges connected to this starting node, including output edges that trigger the connection and input edges that are triggered by the node. Using these valid network edges, find all other network nodes directly connected to the starting node. For example, if the starting node has an output edge pointing to node A, then node A is directly connected to the starting node; if the starting node has an input edge from node B, then node B is also directly connected to the starting node. After finding these directly connected nodes, mark their status as connected network nodes, indicating that they have established a connection with the network through the starting node.

[0071] Step S12783: Using the network node marked as a connected network node as the new starting network node, continue to search for unmarked network nodes through the valid network edges connected by the new starting network node. Mark the found unmarked network nodes as connected network nodes. Repeat the search until no new unmarked network nodes can be found.

[0072] After marking the nodes directly connected to the initial starting node, these newly marked connected nodes are then used as new starting network nodes. For each new starting network node, all its connected valid network edges are examined, and other nodes in the network that have not yet been marked as connected are searched through these edges.

[0073] Once an unmarked node is found, it is immediately marked as a connected network node. Then, starting from these newly marked nodes, the search and marking process is repeated. This iterative expansion, like ripples spreading on water, gradually marks all nodes that can be connected to the initial node through valid network edges as connected, until no more unmarked nodes can be found.

[0074] Step S12784: Count the number of network nodes marked as connected network nodes and compare it with the total number of network nodes in the network node identifier list.

[0075] After completing the iterative search process described above, the total number of nodes currently marked as connected network nodes is counted. Then, this total number is compared with the total number of network nodes recorded in the network node identifier list to determine whether there are any unconnected nodes in the network.

[0076] Step S12785: If the number of network nodes marked as connected network nodes is equal to the total number of network nodes, then keep the network structure unchanged.

[0077] If the number of connected network nodes obtained from the statistics is equal to the total number of nodes in the network node identifier list, this indicates that all nodes in the network are interconnected through valid network edges, there are no isolated nodes, and the network structure is fully connected. In this case, the current network structure remains unchanged.

[0078] Step S12786: If the number of network nodes marked as connected network nodes is less than the total number of network nodes, then the unmarked network nodes are determined to be network nodes without any valid network edge connections, and the identifier of the unmarked network nodes and the corresponding hydrological information type are recorded.

[0079] If the number of connected network nodes is less than the total number of nodes, it indicates that there are nodes in the network that have not been marked as connected. These unmarked nodes are isolated nodes without any valid network edge connections. Record the unique identifier of these isolated nodes and the corresponding hydrological information type (such as upstream water inflow time series information, reservoir water level fluctuation information, etc.) to facilitate subsequent analysis of the reasons why they became isolated nodes.

[0080] Step S12787: Analyze the state change records of the hydrological information type corresponding to the network node without any valid network edge connection in the historical time period, and check whether the absence of any valid network edge connection is due to the lack of significant state change of the hydrological information in the current time period.

[0081] For the isolated nodes that have been recorded, it is necessary to analyze the status change records of their corresponding hydrological information types in various historical periods in order to explore the possible reasons why they became isolated nodes.

[0082] One common reason is that the hydrological information did not undergo significant state changes during the currently analyzed time period. For example, downstream water demand might be very stable during a certain period, with the demand remaining essentially unchanged and showing no obvious increasing or decreasing trend. Because its own state change is not significant, it is difficult to form triggering relationships with other hydrological information, resulting in no effective network edges connecting it. By examining the state changes of this type of hydrological information in other historical periods, if it is found that it usually forms effective network edges with other hydrological information during periods of significant state change, but the state change is not significant in the current period, it can be preliminarily determined that the node is isolated because there is no significant state change in the current period.

[0083] Step S12788: If there are no effective network edge connections due to the lack of significant state changes in the hydrological information during the current time period, then add weakly correlated network edges between the hydrological information and other hydrological information. The temporal continuity attribute and influence range attribute of the weakly correlated network edges are taken as the historical average level, and the network node connectivity check is performed again.

[0084] If an isolated node is determined to be isolated due to the lack of significant state change in its corresponding hydrological information within the current time period, then it is considered to supplement it with weakly correlated network edges. Weakly correlated network edges are a special type of network edge used to represent potential connections under conditions of insignificant state change. The supplemented weakly correlated network edges can be determined based on the historical associations formed between this type of hydrological information and other hydrological information types when the state change was relatively small. Its temporal continuity and influence range attributes can be taken as the historical average level of this type of association. After supplementing the weakly correlated network edges, the network nodes are re-checked according to steps S12781 to S12787 to see if the previously isolated node can establish connections with other nodes through the supplemented weakly correlated network edges.

[0085] Step S12789: If the network node is still a network node without any valid network edge connections after adding weakly associated network edges, or if it has no valid network edge connections for other reasons, then delete the network node from the network structure.

[0086] If, after adding weakly connected network edges, the node still cannot establish any effective connection with other nodes (including the failure of weakly connected network edges to connect successfully), or if the reason for the node becoming an isolated node is not that the current state changes are not significant (for example, the hydrological information type itself rarely interacts with other types), then it will no longer be retained and the isolated node will be completely removed from the network structure.

[0087] Step S127810: After deleting network nodes without any valid network edge connections, reorganize the network node identifier list and renumber the valid network edges of the remaining network nodes so that the relationship between network nodes and network edges in the network structure can be clearly presented through renumbering.

[0088] After deleting orphaned nodes, the number of network nodes changes. Therefore, it's necessary to reorganize the network node identifier list, removing the identifiers of deleted nodes to ensure the list only contains the remaining nodes in the current network. Simultaneously, the valid network edges between the remaining nodes should be renumbered for easier management and subsequent analysis. This renumbering should follow certain rules so that the edge numbers intuitively reflect information such as the nodes they connect to and their directions, thus making the relationships between nodes and edges in the network structure clearer.

[0089] Step S127811: Perform network node connectivity checks again to confirm that the remaining network nodes are all marked as connected network nodes, thus forming a network structure without isolated network nodes.

[0090] After completing the node deletion and organizing of the list and edge numbers, a network node connectivity check needs to be performed again to ensure that there are no more isolated nodes in the network structure. Repeat steps S12782 to S12785 to confirm that all remaining network nodes have been marked as connected network nodes.

[0091] Step S1279: Integrate effective network edges, non-isolated network nodes and their corresponding attribute records to form a hydrological information interaction and transmission network.

[0092] The effective network edges and non-isolated network nodes obtained after all the above steps of screening, verification, and optimization, as well as the temporal continuity and influence range attributes of these network edges and the state change records of network nodes, are integrated. The integrated information constitutes the final hydrological information interaction and transmission network, which can accurately and stably reflect the state changes of different types of hydrological information in different time periods, as well as their triggering relationships and temporal characteristics.

[0093] Step S130: Extract key transmission nodes and corresponding network edge attributes related to power generation from the hydrological interaction transmission network, and generate a power response correlation factor sequence. Each power response correlation factor in the power response correlation factor sequence contains the state change record of the key transmission node and the attribute record of the corresponding network edge.

[0094] After the hydrological information interaction and transmission network is constructed, its nodes and edges reflect the complex interactions between hydrological information. However, not all nodes have a significant impact on power generation. Therefore, it is necessary to extract the key transmission nodes and their associated network edge attributes that are closely related to changes in power generation. These key nodes are nodes on the critical path from hydrological information transmission to power generation, and their state changes and interactions have a decisive impact on power output. After extracting the key transmission nodes, the state change records of each key node are combined with their associated network edge attribute records to form power response correlation factors that reflect the characteristics of the node's impact on power generation. Arranging these factors in chronological order yields the power response correlation factor sequence.

[0095] Step S131: Collect the historical power generation time series records of the hydropower station, and divide the historical power generation time series records into historical power change segments with the same time span as the state segments of the hydrological dynamic interaction information set. Each historical power change segment contains the power start value, power end value and power change process description within the corresponding time period.

[0096] To identify which nodes in the hydrological information transmission network are related to power generation, it is necessary to utilize the historical power generation data of the hydropower station. Time-series records of the power generation of the hydropower station over a relatively long period are collected, arranged chronologically and containing power generation values ​​at different times.

[0097] The aforementioned historical power generation time-series records are divided into multiple historical power change segments according to the same time span as the state segments of the hydrological dynamic interaction information set. Each historical power change segment corresponds to a specific time period, with a one-to-one correspondence between the time periods of the hydrological state segments. Each historical power change segment contains three basic elements: the power generation value at the beginning of the time period, i.e., the initial power value; the power generation value at the end of the time period, i.e., the final power value; and a description of the entire change process of power generation from the initial value to the final value within the time period, such as whether the power is continuously rising, continuously falling, rising first and then falling, fluctuating, or basically remaining stable. This division ensures that the hydrological state segments and the power generation change segments are consistent in time scale, facilitating the correlation analysis between the two.

[0098] Step S132: Perform time-series coupling analysis on the state segment corresponding to each network node in the water information interaction transmission network and the historical power change segment of the same period, and observe whether there are synchronous change characteristics between the change process description of the network node state segment and the change process description of the historical power change segment.

