A method and system for optimizing energy efficiency of an air separation unit based on multivariate data modeling

Abstract

The application discloses a kind of air separation device energy efficiency optimization method and system based on multivariate data modeling, it is related to air separation device operation optimization control technical field;Its technical points are: the present application associates the wet load bearing and regeneration recovery gap of molecular sieve previous period with the subsequent residual of cold box side, identifies the delay transmission of front-end purification disturbance to cold box energy efficiency data, whereby, the cold box energy efficiency sample affected by purification disturbance but surface operation index has recovered does not directly enter multivariate energy efficiency model update chain, and needs to be first subjected to sample admission state judgment and permission door control processing, avoid model to main heat exchanger end difference, cold box pressure drop and the lagging shift of unit product power consumption are mislearned as stable working condition characteristics, this processing can improve the reliability of energy efficiency model sample source, make subsequent expansion, flowback, tower pressure and compressor load etc. Control instruction is more suitable for real cold box operating state, to reduce the optimization deviation caused by purification cycle disturbance is misattributed.

Description

Technical Field

[0001] This invention relates to the field of air separation unit operation optimization and control technology, specifically to an air separation unit energy efficiency optimization method and system based on multivariate data modeling. Background Technology

[0002] Air separation units typically rely on industrial control systems to regulate the continuous gas production process. After compression and precooling, air enters the molecular sieve purification system, where molecular sieve adsorbers remove moisture and carbon dioxide from the air. The air then enters a cold box for heat exchange and distillation separation. Taking a continuous oxygen supply scenario in a chemical industrial park as an example, during nighttime operation in the rainy season, the ambient air humidity remains consistently high, causing short-term fluctuations in the cooling and drainage status of the air compressor. When the currently operating molecular sieve adsorber approaches the end of its adsorption phase, it switches to another adsorber. After the switch is complete, the inlet pressure, product flow rate, and key purity indicators on the control screen usually return to normal relatively quickly. Within acceptable limits, operators can judge whether the device has entered an optimizable stage based on these surface conditions. However, if the wet load is high in the previous adsorption cycle and the regeneration heating and cooling recovery are insufficient, the risks of residual moisture and carbon dioxide will not be eliminated synchronously with the switching action. Their impact will be gradually transmitted after the purified air enters the cold box, and will manifest as a slow increase in the terminal difference of the main heat exchanger, changes in local pressure drop in the cold box, a prolonged tower pressure recovery time, and a shift in unit product power consumption. Such phenomena are not prominent at a single sampling moment, and it is often necessary to combine the front-end purification cycle and the subsequent response process of the cold box to identify their source.

[0003] Existing air separation automatic control and energy efficiency optimization schemes can typically establish multivariate control models based on historical DCS data and generate energy-saving control recommendations based on operating data such as compressor power, cold box temperature status, expansion unit cooling capacity, tower pressure, product flow rate, and product purity. For periodic disturbances caused by molecular sieve switching, engineering often employs switching protection time, alarm limit monitoring, or abnormal sampling point removal to reduce the impact of significant fluctuations on the control model. This approach can adapt to operating conditions where the duration of switching disturbances is short and the subsequent state recovers quickly. However, when high humidity loads, insufficient regeneration recovery, and fluctuations in post-cooling drainage are superimposed, the impact of the front-end purification status on the cold box side energy efficiency data has a delayed transmission characteristic. If the control system only opens the energy efficiency sample write-back based on the recovery status of inlet pressure, product flow rate, and purity indicators after switching, subsequent operating samples affected by front-end purification disturbances may be classified as stable conditions. This causes the energy efficiency model to interpret the deviations in cold box end difference and unit power consumption as changes in cold box efficiency or distillation control deviations, thereby affecting the accuracy of optimized control of expansion capacity, return flow rate, and compressor load. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: An energy efficiency optimization method for air separation units based on multivariate data modeling includes the following steps: Acquire the purification cycle operation record of the air separation unit, generate purification cycle event record according to the cycle operation boundary of the molecular sieve adsorber, perform pairing processing of adsorption load carrying result and regeneration recovery gap, and generate purification disturbance state. After the molecular sieve switching is completed, the cold box side operation record is obtained. According to the purification disturbance state, the observation window associated with the molecular sieve switching event is configured. The cold box side operation record in the observation window is temporarily stored as a cold box energy efficiency sample. The cold box energy efficiency sample is compared with the baseline of similar operating conditions without purification disturbance, and the cold box operation residual is generated. The cold box operation residual is subjected to lag assignment processing. The sample admission state is generated with the cold box residual direction corresponding to the purification disturbance state as a reference. The sample admission state includes the disturbance suspension state and the stable admission state. The sample admission state is generated based on the lag assignment result of the cold box operation residual relative to the cold box residual direction and the assignment judgment result of the corresponding purification disturbance state. The permission gating process is performed according to the sample access status. When the sample access status is in the disturbed and suspended state, the corresponding cold box energy efficiency sample is suspended from participating in the multivariate energy efficiency model update. When the sample access status is in the stable access state, the corresponding cold box energy efficiency sample is allowed to participate in the multivariate energy efficiency model update, and energy efficiency control commands for adjusting the operating control quantity of the air separation unit are output based on the multivariate energy efficiency model.

[0005] An energy efficiency optimization system for a space separation unit based on multivariate data modeling includes: The purification cycle processing module acquires the purification cycle operation records of the air separation unit, generates purification cycle event records according to the cycle operation boundary of the molecular sieve adsorber, performs pairing processing of adsorption load carrying results and regeneration recovery gap, and generates purification disturbance state. The cold box residual processing module acquires the cold box side operation record after the molecular sieve switching is completed, configures the observation window associated with the molecular sieve switching event according to the purification disturbance state, temporarily stores the cold box side operation record in the observation window as a cold box energy efficiency sample, calls the similar working condition baseline without purification disturbance to compare the cold box energy efficiency sample, and generates cold box operation residual. The sample admission processing module performs lag attribution processing on the cold box operation residuals. It generates sample admission states with the direction of the cold box residuals corresponding to the purification disturbance state as a reference. The sample admission states include the disturbance suspension state and the stable admission state. The sample admission states are generated based on the lag attribution result of the cold box operation residuals relative to the direction of the cold box residuals and the attribution judgment result of the corresponding purification disturbance state. The permission gating processing module performs permission gating processing according to the sample admission status. When the sample admission status is in the disturbed and suspended state, it suspends the corresponding cold box energy efficiency sample from participating in the multivariate energy efficiency model update. When the sample admission status is in the stable admission state, it allows the corresponding cold box energy efficiency sample to participate in the multivariate energy efficiency model update and outputs energy efficiency control commands based on the multivariate energy efficiency model to adjust the operating control quantity of the air separation unit.

[0006] This invention provides a method and system for optimizing the energy efficiency of an air separation unit based on multivariate data modeling, which has the following beneficial effects: This invention correlates the wet load carrying capacity and regeneration recovery gap of the molecular sieve in the previous cycle with the subsequent residuals on the cold box side, identifying the delayed transmission of front-end purification disturbances to cold box energy efficiency data. As a result, cold box energy efficiency samples affected by purification disturbances but whose surface operating indicators have recovered will not directly enter the multivariate energy efficiency model update chain, but must first undergo sample admission status judgment and permission gating processing. This avoids the model mislearning the lag offset of the main heat exchanger terminal difference, cold box pressure drop, and unit product power consumption as stable operating condition characteristics. This processing can improve the reliability of the energy efficiency model sample source, making subsequent control commands such as expansion, return flow, tower pressure, and compressor load more consistent with the actual cold box operating state, thereby reducing the optimization deviation caused by the misattribution of purification cycle disturbances. Attached Figure Description

[0007] Figure 1 This is a flowchart of the method of the present invention;

[0008] Figure 2 This is a schematic diagram of the response process of typical purification cycle events and key indicators on the cold box side in this invention; Figure 3 This is a schematic diagram of the distribution of continuous attribution quantities within the observation window under different purification disturbance states in this invention. Detailed Implementation

[0009] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0010] Please see Figure 1 , Figure 2 and Figure 3 This embodiment provides a method for optimizing the energy efficiency of a space separation unit based on multivariate data modeling, including the following specific steps: S1, acquire the purification cycle operation record of the air separation unit, generate purification cycle event record according to the cycle operation boundary of the molecular sieve adsorber, perform pairing processing of adsorption load carrying result and regeneration recovery gap, and generate purification disturbance state.

