AI-driven water pump station energy efficiency intelligent adjustment method

Abstract

The present application relates to the technical field of data mining software, in particular to a water pump station energy efficiency intelligent adjustment method based on AI driving, comprising the following steps: obtaining water pump station flow, pressure, single pump frequency, power, current, rotating speed and start-stop state to form operation records, matching daily type flow curve and determining demand pressure boundary, generating dynamic supply pressure constraint according to outlet pressure wave, inputting offline trained reinforcement learning strategy network to generate candidate frequency and start-stop instruction, and reconstructing executable energy saving instruction after hardware constraint checking and virtual operation prediction. In the present application, the demand pressure boundary and the pressure checking boundary are cooperatively constrained, so that the supply pressure target remains stable with load change, overpressure water supply and pressure fluctuation are reduced, through strategy reasoning, hardware legal checking and energy saving operation shell domain reconstruction, the frequency and start-stop control are adapted to the device boundary and the water pump high efficiency interval, invalid adjustment and shaft power consumption are reduced, and the water pump station operation energy efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of data mining software technology, and in particular to an AI-driven intelligent energy efficiency regulation method for water pumping stations. Background Technology

[0002] The field of data mining software technology involves computer data processing processes such as the collection, storage, indexing, retrieval, and association rule extraction of massive operational data. Among these, the traditional AI-driven intelligent energy efficiency regulation method for water pump stations refers to a data processing method that receives operational records of water pump stations, such as flow rate, pressure, liquid level, motor power, and start / stop status, in a server or industrial computer. This data is then written into a database table or data warehouse, indexed according to timestamps, equipment numbers, and operating parameters, and finally, through batch processing tasks or query engines, extracts load change characteristics, energy consumption statistics, and equipment operating status to generate water pump regulation commands.

[0003] The existing pump station regulation process relies on servers or industrial computers to receive operation records, establish indexes by timestamps and equipment numbers, and then extract load characteristics and energy consumption statistics through batch processing tasks. The data processing chain is biased towards post-event statistics and rule queries, making it difficult to synchronously calibrate the supply pressure demand with flow fluctuations. The regulation commands are not sufficiently connected with real-time pressure boundaries and the high-efficiency range of pump sets, which can easily lead to overpressure water supply, inefficient operation, frequent start-stop and shutdown, and energy consumption redundancy. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides an AI-driven intelligent energy efficiency regulation method for water pumping stations.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an AI-driven intelligent energy efficiency regulation method for water pumping stations, comprising the following steps: The system acquires data on the main outlet flow rate, outlet pressure, single pump frequency, single pump power, motor current, motor speed, and pump start / stop status of the pumping station to form a record of the pumping station's operating status. Based on the acquisition time of the pump station operation status record, the daily type flow curve is matched, and the demand pressure boundary is determined according to the flow-pressure relationship corresponding to the flow rate of the main outlet pipe. The pressure verification boundary is determined based on the fluctuation characteristics of the outlet pressure within the continuous acquisition period. The demand pressure boundary is used as the pressure constraint benchmark to generate dynamic supply pressure constraint conditions. The pump station operation status record and the dynamic pressure supply constraint are input into a reinforcement learning strategy network trained offline in a virtual operating environment to generate candidate frequency commands and candidate start / stop commands. Based on the frequency range of the frequency converter and the number of pumps, a hardware-valid candidate command is formed. Based on the hardware's valid candidate instructions, the predicted outlet pressure, single pump distribution flow, head value, and shaft power value are determined in the virtual operating environment. Instructions that do not meet the pressure verification boundary or are outside the energy-saving operating shell are reconstructed into executable energy-saving instructions that meet the pressure verification boundary and the energy-saving operating shell, and the executable energy-saving instructions are output.

[0006] As a further aspect of the present invention, the step of forming the pumping station operation status record includes: The system acquires the main outlet flow rate, outlet pressure, single pump frequency, single pump power, motor current, motor speed, and pump start / stop status uploaded by the pump station data acquisition interface. It then uses data mining software to establish the correlation between each piece of operational data and the pump number at the same acquisition time and marks each piece of operational data whose acquisition time interval does not meet the preset resampling interval as data to be corrected. Based on time alignment of similar operating data collected at adjacent times before and after the data to be corrected, records that are inconsistent between the pump start / stop status and the motor current and the single pump frequency are marked as status conflict records, and written into the operating status table according to the pump station identifier, collection time and pump number to generate the pump station operating status record.

[0007] As a further aspect of the present invention, the step of determining the demand pressure boundary includes: The data mining software obtains the collection date, collection time, and main outlet flow rate from the pump station's operating status record. Based on the calendar configuration table, the data mining software maps the collection date to a day type, such as weekday, weekend, or holiday. It then locates the time period curve segment that matches the collection time in the candidate flow curves corresponding to the day type. The flow rate of the main outlet pipe is compared with the flow range in the time period curve segment. When the flow rate of the main outlet pipe falls into the flow range, the water supply pressure value and pressure redundancy value associated with the flow range are read. The demand pressure boundary is determined based on the water supply pressure value and the pressure redundancy value.

[0008] As a further aspect of the present invention, the calendar configuration table is generated by data mining software based on the original operating date, statutory holiday dates and manually marked dates, and the candidate flow curve is established based on the original main outlet flow record of the same time period under the same day type, and the collected values ​​in the original main outlet flow record that exceed the preset multiple of the rated range of the flow meter are excluded. The flow range is determined according to the boundary value of adjacent flow segments in the candidate flow curve. When the flow rate of the main outlet pipe is located at the junction of two adjacent flow ranges, the flow range with the higher pressure value is selected as the matching object.