[0099] For each network node (i.e., each hydrological state segment) in the hydrological interaction transmission network, a time-series coupling analysis is performed between it and a historical power change segment from the same time period. Here, "same time period" means that the corresponding time spans are exactly the same. The focus of the analysis is to observe whether there are synchronous change characteristics between the description of the change process of the network node's state segment and the description of the change process of the historical power change segment. For example, if the change process of the network node's state segment is described as "water flow velocity gradually increases," while the change process of the historical power change segment from the same time period is described as "power generation gradually increases," then they can be considered to have synchronous growth characteristics. Similarly, if the network node's state segment is described as "reservoir water level first rises then falls," while the historical power change segment from the same time period is described as "power generation first increases then decreases," this also indicates synchronous change characteristics. Synchronous change characteristics indicate that the changes in hydrological state and power generation have a consistent temporal trend, suggesting that the network node may be related to power generation.

[0100] Step S133: If the description of the change process of the network node state segment and the description of the change process of the historical power change segment have synchronous change characteristics, then mark the network node as a candidate transmission node, and record the number of synchronous change periods and the characteristic type of synchronous change.

[0101] If time-series coupling analysis reveals that the state segment change process description of a network node exhibits synchronous change characteristics with the historical power change process description during the same period, then this network node is marked as a candidate transmission node. Candidate transmission nodes are potential key nodes related to power generation. Simultaneously, it is necessary to record the number of time periods during which the candidate transmission node and the historical power change process exhibit synchronous changes, i.e., the total number of different time periods where the aforementioned synchronous characteristics appear. Furthermore, it is also necessary to record the characteristic type of the synchronous change, such as whether it is a "synchronous rise," "synchronous fall," "synchronous rise followed by fall," or "synchronous fluctuation," etc., specifically the synchronous pattern.

[0102] Step S134: Extract all network edges associated with each candidate propagation node, obtain the temporal continuity attribute and influence range attribute of these network edges, and form a set of candidate node association attributes.

[0103] Each candidate transmission node may be connected to other nodes in the hydrological interaction transmission network by multiple network edges. These network edges reflect the triggering relationships between the candidate node and other hydrological nodes. To fully understand the impact pathways and characteristics of candidate transmission nodes on power generation, it is necessary to extract all network edges associated with the candidate transmission node.

[0104] For each associated network edge, its temporal continuity attribute and influence range attribute values ​​are obtained. These attribute values ​​of all associated network edges of a candidate transmission node are summarized to form the association attribute set of the candidate node, which reflects the transmission characteristics of the candidate node in the hydrological network.

[0105] Step S135: For each candidate transmission node, count the frequency of synchronous changes of the candidate transmission node with historical power change segments in all historical time periods, screen out candidate transmission nodes whose synchronous change frequency meets the preset frequency requirements, and mark the screened candidate transmission nodes as key transmission nodes related to power generation.

[0106] The synchronous changes between candidate transmission nodes and power generation may not occur in all historical periods, and the frequency of these changes is an important indicator of the strength of the correlation between the node and power generation. Therefore, it is necessary to statistically analyze the frequency of synchronous changes between each candidate transmission node and historical power change segments across all historical periods.

[0107] A preset frequency requirement is established, such as the proportion of synchronous changes to a certain threshold in the total number of historical periods, or the absolute number of changes reaching a certain threshold. The synchronous change frequency of each candidate transmission node is compared with this preset frequency requirement, and candidate transmission nodes that meet the synchronous change frequency requirement are selected. These nodes are considered to have a strong and stable correlation with power generation and are marked as key transmission nodes related to power generation.

[0108] Step S1351: Establish a synchronous change statistics table for each candidate conduction node. The synchronous change statistics table includes the type of candidate conduction node, the corresponding historical time period number, and a mark indicating whether it changes synchronously with the historical power change segment of the same period.

[0109] To systematically analyze the frequency of synchronization changes among candidate transmission nodes, a synchronization change statistics table should be established for each candidate node. The columns should include at least: the type of the candidate transmission node (i.e., its corresponding hydrological information type), the historical time period number corresponding to the node (each time period has a unique number), and a flag indicating whether the node synchronized with historical power change segments during that time period (e.g., "yes" or "no"). This allows for tracking the synchronization of each candidate transmission node with changes in power generation across different historical time periods.

[0110] Step S1352: Traverse all historical time periods. For each candidate conduction node, check whether the description of the state change process of the candidate conduction node in the historical time period and the description of the state change process of the historical power change segment in the same period meet the synchronous change condition. The synchronous change condition is that the difference between the start time of the state change of the candidate conduction node and the start time of the state change of the historical power change segment does not exceed the preset time difference threshold, and the state change trend direction of the candidate conduction node is consistent with the state change trend direction of the historical power change segment.

[0111] Iterate through all historical time periods, and for each candidate propagation node, check the synchronization change conditions in each historical time period. Synchronization change conditions include two aspects:

[0112] First, there is the temporal synchronization, specifically whether the difference between the start time of the state change of the candidate transmission node and the start time of the state change of a historical power change segment during the same period is within a preset time difference threshold. This threshold takes into account the delay time in the transmission of water condition changes to power generation changes, ensuring that the start of the changes in both are sufficiently close in time.

[0113] Second, there must be consistency in trend direction, meaning whether the trend direction of the candidate transmission node's state change (e.g., increase, decrease, increase followed by decrease, etc.) is consistent with the trend direction of the state change in the historical power change segment. Only when the two directions are consistent is the trend synchronization condition considered met. Only when both conditions are met simultaneously is it determined that the candidate transmission node has undergone synchronous changes with the power generation change segment in that historical period.

[0114] Step S1353: If the synchronous change condition is met, mark the corresponding time period number in the synchronous change statistics table as synchronous; if the synchronous change condition is not met, mark the corresponding time period number in the synchronous change statistics table as asynchronous.

[0115] Based on the inspection results of step S1352, each historical time period is marked in the synchronization change statistics table. If the synchronization change conditions are met, it is marked as "synchronous state" in the row corresponding to the time period number; otherwise, it is marked as "asynchronous state". Through the above marking, it is possible to intuitively see in which time periods each candidate transmission node changes synchronously with the power generation and in which time periods it is asynchronous.

[0116] Step S1354: Count the number of times each candidate transmission node is marked as synchronized in all historical time periods to obtain the total frequency of synchronization changes of the candidate transmission node.

[0117] After marking all historical time periods, the synchronization change statistics table for each candidate transmission node is compiled. The number of times marked as "synchronization state" in the table is counted; this count is the total synchronization change frequency of that candidate transmission node.

[0118] Step S1355: Calculate the proportion of the total frequency of synchronous changes of each candidate transmission node to the total number of historical periods, and obtain the proportion of synchronous changes.

[0119] Dividing the total frequency of synchronous changes at each candidate transmission node by the total number of historical periods (i.e., the total number of statistical periods) yields the percentage of synchronous changes relative to the total number of historical periods, i.e., the synchronous change ratio. This ratio reflects the frequency with which the candidate transmission node and its power generation change synchronously.

[0120] Step S1356: Set the preset frequency requirement as follows: the total frequency of synchronous changes is not less than the preset frequency threshold, and the proportion of synchronous changes is not less than the preset proportion threshold.

[0121] The preset frequency requirement is the standard for screening key transmission nodes. It includes two aspects: First, the total frequency of synchronous changes cannot be lower than a preset frequency threshold to ensure that there are enough instances to prove its synchronicity; second, the proportion of synchronous changes cannot be lower than a preset proportion threshold to ensure that the occurrence of synchronous changes is not a random phenomenon, but has a certain degree of universality.

[0122] The setting of these two thresholds needs to be determined comprehensively based on factors such as the specific situation of the hydropower station, the amount of historical data, and the requirements for prediction accuracy.

[0123] Step S1357: Mark the candidate transmission nodes that meet the preset frequency requirements for both the total frequency of synchronous changes and the proportion of synchronous changes as preliminary key transmission nodes.

[0124] The total frequency of synchronous changes for each candidate transmission node is compared with a preset frequency threshold, and the proportion of synchronous changes is compared with a preset proportion threshold. Only when both requirements are met is the candidate transmission node marked as a preliminary key transmission node. This step filters out nodes that perform well in both the frequency and proportion of synchronous changes.

[0125] Step S1358: Perform cross-time period consistency analysis on the preliminary key transmission nodes, select multiple discontinuous historical time periods, and check whether the synchronous change markers of the preliminary key transmission nodes in each time period maintain the continuity that meets the preset requirements.

[0126] Although the initial key transmission nodes meet the requirements in terms of overall frequency and proportion, it is still necessary to check the stability of their synchronous changes, that is, whether the above-mentioned synchronicity can be maintained in different historical periods, rather than being concentrated in a specific period.

[0127] When conducting cross-time period consistency analysis, multiple discontinuous historical time periods can be selected. For example, time periods can be divided by season, or by high water period, normal water period, and low water period. Alternatively, multiple discontinuous time periods can be randomly selected.