[0011] In this embodiment, S1 processes the periodic status of the front-end purification side as an event, so that the subsequent cold box side operation record can correspond to the molecular sieve adsorber that actually supplies gas to the cold box air inlet channel. The air separation unit is equipped with two or more molecular sieve adsors that operate alternately. The molecular sieve adsorber switches between adsorption, regeneration, cooling and standby states. S1 uses the molecular sieve adsorber that is put into the cold box air inlet channel after this switch as the target molecular sieve adsorber.

[0012] Once the target molecular sieve adsorber is determined, the target molecular sieve adsorber number and the current commissioning boundary time are used as the associated fields. The subsequent adsorption load carrying results, regeneration recovery gap and purification disturbance status are written into the same purification cycle event record. This process ensures that the current cold box side operation record corresponds to the target molecular sieve adsorber that is actually supplied with gas, and avoids including the regeneration data of the molecular sieve adsorber that has just exited the adsorption state in the current cold box side operation record.

[0013] The purification cycle operation record is read from the DCS historical database of the air separation unit, the molecular sieve switching sequence control record, and the purification system operation log. The cycle operation boundary is used to define the object of the purification cycle event record; the inlet humidity state and the processed gas volume are used to form the adsorption load carrying result; the regeneration recovery state is used to form the regeneration recovery gap; the commissioning confirmation state and the regeneration completion confirmation state are used to verify the cycle continuity. The historical benchmark, historical boundary, and stable operation archive used for judging the current purification cycle event record all use data that has been formed and verified for effectiveness before the commissioning boundary time of this operation, and do not use the cold box side response data after the commissioning boundary time of this operation to correct the current purification disturbance state.

[0014] The DCS sampling period and historical archiving period preferentially follow the existing control system configuration of the device. In a preferred implementation, the DCS sampling period is 1 to 30 seconds and the historical archiving period is 10 to 60 seconds. This range of values ​​comes from the industrial control system's conventional acquisition capability of sequential control status, valve position status, pressure status and dew point status, and is used to establish the time correlation between molecular sieve switching events and cold box side operation records.

[0015] When the archiving periods of different variables are inconsistent, the event time recorded by the molecular sieve switching control is used as the main time stamp. Adjacent archiving points are aligned according to the most recent valid value. Segments with an alignment deviation exceeding one archiving period are configured as insufficient credibility segments. Insufficient credibility segments are not used to form the baseline moisture content level, carrying capacity boundary and stable operation reference boundary. The corresponding period is treated as an observational purification disturbance state.

[0016] When generating a purification cycle event record, the most recent purification cycle operation record of the target molecular sieve adsorber before this use is read. This purification cycle operation record takes the last time the target molecular sieve adsorber entered the adsorption use state as the adsorption use boundary, the last time the target molecular sieve adsorber exited the adsorption use state as the adsorption exit boundary, the target molecular sieve adsorber completed regeneration and reached the ready-to-use condition as the regeneration completion boundary, and the target molecular sieve adsorber entering the cold box inlet channel after this switch as the current use boundary. After the four cycle operation boundaries are encapsulated in chronological order, a purification cycle event record is generated.

[0017] The purification cycle event record includes the target molecular sieve adsorber number, adsorption start-up boundary, adsorption exit boundary, regeneration completion boundary, current start-up boundary, boundary source marker, and cycle validity marker. The boundary source marker is used to record the specific source of the cycle operation boundary from the sequential control event record, valve position status record, differential pressure status record, and regeneration completion confirmation status. The cycle validity marker is used to distinguish between valid cycles and cycles with insufficient reliability.

[0018] During the periodic encapsulation process, if short-term valve position jitter, sampling interruption, or repeated triggering of switching signals occurs, a periodic boundary check is performed. The periodic boundary check is completed based on the commissioning confirmation status and regeneration completion confirmation status of the target molecular sieve adsorber. The commissioning confirmation status is formed by the target molecular sieve adsorber outlet valve being continuously in the commissioning position, the molecular sieve inlet and outlet pressure difference being within the allowable range, and the cold box inlet channel switching completion status. The regeneration completion confirmation status is formed by the regeneration stage end signal, the cooling end status, and the bed pressure difference recovery status.

[0019] The preferred continuous holding time for commissioning confirmation is 30 to 180 seconds, and is calibrated based on the full stroke action time of the target molecular sieve adsorber outlet valve, the effective DCS scan cycle, and the stabilization time of the molecular sieve inlet and outlet pressure difference. In one preferred implementation, the continuous holding time is not less than the full stroke action time of the target molecular sieve adsorber outlet valve and not less than two effective DCS scan cycles. When the required continuous holding time after calibration according to the equipment sequential control parameters exceeds 180 seconds, the commissioning confirmation time in the equipment sequential control parameter table is used as the specific boundary of the device, and the specific boundary mark of the device is written in the purification cycle event record. When no effective calibration record is formed during the device commissioning phase, the continuous holding time is taken as 60 seconds as the initial value. After 10 subsequent stable switching records are completed, only the continuous holding time of the subsequent purification cycle is recalibrated, and the already generated purification cycle event records are not modified.

[0020] If a cycle still fails to form a complete and continuous time sequence after the cycle boundary review, it is configured as a cycle with insufficient credibility, and an observational purification disturbance state for that cycle is generated. This observational purification disturbance state is output along with the purification cycle event record, so that subsequent steps will include the corresponding cold box side operation record in the temporary storage and residual verification process.

[0021] After the purification cycle event record is effectively formed, the adsorption load carrying process is executed. The adsorption load carrying process reads the inlet air humidity and the change in the processed gas volume of the target molecular sieve adsorber in the most recent adsorption cycle before this commissioning. For each sampling point, the inlet humidity is first calculated using the wet air enthalpy calculation method or the saturated water vapor pressure lookup table conversion method based on the inlet dew point, inlet temperature and inlet pressure. Then, the positive difference between the inlet humidity and the baseline humidity level is calculated. This positive difference is multiplied by the processed gas volume in the same sampling interval and accumulated in the adsorption cycle to obtain the additional wet load.

[0022] The inlet moisture content and the reference moisture content level use the same dimensions, preferably uniformly defined as the water content in a unit standard volume of dry air. The treated gas volume adopts the standard volume flow rate that matches the inlet moisture content. When the output of the field flow meter is the operating volume flow rate, it is first converted into the standard volume flow rate according to the inlet temperature and inlet pressure, and then included in the additional moisture load accumulation. The additional moisture load obtained is the amount of additional moisture carried by the target molecular sieve adsorber in one adsorption cycle.

[0023] When inlet dew point sampling is missing, the pre-cooled outlet dew point and post-cooled drainage status are used for verification. If a reliable inlet moisture content cannot be obtained after verification, it is configured as a period with insufficient reliability, and an observation-type purification disturbance state is generated. When the target molecular sieve adsorber experiences an outlet dew point alarm in the most recent adsorption cycle before this commissioning, the alarm is written into the purification cycle event record, and an observation-type purification disturbance state is generated. This outlet dew point alarm is not used as the basis for screening stable benchmark samples.

[0024] The baseline humidity level is formed by the purification cycle operation records of the same air separation unit under similar environmental operating conditions. Similar environmental operating conditions are divided according to season, inlet air temperature range, and inlet air humidity range. In a preferred implementation, purification cycles with an inlet air temperature difference of no more than 3°C and an inlet relative humidity difference of no more than 10 percentage points are classified into the same environmental operating condition level. This grouping condition is used as the initial preferred condition and is calibrated based on the stable purification cycles formed by the same unit in different seasons.