[0009] As a further aspect of the present invention, the step of generating the dynamic pressure supply constraint conditions includes: The outlet pressure and the demand pressure boundary are obtained within a continuous acquisition period. The pressure change amplitude between adjacent outlet pressures is calculated according to the acquisition time sequence. The pressure fluctuation characteristics are determined based on the pressure change amplitude and the average pressure change amplitude within the same verification window. Using the demand pressure boundary as the pressure constraint benchmark, the upper and lower limits of the demand pressure boundary are expanded or tightened according to the pressure fluctuation characteristics. When the pressure fluctuation characteristics exceed the preset fluctuation threshold, the lower limit protection amount is optimized, a pressure verification boundary is generated, and the dynamic supply pressure constraint conditions are generated based on the pressure verification boundary.

[0010] As a further aspect of the present invention, the process of determining the pressure fluctuation characteristics specifically includes: Data mining software removes outlet pressure records with missing acquisition times, abnormal pressure sensor status, or switching of the water pump start / stop status within the same verification window, and establishes a pressure sequence based on the chronological order of the remaining outlet pressure records. The pressure sequence is statistically analyzed for adjacent pressure differences and the number of pressure direction changes. When the number of pressure direction changes exceeds a preset threshold, the average pressure change amplitude and the pressure change amplitude are used together as pressure fluctuation feature parameters to generate the pressure fluctuation feature filtered by start-stop disturbances.

[0011] As a further aspect of the present invention, the step of forming the hardware valid candidate instruction includes: The pump station operation status record and the dynamic pressure supply constraint are obtained. The main outlet flow rate, outlet pressure, single pump frequency, single pump power, motor current, motor speed, pump start / stop status and demand pressure boundary are converted into policy input vectors. The policy input vectors are then input into the reinforcement learning policy network trained offline in the virtual operating environment. Read the candidate frequency value and candidate start / stop flag output by the reinforcement learning policy network, generate the candidate frequency command and the candidate start / stop command respectively, compare the candidate frequency command with the frequency range of the frequency converter, and compare the candidate start / stop command with the water pump number constraint, filter out the commands that do not meet the hardware constraints, and generate the hardware valid candidate command.

[0012] As a further aspect of the present invention, the process of generating the policy input vector specifically includes: The data mining software arranges the single pump frequency, single pump power, motor current, motor speed and pump start / stop status in order of pump number, and inserts the lower and upper limits of the demand pressure boundary and the pressure constraint benchmark in the dynamic pressure supply constraint conditions into the input field at the same acquisition time. When the operating data corresponding to any pump number is missing, a stop placeholder value or missing marker value is written according to the pump start / stop status, and the number of pump number fields is kept consistent with the number of pumps to generate the strategy input vector with fixed input dimensions.

[0013] As a further aspect of the present invention, the step of outputting the executable energy-saving command includes: The hardware legal candidate instructions, the pressure verification boundary and the energy-saving operation shell domain are obtained. The data mining software inputs the hardware legal candidate instructions into the virtual operation environment, determines the predicted outlet pressure according to the pump group response relationship, and determines the single pump allocation flow according to the pump start-stop combination and frequency allocation relationship. Based on the single-pump allocated flow rate and the corresponding pump frequency value, the head value and shaft power value are determined. The predicted outlet pressure is compared with the pressure verification boundary, and the single-pump allocated flow rate, the head value, and the shaft power value are matched with the energy-saving operation shell domain. For candidate instructions that fail the comparison or matching, frequency backoff and number replacement are performed, and the executable energy-saving instruction is output.

[0014] As a further aspect of the present invention, the step of determining the energy-saving operation shell domain includes: The data mining software reads the single pump characteristic table, which records the correspondence between pump number, frequency segment, flow segment, head range and shaft power range. Applicable characteristic records are then filtered based on the pump number and candidate frequency value in the valid candidate hardware instructions. The single pump's allocated flow rate is compared with the flow rate segments in the applicable characteristic record. The matched head range and shaft power range are read, and the flow rate segments, head range, and shaft power range are associated with the corresponding pump's energy-saving operating domain.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, the demand pressure boundary is determined by matching the operating status record with the daily type flow curve and combining the flow and pressure relationship. This ensures that the pressure supply target matches the load change and suppresses overpressure energy consumption. The pressure verification boundary is formed based on the outlet pressure fluctuation within the continuous acquisition period, which enables the regulation process to have pressure stability constraints and reduces the impact of abnormal fluctuations. The operating status and dynamic pressure supply constraints are input into the offline training strategy network and hardware legality verification is performed to adapt the frequency and start / stop commands to the equipment boundary and reduce invalid actions. The pressure is predicted through the virtual operating environment, and the flow rate, head, and shaft power are allocated. Over-limit commands are reconstructed so that the output commands take into account both pressure supply safety and efficient operation and reduce overall energy consumption. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is the main flow chart of the intelligent energy efficiency regulation of the water pumping station according to the present invention; Figure 2 This is a schematic diagram illustrating the application scenario of the mobile data communication service link of the present invention; Figure 3 This is a schematic diagram illustrating the effect of generating dynamic pressure supply constraint conditions according to the present invention. Figure 4 This is a schematic diagram illustrating the effect of instruction verification and reconstruction in the virtual runtime environment of the present invention; Figure 5 This is a schematic diagram illustrating the data flow closed-loop effect of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0019] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0020] Please see Figures 1 to 5 This invention provides an AI-driven intelligent energy efficiency regulation method for water pumping stations. In practical applications, such as during the continuous water distribution operation of urban water supply pumping stations, the pumping station edge acquisition gateway transmits the main outlet flow rate, outlet pressure, and pump operation data to a remote data processing node via a mobile data communication service link. The remote data processing node generates energy-saving control commands based on the pumping station's historical operation records, daily flow curves, pressure constraint rules, and the virtual operating environment. After receiving the commands, the pumping station control side adjusts the inverter frequency and pump start / stop status in a coordinated manner. The method includes the following processing steps: S1: Obtain the main outlet flow rate, outlet pressure, single pump frequency, single pump power, motor current, motor speed, and pump start / stop status of the water pump station to form a record of the water pump station's operating status.