[0128] For each selected time period, check the synchronization change markings of the initial key transmission nodes within that time period to see if they can maintain the synchronization change markings in multiple consecutive time periods within that time period, i.e., whether they have the continuity requirements. For example, it is required that at least a certain number of consecutive time periods within a time period be marked as synchronized.

[0129] Step S1359: If the initial key transmission node maintains continuity that meets the preset requirements in multiple time periods, then the initial key transmission node is finally marked as a key transmission node related to power generation; if the continuity of the initial key transmission node does not meet the requirements in any time period, then the synchronous change statistics of the initial key transmission node are rechecked, and if the data is correct, the initial key transmission node is removed from the initial key transmission nodes.

[0130] If a preliminary critical transmission node can maintain synchronous change continuity that meets the preset requirements in all or most of the selected time periods, it indicates that the synchronous change of the node has good cross-time stability and is a reliable node related to power generation. Finally, it is marked as a critical transmission node.

[0131] If the continuity of synchronization changes at a certain initial critical transmission node does not meet the preset requirements within a certain period, such as intermittent synchronization changes during a certain season, failing to form continuous synchronization, then the synchronization change statistics of that node need to be rechecked for errors. If the data is correct, it indicates that the synchronization of that node is not stable enough and may lose its correlation with power generation in certain situations; therefore, it should be removed from the initial critical transmission node list.

[0132] Step S13510: Organize all the final marked key transmission nodes, record the type, total frequency of synchronous changes, proportion of synchronous changes and cross-time consistency results of each key transmission node, and form a list of key transmission nodes.

[0133] All nodes that passed the final screening and were marked as key transmission nodes were organized, and the corresponding hydrological information type, total frequency of synchronous changes, proportion of synchronous changes, and cross-time period consistency analysis results (such as in which time periods they were stable) were recorded for each key transmission node. The above information was summarized to form a list of key transmission nodes, which showed the basic information of each key transmission node and the intensity and stability characteristics associated with power generation.

[0134] Step S136: Combine the state change record of each key transmission node with the attribute record of the corresponding network edge to form a single power response correlation factor. The single power response correlation factor includes the state start value, state end value, state change process description of the key transmission node, as well as the temporal continuity attribute and influence range attribute of the corresponding network edge.

[0135] For each key transmission node in the list of key transmission nodes, extract its complete state change record from the hydrological information interaction transmission network, including the node's initial state value, final state value, and detailed state change process description for the corresponding time period.

[0136] Simultaneously, the attribute records of all valid network edges associated with this key transmission node in the hydrological information interaction transmission network are extracted, namely the temporal continuity attribute value and the influence range attribute value of each associated network edge.

[0137] By organically combining the state change records of the aforementioned key transmission nodes with the attribute records of the associated network edges, a single power response correlation factor with a complete structure is formed. This factor acts like an information package, containing the state change information of the key transmission node itself and the temporal characteristics of its interaction with other nodes, comprehensively reflecting the node's potential impact on power generation.

[0138] Step S137: Arrange all individual power response correlation factors in chronological order according to the time span to form an initial power response correlation factor sequence.

[0139] After generating all individual power response correlation factors, they need to be organized in chronological order. Specifically, this means arranging them according to the chronological order of the time spans corresponding to each power response correlation factor. For example, the power response correlation factor corresponding to the earliest time span is placed at the beginning of the sequence, followed by the factor for the next time span, and so on, until the factor for the most recent time span. If multiple key transmission nodes exist within the same time span, resulting in multiple power response correlation factors, these factors can be arranged according to the type priority of the key transmission nodes or other reasonable order at that time span position. Through the above chronological sorting, an initial sequence of power response correlation factors is formed, which shows the evolution of the influence characteristics of key transmission nodes on power generation from a temporal perspective.

[0140] Step S138: Perform a time period coverage check on the initial power response correlation factor sequence, supplement the power response correlation factors corresponding to the missing time spans, form a sequence covering all preset time spans, and obtain the power response correlation factor sequence.

[0141] The initial power response correlation factor sequence may contain missing power response correlation factors for certain time spans. This could be because no key transmission nodes were identified within certain time spans, or although key transmission nodes existed, corresponding power response correlation factors could not be generated for some reason.

[0142] To ensure the integrity of the power response correlation factor sequence, a time period coverage check is required. This involves iterating through all preset time spans (i.e., all time spans from the start to the end of the analysis) and checking whether each time span has a corresponding power response correlation factor in the initial sequence.

[0143] If a factor is found to be missing for a certain time span, it needs to be supplemented. Supplementation can be done by making reasonable inferences based on the trend of power response correlation factors before and after that time span, or by re-evaluating whether a candidate transmission node should be included and generating its corresponding factor if a candidate node exists within that period but was not selected as a key node. After supplementation, a complete sequence of power response correlation factors covering all preset time spans is formed.

[0144] Step S140: Based on the power response correlation factor sequence and the historical power generation time series record of the hydropower station, construct a power response correlation rule base, which contains the correspondence between different power response correlation factor combinations and power generation change patterns.

[0145] The power response correlation factor sequence reflects the state changes and interaction characteristics of key transmission nodes, while the historical power generation time series records reflect the actual power generation changes. The purpose of constructing a power response correlation rule base is to establish the mapping relationship between these two, that is, what power generation change patterns correspond to different combinations of power response correlation factors.

[0146] By analyzing the various combinations of power response correlation factor sequences in historical data and the corresponding power generation change patterns when they occur, a series of correlation rules can be summarized. These rules are stored in a rule base. When it is necessary to predict the power generation in the current period, it is only necessary to match the current power response correlation factor combination with the rules in the rule base to obtain the corresponding power generation change pattern prediction.

[0147] Step S141: Divide the power response correlation factor sequence into a training power response correlation factor sequence and a test power response correlation factor sequence. At the same time, divide the corresponding historical power generation time series record of the hydropower station into a training power record and a test power record. The time span of the training power response correlation factor sequence and the training power record are completely corresponding, and the time span of the test power response correlation factor sequence and the test power record are completely corresponding.

[0148] To construct and validate a power response association rule base, existing power response association factor sequences and corresponding historical power generation time-series records need to be divided into training and testing sets. The training set is used to learn and summarize association rules from the data, including training power response association factor sequences and corresponding training power records. The testing set is used to evaluate the generalization ability and prediction accuracy of the learned rules, including test power response association factor sequences and corresponding test power records.

[0149] When partitioning, it's crucial to ensure a complete correspondence between the training power response correlation factor sequence and the training power record's time span. This means that each power response correlation factor in the training set has a corresponding training power record segment from the same period. Similarly, the test power response correlation factor sequence and the test power record's time span must also completely correspond. The partition ratio can be determined based on the data size and model training needs. For example, a common approach is to use the majority of the data for training and a smaller portion for testing.

[0150] Step S142: Perform feature classification on each power response correlation factor in the training power response correlation factor sequence. Based on the type of key transmission nodes and the attribute features of the corresponding network edges, divide the power response correlation factors into different power response correlation factor categories. Each power response correlation factor category corresponds to a set of similar state changes and network edge attribute features.

[0151] Each power response correlation factor in the training power response correlation factor sequence has a unique state change record and network edge attribute record. In order to discover patterns, factors with similar characteristics need to be grouped into one category for feature classification.

[0152] The classification is based on two main aspects: First, the type of key transmission nodes. Different types of key nodes (such as upstream water inflow, reservoir water level, etc.) have different impact mechanisms on power generation and should be used as the primary basis for classification. Second, under the same type of key node, the characteristics of its state change (such as the trend type and change magnitude level reflected in the state change process description) and the attribute characteristics of the corresponding network edge (such as the range of time-series continuity attributes and the range of influence attributes).

[0153] Power response correlation factors with the same key node type, similar state change characteristics, and similar network edge attribute characteristics are grouped into the same power response correlation factor category. Each category represents a specific key node influence characteristic pattern.

[0154] Step S143: Analyze the historical power change segments in the training power record. Based on the direction of the difference between the initial power value and the final power value and the description of the change process, divide the historical power change segments into different power generation change patterns. Each power generation change pattern corresponds to a power change trend.

[0155] The analysis of historical power change segments in the training power records aims to classify them into different power generation change patterns. The classification is based primarily on two criteria:

[0156] First, the direction of the difference between the initial power value and the final power value, that is, whether the power generally increases (positive difference) or decreases (negative difference) during this period, or remains basically unchanged (difference close to zero).

[0157] Second, the description of the power change process, that is, the specific change path and characteristics of power within a period of time, such as "continuous and steady increase", "rapid increase followed by stabilization", "slow decrease", "fluctuating increase", "fluctuating decrease", "basically remain constant", etc.

[0158] Combining the two aspects mentioned above, historical power variation segments can be divided into several different power generation variation patterns. These include patterns such as "continuous increase pattern," "decreasing fluctuation pattern," and "stable pattern." Each pattern corresponds to a specific trend in power variation over time.

[0159] Step S144: Statistically analyze the frequency of combinations of different power response correlation factor categories in the training power response correlation factor sequence, and the power generation change pattern corresponding to each power response correlation factor category combination, to form a preliminary power response correlation factor combination-power generation pattern correspondence table.