[0025] The purification cycle operation records used to form the baseline moisture content level should meet the following requirements: no outlet dew point alarm, no abnormal post-cooling drainage status, valid regeneration completion confirmation status, valid bed pressure differential recovery, successful cycle boundary verification, and reliable inlet moisture content status. Preferably, there should be no fewer than 10 historical cycles that meet these conditions. If there are fewer than 10, the calibration benchmarks from the device commissioning and operation phases should be used, and the benchmark source mark should be written into the purification cycle event record. Records with failed cycle boundary verification, unreliable inlet moisture content status, conflicting regeneration recovery signals, or existing outlet dew point alarms in the current cycle should not be included in the formation of the baseline moisture content level.

[0026] The load-bearing capacity boundary is formed by the equipment manufacturer's calibration curve and the most recent maintenance calibration record, and is corrected according to the inlet temperature and inlet pressure of the target molecular sieve adsorber. If the most recent maintenance calibration record is missing, a conservative reduction is made according to the manufacturer's calibration curve. The reduction ratio is preferably 5% to 20%. The specific reduction ratio is formed based on the cumulative operating time of the adsorbent, the most recent replacement record, the historical alarm status of the outlet dew point, and the maintenance calibration deviation.

[0027] When the cumulative operating time of the adsorbent is short and there are no historical alarms for the outlet dew point, the reduction ratio is taken as the lower value of the preferred range; when the adsorbent is close to the replacement cycle, there are continuous historical alarms for the outlet dew point, and the maintenance calibration record is missing, the reduction ratio is taken as the higher value of the preferred range, and a conservative boundary mark is written in the purification cycle event record. Historical alarms for the outlet dew point are used to correct the carrying capacity boundary. The outlet dew point alarm of the target molecular sieve adsorber in the most recent adsorption cycle before this commissioning is treated as an observation-type purification disturbance state.

[0028] The load-bearing capacity classification boundary is jointly defined by the historical wet load boundary and the load-bearing capacity safety boundary. The historical wet load boundary is formed based on the distribution of additional wet load during the historical stable purification cycle under the same environmental condition level. In a preferred implementation, the value between the 80th and 90th percentiles is taken as the initial boundary. The load-bearing capacity safety boundary is formed by multiplying the load-bearing capacity boundary by the safety percentage. The safety percentage is preferably 60% to 85%, and is calibrated based on the outlet dew point alarm history, adsorbent operating time, and maintenance calibration deviation.

[0029] The historical wet load boundary, carrying capacity boundary, and carrying capacity safety boundary are all unified as the cumulative amount of additional wet load within one adsorption cycle, in units of water mass per adsorption cycle. The carrying capacity classification boundary is taken as the lower value between the historical wet load boundary and the carrying capacity safety boundary to retain the purification margin at the end of the adsorption cycle. When the historical stable purification cycle is insufficient, the carrying capacity classification boundary is formed by the manufacturer's calibration curve after conservative reduction, and the conservative boundary mark is written.

[0030] The adsorption load carrying capacity results are divided into normal carrying capacity state and elevated carrying capacity state. When the additional wet load does not reach the carrying capacity classification boundary, the normal carrying capacity state is generated; when the additional wet load reaches the carrying capacity classification boundary, the elevated carrying capacity state is generated. The adsorption load carrying capacity results formed according to the conservative boundary marking are only used for the generation of the purification disturbance state in the current cycle and are not used to update the baseline moisture content level and carrying capacity boundary.

[0031] Subsequently, a regeneration recovery gap treatment is performed. The regeneration recovery gap treatment reads the recovery status of the target molecular sieve adsorber in the most recent regeneration stage before this commissioning, and evaluates the degree of recovery of the additional wet load of the previous adsorption cycle by the regeneration process. The regeneration recovery status is read from the molecular sieve regeneration sequential control record and DCS operation record, including the regeneration heating recovery, regeneration gas purging, cooling end status and bed pressure difference recovery status.

[0032] The regeneration heating recovery is determined based on the duration of the regeneration temperature reaching the set range and the regeneration end temperature. The regeneration gas purging is determined based on the cumulative regeneration gas volume and the purging duration. The cooling end status is determined based on whether the bed outlet temperature has dropped back to the ready-to-use boundary. The bed pressure difference recovery status is determined based on the comparison between the bed pressure difference before commissioning and the stable operating pressure difference boundary. The set range, duration, gas volume boundary, temperature boundary, and pressure difference boundary required for the above judgments are preferentially taken from the equipment manufacturer's commissioning requirements, DCS interlock boundaries, and the stable operation records of the same unit.

[0033] Each regeneration recovery status generates three evaluation results: reaching the stable operation reference boundary, reaching only the reactivation boundary, and not reaching the reactivation boundary. Reaching the stable operation reference boundary means that the regeneration recovery status meets both the reactivation boundary and the stable operation reference boundary; reaching only the reactivation boundary means that the regeneration recovery status meets the reactivation boundary but does not reach the stable operation reference boundary; and not reaching the reactivation boundary means that the regeneration recovery status does not meet the minimum boundary required for the equipment to be put back into operation.

[0034] The re-operation boundary is derived from the equipment manufacturer's operation requirements and the DCS interlock boundary. It is used to determine whether the target molecular sieve adsorber meets the minimum conditions for re-operation in the cold box inlet channel. The stable operation reference boundary is formed by the stable operation records of the same device. It is used to determine whether the target molecular sieve adsorber has reached the stable benchmark level. When the stable operation records are insufficient, the re-operation boundary is used as the priority boundary, and a benchmark insufficient cycle mark is written in the purification cycle event record.

[0035] When regeneration recovery data is missing, the sequential control end signal, cooling end status and adjacent measurement point records are first called for verification. If the cycle still cannot form a regeneration recovery status evaluation result after verification, it is configured as a cycle with insufficient credibility and an observation-type purification disturbance status is generated. If the missing data is accompanied by regeneration insufficiency interlocking restrictions, commissioning confirmation status and regeneration completion confirmation status conflict, a high purification disturbance status is generated.

[0036] The regeneration recovery gap is divided into low gap, medium gap, and high gap states. If the regeneration heating recovery, regeneration gas purging, cooling end state, and bed pressure difference recovery state all reach the stable operation reference boundary, a low gap state is generated. If the regeneration heating recovery, regeneration gas purging, cooling end state, and bed pressure difference recovery state all meet the restart boundary, but at least one of them only meets the restart boundary, a medium gap state is generated. If any of the regeneration heating recovery, regeneration gas purging, cooling end state, and bed pressure difference recovery state does not meet the restart boundary, a high gap state is generated. When the interlocking limit related to insufficient regeneration is triggered, a high gap state is generated. If the regeneration recovery data indicates that the restart boundary is not met, but a restart confirmation state has been generated subsequently, this cycle is treated as a regeneration recovery signal conflict cycle, a high gap state is generated, and a signal conflict flag is written in the purification cycle event record.

[0037] After obtaining reliable adsorption load results and regeneration recovery gap, pairing processing is performed to generate a purification disturbance state. The pairing processing is performed in the order of first processing high-risk conditions, then processing observation conditions, and finally processing low-disturbance conditions. When the generation conditions of observation-type purification disturbance state and high purification disturbance state are met simultaneously in the same cycle, a high purification disturbance state is generated.

[0038] The purification disturbance state is divided into low purification disturbance state, observational purification disturbance state, and high purification disturbance state. If the adsorption load is in a normal load state and the regeneration recovery gap is in a low gap state, a low purification disturbance state is generated. If the adsorption load is in a normal load state and the regeneration recovery gap is in a medium gap state, an observational purification disturbance state is generated. If the adsorption load is in an increased load state and the regeneration recovery gap is in a low gap state, an observational purification disturbance state is generated. If the adsorption load is in a normal load state and the regeneration recovery gap is in a high gap state, a high purification disturbance state is generated. If the adsorption load is in an increased load state and the regeneration recovery gap is in a medium gap state, a high purification disturbance state is generated. If the adsorption load is in an increased load state and the regeneration recovery gap is in a high gap state, a high purification disturbance state is generated.

[0039] The low purification disturbance state corresponds to a combination of the normal load state and the low gap state, indicating that the target molecular sieve adsorber did not have the risk of high humidity load and insufficient regeneration before this commissioning. The observational purification disturbance state corresponds to at least one of the following: unilateral risk and insufficient cycle reliability, indicating that the corresponding cold box side operation record needs to be temporarily stored and residual verified in subsequent steps. The high purification disturbance state corresponds to a combination of high gap state or increased load state and insufficient recovery, indicating that the corresponding cold box side operation record needs to be restricted from directly participating in the multivariate energy efficiency model update in subsequent steps.