[0021] The pump station operation status record is a data object organized by pump station identifier, acquisition time, pump number, and operation fields. It includes fields for main outlet flow rate, outlet pressure, single pump frequency, single pump power, motor current, motor speed, and pump start / stop status. The main outlet flow rate field comes from the acquisition message of the main outlet flow meter; the outlet pressure field comes from the acquisition message of the outlet pressure sensor; the single pump frequency field comes from the inverter status message; the single pump power, motor current, and motor speed fields come from the motor control interface; and the pump start / stop status field comes from the status signal of the pump control cabinet. The mobile data communication service link carries the acquisition messages, edge gateway upload confirmations, and remote data processing node reception receipts. Before entering the data mining software, the acquisition messages are appended with the pump station identifier, acquisition time, interface identifier, and message source identifier. The reception receipts are written to the communication log.

[0022] S101: Obtain the main outlet flow rate, outlet pressure, single pump frequency, single pump power, motor current, motor speed, and pump start / stop status uploaded by the pump station data acquisition interface. Use data mining software to establish the correlation between each piece of operating data and the pump number according to the same acquisition time, and mark each piece of operating data whose acquisition time interval does not meet the preset resampling interval as data to be corrected.

[0023] The preset resampling interval is the time caliber field in the acquisition parameter configuration table, derived from the message upload rules of the pump station acquisition gateway, the mobile data communication service link reception rules, and the remote data processing node database entry rules, and is invoked during the operation status record generation process. The data mining software reads the pump station identifier, interface identifier, acquisition time, pump number, and field name from the acquired messages, attaching the main pipeline type fields to the pump station's common record and attaching the individual pump type fields to the pump record according to the pump number. When the acquisition time is missing, duplicate acquisition time, unrecognizable interface identifier, mismatched field name, missing reception receipt, or field format inconsistent with the acquisition parameter configuration table, the running data enters the correction queue, and a data status identifier is written to the operation status table. The data to be corrected retains the original message content, mobile data communication service link reception identifier, source interface, reception time, and exception type, providing time alignment and policy input filtering invocation.

[0024] S102: Based on the time alignment of similar operating data at adjacent acquisition times before and after the data to be corrected, the records that are inconsistent between the pump start / stop status and the motor current and single pump frequency are marked as status conflict records, and written into the operating status table according to the pump station identifier, acquisition time and pump number to generate the pump station operating status record.

[0025] During time alignment, the data to be corrected is assigned to the corresponding data processing cycle based on the acquisition time of adjacent valid records, while retaining both the original acquisition time field and the aligned acquisition time field. If the pump start / stop status shows as stopped, but the motor current field or single pump frequency field carries operational characteristics, a status conflict flag is written to the record; if the pump start / stop status shows as running, but neither the motor current field nor the single pump frequency field forms a valid operational characteristic, a pending review flag is written to the record. Status conflict records first enter a conflict cache, and the data mining software, combined with the previous valid start / stop status, control cabinet receipt, and inverter status message, determines the available status for the current acquisition cycle. The operating status table stores the original fields, correction fields, conflict flags, received receipt index, and log index. Subsequent daily type matching, strategy input vector generation, and virtual operating environment calls all read fields from the operating status table.

[0026] S2: Based on the data collection time of the pump station operation status record, match the daily type flow curve, and determine the demand pressure boundary according to the flow-pressure relationship corresponding to the flow rate of the main outlet pipe.

[0027] The daily flow rate curve is a flow rate change rule object organized by data mining software according to date and time period attributes. It contains fields for daily type, time period curve segment, flow rate interval, water supply pressure value, and pressure redundancy value. The demand pressure boundary is the pressure range that the current outflow main flow rate needs to meet under the corresponding daily type and time period curve segment. It contains fields for lower demand pressure limit, upper demand pressure limit, and pressure constraint benchmark, and serves as the benchmark input for dynamic supply pressure constraints.

[0028] S201: Obtain the collection date, collection time, and main outlet flow rate from the pump station's operating status record. The data mining software maps the collection date to a day type (weekday, weekend, or holiday) based on the calendar configuration table, and locates the time period curve segment that matches the collection time in the candidate flow curves corresponding to the day type.

[0029] The system verifies whether the collected date matches a manually defined special scheduling day, and secondly, whether it matches a statutory holiday. If neither condition is met, it is classified as a weekday or weekend according to the regular calendar attributes, thus determining the target day type for the collected date. Further, it extracts the outflow data of the main water pipe with the same day type attribute from historical operation records, constructs a baseline flow curve corresponding to that day type, and extracts a matching time segment from this baseline flow curve based on the current collection time. When data is missing at the collection time or cannot accurately match the current time segment, it automatically extracts adjacent valid curve segments as temporary matching objects for replacement. Simultaneously, it generates a time segment supplementation flag in the demand pressure boundary. This flag will be called later when generating the pressure verification boundary to determine the boundary processing and correction strategy.

[0030] S202: Compare the flow rate of the main outlet pipe with the flow range in the time period curve segment. When the flow rate of the main outlet pipe falls into the flow range, read the water supply pressure value and pressure redundancy value associated with the flow range, and determine the demand pressure boundary based on the water supply pressure value and pressure redundancy value.