[0160] After classifying power response correlation factors and dividing power generation change patterns, the correspondence between them is then sought. Each time span in the training power response correlation factor sequence is traversed. For each time span, all power response correlation factor categories appearing within that period are identified. These categories form a power response correlation factor category combination for that period (multiple categories appearing simultaneously). The frequency of each category combination is recorded, i.e., how many times the combination has appeared in the training set in total. Simultaneously, the power generation change pattern corresponding to the historical power change segment each time the category combination appears is recorded.

[0161] The above information is organized into a preliminary table of power response correlation factor combination-power generation mode correspondence. Each row represents a power response correlation factor combination category, and the columns include the composition of the combination, the frequency of occurrence, and the corresponding power generation mode (there may be multiple combinations if the combination corresponds to multiple modes).

[0162] Step S145: Perform validity screening on the preliminary power response correlation factor combination-power generation mode correspondence table, and retain the power response correlation factor combination-power generation mode correspondence that appears more frequently than the preset frequency threshold during the training period and corresponds to the same power generation change mode each time.

[0163] The preliminary correspondence table may contain some unreliable correspondences. For example, some combinations of power response correlation factor categories appear very infrequently, and their corresponding power generation change patterns may be accidental; or some combinations may appear frequently but correspond to multiple different power generation change patterns, indicating that the correlation between the combination and the pattern is not strong.

[0164] Therefore, effectiveness screening is necessary. There are two main screening criteria: first, the frequency of occurrence of the power response correlation factor category combination during the training period must exceed a preset frequency threshold to ensure the combination has a certain degree of representativeness; second, each time the combination appears, the corresponding power generation change pattern must be the same, i.e., it must have a unique corresponding pattern to ensure a deterministic correlation between the combination and the pattern.

[0165] Only correspondences that satisfy both of the above conditions are retained.

[0166] Step S146: Organize the filtered power response correlation factor combination-power generation mode correspondence into rule entries. Each rule entry includes a description of the power response correlation factor category combination, a description of the corresponding power generation change mode, and the number of applicable time periods for the corresponding rule during the training period.

[0167] The power response correlation factor combinations and power generation mode correspondences, after effectiveness screening, are standardized and organized into rule entries. Each rule entry is an independent correlation rule, containing the following core content:

[0168] First, a description of the power response correlation factor category combination is provided, detailing the power response correlation factor category combination to which the rule applies, i.e., which categories of factors must appear simultaneously for the rule to take effect.

[0169] Second, the corresponding description of the power generation change pattern clearly indicates what the expected power generation change pattern is when this combination occurs, including the direction of the change trend and the characteristics of the change process.

[0170] Third, the number of applicable time periods for the corresponding rule during the training period, that is, how many time periods the rule was applied to in the training set in total. This number reflects the empirical support for the rule.

[0171] Step S147: Input all rule entries into the rule base framework, and add a rule number and time validity mark to each rule entry. The time validity mark is determined by the time distance between the applicable time period of the corresponding rule and the current time period.

[0172] All compiled rule entries are entered into a pre-defined rule base framework for management. For ease of referencing and identification, each rule entry is assigned a unique rule number. Furthermore, a timeliness flag is added to each rule entry. The purpose of the timeliness flag is to distinguish the newness or oldness of different rules; its value is primarily determined by the time distance between the rule's applicable period in the training set and the current prediction period. For example, rules with a closer applicable period than the current period are flagged as "high" timeliness, while those with a greater distance are flagged as "low". During subsequent rule matching, rules with higher timeliness can be assigned higher priority.

[0173] Step S148: Input the test power response correlation factor sequence into the rule base framework, check whether the power response correlation factor category combination corresponding to each test power response correlation factor sequence can match the rule entries in the rule base, and observe whether the power generation change pattern corresponding to the matching rule entry is consistent with the actual power change pattern in the test power record; retain the rule entries whose matching accuracy meets the preset accuracy requirement, delete the rule entries whose matching accuracy is lower than the preset accuracy requirement, and form the power response correlation rule base.

[0174] To ensure the constructed rule base has good predictive performance, it is necessary to validate and further filter rule entries using a test set. For example, the test power response correlation factor sequence is input into the rule base framework containing rule entries. For each time span in the test set, the combination of power response correlation factor categories for that period is first determined, and then the rule base is searched for rule entries that match this combination. For the found matching rule entries, their corresponding power generation change pattern descriptions are extracted and compared with the actual power change patterns for that period in the test power records to observe whether they are consistent. The number of successful and unsuccessful matches for each rule entry in the test set is counted, and its matching accuracy is calculated.

[0175] An accuracy requirement is preset. Rule entries that meet or exceed this requirement are retained, while rule entries with accuracy below the requirement are deleted. After the above testing and filtering, a reliable power response association rule base is finally formed.

[0176] Step S1481: Divide the test power response correlation factor sequence into multiple test power response correlation factor segments according to the time span. Each test power response correlation factor segment contains all power response correlation factors within that time span.

[0177] Similar to the processing of the training data, the test power response correlation factor sequence is split into multiple test power response correlation factor segments according to the same time span. Each segment corresponds to a specific time span and contains all the power response correlation factors within that span.

[0178] Step S1482: Determine the power response correlation factor category for each power response correlation factor in each test power response correlation factor segment, and form a power response correlation factor category combination corresponding to the test power response correlation factor segment.

[0179] For each power response correlation factor in each test power response correlation factor segment, its power response correlation factor category is determined according to the power response correlation factor classification criteria determined in step S142.

[0180] The categories of all power response correlation factors in the test segment are summarized to form the power response correlation factor category combination corresponding to the test power response correlation factor segment.

[0181] Step S1483: Search the rule base framework for rule entries that are completely consistent with the combination of power response correlation factor categories. If a unique rule entry is found, record the power generation change pattern corresponding to the rule entry. If multiple rule entries are found, select the rule entry with the strongest timeliness according to the timeliness flag of the rule entry and record the power generation change pattern corresponding to the rule entry. If no matching rule entry is found, mark the test power response correlation factor segment as having no matching rule.

[0182] The test power response association factor category combination is input into the retrieval module of the rule base framework. The retrieval module searches the rule base for a rule entry that exactly matches the combination.

[0183] If the search result is a unique rule entry, the power generation change pattern corresponding to that rule entry is directly recorded. If multiple rule entries match the combination (this may be due to fuzzy matching allowed during rule base construction or the existence of multiple similar rules), the rule entry with the strongest timeliness is selected based on its timeliness flag, and its corresponding power generation change pattern is recorded. If no rule entry matching the combination is found in the rule base, the test power response correlation factor fragment is marked as "no matching rule state".

[0184] Step S1484: Extract the actual power change segment from the test power record that is contemporaneous with the factor segment associated with the test power response, and determine the actual power change pattern.

[0185] Extract the corresponding actual power change segment from the test power record, which corresponds to the factor segment associated with the current test power response. Based on the power start value, end value, and change process description of this actual power change segment, determine its corresponding actual power generation change pattern.

[0186] Step S1485: Compare the recorded power generation change pattern with the actual power change pattern. If the trend direction and range of the power generation change pattern and the actual power change pattern both meet the preset consistency requirements, then mark the matching of the corresponding rule entry as a successful match; otherwise, mark the matching of the corresponding rule entry as a failed match.

[0187] The predicted power generation change pattern is compared in detail with the actual power generation change pattern. The comparison focuses on whether the direction of the change trend is consistent (e.g., both are increasing, both are decreasing, or both are stable) and whether the range of change is within the preset consistency requirements (e.g., whether the predicted change magnitude is the same as the actual change magnitude). If both meet the preset consistency requirements in both trend direction and range, the rule match is marked as a "successful match." Otherwise, it is marked as a "failed match."

[0188] Step S1486: For each rule entry in the rule base framework, count the number of successful matches and the number of failed matches of the corresponding rule entry in all test power response correlation factor fragment matches, and determine the proportion of successful matches of the corresponding rule entry.

[0189] Iterate through each rule entry in the rule base, counting the total number of times it was invoked during all test power response association factor fragment matching processes (i.e., how many times the test fragment combination matched the rule), as well as the number of successful matches and the number of failed matches. Calculate the proportion of successful matches to the total number of matches to obtain the accuracy rate of that rule entry.

[0190] Step S1487: Set the preset accuracy requirement to be that the percentage of successful matches is not less than the preset success rate threshold.

[0191] Based on the required prediction accuracy, a preset success rate threshold is set. This preset success rate threshold is the standard for judging whether a rule entry has sufficient predictive accuracy.

[0192] Step S1488: Filter out rule entries whose success rate meets the preset accuracy requirement, and retain rule entries whose success rate meets the preset accuracy requirement in the rule base.

[0193] The percentage of successful matches for each rule entry is compared to a preset success rate threshold. Rule entries with a success rate that reaches or exceeds this threshold are retained; these rules are considered to have good generalization and predictive ability.