[0040] This step outputs the purification cycle event record and purification disturbance status. The purification cycle event record includes at least the target molecular sieve adsorber number, the current commissioning boundary time, the cycle validity mark, the adsorption load carrying capacity result, the regeneration recovery gap, the purification disturbance status, and the boundary source mark. The purification disturbance status is used to configure the observation window after molecular sieve switching in subsequent steps and serves as the front-end admission basis for whether the cold box energy efficiency sample enters the temporary storage process. S1 does not judge the change in cold box efficiency, nor does it update the multivariate energy efficiency model. When the cold box operating residual is inconsistent with the purification disturbance status, the attribution verification result of the cold box operating residual is used as the basis for forming the sample admission status.

[0041] S2. After the molecular sieve switching is completed, the cold box side operation record is obtained. According to the purification disturbance state, the observation window associated with the molecular sieve switching event is configured. The cold box side operation record in the observation window is temporarily stored as a cold box energy efficiency sample. The similar operating condition baseline without purification disturbance is called to compare the cold box energy efficiency sample and generate the cold box operation residual.

[0042] Following the S1 purification cycle event record, S1 has determined the target molecular sieve adsorber to be put into the cold box inlet channel after this switch, and writes the target molecular sieve adsorber number, the boundary of this use, the adsorption load carrying result, the regeneration recovery gap and the purification disturbance status into the purification cycle event record. S2 uses this purification cycle event record as input, configures the cold box side observation window after the molecular sieve switch according to the purification disturbance status, and temporarily stores and compares the cold box side operation record in the observation window with the baseline.

[0043] The time corresponding to the current commissioning boundary is recorded as t0. t0 comes from the purification cycle event record generated by S1 and is used as the time anchor point between the molecular sieve switching event and the cold box side operation record. When S2 configures the observation window, it calls the historical operation segment that has been formed and completed the validity verification before t0. It does not use the cold box side operation response after t0 to correct the current observation window. When the purification cycle event record lacks the current commissioning boundary, S2 will temporarily store the corresponding cold box side operation record and write a time anchor point missing mark in the cold box energy efficiency sample.

[0044] The cold box side operation records are read from the air separation unit's DCS historical database, cold box control records, and energy consumption metering records, and unified to the same time axis according to t0. S2 divides the cold box side operation records into operating condition parameters and cold box side response quantities. Operating condition parameters are used to screen similar operating conditions without purification disturbances, and cold box side response quantities are used to generate cold box operation residuals. Operating condition parameters include purified air flow rate, product flow rate, product extraction quantity, product purity setpoint, tower pressure setpoint, expansion quantity setpoint, return flow setpoint, and ambient temperature. Cold box side response quantities include main heat exchanger terminal difference, cold box pressure drop, tower pressure deviation, and unit product power consumption. Tower pressure setpoint belongs to operating condition parameters, and tower pressure deviation belongs to cold box side response quantities. Tower pressure deviation is formed by the difference between the measured tower pressure value and the tower pressure setpoint. Unit product power consumption is calculated by the total power consumption of the air separation unit in the corresponding metering period and the product conversion quantity. The product conversion quantity is formed according to the product metering rules and conversion coefficients of the same unit.

[0045] Before configuring the observation window, S2 calculates the gas path transmission delay between the outlet of the target molecular sieve adsorber and the response measurement point on the cold box side. The gas path transmission delay is used to determine the starting point of the observation window so that the temporary storage starting point of the cold box side operation record corresponds to the response stage after the purified air arrives at the response measurement point on the cold box side.

[0046] The gas transmission delay is denoted as The calculation method is as follows: Divide the equivalent standard gas volume between the outlet of the target molecular sieve adsorber and the response measurement point on the cold box side by the time-weighted standard volumetric flow rate of purified air within the stable flow confirmation segment to obtain the gas transmission time; superimpose the gas transmission time on the measurement response time of the response measurement point on the cold box side to obtain the initial gas path transmission delay; limit the initial gas path transmission delay using the lower limit and upper limit of the gas path transmission delay to obtain... .

[0047] It can be represented as: ;in, This refers to the gas transmission delay, expressed in minutes. The equivalent standard gas volume between the outlet of the target molecular sieve adsorber and the response measurement point on the cold box side is in Nm³, and is taken from the device pipeline volume table, the cold box inlet pipe section volume and the on-site commissioning record. The time-weighted standard volumetric flow rate of purified air during the stable flow confirmation segment after the commissioning confirmation state is formed is Nm³ / min, taken from the DCS flow record and converted to the standard state according to the inlet pressure and inlet temperature. The measurement response time of the cold box side response measurement point is in minutes and is taken from the measurement point calibration record and instrument response parameters. The lower limit of the gas path transmission delay is preferably 5 to 15 minutes; The upper limit of the gas path transmission delay is preferably 30 to 60 minutes.

[0048] and The data comes from the same device's pipeline volume, purified air flow range, and cold box-side response measurement point calibration records. Used to limit the instantaneous disturbance of the molecular sieve switching valve from being prematurely covered by the observation window. To limit excessive shift of the observation window caused by a short-term decrease in flow rate, the stable flow confirmation segment is captured after the activation confirmation state is established. The preferred length is 3 to 10 minutes, and the purified air flow rate within this segment must be within the effective range of the DCS. This segment does not include cold box inlet interlock restrictions, main valve position reswitching, or flow measurement point fault markers. If the stable flow confirmation segment cannot be formed, or if flow measurement point fault markers are present, S2... A conservative observation window starting point is established, and a time-delay conservative marker is written into the cold box energy efficiency sample; it is able to form At that time, write the normal latency flag.

[0049] The observation window is denoted as W, and the starting point of W is t0+. S2 sets the basic window length T0, which is preferably 30 min to 90 min. The value is based on the response time of the cold box side response measurement point, the tower pressure adjustment cycle, and the unit product power consumption metering cycle. When the purification disturbance state is low, the observation window length is T0; when the purification disturbance state is observation-type purification disturbance state, the observation window length is Ko×T0; when the purification disturbance state is high, the observation window length is Kh×T0. Ko is the first amplification factor, preferably 1.2 to 1.8; Kh is the second amplification factor, preferably 2.0 to 3.0. The observation window length is only used to limit the temporary storage range of the cold box side operation record and does not change the purification disturbance state output by S1.

[0050] Ko and Kh are formed based on the duration of the persistent impact of historical cold box operation residuals archived before t0. The duration of the persistent impact of historical cold box operation residuals refers to the time length within the observation window corresponding to the historical molecular sieve switching event, after at least one of the following—main heat exchanger terminal difference residual, cold box pressure drop residual, tower pressure deviation residual, and unit product power consumption residual—reaches the residual manifestation boundary, until all cold box operation residuals fall back below the residual fallback boundary and remain continuously below it for no less than two sample segments. The residual fallback boundary is a dimensionless boundary used to represent the boundary where historical cold box operation residuals fall back from identifiable deviations to the normal fluctuation range, and its value is identical to the residual in S3. To maintain consistency with the boundary, the residual manifestation boundary is preferably 0.5 to 0.8, and is formed by the values ​​between the 90th and 95th percentiles of the residual distribution of the historical stable operation segments of the same device. For historical events corresponding to observational purification disturbance states, Ko is formed according to the ratio of the duration of the historical cold box operation residual to the basic window length T0; for historical events corresponding to high purification disturbance states, Kh is formed according to the ratio of the duration of the historical cold box operation residual to the basic window length T0. Ko is limited to 1.2 to 1.8, and Kh is limited to 2.0 to 3.0; when there are insufficient historical events, Ko is 1.8 and Kh is 3.0.