[0031] The flow range is a lookup field formed by the boundaries of adjacent flow segments in the candidate flow curves. The water supply pressure value is the basic supply pressure field corresponding to the flow range, and the pressure redundancy value is the pump station network pressure protection field. All fields are stored in the flow-pressure relationship table. The comparison process is executed in the order of flow range boundaries. After the main outlet flow matches a flow range, the water supply pressure value field and the pressure redundancy value field are read to generate the lower limit of demand pressure, the upper limit of demand pressure, and the pressure constraint benchmark. If the main outlet flow does not match any flow range, the nearest valid flow range is read and the range extension identifier is written. The range extension identifier enters the dynamic supply pressure constraint condition, triggering the protection processing of the pressure verification boundary. When there is a boundary match between flow ranges, the flow range with the higher pressure field is selected as the matching object.

[0032] S203: The calendar configuration table is generated by data mining software based on the original operating date, statutory holiday dates, and manually marked dates. The candidate flow curve is established based on the original main outlet flow records of the same time period under the same day type, and the collected values ​​in the original main outlet flow records that exceed the preset abnormal flow rule field are excluded.

[0033] The preset abnormal flow rule fields are derived from flow meter range configuration, historical operational data quality rules, and pump station manual confirmation records, and are stored in the flow cleaning rule table. Before generating candidate flow curves, original flow records corresponding to missing data acquisition times, missing mobile data communication service link reception receipts, abnormal flow meter status, and start / stop switching disturbances are excluded. Records are retained and grouped by daily type and time period curve segment to form the effective data source for candidate flow curves. Flow intervals are determined according to the boundary values ​​of adjacent flow segments in the candidate flow curves. When the flow rate in the main outlet pipe is located at the boundary of adjacent flow intervals, the flow interval with the higher pressure value is selected as the matching object. The demand pressure boundary is written to the pressure boundary cache, carrying the daily type identifier, curve segment identifier, flow interval identifier, and abnormal cleaning identifier, for use in pressure fluctuation verification and strategy input vector generation.

[0034] S3: Determine the pressure verification boundary based on the fluctuation characteristics of the outlet pressure within the continuous acquisition period, use the demand pressure boundary as the pressure constraint benchmark, and generate dynamic pressure supply constraint conditions.

[0035] Pressure fluctuation characteristics are state objects formed by the effective outlet pressure sequence within the verification window, including fields for pressure change amplitude, average pressure change amplitude, pressure direction change count, and start / stop disturbance filtering. Pressure verification boundaries are verification range objects formed by combining the demand pressure boundary with pressure fluctuation characteristics, including fields for lower pressure verification limit, upper pressure verification limit, lower limit protection quantity, and boundary status. Dynamic pressure supply constraints are constraint objects jointly invoked by the reinforcement learning strategy network and the virtual operating environment, including pressure constraint benchmarks, pressure verification boundaries, boundary source identifiers, fluctuation status identifiers, and anomaly handling identifiers.

[0036] S301: Obtain the outlet pressure and demand pressure boundaries within the continuous acquisition period, calculate the pressure change amplitude between adjacent outlet pressures according to the acquisition time sequence, and determine the pressure fluctuation characteristics based on the pressure change amplitude and average pressure change amplitude within the same verification window.

[0037] The continuous acquisition period is determined by the verification window field in the acquisition parameter configuration table. This verification window field originates from the pump station control stability rules and pressure sensor acquisition rules, and is stored in the pressure verification configuration table. Before the outlet pressure enters the pressure sequence, a validity check is performed. Records with missing acquisition times, abnormal pressure sensor states, missing mobile data communication service link reception receipts, inconsistent pressure field formats, or overlap with start / stop switching events are written to the pressure anomaly cache and are not included in pressure fluctuation feature generation. The pressure change amplitude field indicates the degree of change between adjacent valid outlet pressures, the average pressure change amplitude field indicates the pressure change state within the verification window, and the pressure direction change count field indicates the switching state of the pressure change direction within the verification window. After pressure fluctuation features are generated, they are written to the pressure fluctuation cache and associated with the pressure constraint benchmark at the demand pressure boundary.

[0038] S302: Using the demand pressure boundary as the pressure constraint benchmark, the upper and lower limits of the demand pressure boundary are expanded or tightened according to the pressure fluctuation characteristics. When the pressure fluctuation characteristics exceed the preset fluctuation threshold, the lower limit protection amount is optimized, a pressure verification boundary is generated, and dynamic supply pressure constraint conditions are generated based on the pressure verification boundary.

[0039] The preset fluctuation threshold is the fluctuation state judgment field in the pressure verification configuration table, derived from the pump station's historical pressure stability records, pressure sensor range configuration, and water supply network operation rules. When the pressure fluctuation characteristic falls into the stable state category, the pressure verification boundary follows the benchmark caliber of the demand pressure boundary; when the pressure fluctuation characteristic falls into the fluctuation state category, the pressure verification boundary is adjusted according to the boundary state field. The lower limit protection quantity field is derived from the pressure boundary configuration table and is updated after changes in pressure fluctuation state, daily type, flow range, or sensor state recovery. After dynamic pressure supply constraints are generated, they are written to the constraint cache, serving as the source of the pressure field for the strategy input vector and as the basis for pressure verification during candidate instruction reconstruction.

[0040] S303: Data mining software removes outlet pressure records with missing acquisition times, abnormal pressure sensor status, or pump start / stop status switching within the same verification window, and establishes a pressure sequence based on the chronological order of the remaining outlet pressure records.

[0041] Pressure sensor status anomalies include sensor self-test failure, interface return anomalies, message parsing failure, missing data acquisition fields, and status indicators inconsistent with the previous valid status. When the pump start-up / shutdown status switches, the outlet pressure is affected by the pump unit's response process, and a start-up / shutdown disturbance filtering flag is written to the corresponding acquisition cycle. During the start-up phase, if a complete pressure sequence is not formed, a temporary sequence is established in the first cycle using the initial pressure benchmark from the acquisition configuration table and the current valid outlet pressure. Subsequent cycles supplement this sequence using already obtained valid pressure records. Once stable operating conditions are reached, the system switches to the complete verification window rule. The pressure sequence stores the valid pressure field, acquisition time field, data source interface, received receipt status, and filtering flag. Subsequent pressure direction change statistics and pressure verification boundary generation are both read from this sequence.