[0194] Step S1489: For rule entries where the percentage of successful matches is lower than the preset accuracy requirement, analyze the reasons for the failed matches of the corresponding rule entries. If the failed matches are due to special state changes of the test power response associated factor fragment, supplement the rule supplement clauses corresponding to the special state and redetermine the percentage of successful matches after the supplement. If the percentage of successful matches after the supplement still does not meet the requirements, delete the corresponding rule entries.

[0195] For rule entries where the success rate is lower than the preset accuracy requirement, it is necessary to analyze the specific reasons for the failed matches. If the failed matches are due to the presence of special state change features in the test power response correlation factor fragment that have not appeared in the training set, causing the rule to fail to match accurately, then it is advisable to add supplementary rule clauses for the aforementioned special states to the rule entry.

[0196] After adding the supplementary clause, recalculate the percentage of successful matches for that rule entry (including the supplementary clause) on the test set. If the percentage after addition meets the preset accuracy requirement, retain the rule entry and its supplementary clause. If the requirement is still not met after addition, or if the failure reason is not a special state change, delete the rule entry from the rule base.

[0197] Step S14810: Integrate the retained rule entries and supplementary clauses to form a power response association rule base.

[0198] The rules that have been tested and filtered and retained, along with the supplementary clauses added to some rules, are integrated to form a complete and reliable power response association rule library.

[0199] Step S150: Match the power response correlation factor sequence generated from the currently acquired dynamic hydrological information set with the power response correlation rule base to generate the power generation prediction result of the hydropower station for the current period.

[0200] When it is necessary to predict the power generation of a hydropower station in the current period, the first step is to obtain the set of dynamic hydrological interaction information for the current period. Following the same process as historical data processing, the hydrological interaction transmission network for the current period is constructed, key transmission nodes are extracted, and a power response correlation factor sequence is generated.

[0201] Then, the current power response correlation factor sequence is matched against the established power response correlation rule base. For each time span in the sequence, matching rule entries are searched based on the combination of its power response correlation factor categories to obtain the predicted power generation change pattern. Finally, a complete power generation forecast for the current period is generated based on these patterns.

[0202] Step S151: Obtain the set of dynamic water situation interaction information for the current time period, and construct the water situation interaction transmission network for the current time period. The water situation interaction transmission network includes the state change records of each water situation information for the current time period as network nodes, and the mutual triggering relationship between each network node as network edges.

[0203] Obtain a set of dynamic interactive water situation information for the current forecast period (e.g., the next 24 hours, 48 ​​hours, etc.). This information may come from real-time monitoring data, short-term weather forecast data, downstream water use plan declaration data, etc.

[0204] Following the same method and process as in steps S121 to S1279 for constructing the historical hydrological information interaction and transmission network, the current time period's dynamic hydrological information interaction set is processed to construct the current time period's hydrological information interaction and transmission network. This hydrological information interaction and transmission network also uses the state change records of each hydrological information in the current time period as network nodes, and the mutual triggering relationships between each node as network edges, and undergoes optimization steps such as time sequence validity verification and node connectivity checks.

[0205] Step S152: Extract key transmission nodes and corresponding network edge attributes from the hydrological interaction transmission network of the current time period, and generate the power response correlation factor sequence of the current time period. The power response correlation factor sequence contains the power response correlation factor corresponding to each time span within the current time period.

[0206] Referring to the method in steps S130 to S138 for extracting key transmission nodes from the historical hydrological interaction transmission network and generating power response correlation factor sequences, the hydrological interaction transmission network for the current period is processed.

[0207] Key transmission nodes in the network are extracted, and these nodes are determined based on methods such as the current water situation and the similarity of historical key node features. Then, the state change records of each key transmission node are combined with the corresponding network edge attributes to generate a single power response correlation factor for the current time period, and these factors are arranged in chronological order to form a sequence of power response correlation factors for the current time period.

[0208] Step S153: Divide the power response correlation factor sequence of the current time period into multiple current power response correlation factor segments, each current power response correlation factor segment containing all power response correlation factors within the time span.

[0209] The power response correlation factor sequence for the current time period is divided into multiple current power response correlation factor segments according to the same time span as the historical processing. Each segment corresponds to a current time span and contains all power response correlation factors within that span.

[0210] Step S154: Perform power response correlation factor category determination for each power response correlation factor in each current power response correlation factor segment, determine the power response correlation factor category to which each power response correlation factor belongs, and form a power response correlation factor category combination corresponding to the current power response correlation factor segment.

[0211] According to the power response correlation factor classification criteria determined in step S142, each power response correlation factor in the current power response correlation factor segment is classified to determine its corresponding power response correlation factor category. All factor categories in the same current power response correlation factor segment are combined to form the current power response correlation factor category combination corresponding to that segment.

[0212] Step S155: Retrieve rule entries that match the current power response association factor category combination in the power response association rule base, prioritizing the retrieval of rule entries with the latest timeliness marker.

[0213] The current power response correlation factor category combination is input into the power response correlation rule base for retrieval. During the retrieval process, the most recent rule entries with the most timely tags are given priority to utilize the latest empirical knowledge.

[0214] Step S156: If a matching rule entry is found, extract the power generation change pattern description corresponding to the rule entry. The power generation change pattern description includes the direction of change trend, the range of change magnitude, and the characteristics of the change process.

[0215] If a rule entry matching the current power response associated factor category combination is found in the rule base, the power generation change pattern description corresponding to that rule entry is extracted. This description details the expected power generation change trend direction (increase, decrease, stable), change range (small, medium, large), and change process characteristics (continuous, fluctuating, fast at first and slow later, etc.).

[0216] Step S157: Based on the description of the power generation change pattern and combined with the time span length corresponding to the current power response correlation factor segment, generate a power generation prediction segment within the time span. The power generation prediction segment includes the predicted power start value, the predicted power end value, and a description of the predicted power change process.

[0217] Based on the extracted description of power generation change patterns and the time span corresponding to the current power response correlation factor segment, and referring to the specific numerical range and process characteristics of power changes under the same historical patterns, a power generation prediction segment within this time span is generated.

[0218] The prediction segment includes the predicted starting power value (which can be based on the predicted ending value of the previous period or the current initial power estimate), the predicted ending power value, and a description of the predicted power change process within that period, so that the prediction results have both specific numerical values ​​and process characteristics.

[0219] Step S158: If no completely matching rule entry is found, find the rule entry with the highest similarity to the current power response associated factor category combination, extract the power generation change pattern description of the corresponding similar rule entry, and adjust the pattern in combination with the special state changes of the current hydrological information to generate the adjusted power generation prediction segment.

[0220] If no rule entry is found in the rule base that perfectly matches the current power response associated factor category combination, a similarity rule search mechanism is initiated. This mechanism calculates the similarity between the current combination and all rule entry combinations in the rule base, and identifies the rule entry with the highest similarity.

[0221] The power generation change pattern description of the similarity rule entry is extracted, and then the current hydrological information is analyzed to determine whether there are special state changes that differ from the applicable conditions of the similarity rule. Based on these special state changes, the pattern description of the similarity rule is appropriately adjusted, such as adjusting the change magnitude and correcting the change process characteristics, and finally, an adjusted power generation prediction segment is generated.

[0222] Step S159: Segment all power generation prediction segments corresponding to all time spans in chronological order to form a preliminary power generation prediction sequence for the current time period.

[0223] The power generation forecast segments corresponding to all time spans within the current period are spliced ​​together sequentially according to time. The ending value of the forecast segment of the previous time span is used as the starting value of the forecast segment of the next time span, thus forming a continuous preliminary power generation forecast sequence for the current period.

[0224] Step S1510: Perform a trend coherence check on the preliminary power generation prediction sequence to ensure that the direction and magnitude of power change trends between power generation prediction segments in adjacent time spans meet the preset coherence requirements.

[0225] Preliminary power generation forecast sequences may exhibit inconsistencies in trends between adjacent forecast segments. For example, a segment might predict a continuous increase in power generation, while a subsequent segment might suddenly predict a significant decrease, without any reasonable basis in hydrological changes to support such abrupt changes. Therefore, a trend consistency check is necessary for the preliminary forecast sequence. This check examines whether the direction of power change trends between adjacent time spans is consistent or whether there is a reasonable transition, and whether the increase or decrease in the magnitude of the change is smooth, conforming to the general laws governing the impact of hydrological changes on power generation and the pre-set consistency requirements.

[0226] For example, step S15101: Extract power generation prediction segments from two adjacent time spans in the preliminary power generation prediction sequence, and denot them as the preceding power generation prediction segment and the following power generation prediction segment, respectively.

[0227] Two adjacent power generation prediction segments are selected from the preliminary power generation prediction sequence. The segment that is earlier is called the preceding power generation prediction segment, and the segment that is later is called the following power generation prediction segment.

[0228] Step S15102: Obtain the predicted power termination value and the predicted power change trend direction of the preceding power generation prediction segment; obtain the predicted power starting value and the predicted power change trend direction of the subsequent power generation prediction segment.

[0229] Extract the predicted power termination value of the preceding power generation prediction segment, i.e., the predicted power value at the end of the preceding power generation prediction segment; and the direction of the predicted power change trend of the preceding power generation prediction segment, such as "increasing", "decreasing", or "stable". Similarly, extract the predicted power starting value of the subsequent power generation prediction segment, i.e., the predicted power value at the beginning of the segment; and the direction of the predicted power change trend of the segment.