[0051] S2 reads the cold box side operation record in the observation window W and encapsulates it into cold box energy efficiency samples according to sample segments. The length of the sample segment is denoted as Ts. Ts is preferably 5 min to 15 min. The value is based on the DCS archiving cycle, the unit product power consumption metering cycle, and the cold box side adjustment action completion time. When the effective metering cycle of unit product power consumption is greater than 15 min, Ts is not shorter than the effective metering cycle of unit product power consumption. The segment source mark is written into the cold box energy efficiency sample. Each cold box energy efficiency sample is written with the sample start and end time, target molecular sieve adsorber number, current commissioning boundary, purification disturbance state, operating parameters, cold box side response quantity, sample integrity mark, and time delay mark. The cold box energy efficiency sample is temporarily stored in S2 and is not written into the multivariate energy efficiency model.

[0052] S2 forms a sample integrity mark based on the validity of the cold box side operation record. When there is a missing measurement in the cold box side operation record, S2 reviews it based on the adjacent valid sampling points of the same measurement point and the DCS event record. The operating parameters that are still missing after review are not included in the calculation of the operating condition fit difference. When the missing measurement duration of non-core records does not exceed the allowable percentage of missing measurements of one sample segment length, a missing measurement mark is written on the missing measurement location, and the cold box energy efficiency sample is retained for comparison. The allowable percentage of missing measurements is preferably 20%, and can be calibrated within the range of 10% to 30% based on the DCS archiving quality and measurement point verification records.

[0053] The main heat exchanger terminal difference, cold box pressure drop, tower pressure deviation, and unit product power consumption are core response quantities. The effective coverage rate of the core response quantities is formed by the ratio of the number of effective sampling points to the number of sampling points within the sample segment, preferably 80%, and can be calibrated within the range of 70% to 90% based on the measurement point calibration records and DCS archiving quality. When the main heat exchanger terminal difference, cold box pressure drop, tower pressure deviation, and unit product power consumption all meet the corresponding effective coverage rate requirements, and the number of operating condition parameters participating in the calculation of the operating condition fit difference reaches more than 70% of the number of candidate operating condition parameters, the sample integrity is marked as reliable; if the above conditions are not met, the sample integrity is marked as low reliability.

[0054] Subsequently, S2 establishes a similar operating condition baseline without purification disturbance. This similar operating condition baseline is generated from historical observation segments archived before t0 and in a low purification disturbance state. When a historical observation segment enters the baseline candidate library, it should meet the following requirements: the corresponding front-end purification state is in a low purification disturbance state, and no corresponding historical target molecular sieve adsorber outlet dew point alarm, cold box interlock restriction, product purity exceeding the limit, core response quantity missing measurement, or cold box side control mode switching has occurred within the historical observation segment. The cold box side control mode is confirmed through DCS control mode marking, interlock restriction status, product purity control status, tower pressure control status, expansion control status, and reflux control status. The baseline candidate library does not call data after t0.

[0055] In this step, historical stable operating segments are used to form the fitting weight, fitting difference limit, fitting stability term, engineering response limit, and window magnification factor. The historical stable operating segments should meet the following requirements: the purification disturbance state is low, the cold box side control mode is consistent, no cold box interlocking restrictions occur, product purity exceeds the limit, and core response quantity is missing. If the historical stable operating segments are insufficient, subsequent processing is formed according to the initial values ​​and conservative boundaries of each parameter, and the parameter source mark is written into the cold box energy efficiency sample.

[0056] For any historical observation segment in the baseline candidate library S2 calculates the difference in operating condition fit between the historical observation segment and the current cold box energy efficiency sample. During the calculation, the first value in the current cold box energy efficiency sample is obtained. Representative values ​​of individual operating parameters, and historical observation segments. The Middle The segment representative value of each working condition parameter; calculate the absolute value of the difference between the two; divide this absolute value by the first... The allowable variation range of each working condition parameter is determined to obtain the dimensionless difference; the dimensionless differences corresponding to each working condition parameter are then weighted and summed according to the fitting weight to obtain... .

[0057] It can be represented as: ;in, The difference in the fit between historical observation segment b and the current cold box energy efficiency sample is dimensionless. The number of operating condition parameters to be included in the calculation is preferably 5 to 9. The flow rate of the same type of product and the extraction volume of the same type of product can be combined into a representative item of the same type of operating condition parameter. The first in the current cold box energy efficiency sample A segment representing the value of a working condition parameter; Historical observation fragments The Middle A segment representing the value of a working condition parameter; For the first The allowable variation range of the engineering parameters for each operating condition, and related to the first... All operating parameters maintain the same dimensions; For the first The fitting weights for each operating condition parameter range from 0.05 to 0.40, and the sum of all fitting weights is 1. The representative value of a segment is the time-weighted running value within the sample segment, and the weights are formed by the effective sampling duration. Taken from DCS operating limits, product scheduling allowable fluctuation range, and equipment design operating range. Missing 0 and When the dimensions of the operating condition parameter are inconsistent with those of the corresponding operating condition parameter, the operating condition parameter will not be included in the calculation of the operating condition fit difference, and a parameter limit missing mark will be written in the cold box energy efficiency sample. When the number of operating condition parameters involved in the calculation is less than 70% of the number of candidate operating condition parameters, the baseline validity mark will be set to low confidence.

[0058] Fit weight Formed by the control sensitivity of historically stable operating segments, for the first For each operating condition parameter, a historical stable operating segment is selected where the operating condition parameter undergoes a controlled change while other operating condition parameters remain within the allowable range of engineering variation. The engineering response amplitude of the unit product power consumption before and after the change of the operating condition parameter is calculated, and the engineering response amplitudes of each operating condition parameter are summed and normalized to obtain the fitting weight. If the normalization result does not meet the value range of 0.05 to 0.40, first merge the flow rate of the same type of product, the extraction quantity of the same type of product, and the setting parameters of the same type, and then recalculate the fitting weight. If the value range still cannot be met after recalculation, the baseline validity mark is set to low confidence. If there are insufficient historical stable operation segments, the initial fitting weight is configured with equal weights according to the merged 5 to 9 operating condition parameters. If 5 to 9 operating condition parameters cannot be formed, the baseline validity mark is set to low confidence. After the subsequent historical stable operation segments reach 10, the fitting weight is recalibrated for the subsequent cold box energy efficiency samples. The recalibration does not reverse the already generated cold box operation residuals.

[0059] when Historical observation segments when the fit difference is not greater than the limit Dlim When entering the set of similar segments, Dlim is preferably between 0.20 and 0.45. Its value is formed by the distribution of the operating condition fit difference of historical stable operating segments. When there are enough historical stable operating segments, the value between the 80th and 90th percentiles of the corresponding distribution is taken as Dlim. When there are not enough historical stable operating segments, Dlim is taken as 0.35 as the initial value. After there are 10 historical stable operating segments, the subsequent cold box energy efficiency samples are recalibrated. When there are fewer than 3 similar segments, S2 increases Dlim in the range of 0.20 to 0.45 with a step size of 0.05. After each increase, similar segments are re-selected. When there are 3 similar segments, Dlim is stopped from being increased. The relaxed similar segments must still meet the admission conditions of no purification disturbance, no interlocking restrictions, qualified effective coverage of core response quantity, and consistent control mode on the cold box side. When there are still fewer than 3 similar segments after Dlim reaches 0.45, the baseline validity mark is set to low confidence.

[0060] After the set of similar segments is formed, S2 generates a baseline of similar operating conditions without cleanup disturbances according to the time phase alignment method. The phase alignment is based on the relative time between the start of the current observation window and the start of the historical observation window. The start of the current observation window is t0+. The starting point of the historical observation window is formed by the time corresponding to the historical commissioning boundary and the historical gas path transmission delay. When the effective coverage rate of the core response quantity of the historical observation segment at the corresponding relative time position is lower than the corresponding effective coverage rate requirement, it will not participate in the baseline generation at that relative time position. When the number of historical observation segments participating in the baseline generation at that relative time position is less than 3, the baseline validity mark is set to low confidence.

[0061] For the The response quantity on the cold box side, within the observation window. S2 generates baseline values ​​for similar operating conditions without purification disturbances. During calculation, it reads the first historical observation segment from the similar segment set at the same relative time position. The cold box side response is measured; baseline weights are formed based on the difference in operating condition fit and data completeness of each historical observation segment; the cold box side response of each historical observation segment is weighted and averaged to obtain the baseline value for similar operating conditions.