[0042] S304: Calculate the pressure change amplitude and pressure change amplitude field statistics for the pressure sequence. When the pressure direction change count field exceeds the preset direction change rule field, the average pressure change amplitude and the pressure change amplitude are used together as pressure fluctuation feature parameters to generate pressure fluctuation features filtered by start-stop disturbances.

[0043] The preset direction change rule field is derived from the pressure verification configuration table, and the engineering boundary is determined by the pump station pipeline stability record and pressure sensor status rules. When the pressure direction change count field does not trigger a fluctuation state, the pressure change amplitude is used to confirm the stability category of the pressure sequence; when the pressure direction change count field triggers a fluctuation state, the average pressure change amplitude is used to describe the overall pressure change within the verification window, and the pressure change amplitude is used to describe the local changes in adjacent acquisition cycles. The pressure fluctuation characteristics filtered by start-stop disturbances carry the source of the valid sequence, the reason for filtering, and the fluctuation state identifier, and are output to the dynamic pressure supply constraint generation process, and serve as the basis for pressure boundary adjustment during the candidate instruction verification stage.

[0044] S4: Input the pump station operation status record and dynamic pressure supply constraints into the reinforcement learning strategy network trained offline in the virtual operation environment to generate candidate frequency commands and candidate start / stop commands, and form hardware-valid candidate commands based on the frequency range of the frequency converter and the number of pumps.

[0045] The virtual operating environment serves as a data simulation object to reproduce the pump station's operational response, encompassing pump group response relationships, single pump characteristic tables, start-stop combination relationships, frequency allocation relationships, pressure boundary rules, and control feedback rules. The reinforcement learning policy network is a policy mapping network trained offline. Internally, it processes operating states and pressure boundaries sequentially through an input mapping layer, a state encoding layer, a constraint fusion layer, an action generation layer, and an action verification layer. The input mapping layer converts flow rate, pressure, frequency, power, current, speed, start-stop state, and pressure boundary fields. The state encoding layer extracts operational correlations through a rectified linear activation mechanism. The constraint fusion layer embeds pressure verification boundaries and pressure constraint benchmarks. The action generation layer outputs frequency action fields and start-stop action fields. The action verification layer aligns the action fields with the hardware constraint fields in the virtual operating environment. During offline training, historical operating state records, dynamic pressure supply constraints, and pump station control feedback records are input into the virtual operating environment. Predicted actions are compared with historical stable water supply actions, energy consumption states, and pressure boundary states. Deviation feedback is used to adjust network layer connection parameters. The trained policy network parameters are stored in a policy version repository, and the policy version identifier is written to the candidate instruction log.

[0046] S401: Obtain the operating status record and dynamic pressure supply constraints of the water pumping station, convert the main outlet flow rate, outlet pressure, single pump frequency, single pump power, motor current, motor speed, water pump start / stop status and demand pressure boundary into a strategy input vector, and input the strategy input vector into a reinforcement learning strategy network trained offline in a virtual operating environment.

[0047] The strategy input vector consists of data objects with a fixed field order. Field sources include the operating status table, pressure boundary cache, and constraint cache. The main outlet flow rate and outlet pressure are located in the pump station's common status field. Single pump frequency, single pump power, motor current, motor speed, and pump start / stop status are entered into the pump group status field in pump number order. The lower pressure verification limit, upper pressure verification limit, and pressure constraint benchmark are entered into the constraint status field. During field conversion, the field source and acquisition time are verified first, followed by the field status identifier. Then, the input fields are assembled according to the order recorded in the strategy version library. If any field has an unconfirmed communication, inconsistent format, or status conflict identifier, the previous valid operating status or a shutdown placeholder value is read to fill the gap, and the filler source identifier is written into the strategy input vector. After the strategy input vector is written to the strategy call log, it enters the reinforcement learning strategy network. The output result and the strategy version identifier are entered into the candidate instruction cache.

[0048] S402: Read the candidate frequency value and candidate start / stop flag output by the reinforcement learning policy network, generate candidate frequency command and candidate start / stop command respectively, compare the candidate frequency command with the frequency range of the inverter, and compare the candidate start / stop command with the water pump number constraint, filter out the commands that do not meet the hardware constraints, and generate the hardware valid candidate command.

[0049] Candidate frequency commands are data objects containing pump number, candidate frequency value, strategy version identifier, and generation time. Candidate start / stop commands are data objects containing pump number, candidate start / stop identifier, start / stop combination relationship, and generation time. The inverter frequency range is derived from the inverter configuration table, and the pump quantity constraint is derived from the pump station equipment ledger and pump group control cabinet configuration table. The comparison process first verifies whether the pump number exists in the equipment ledger, then verifies whether the candidate frequency value falls within the allowable range of the corresponding inverter, and then verifies whether the candidate start / stop combination conforms to the operable pump group relationship. Candidate frequency commands that fail the inverter range verification are marked with a frequency out-of-bounds flag, and candidate start / stop commands that fail the pump quantity constraint verification are marked with a combination out-of-bounds flag. Valid hardware candidate commands retain the verified pump number, frequency action, start / stop action, pressure constraint source, and strategy version identifier, and are output to the virtual operating environment prediction stage.

[0050] S403: Data mining software arranges the single pump frequency, single pump power, motor current, motor speed, and pump start / stop status in order of pump number, and inserts the lower and upper limits of the demand pressure boundary and the pressure constraint benchmark in the dynamic pressure supply constraint conditions into the input fields at the same acquisition time.