[0230] Step S15103: Calculate the difference between the predicted power termination value of the preceding power generation prediction segment and the predicted power start value of the subsequent power generation prediction segment to obtain the power transition difference.

[0231] The difference between the termination value of the preceding segment and the starting value of the following segment is calculated. This difference reflects the power value difference between the two adjacent segments at the connection point, i.e., the power transition difference.

[0232] Step S15104: Determine whether the predicted power change trend direction of the preceding power generation prediction segment is consistent with the predicted power change trend direction of the subsequent power generation prediction segment. If they are consistent, check whether the absolute value of the power transition difference is less than the preset same-direction transition threshold. If they are inconsistent, check whether the absolute value of the power transition difference is within the preset reverse transition threshold range.

[0233] Compare the predicted power change trends of the preceding and following segments to see if they are consistent. If they are consistent (e.g., both increasing or both decreasing), check if the absolute value of the power transition difference is less than a preset same-direction transition threshold. This same-direction transition threshold sets the maximum allowable power value jump under the same-direction trend. If they are inconsistent (e.g., the preceding segment increases, the following segment decreases), check if the absolute value of the power transition difference is within a preset reverse transition threshold range. This reverse transition threshold range sets the reasonable range for power value change when the trend reverses.

[0234] Step S15105: If the power transition difference meets the corresponding threshold requirement, and the change process characteristics of the preceding power generation prediction segment do not have obvious conflicts with the change process characteristics of the subsequent power generation prediction segment, then mark the transition of the adjacent power generation prediction segment as a continuous transition.

[0235] If the power transition difference meets the corresponding threshold requirements (less than the same-direction threshold when in the same direction, and within the reverse range when in the opposite direction), and there is no obvious logical conflict between the change process characteristics of the preceding segment (such as "continuous and steady increase") and the change process characteristics of the following segment (such as "continuous and steady increase" or "increase with a slowing growth rate") (e.g., the preceding segment is "violent fluctuations" and the following segment suddenly becomes "completely stable" without a reasonable explanation), then the transition between the adjacent predicted segments is marked as a "coherent transition".

[0236] Step S15106: If the power transition difference does not meet the corresponding threshold requirement, or if there is a significant conflict between the change process characteristics of the preceding power generation prediction segment and the change process characteristics of the subsequent power generation prediction segment, then mark the transition of the adjacent power generation prediction segment as a discontinuous transition and record the type of discontinuous transition.

[0237] If the power transition difference does not meet the corresponding threshold requirement, or if there is a significant conflict in the change process characteristics between the preceding and following segments, the adjacent predicted segment is marked as an "incoherent transition." At the same time, the specific type of incoherent transition is recorded, such as "excessive transition difference in the same trend," "transition difference in the opposite trend exceeding the range," or "conflicting change process characteristics," for subsequent targeted adjustments.

[0238] Step S15107: Traverse all adjacent power generation prediction segments in the preliminary power generation prediction sequence and count the number of discontinuous transitions and their corresponding positions.

[0239] Traverse the entire preliminary power generation prediction sequence, perform the above-mentioned coherence check on all adjacent power generation prediction segments, count the total number of discontinuous transitions in the entire sequence, and record the specific location of each discontinuous transition (i.e., between which two adjacent time spans).

[0240] Step S15108: If the number of discontinuous transitions is zero, then the preliminary power generation prediction sequence is confirmed to meet the trend consistency requirement.

[0241] If, after inspection, there are no discontinuous transitions in the preliminary power generation prediction sequence, that is, all adjacent segment transitions are coherent transitions, then the preliminary prediction sequence is confirmed to meet the trend coherence requirement and can be directly used as the subsequent result.

[0242] Step S15109: If the number of discontinuous transitions is not zero, then for each discontinuous transition position, find the power response correlation factor sequence and matching rule entries that the adjacent power generation prediction segment corresponding to the discontinuous transition position depends on.

[0243] If discontinuous transitions exist, each discontinuous transition location is analyzed in detail. The preceding and following power generation prediction segments corresponding to that location are identified based on which power response correlation factor sequences and which rule entries were matched.

[0244] Step S151010: Analyze whether the rule matching deviation is caused by the sudden change in the current hydrological information. If there is a sudden change in the current hydrological information, supplement the temporary rule clause corresponding to the sudden state, and regenerate the power generation prediction segment of the discontinuous transition position based on the temporary rule clause.

[0245] Analyzing the causes of discontinuous transitions reveals that they are due to sudden, abnormal changes in the current hydrological information (such as a sudden rainstorm causing a sharp increase in upstream water flow, a situation rarely seen in historical data, and lacking corresponding rules in the rule base). This leads to rule matching deviations, resulting in discontinuous transitions in the prediction segments. In this case, based on domain expert knowledge or emergency response experience for similar emergencies, temporary rule clauses are added to address this sudden change. Based on these temporary rule clauses, the preceding or following power generation prediction segments for the discontinuous transition location are regenerated.

[0246] Step S151011: If there is no sudden change in the current hydrological information, adjust the predicted power value of the adjacent power generation prediction segment so that the power transition difference meets the threshold requirement, while keeping the predicted power change trend direction consistent with the rule entry description.

[0247] If the discontinuous transition is not caused by a sudden change in state, but by a problem with the predicted values ​​themselves, then the starting or ending values ​​of the predicted power of adjacent predicted segments shall be fine-tuned, provided that the trend of power generation change in each predicted segment is consistent with the description of the rule entries.

[0248] For example, by appropriately increasing or decreasing the starting value of the subsequent segment, or adjusting the ending value of the preceding segment, the power transition difference between the two can meet the preset threshold requirement, thereby eliminating the inconsistency of the transition.

[0249] Step S151012: After adjustment, perform trend coherence check again until all adjacent power generation prediction segments are marked as coherent transitions, and obtain a power generation prediction sequence that meets the trend coherence requirements.

[0250] After adjusting for discontinuous transitions, the entire power generation prediction sequence is re-checked for trend consistency. Steps S15101 to S151011 are repeated until all transitions between adjacent power generation prediction segments are marked as coherent transitions. The resulting power generation prediction sequence then meets the trend consistency requirement.

[0251] Step S151013: Use the power generation prediction sequence that meets the trend continuity requirement as the power generation prediction result of the hydropower station for the current period.

[0252] Ultimately, the power generation forecast sequence, after trend consistency adjustment and verification, is determined as the power generation forecast result for the hydropower station in the current period.

[0253] Figure 2 This illustration shows exemplary hardware and software components of a hydropower station power generation prediction system 100 based on hydrological information, which can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used on the hydropower station power generation prediction system 100 based on hydrological information and to perform the functions in this application. For example, the hydropower station power generation prediction system 100 based on hydrological information may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the hydropower station power generation prediction system 100 based on hydrological information may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The hydropower station power generation prediction system 100 based on hydrological information also includes an I / O interface 150 between the computer and other input / output devices.

[0254] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for predicting the power generation capacity of a hydropower station based on hydrological information, characterized in that, The method includes: A set of dynamic hydrological interaction information for a hydropower station is obtained. The set of dynamic hydrological interaction information includes upstream water inflow time sequence information, reservoir water level fluctuation information, water flow velocity change information, and downstream water demand response information. Each piece of information in the set of dynamic hydrological interaction information is marked with a continuous time series, and each piece of dynamic hydrological interaction information contains state change records for multiple time periods. A hydrological interaction transmission network is constructed based on the aforementioned set of dynamic hydrological information. The hydrological interaction transmission network uses the state change records of each hydrological information as network nodes and the mutual triggering relationships between the state changes of each hydrological information as network edges. The attributes of the network edges are determined by the temporal continuity and influence range of the triggering relationships. From the hydrological interaction transmission network, key transmission nodes and corresponding network edge attributes related to power generation are extracted to generate a power response correlation factor sequence. Each power response correlation factor in the power response correlation factor sequence contains the state change record of the key transmission node and the attribute record of the corresponding network edge. Based on the power response correlation factor sequence and the historical power generation time series record of the hydropower station, a power response correlation rule base is constructed. The power response correlation rule base contains the correspondence between different power response correlation factor combinations and power generation change patterns. The power response correlation factor sequence generated by the currently acquired dynamic hydrological information set is matched with the power response correlation rule base to generate the power generation prediction result of the hydropower station for the current period.