[0062] It can be represented as: ; ; in, For the first The response of the cold box at time... The baseline value of similar operating conditions; The number of historical observation segments that enter the set of similar segments and have a valid core response at the corresponding relative time position; Historical observation fragments At the same relative time position The response quantity on the cold box side; Historical observation fragments Baseline weights; Historical observation fragments Data integrity coefficient; This is a stability term for fit, with a value range of 0.05 to 0.15; Historical observation fragments The difference in operating condition fit between the current cold box energy efficiency sample and the sample. The baseline weight is determined by the ratio of the number of valid sampling points to the number of sampling points within a historical observation segment, and is included in the baseline weight calculation. Set the corresponding effective coverage requirement to 1.00; When the coverage rate is lower than the corresponding effective coverage requirement, historical observation segments Not involved in the generation of baselines for similar operating conditions. Calibration is performed based on the condition fit distribution of historical stable operating segments; if historical stable operating segments are insufficient, Take 0.10.

[0063] After a baseline of similar operating conditions without purification disturbance is formed, S2 generates a baseline validity marker. The baseline validity marker is considered reliable when there are more than 3 similar segments, all historical observation segments involved in the baseline generation meet the data integrity requirements, the effective coverage of the core response quantity is qualified, and the cold box side control mode is consistent. The baseline validity marker is considered low reliability when any of the following conditions are met: insufficient number of similar segments, insufficient effective coverage of the core response quantity, or inconsistent cold box side control mode.

[0064] Engineering response limit Based on the response quantity on the cold box side, the differential pressure at the main heat exchanger terminals is calculated separately. Take the larger value among the instrument's permissible error, the DCS control dead zone, the equipment design's permissible terminal differential fluctuation, and the upper limit of the terminal differential offset in historical stable operating segments; for cold box pressure drop, Take the larger value among the allowable error of the differential pressure measuring point, the DCS control dead zone, the design allowable pressure drop fluctuation, and the upper limit of the pressure drop deviation in historical stable operating segments; for the tower pressure deviation... Take the larger of the pressure measurement point allowable error, tower pressure control dead zone, and tower pressure deviation from the upper limit during historical stable operation segments; for unit product power consumption, Take the larger of the following: metering allowable error, the upper limit of normal fluctuation within the energy consumption statistical period, and the upper limit of unit product power consumption deviation during historical stable operation periods. If historical stable operation periods are insufficient... The larger of the allowable fluctuation in equipment design and the allowable error of the instrument is used. If the value is 0, missing, or inconsistent with the dimensions of the corresponding cold box side response, the cold box operation residual corresponding to that cold box side response will not be generated.

[0065] After generating a baseline of similar operating conditions without purification disturbances and establishing an engineering response limit, S2 performs a comparison process on the current cold box energy efficiency sample to generate the cold box operating residual. For the first... The cold box side response quantity will be the current cold box energy efficiency sample at time 10:00. The The response offset is obtained by subtracting the baseline value of similar operating conditions at the corresponding moment from the cold box side response value; using the first... The engineering response limit of each cold box side response quantity is normalized to obtain the cold box operation residual.

[0066] It can be represented as: ;in, For the first The response of the cold box at time... The residuals of cold box operation are dimensionless; For the current energy efficiency sample of the cold box at time The The response quantity on the cold box side; This represents the baseline value for similar operating conditions at the corresponding time. For the first The engineering response limit of the cold box side response quantity, and the first The response quantities on each cold box side maintain the same dimensions.

[0067] In one specific implementation, S1 has determined that the target molecular sieve adsorber is adsorber A, the time corresponding to the current commissioning boundary is 10:00, the purification disturbance state is the observation type purification disturbance state, the equivalent standard gas volume between the outlet of the target molecular sieve adsorber and the response measurement point on the cold box side is 1800 Nm³, the standard volumetric flow rate of purified air in the stable flow confirmation segment is 120 Nm³ / min, the measurement response time of the response measurement point on the cold box side is 4 min, the lower limit of the gas path transmission delay is 5 min, the upper limit of the gas path transmission delay is 60 min, then the gas transmission time is 15 min, the initial gas path transmission delay is 19 min, the starting point of the observation window is 10:19, the basic window length is 60 min, the first amplification factor is 1.5, and the observation window is from 10:19 to 11:49. S2 encapsulates the cold box side operation record in 10-minute sample segments within this observation window to form 9 cold box energy efficiency samples.

[0068] For the cold box energy efficiency sample from 10:19 to 10:29, if the operating condition fit differences of historical observation segments B1, B2, and B3 are 0.18, 0.24, and 0.31 respectively, and none of them exceed the fit difference limit of 0.35, then B1, B2, and B3 are included in the similar segment set. Taking the main heat exchanger terminal temperature difference as an example, if the main heat exchanger terminal temperature differences of B1, B2, and B3 at the same relative time position are 1.7℃, 1.8℃, and 1.9℃ respectively, the data is complete. With integer coefficients of 0.98, 0.95, and 0.92, and a fit stability term of 0.10, the baseline weights are 3.50, 2.79, and 2.24, respectively. After weighted averaging, the baseline value of the main heat exchanger terminal difference under similar operating conditions is approximately 1.79℃. If the current cold box energy efficiency sample has a main heat exchanger terminal difference of 2.2℃ at the same relative time position, and the engineering response limit of the main heat exchanger terminal difference is 0.5℃, then the residual of the main heat exchanger terminal difference is 0.82.

[0069] The example data for S2 in this specific embodiment is shown in the table below. The example data listed is used to illustrate the calculation process. The specific engineering values ​​are formed by the DCS configuration, equipment information, debugging records and historical stable segment calibration of the same device.

[0070] Table 1. Example data for S2 and corresponding technical meanings:

[0071] S2 controls the output of cold box operation residuals according to sample integrity label, time delay label, and baseline validity label. When the sample integrity label is low confidence, no corresponding cold box operation residual is generated; when the time delay label is conservative, no corresponding cold box operation residual is generated; when the baseline validity label is low confidence, no corresponding cold box operation residual is generated. Cold box energy efficiency samples for which no cold box operation residuals are generated remain in a temporary state and are not entered into the lag assignment process of S3, nor are they written into the multivariate energy efficiency model.

[0072] The cold box operation residuals are output in the form of residual records. The residual records include at least the target molecular sieve adsorber number, the current commissioning boundary, the observation window number, the cold box energy efficiency sample number, the number of similar segments, the baseline validity mark, the main heat exchanger terminal difference residual, the cold box pressure drop residual, the tower pressure deviation residual, and the unit product power consumption residual. The cold box energy efficiency sample output in this step is used to save the cold box side operation record in the observation window after the molecular sieve switching. The cold box operation residuals are used to represent the deviation of the current cold box energy efficiency sample from the similar operating condition without purification disturbance. S2 does not generate the sample admission status; the sample admission status is formed by the subsequent S3 based on the lag relationship between the cold box operation residuals and the purification disturbance status.

[0073] S3. Perform lag assignment processing on the cold box operation residuals. Generate sample admission states with the cold box residual direction corresponding to the purification disturbance state as a reference. The sample admission states include the disturbance-prone temporary state and the stable admission state. The sample admission states are generated based on the lag assignment results of the cold box operation residuals relative to the cold box residual direction and the assignment judgment results of the corresponding purification disturbance state.

[0074] Following the cold box operation residuals from S2, S2 has already formed an observation window based on this molecular sieve switching event, and generated cold box energy efficiency samples and cold box operation residuals within the observation window. S3 does not reconfigure the observation window, does not re-establish a similar operating condition baseline, and does not change the purification disturbance state output by S1. S3 uses the cold box energy efficiency samples for which S2 has already generated cold box operation residuals as the processing object, reads the target molecular sieve adsorber number, the current commissioning boundary, the observation window number, the cold box energy efficiency sample number, the purification disturbance state, and the cold box operation residuals from the corresponding residual record, and performs hysteresis assignment processing.