[0051] The pump number sequence is derived from the pump station equipment ledger, which records the correspondence between pump number, control cabinet port, inverter identifier, and operating status field. The upper and lower limits of the demand pressure boundary and the pressure constraint benchmark are read from the dynamic supply pressure constraint conditions and written into the constraint status field of the strategy input vector. When the acquisition time is inconsistent, the acquisition time of the common status field in the operating status table is used as the benchmark to align the time of the single pump field and the pressure boundary field; fields that cannot be aligned are marked as pending review and supplemented by the previous valid field. After the strategy input vector is completed, it is written to the vector cache. The vector cache records the field source, status identifier, pump number sequence, and pressure boundary version for strategy network calls and subsequent audit traceability.

[0052] S404: When the operating data corresponding to any pump number is missing, write a stop placeholder value or missing flag value according to the pump start / stop status, and keep the number of pump number fields consistent with the number of pumps to generate a strategy input vector with fixed input dimensions.

[0053] The shutdown placeholder value is the shutdown status field recorded in the policy version library, used to indicate that the pump number is not participating in operation at the current data collection time; the missing flag value is the abnormal status field in the data quality rule table, used to indicate that the operation data of the pump number cannot be confirmed. When the pump start / stop status is shutdown, the single pump frequency, single pump power, motor current, and motor speed fields are written to the shutdown placeholder value; when the pump start / stop status is running and the operation data is missing, the relevant fields are written to the missing flag value, and the data quality log is triggered. The fixed input dimension comes from the pump number field configuration in the pump station equipment ledger, and the policy network only accepts field structures consistent with the pump station configuration. The missing flag in the policy input vector enters the action verification layer, prompting candidate instructions to undergo dual verification of pressure boundaries and hardware constraints in the virtual operating environment.

[0054] S5: Based on the valid hardware candidate instructions, determine the predicted outlet pressure, single pump distribution flow, head value and shaft power value in the virtual operating environment, reconstruct instructions that do not meet the pressure verification boundary or are out of the energy-saving operation shell into executable energy-saving instructions that meet the pressure verification boundary and energy-saving operation shell, and output the executable energy-saving instructions.

[0055] Executable energy-saving commands are data objects that the control side can receive, carrying the pump number, target frequency, start / stop action, command source, pressure verification status, energy-saving operation shell domain matching status, and control confirmation requirements. The energy-saving operation shell domain is a feasible operating range object generated from a single pump characteristic table, carrying the correspondence between the pump number, frequency segment, flow segment, head range field, and shaft power range field. It is used to determine whether the single pump's allocated flow rate, head status, and shaft power status are within the energy-saving operation zone. After receiving valid hardware candidate commands, the virtual operating environment first performs single-pump characteristic matching, then pump group response prediction, and subsequently pressure verification and energy-saving shell domain matching. Candidate commands that fail verification are entered into the command reconstruction cache.

[0056] S501: The data mining software reads the single pump characteristic table, which records the correspondence between pump number, frequency segment, flow segment, head range and shaft power range, and filters applicable characteristic records based on the pump number and candidate frequency value in the valid hardware candidate instructions.

[0057] The single-pump characteristic table is derived from the pump's factory characteristics, pump station commissioning records, historical operating status records, and maintenance confirmation records. It is stored in a characteristic parameter database and linked to the pump station equipment ledger via version identifiers. After a valid hardware candidate command enters the characteristic matching process, it first filters the corresponding pump by pump number, then matches the applicable frequency segment by candidate frequency value, and subsequently reads the flow segment, head range field, and shaft power range field associated with that frequency segment. If an applicable characteristic record is missing, a characteristic missing flag is written to the candidate command, and the command does not enter the control output, instead proceeding to the frequency rollback and unit replacement process. If a characteristic record has an invalid version, unconfirmed maintenance status, or inconsistent pump numbers, the previous valid characteristic version is read, and a characteristic version inheritance flag is written to the command log.

[0058] S502: Compare the single pump's distributed flow rate with the flow segment in the applicable characteristic record, read the hit head range and shaft power range, and associate the flow segment, head range, and shaft power range with the corresponding pump's energy-saving operating shell domain.

[0059] The single-pump allocated flow rate is a field generated by the virtual operating environment based on the pump set start / stop combination, frequency allocation relationship, and the flow requirement of the main outlet pipe, and participates in the energy-saving operation shell domain matching. After the flow segment is matched, the head range field and shaft power range field serve as the operating boundaries of that pump number at the current candidate frequency. When the single-pump allocated flow rate is located at the boundary of adjacent flow segments, the flow segment that matches both the pressure verification boundary and the shaft power range is selected, and the boundary status is written to the shell domain matching log. When the single-pump allocated flow rate does not match the applicable characteristic record, the candidate instruction is written to the shell domain departure flag, and the pump number, candidate frequency value, allocated flow rate field, and departure reason are sent to the instruction reconstruction cache.

[0060] S503: Obtain valid hardware candidate instructions, pressure verification boundaries, and energy-saving operation shell domains. The data mining software inputs the valid hardware candidate instructions into the virtual operating environment, determines the predicted outlet pressure according to the pump group response relationship, and determines the single pump distribution flow according to the pump start-stop combination and frequency distribution relationship.