2. The method for predicting the power generation of a hydropower station based on hydrological information according to claim 1, characterized in that, The construction of the hydrological information interaction transmission network based on the hydrological dynamic interaction information set includes: The state change records of the upstream water inflow time sequence information, reservoir water level fluctuation information, water flow velocity change information and downstream water demand response information of the hydropower station in the hydrological dynamic interaction information set are divided into multiple continuous state segments. Each state segment corresponds to the same time span, and each state segment contains the state start value, state end value and state change process description within the corresponding time period. Select any two different types of hydrological information state segments as the trigger node and the triggered node, respectively, and analyze the temporal relationship between the state change process description of the trigger node and the state change process description of the triggered node to determine whether the trigger node changes state before the triggered node. If the triggering node changes state before the triggered node, further analyze the correlation trend between the state change magnitude of the triggering node and the state change magnitude of the triggered node to determine whether there is a unidirectional or inverse correlation between the triggering node and the triggered node. If there is a change association in the same or opposite direction between the triggering node and the triggered node, a network edge is established between the triggering node and the triggered node. The time interval between the triggering node and the triggered node is used as the temporal continuity attribute of the network edge, and the length of the coverage period of the change association is used as the influence range attribute of the network edge. Repeatedly execute the steps of state fragment splitting, temporal relationship analysis, change correlation judgment and network edge establishment, traverse all state fragment combinations of different types of hydrological information, and establish multiple sets of network edges; All state fragments are treated as network nodes, and all established network edges are connected according to the network node association relationship to form an initial hydrological information interaction and transmission network. The temporal validity of the network edges in the initial hydrological information interaction and transmission network is verified. Network edges that can maintain stable temporal continuity and influence range attributes over multiple consecutive time spans are retained, while network edges that exist only in a single or two time spans are deleted, thus obtaining the hydrological information interaction and transmission network.

3. The hydropower generation prediction method based on hydrological information according to claim 2, characterized in that, The process involves verifying the temporal validity of network edges in the initial hydrological information interaction and transmission network. Network edges that maintain stable temporal continuity and influence range attributes across multiple consecutive time spans are retained, while those existing only in a single or two time spans are deleted. This process yields the hydrological information interaction and transmission network, including: Extract the temporal continuity attribute records and influence range attribute records of each network edge in the initial hydrological interaction transmission network across all time spans to form the attribute temporal sequence corresponding to each network edge; Perform continuous stability analysis on the attribute time series of each network edge to check whether there are attribute records with multiple consecutive time spans, wherein the change amplitude of the time series continuity attribute does not exceed the preset continuity change threshold, and the change amplitude of the influence range attribute does not exceed the preset range change threshold. If there are attribute records with multiple consecutive time spans that satisfy the stability conditions of the change amplitude of the temporal continuity attribute and the change amplitude of the influence range attribute, then the network edge is marked as a candidate valid network edge, and the length of the longest consecutive stable time span is recorded. The longest continuous stable time span of the candidate valid network edges is used to verify the correlation, and the state change process descriptions of the corresponding triggering node and the triggered node within the corresponding time period are analyzed to see if the change correlation type remains consistent. If the state change process descriptions of the triggering node and the triggered node always maintain the same change association type, then the candidate valid network edge is confirmed as a valid network edge; if the state change process descriptions of the triggering node and the triggered node change the change association type within a continuous stable time span, then the valid time period of the network edge is truncated into a sub-time period with the same change association type, and the sub-time period is re-checked to see if it meets the requirement of multiple consecutive time spans. If the truncated sub-time period still meets the requirement of multiple consecutive time spans, then the network edge is confirmed as a valid network edge, and the valid time period of the network edge is marked; if the truncated sub-time period does not meet the requirement of multiple consecutive time spans, then the network edge is marked as an invalid network edge. Collect all confirmed valid network edges and the network nodes corresponding to the valid network edges, and reconstruct the network structure according to the relationship between network nodes and network edges; Perform a network node connectivity check on the reconstructed network structure, find at least one propagation path for all network nodes through valid network edges, and delete any network nodes that are not connected by any valid network edges. By integrating effective network edges, non-isolated network nodes, and their corresponding attribute records, a hydrological information interaction and transmission network is formed.

4. The method for predicting the power generation capacity of a hydropower station based on hydrological information according to claim 3, characterized in that, The process of checking network node connectivity in the reconstructed network structure involves finding at least one propagation path for each network node using valid network edges. If a network node exists that is not connected by any valid network edge, then that node is deleted. This includes: Assign a unique network node identifier to each network node in the reconstructed network structure, and establish a list of network node identifiers; Select the first network node in the list of network node identifiers as the starting network node. Through the valid network edges connected to the starting network node, find other network nodes that are directly connected to the starting network node and mark the found network nodes as connected network nodes. Using the network node marked as a connected network node as the new starting network node, continue to search for unmarked network nodes through the valid network edges connected by the new starting network node. Mark the found unmarked network nodes as connected network nodes, and repeat the search until no new unmarked network nodes can be found. The number of network nodes marked as connected is counted and compared with the total number of network nodes in the network node identifier list; If the number of network nodes marked as connected network nodes is equal to the total number of network nodes, then the network structure remains unchanged; If the number of network nodes marked as connected network nodes is less than the total number of network nodes, then the unmarked network nodes are determined to be network nodes without any valid network edge connections, and the identifier of the unmarked network nodes and the corresponding hydrological information type are recorded. Analyze the status change records of the hydrological information type corresponding to the network node without any effective network edge connection in the historical time period, and check whether the absence of any effective network edge connection is due to the lack of significant status change of the hydrological information in the current time period. If there are no effective network edge connections because the hydrological information has no significant state change in the current time period, then add weak network edges between the hydrological information and other hydrological information. The temporal continuity attribute and the scope of influence attribute of the weak network edges are taken as the historical average level, and the network node connectivity check is performed again. If a network node still has no valid network edge connections after adding weakly associated network edges, or if it has no valid network edge connections for other reasons, then the network node will be deleted from the network structure. After deleting network nodes that have no valid network edge connections, the network node identifier list is reorganized, and the valid network edges of the remaining network nodes are renumbered so that the relationship between network nodes and network edges in the network structure can be clearly presented through renumbering. A second network node connectivity check was performed to confirm that all remaining network nodes were marked as connected, thus forming a network structure without any isolated network nodes.

5. The method for predicting the power generation of a hydropower station based on hydrological information according to claim 1, characterized in that, The step of extracting key transmission nodes and corresponding network edge attributes related to power generation from the hydrological interaction transmission network to generate a power response correlation factor sequence includes: Collect historical power generation time-series records of hydropower stations, and break down the historical power generation time-series records into historical power change segments with the same time span as the state segments of the hydrological dynamic interaction information set. Each historical power change segment contains the power start value, power end value and power change process description within the corresponding time period. The state segment corresponding to each network node in the hydrological interactive transmission network is subjected to time-series coupling analysis with the historical power change segment of the same period to observe whether there are synchronous change characteristics between the change process description of the network node state segment and the change process description of the historical power change segment. If the description of the change process of a network node state segment and the description of the change process of a historical power change segment show synchronous change characteristics, then the network node is marked as a candidate transmission node, and the number of synchronous change periods and the characteristic type of synchronous change are recorded. Extract all network edges associated with each candidate propagation node, obtain the temporal continuity attribute and influence range attribute of these network edges, and form a set of candidate node association attributes; For each candidate transmission node, the frequency of synchronous changes between the candidate transmission node and historical power change segments in all historical time periods is counted. Candidate transmission nodes whose synchronous change frequency meets the preset frequency requirements are selected and marked as key transmission nodes related to power generation. The state change records of each key transmission node are combined with the attribute records of the corresponding network edge to form a single power response correlation factor. The single power response correlation factor includes the state start value, state end value, state change process description of the key transmission node, as well as the temporal continuity attribute and influence range attribute of the corresponding network edge. Arrange all individual power response correlation factors sequentially according to the time span to form an initial power response correlation factor sequence; The initial power response correlation factor sequence is checked for time period coverage, and the power response correlation factors corresponding to the missing time spans are supplemented to form a sequence covering all preset time spans, thus obtaining the power response correlation factor sequence.

6. The method for predicting the power generation of a hydropower station based on hydrological information according to claim 5, characterized in that, For each candidate transmission node, the frequency of its synchronous changes with historical power variation segments across all historical time periods is counted. Candidate transmission nodes whose synchronous change frequency meets a preset frequency requirement are selected, and these selected candidate transmission nodes are marked as key transmission nodes related to power generation, including: A synchronous change statistics table is established for each candidate conduction node. The synchronous change statistics table includes the type of candidate conduction node, the corresponding historical time period number, and a mark indicating whether it changes synchronously with the historical power change segment of the same period. Traverse all historical time periods. For each candidate transmission node, check whether the description of the state change process of the candidate transmission node in the historical time period and the description of the change process of the historical power change segment in the same period meet the synchronous change condition. The synchronous change condition is that the difference between the start time of the state change of the candidate transmission node and the start time of the state change of the historical power change segment does not exceed the preset time difference threshold, and the state change trend direction of the candidate transmission node is consistent with the state change trend direction of the historical power change segment. If the conditions for synchronous change are met, the corresponding time period number in the synchronous change statistics table will be marked as synchronous; if the conditions for synchronous change are not met, the corresponding time period number in the synchronous change statistics table will be marked as asynchronous. The total frequency of synchronization changes of each candidate transmission node is obtained by counting the number of times each candidate transmission node is marked as synchronized in all historical time periods. Calculate the proportion of synchronous changes in the total frequency of each candidate transmission node relative to the total number of historical periods to obtain the proportion of synchronous changes; The preset frequency requirement is set so that the total frequency of synchronous changes is not less than the preset frequency threshold, and the proportion of synchronous changes is not less than the preset proportion threshold. Candidate transmission nodes whose total frequency of synchronous changes and the proportion of synchronous changes both meet the preset frequency requirements are marked as preliminary key transmission nodes. A cross-period consistency analysis was conducted on the initial key transmission nodes. Multiple discontinuous historical periods were selected to check whether the synchronous change markers of the initial key transmission nodes maintained the continuity that met the preset requirements in each period. If the initial critical transmission node maintains continuity that meets the preset requirements in multiple time periods, then the initial critical transmission node is finally marked as a critical transmission node related to power generation. If the continuity of the initial critical transmission node does not meet the requirements in any time period, then the synchronous change statistics of the initial critical transmission node are re-examined. If the data is correct, then the initial critical transmission node is removed from the list of initial critical transmission nodes. Organize all the final marked key transmission nodes, record the type, total frequency of synchronous changes, percentage of synchronous changes, and consistency results across time periods for each key transmission node, and form a list of key transmission nodes.