[0075] S3 is used to distinguish between the normal adjustment fluctuations on the cold box side after molecular sieve switching and the delayed effects of front-end purification disturbances. After the molecular sieve switching is completed, the inlet pressure, purified air flow rate, and product purity may recover and stabilize before the cold box side. However, the residual effects caused by the high adsorption load and insufficient regeneration recovery will continue to manifest as an increase in the main heat exchanger terminal difference, an increase in cold box pressure drop, an increase in tower pressure deviation, and an increase in unit product power consumption after the purified air enters the cold box. Therefore, S3 matches the cold box operating residual with the direction of the cold box residual corresponding to the purification disturbance state, and generates the sample access state by combining the lag position and continuity within the observation window.

[0076] S3 first determines the cold box residual direction based on the purification disturbance state. When the purification disturbance state is an observational state, it indicates that S1 has identified a possible purification disturbance in the target molecular sieve adsorber, but the disturbance intensity still needs to be confirmed in conjunction with the cold box side response. When the purification disturbance state is a high purification disturbance state, it indicates that the target molecular sieve adsorber already had a high risk of purification disturbance before this commissioning. For observational and high purification disturbance states, S3 configures the cold box residual direction so that the cold box operating residual is located on the side unfavorable to the cold box energy efficiency, including... The residual values ​​of the main heat exchanger terminal difference, cold box pressure drop, tower pressure deviation, and unit product power consumption all increase positively. The positive increase of the tower pressure deviation residual refers to the change of the residual formed by the tower pressure deviation as the cold box side response quantity in S2 along the direction of increasing deviation amplitude. The tower pressure deviation is formed according to the deviation amplitude of the measured tower pressure value relative to the tower pressure set value. When the purification disturbance state is a low purification disturbance state, S3 does not configure the purification lag attribution direction, and the corresponding cold box energy efficiency sample does not generate a disturbed temporary state due to purification disturbance.

[0077] Subsequently, S3 performs directional screening on the cold box operation residuals in the cold box energy efficiency samples. The objects of directional screening are the main heat exchanger terminal difference residual, cold box pressure drop residual, tower pressure deviation residual, and unit product power consumption residual output by S2. For each cold box operation residual, S3 determines whether the residual is located on the cold box residual direction side and whether it exceeds the residual no-action boundary. Only cold box operation residuals that are located on the cold box residual direction side and exceed the residual no-action boundary will enter the formation process of directional assignment quantity. If either the case of not being located on the cold box residual direction side or not exceeding the residual no-action boundary is met, the corresponding cold box operation residual will not participate in the formation of directional assignment quantity.

[0078] The residual actionless boundary is a dimensionless boundary, preferably between 0.10 and 0.30. This boundary matches the normalized caliber of the cold box operation residual in S2 and is used to exclude the allowable error of the cold box side measurement point, the DCS control dead zone, the normal fluctuation within the unit product power consumption metering cycle, and the baseline error of similar operating conditions. The actual value is extracted from the historical stable operation segments that have been archived before this commissioning boundary. It is preferred to take the value between the 80th and 90th percentiles of the corresponding cold box operation residual distribution in the historical stable operation segments, and limit it to between 0.10 and 0.30. If there are insufficient historical stable operation segments, 0.20 is taken as the initial value.

[0079] After completing the directional screening, S3 generates a directional assignment quantity based on the screened cold box operating residuals. The directional assignment quantity is used to represent the degree of agreement between the cold box operating residuals in the current cold box energy efficiency sample and the cold box residuals corresponding to the purification disturbance. When generating the directional assignment quantity, S3 subtracts the corresponding residual no-action boundary from each screened cold box operating residual to obtain the remaining quantity exceeding the boundary. The part less than 0 after subtraction is treated as 0. Then, the remaining quantities are synthesized according to the contribution weight of the cold box side response quantity to obtain the directional assignment quantity. This processing ensures that the directional assignment quantity only reflects the effective residuals exceeding the normal fluctuation boundary, avoiding the inclusion of low-amplitude measurement point fluctuations in the purification disturbance assignment result. The directional assignment quantity only represents the degree of directional agreement of the current cold box energy efficiency sample and is not directly used as the result of the sample admission status determination.

[0080] The contribution weights of the cold box side response quantities are used to distinguish the sensitivity of different cold box side response quantities to the hysteresis propagation of purification disturbances. The main heat exchanger terminal difference and cold box pressure drop are used to characterize the changes in cold box heat exchange load and flow resistance. The unit product power consumption is used to characterize the energy efficiency results. The tower pressure deviation is used to characterize the pressure regulation status of the distillation system. The preferred value range of the contribution weights is 0.10 to 0.50, and the sum of all contribution weights is 1. The contribution weights are formed by historical purification disturbance events that have been archived before the current commissioning boundary. Specifically, during the formation, the contribution weights are formed within the observation window corresponding to the historical observation type purification disturbance state and the historical high purification disturbance state. The residual of each cold box operation is calculated separately, and the remaining amount after it is located on the side of the cold box residual direction and exceeds the residual no-action boundary is accumulated according to continuous sample segments. Then, the accumulated residual amount corresponding to each cold box side response quantity is summed and normalized to obtain the initial contribution weight. When the initial contribution weight is between 0.10 and 0.50, the initial contribution weight is used as the contribution weight of the corresponding cold box side response quantity. If any of the following situations exist, such as the initial contribution weight exceeds this range, there are fewer than 5 historical purification disturbance events, the sum of the accumulated residual amount is 0, or the core response quantity in the historical events is incomplete, the contribution weight is configured equally according to the 4 cold box side response quantities.

[0081] After the directional assignment quantity is formed, S3 performs lag processing on the directional assignment quantity based on the current position of the cold box energy efficiency sample within the observation window, generating a lag assignment quantity. S3 uses the observation window start point, basic window length, and cold box energy efficiency sample segment length already formed in S2, and reads the midpoint time of the current cold box energy efficiency sample segment. If the midpoint time of the current cold box energy efficiency sample segment is earlier than the observation window start point, the directional assignment quantity does not enter the continuity determination. If the midpoint time of the segment is between the observation window start point and the end point of the lag rise range, S3 reduces the directional assignment quantity according to the ratio between the time distance of the midpoint time of the segment relative to the observation window start point and the lag rise range. If the midpoint time of the segment is not earlier than the end point of the lag rise range, the directional assignment quantity enters the continuity determination according to the complete contribution. The result obtained after this lag processing is the lag assignment quantity, which is used for the generation of subsequent continuous assignment quantities.

[0082] The hysteresis rise range is formed by the hysteresis rise coefficient corresponding to the purification disturbance state and the base window length. When the purification disturbance state is an observational purification disturbance state, the hysteresis rise coefficient is preferably 0.30 to 0.60; when the purification disturbance state is a high purification disturbance state, the hysteresis rise coefficient is preferably 0.20 to 0.50. The hysteresis rise coefficient is formed by historical purification disturbance events that have been archived before the current commissioning boundary. The formation method is as follows: calculate the ratio between the time when the cold box operating residual first continuously reaches the residual manifestation boundary in the historical purification disturbance event and the corresponding base window length, and then set this ratio... The value is limited to the range of the hysteresis rise coefficient corresponding to the purification disturbance state. The residual manifestation boundary adopts the boundary used in S2 to form the duration of the continuous influence of the historical cold box operation residual, preferably 0.50 to 0.80, and is formed by the residual distribution of the historical stable operation segment of the same device. This residual manifestation boundary is only used for the calibration of the historical hysteresis rise coefficient and does not directly replace the boundary for the attribution of the current cold box energy efficiency sample. When there are less than 5 historical purification disturbance events, the hysteresis rise coefficient corresponding to the observed purification disturbance state is 0.50, and the hysteresis rise coefficient corresponding to the high purification disturbance state is 0.35.

[0083] S3 further reads the hysteresis assignments of the current cold box energy efficiency sample and its completed predecessor cold box energy efficiency samples, generates continuous assignments, and records the number of cold box-side response quantities that satisfy the cold box residual direction. The continuous assignments are used to indicate the degree to which the cold box operating residual continues to exist along the cold box residual direction, avoiding the triggering of a disturbance pause state by short-term fluctuations of a single sample segment. The continuous assignments are formed by averaging the hysteresis assignments of the current cold box energy efficiency sample and its completed predecessor cold box energy efficiency samples. The continuity determination only uses the current cold box energy efficiency sample and its completed predecessor cold box energy efficiency samples, does not call the cold box operating residual that has not yet been formed after the current cold box energy efficiency sample, and does not use subsequent cold box energy efficiency samples to modify the already output sample admission state. When the number of completed sample segments is less than the continuity determination length, S3 does not generate a sample admission state for the current cold box energy efficiency sample. The cold box energy efficiency sample continues to be temporarily stored and is saved as a completed predecessor sample when the continuity determination of subsequent cold box energy efficiency samples is performed, without entering S4.