[0061] The pump set response relationship is a rule object recorded in the virtual operating environment between input actions and outlet pressure responses, derived from historical operating status records, control feedback records, and pump station commissioning data. The frequency allocation relationship is the correspondence rule between pump number, candidate frequency value, and start / stop combination, stored in the pump set control configuration table. The virtual operating environment first reads the current outlet main flow rate, outlet pressure, and dynamic supply pressure constraints, then loads the start / stop combinations and frequency allocation relationships corresponding to the valid hardware candidate instructions, subsequently generating a predicted outlet pressure field and a single pump allocated flow field. The predicted outlet pressure field is output to the pressure verification process, and the single pump allocated flow field is output to the energy-saving operation shell domain matching process. When the virtual operating environment interface returns an exception, the pump set response relationship is missing, or the control feedback rule is not matched, the candidate instruction is written to a simulation failure flag, and the previous valid energy-saving instruction retention rule is triggered; the control side does not accept unacknowledged instructions.

[0062] S504: Determine the head and shaft power values ​​based on the single pump distribution flow and the corresponding pump frequency value, compare the predicted outlet pressure with the pressure verification boundary, and match the single pump distribution flow, head, and shaft power values ​​with the energy-saving operation shell domain. For candidate instructions that fail the comparison or matching, perform frequency backoff and number replacement, and output executable energy-saving instructions.

[0063] The head and shaft power values ​​are operating status fields read by the virtual operating environment from the single-pump characteristic table, originating from the boundary fields associated with pump number, frequency segment, and flow segment in the applicable characteristic record. When the predicted outlet pressure is lower than the pressure verification lower limit, the candidate command enters the frequency increase correction process. The rollback process reads the adjacent executable frequency action from the inverter configuration table and re-enters the virtual operating environment for prediction. When the predicted outlet pressure is higher than the pressure verification upper limit, the candidate command enters the pressure decrease correction process. The pressure decrease correction prioritizes adjusting the frequency action and then verifies the start-stop combination. When the single pump's allocated flow, head, or shaft power value leaves the energy-saving operating shell domain, the candidate command enters the unit replacement process. The unit replacement reads the available pump number and start-stop combination from the pump group control configuration table and regenerates the candidate action in conjunction with the single-pump characteristic table. The reconstructed command re-enters the pressure verification and shell domain matching. After successful matching, an executable energy-saving command is generated. The executable energy-saving command returns to the pump station edge control gateway via the mobile data communication service link. The control gateway issues commands to the inverter and pump group control cabinet and writes the control confirmation receipt, execution status, and abnormal receipt to the command log. If the control confirmation receipt is not returned or the returned status is inconsistent with the executable energy-saving command, an unconfirmed flag is written to the command log, the data processing node retains the status of the previous valid command, and regenerates the operating status record and dynamic pressure supply constraints in the next acquisition cycle.

[0064] The pump station's operating status record, demand pressure boundary, pressure fluctuation characteristics, dynamic pressure supply constraints, strategy input vector, valid hardware candidate instructions, energy-saving operation shell domain, and executable energy-saving instructions sequentially form a closed data flow loop. Acquired messages enter the remote data processing node via the mobile data communication service link, undergoing time alignment, status conflict identification, daily type curve matching, pressure boundary verification, strategy network output, hardware constraint filtering, virtual operating environment prediction, and instruction reconstruction before being returned to the pump station control side. The frequency actions and start / stop actions received by the control side all carry pressure verification sources, hardware constraint sources, and energy-saving operation shell domain matching sources, forming a closed loop of acquisition, constraint, prediction, reconstruction, and feedback.

[0065] The above embodiments illustrate preferred embodiments of the present invention. Any equivalent adjustments to the technical solution based on software engineering methods are within the scope of protection, including but not limited to: implementing algorithm logic using different programming languages, refactoring functional modules into services, adjusting data interaction protocols, and optimizing resource scheduling strategies. Any implementation scheme derived from reasonable modifications to the data processing flow, service call chain, or system architecture layer without departing from the core technology of the present invention should be considered within the protection scope defined by the technical solution of the present invention.

Claims

1. An AI-driven intelligent energy efficiency regulation method for water pumping stations, characterized in that, Includes the following steps: The system acquires data on the main outlet flow rate, outlet pressure, single pump frequency, single pump power, motor current, motor speed, and pump start / stop status of the pumping station to form a record of the pumping station's operating status. Based on the acquisition time of the pump station operation status record, the daily type flow curve is matched, and the demand pressure boundary is determined according to the flow-pressure relationship corresponding to the flow rate of the main outlet pipe. The pressure verification boundary is determined based on the fluctuation characteristics of the outlet pressure within the continuous acquisition period. The demand pressure boundary is used as the pressure constraint benchmark to generate dynamic supply pressure constraint conditions. The pump station operation status record and the dynamic pressure supply constraint are input into a reinforcement learning strategy network trained offline in a virtual operating environment to generate candidate frequency commands and candidate start / stop commands. Based on the frequency range of the frequency converter and the number of pumps, a hardware-valid candidate command is formed. Based on the hardware's valid candidate instructions, the predicted outlet pressure, single pump distribution flow, head value, and shaft power value are determined in the virtual operating environment. Instructions that do not meet the pressure verification boundary or are outside the energy-saving operating shell are reconstructed into executable energy-saving instructions that meet the pressure verification boundary and the energy-saving operating shell, and the executable energy-saving instructions are output.

2. The AI-driven intelligent energy efficiency regulation method for water pumping stations according to claim 1, characterized in that, The steps for generating the pumping station operation status record include: The system acquires the main outlet flow rate, outlet pressure, single pump frequency, single pump power, motor current, motor speed, and pump start / stop status uploaded by the pump station data acquisition interface. It then uses data mining software to establish the correlation between each piece of operational data and the pump number at the same acquisition time and marks each piece of operational data whose acquisition time interval does not meet the preset resampling interval as data to be corrected. Based on time alignment of similar operating data collected at adjacent times before and after the data to be corrected, records that are inconsistent between the pump start / stop status and the motor current and the single pump frequency are marked as status conflict records, and written into the operating status table according to the pump station identifier, collection time and pump number to generate the pump station operating status record.