7. The method for predicting the power generation of a hydropower station based on hydrological information according to claim 1, characterized in that, The step of constructing a power response association rule base based on the power response association factor sequence and the historical power generation time series records of the hydropower station includes: The power response correlation factor sequence is divided into a training power response correlation factor sequence and a test power response correlation factor sequence. At the same time, the corresponding historical power generation time series records of the hydropower station are divided into training power records and test power records. The time span of the training power response correlation factor sequence and the training power records are completely corresponding, and the time span of the test power response correlation factor sequence and the test power records are completely corresponding. Each power response correlation factor in the training power response correlation factor sequence is classified by feature. Based on the type of key transmission node and the attribute features of the corresponding network edge, the power response correlation factors are divided into different power response correlation factor categories. Each power response correlation factor category corresponds to a set of similar state changes and network edge attribute features. Analyze the historical power change segments in the training power records, and divide the historical power change segments into different power generation change patterns according to the direction of the difference between the power start value and the power end value and the change process. Each power generation change pattern corresponds to a power change trend. The frequency of different power response correlation factor categories in the training power response correlation factor sequence was statistically analyzed, and the power generation change pattern corresponding to each power response correlation factor category combination was analyzed to form a preliminary power response correlation factor combination-power generation pattern correspondence table. The initial power response correlation factor combination-power generation mode correspondence table is filtered for effectiveness. The power response correlation factor combination-power generation mode correspondence table that appears more frequently than the preset frequency threshold during the training period and corresponds to the same power generation change mode each time is retained. The filtered power response correlation factor combination-power generation mode correspondence is organized into rule entries. Each rule entry includes a description of the power response correlation factor category combination, a description of the corresponding power generation change mode, and the number of applicable time periods for the corresponding rule during the training period. Enter all rule entries into the rule base framework, and add a rule number and time validity mark to each rule entry. The time validity mark is determined by the time distance between the applicable time period of the corresponding rule and the current time period. Input the test power response correlation factor sequence into the rule base framework, check whether the power response correlation factor category combination corresponding to each test power response correlation factor sequence can match the rule entries in the rule base, and observe whether the power generation change pattern corresponding to the matching rule entries is consistent with the actual power change pattern in the test power record. Retain rule entries that meet the preset accuracy requirements and delete rule entries that have a lower accuracy requirements to form a power response association rule base.

8. The method for predicting the power generation of a hydropower station based on hydrological information according to claim 7, characterized in that, The test power response correlation factor sequence is input into the rule base framework. It is checked whether the power response correlation factor category combination corresponding to each test power response correlation factor sequence can match the rule entries in the rule base, and it is observed whether the power generation change pattern corresponding to the matching rule entries is consistent with the actual power change pattern in the test power record. Retain rule entries that meet the preset accuracy requirement, and delete rule entries with a matching accuracy lower than the preset accuracy requirement to form a power response association rule base, including: The test power response correlation factor sequence is divided into multiple test power response correlation factor segments according to the time span. Each test power response correlation factor segment contains all power response correlation factors within that time span. For each power response correlation factor in each test power response correlation factor segment, the power response correlation factor category is determined, and the power response correlation factor category to which each power response correlation factor belongs is determined, forming the power response correlation factor category combination corresponding to the test power response correlation factor segment; Search the rule base framework for rule entries that are completely consistent with the combination of power response correlation factor categories. If a unique rule entry is found, record the power generation change pattern corresponding to the rule entry. If multiple rule entries are found, select the rule entry with the strongest timeliness based on the timeliness flag of the rule entry and record the power generation change pattern corresponding to the rule entry. If no matching rule entry is found, mark the test power response correlation factor segment as having no matching rule. Extract the actual power change segment from the test power record that is contemporaneous with the factor segment associated with the test power response, and determine the actual power change pattern; Compare the recorded power generation change pattern with the actual power change pattern. If the trend direction and range of the power generation change pattern meet the preset consistency requirements, then mark the matching of the corresponding rule entry as a successful match; otherwise, mark the matching of the corresponding rule entry as a failed match. For each rule entry in the rule base framework, the number of successful matches and the number of failed matches for the corresponding rule entry in all test power response correlation factor fragment matches are counted to determine the proportion of successful matches for the corresponding rule entry. The preset accuracy requirement is set so that the percentage of successful matches is not less than the preset success rate threshold. Filter out rule entries whose success rate meets the preset accuracy requirement, and retain rule entries whose success rate meets the preset accuracy requirement in the rule base; For rule entries where the percentage of successful matches is lower than the preset accuracy requirement, analyze the reasons for the failed matches of the corresponding rule entries. If the failed matches are due to special state changes of the test power response related factor segments, supplement the rule clauses corresponding to the special state changes and redetermine the percentage of successful matches after the supplementation. If the percentage of successful matches after the supplementation still does not meet the requirements, delete the corresponding rule entries. The retained rule entries and supplementary clauses are integrated, each rule entry is reassigned a rule number, and the timeliness flags of the rule base are updated to form a power response association rule base.

9. The method for predicting the power generation of a hydropower station based on hydrological information according to claim 1, characterized in that, The step of matching the power response correlation factor sequence generated from the currently acquired dynamic hydrological information set with the power response correlation rule base to generate the power generation prediction result of the hydropower station for the current period includes: Obtain the set of dynamic water situation interaction information for the current time period, and construct the water situation interaction transmission network for the current time period. This water situation interaction transmission network includes the state change records of each water situation information for the current time period as network nodes, and the mutual triggering relationship between each network node as network edges. Extract key transmission nodes and corresponding network edge attributes from the hydrological interaction transmission network for the current time period to generate a power response correlation factor sequence for the current time period. This power response correlation factor sequence contains the power response correlation factors corresponding to each time span within the current time period. The power response correlation factor sequence for the current time period is divided into multiple current power response correlation factor segments. Each current power response correlation factor segment corresponds to a time span, and each current power response correlation factor segment contains all power response correlation factors within that time span. For each power response correlation factor in each current power response correlation factor segment, determine the power response correlation factor category to which each power response correlation factor belongs, and form the current power response correlation factor category combination corresponding to the current power response correlation factor segment; Retrieve rule entries that match the current power response association factor category combination in the power response association rule base, prioritizing the retrieval of rule entries with the latest timeliness marker; If a matching rule entry is found, the power generation change pattern description corresponding to the rule entry is extracted. The power generation change pattern description includes the direction of change trend, the range of change magnitude, and the characteristics of the change process. Based on the description of the power generation change pattern, and combined with the time span length corresponding to the current power response correlation factor segment, a power generation prediction segment within the time span is generated. The power generation prediction segment includes the predicted power start value, the predicted power end value, and a description of the predicted power change process. If no perfectly matching rule entry is found, the rule entry with the highest similarity to the current power response associated factor category combination is searched, the power generation change pattern description of the corresponding similar rule entry is extracted, and the pattern is adjusted in combination with the special state changes of the current hydrological information to generate an adjusted power generation prediction segment. All power generation prediction segments corresponding to all time spans are spliced ​​together in chronological order to form a preliminary power generation prediction sequence for the current period. A trend consistency check is performed on the preliminary power generation forecast sequence to ensure that the direction and magnitude of power change trends between power generation forecast segments in adjacent time spans meet the preset consistency requirements. If there are discontinuous power generation prediction segments in the transition, the predicted power value during the transition period is adjusted according to the correlation of water condition changes between adjacent power generation prediction segments, so that the overall power generation prediction sequence maintains a consistent trend. The integrated and adjusted power generation forecast segments form a power generation forecast result that includes all time spans of the current period. This power generation forecast result includes the initial forecast power value, the final forecast power value, a description of the forecast power change process, and a summary of the overall forecast trend for each time span.

10. A hydropower station power generation prediction system based on hydrological information, characterized in that, The hydropower generation prediction system based on hydrological information includes a processor and a memory, the memory and the processor are connected, the memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the hydropower generation prediction method based on hydrological information as described in any one of claims 1-9.

Images