3. The AI-driven intelligent energy efficiency regulation method for water pumping stations according to claim 1, characterized in that, The steps for determining the demand pressure boundary include: The data mining software obtains the collection date, collection time, and main outlet flow rate from the pump station's operating status record. Based on the calendar configuration table, the data mining software maps the collection date to a day type, such as weekday, weekend, or holiday. It then locates the time period curve segment that matches the collection time in the candidate flow curves corresponding to the day type. The flow rate of the main outlet pipe is compared with the flow range in the time period curve segment. When the flow rate of the main outlet pipe falls into the flow range, the water supply pressure value and pressure redundancy value associated with the flow range are read. The demand pressure boundary is determined based on the water supply pressure value and the pressure redundancy value.

4. The AI-driven intelligent energy efficiency regulation method for water pumping stations according to claim 3, characterized in that, The calendar configuration table is generated by data mining software based on the original operating date, statutory holiday dates and manually marked dates. The candidate flow curve is established based on the original main outlet flow record of the same time period under the same day type, and the collected values ​​in the original main outlet flow record that exceed the preset multiple of the rated range of the flow meter are excluded. The flow range is determined according to the boundary value of adjacent flow segments in the candidate flow curve. When the flow rate of the main outlet pipe is located at the junction of two adjacent flow ranges, the flow range with the higher pressure value is selected as the matching object.

5. The AI-driven intelligent energy efficiency regulation method for water pumping stations according to claim 1, characterized in that, The steps for generating the dynamic pressure supply constraints include: The outlet pressure and the demand pressure boundary are obtained within a continuous acquisition period. The pressure change amplitude between adjacent outlet pressures is calculated according to the acquisition time sequence. The pressure fluctuation characteristics are determined based on the pressure change amplitude and the average pressure change amplitude within the same verification window. Using the demand pressure boundary as the pressure constraint benchmark, the upper and lower limits of the demand pressure boundary are expanded or tightened according to the pressure fluctuation characteristics. When the pressure fluctuation characteristics exceed the preset fluctuation threshold, the lower limit protection amount is optimized, a pressure verification boundary is generated, and the dynamic supply pressure constraint conditions are generated based on the pressure verification boundary.

6. The AI-driven intelligent energy efficiency regulation method for water pumping stations according to claim 5, characterized in that, The process of determining the pressure fluctuation characteristics specifically includes: Data mining software removes outlet pressure records with missing acquisition times, abnormal pressure sensor status, or switching of the water pump start / stop status within the same verification window, and establishes a pressure sequence based on the chronological order of the remaining outlet pressure records. The pressure sequence is statistically analyzed for adjacent pressure differences and the number of pressure direction changes. When the number of pressure direction changes exceeds a preset threshold, the average pressure change amplitude and the pressure change amplitude are used together as pressure fluctuation feature parameters to generate the pressure fluctuation feature filtered by start-stop disturbances.

7. The AI-driven intelligent energy efficiency regulation method for water pumping stations according to claim 1, characterized in that, The steps for forming the hardware valid candidate instructions include: The pump station operation status record and the dynamic pressure supply constraint are obtained. The main outlet flow rate, outlet pressure, single pump frequency, single pump power, motor current, motor speed, pump start / stop status and demand pressure boundary are converted into policy input vectors. The policy input vectors are then input into the reinforcement learning policy network trained offline in the virtual operating environment. Read the candidate frequency value and candidate start / stop flag output by the reinforcement learning policy network, generate the candidate frequency command and the candidate start / stop command respectively, compare the candidate frequency command with the frequency range of the frequency converter, and compare the candidate start / stop command with the water pump number constraint, filter out the commands that do not meet the hardware constraints, and generate the hardware valid candidate command.

8. The AI-driven intelligent energy efficiency regulation method for water pumping stations according to claim 7, characterized in that, The process of generating the policy input vector specifically includes: The data mining software arranges the single pump frequency, single pump power, motor current, motor speed and pump start / stop status in order of pump number, and inserts the lower and upper limits of the demand pressure boundary and the pressure constraint benchmark in the dynamic pressure supply constraint conditions into the input field at the same acquisition time. When the operating data corresponding to any pump number is missing, a stop placeholder value or missing marker value is written according to the pump start / stop status, and the number of pump number fields is kept consistent with the number of pumps to generate the strategy input vector with fixed input dimensions.

9. The AI-driven intelligent energy efficiency regulation method for water pumping stations according to claim 1, characterized in that, The steps for outputting the executable energy-saving command include: The hardware legal candidate instructions, the pressure verification boundary and the energy-saving operation shell domain are obtained. The data mining software inputs the hardware legal candidate instructions into the virtual operation environment, determines the predicted outlet pressure according to the pump group response relationship, and determines the single pump allocation flow according to the pump start-stop combination and frequency allocation relationship. Based on the single-pump allocated flow rate and the corresponding pump frequency value, the head value and shaft power value are determined. The predicted outlet pressure is compared with the pressure verification boundary, and the single-pump allocated flow rate, the head value, and the shaft power value are matched with the energy-saving operation shell domain. For candidate instructions that fail the comparison or matching, frequency backoff and number replacement are performed, and the executable energy-saving instruction is output.

10. The AI-driven intelligent energy efficiency regulation method for water pumping stations according to claim 1, characterized in that, The steps for determining the energy-saving operating shell domain include: The data mining software reads the single pump characteristic table, which records the correspondence between pump number, frequency segment, flow segment, head range and shaft power range. Applicable characteristic records are then filtered based on the pump number and candidate frequency value in the valid candidate hardware instructions. The single pump's allocated flow rate is compared with the flow rate segments in the applicable characteristic record. The matched head range and shaft power range are read, and the flow rate segments, head range, and shaft power range are associated with the corresponding pump's energy-saving operating domain.

Images