[0084] The continuity judgment length is configured according to the purification disturbance state. When the purification disturbance state is an observational purification disturbance state, the continuity judgment length is preferably 2 to 3 sample segments. When the purification disturbance state is a high purification disturbance state, the continuity judgment length is preferably 2 sample segments. The sample segment length follows the cold box energy efficiency sample packaging granularity in S2, preferably 5 min to 15 min. Therefore, the actual time length corresponding to the continuity judgment is preferably 10 min to 45 min. This time length comes from the engineering matching of the cold box side measurement point response time, tower pressure adjustment cycle, and unit product power consumption metering cycle. When the historical operation records are insufficient to support the differentiated configuration, 2 sample segments are taken as the initial continuity judgment length for both the observational purification disturbance state and the high purification disturbance state.

[0085] The number of cold box-side response quantities that satisfy the direction of the cold box residual is formed by the direction screening results in the current cold box energy efficiency sample. For the observed purification disturbance state, when the number of cold box-side response quantities that satisfy the direction of the cold box residual is not less than 2, it means that there are at least 2 cold box-side response quantities in the current sample that are consistent with the direction of the hysteresis effect of the front-end purification disturbance. For the high purification disturbance state, in addition to the number of cold box-side response quantities that satisfy the direction of the cold box residual being not less than 2, it is also required that at least one of the main heat exchanger terminal differential residual and the cold box pressure drop residual satisfies the direction of the cold box residual, so as to avoid the disturbance suspension state being triggered only by the short-term fluctuation of unit product power consumption and the short-term fluctuation of tower pressure control.

[0086] S3 generates sample admission status based on continuous attribution quantities, the number of cold box-side response quantities satisfying the direction of cold box residuals, and the attribution determination boundary. The attribution determination boundary is a dimensionless boundary. When the purification disturbance state is an observational purification disturbance state, the attribution determination boundary is preferably 0.25 to 0.45; when the purification disturbance state is a high purification disturbance state, the attribution determination boundary is preferably 0.20 to 0.40. The attribution determination boundary is formed by the historical sample admission status records that have been archived before the current commissioning boundary. The historical sample admission status records should correspond to the cold box energy efficiency samples for which cold box operating residuals have been generated. When there are no less than 5 disturbed temporary state samples and no less than 5 stable admission state samples in the historical sample admission status records, and the total number of the two types of samples is no less than 10, the median value between the continuous attribution quantity boundaries of the two types of samples is taken, and the obtained value is limited to the range of the attribution determination boundary of the corresponding purification disturbance state. When there are insufficient historical sample admission status records, the attribution determination boundary corresponding to the observational purification disturbance state is taken as 0.35, and the attribution determination boundary corresponding to the high purification disturbance state is taken as 0.30.

[0087] In this step, the residual action-free boundary, contribution weight, lag rise coefficient, continuity determination length, and attribution determination boundary are all formed from historical fragments or historical records that have been archived before the current deployment boundary; when historical data is insufficient, it is processed according to the corresponding initial value or equal weighting method; after subsequent historical data meet the calibration conditions, only the subsequent cold box energy efficiency samples are recalibrated, and the sample admission status already formed in the current observation window is not changed back.

[0088] When the purification disturbance state is an observation-type purification disturbance state, the continuous attribution quantity reaches the attribution judgment boundary corresponding to the observation-type purification disturbance state, and the number of cold box-side response quantities in the direction of the cold box residual is not less than 2, S3 generates a disturbed temporary state. When the purification disturbance state is a high purification disturbance state, the continuous attribution quantity reaches the attribution judgment boundary corresponding to the high purification disturbance state, the number of cold box-side response quantities in the direction of the cold box residual is not less than 2, and at least one of the main heat exchanger terminal differential residual and the cold box pressure drop residual satisfies the direction of the cold box residual, S3 generates a disturbed temporary state. When the purification disturbance state is an observation-type... If either the purification disturbance state or the high purification disturbance state is not met, and the corresponding disturbance suspension state generation condition is not satisfied, S3 generates a stable access state. When the purification disturbance state is a low purification disturbance state, S3 does not configure the purification lag attribution direction, and the corresponding cold box energy efficiency sample will not generate a disturbance suspension state due to purification disturbance. If the corresponding cold box energy efficiency sample has generated a cold box operation residual, S3 generates a stable access state. If no cold box operation residual has been generated, it continues to maintain the temporary state. The stable access state indicates that the corresponding cold box energy efficiency sample does not meet the purification disturbance lag attribution condition, and subsequent permission gating processing will still be performed by S4.

[0089] In the aforementioned S2 specific implementation, the target molecular sieve adsorber is adsorber A, the time corresponding to the current commissioning boundary is 10:00, the purification disturbance state is the observation type purification disturbance state, the observation window formed by S2 is from 10:19 to 11:49, and the cold box side operation record is packaged according to 10-minute sample segments, and S3 directly calls the observation window, cold box energy efficiency sample and cold box operation residual formed by S2.

[0090] For the cold box energy efficiency sample from 10:19 to 10:29, S2 has already generated a main heat exchanger terminal differential residual of 0.82. After entering S3, the same cold box energy efficiency sample continues to read the cold box pressure drop residual, tower pressure deviation residual, and unit product power consumption residual. In this embodiment, the four cold box operating residuals corresponding to this cold box energy efficiency sample are 0.82, 0.41, 0.28, and 0.35, respectively. After S3 reads the cold box residual direction corresponding to the observed purification disturbance state, the four cold box operating residuals are all filtered according to the cold box residual direction side. If the residuals of the four cold box side response quantities have no action boundary, they are all taken as 0.20. If the residuals of the main heat exchanger terminal difference, cold box pressure drop, tower pressure deviation, and unit product power consumption exceed the residual action boundary, the values ​​of 0.62, 0.21, 0.08, and 0.15, respectively. If the contribution weights of the residuals of the main heat exchanger terminal difference, cold box pressure drop, tower pressure deviation, and unit product power consumption are 0.35, 0.25, 0.15, and 0.25, respectively, then the directional assignment value formed by the cold box energy efficiency sample is 0.319. This directional assignment value only indicates the degree of directional agreement of the current sample and needs to be processed by hysteresis to form a hysteresis assignment value before participating in the generation of continuous assignment values.

[0091] For the cold box energy efficiency sample from 10:19 to 10:29, the midpoint of the segment is 10:24, which is 5 minutes after the start of the observation window. If the basic window length is 60 minutes and the hysteresis rise coefficient corresponding to the observed purification disturbance state is 0.50, then the hysteresis rise range is 30 minutes. The midpoint of the segment for this sample has not yet reached the end of the hysteresis rise range. S3 reduces the direction assignment quantity according to the ratio between the time distance of the midpoint of the segment relative to the start of the observation window and the hysteresis rise range to form the hysteresis assignment quantity. For subsequent samples within the same observation window, S3 generates the hysteresis assignment quantity one by one in the same way, and only uses the current sample and its completed preceding samples to generate continuous assignment quantities.

[0092] For example, for the cold box energy efficiency sample from 10:49 to 10:59, the midpoint of the segment is 10:54, which is 35 minutes after the start of the observation window. This exceeds the 30-minute lag rise range. The directional assignment quantity forms the lag assignment quantity according to the complete contribution. If the continuous assignment quantity formed by this sample and the previous completed sample is 0.38, the assignment judgment boundary is 0.35, and the number of cold box side response quantities in the direction of the cold box residual is not less than 2, then the cold box energy efficiency sample meets the generation conditions of the disturbed temporary state corresponding to the observed purification disturbance state. S3 generates the disturbed temporary state. If the continuous assignment quantity of a cold box energy efficiency sample does not reach 0.35, or if the number of cold box side response quantities in the direction of the cold box residual is less than 2, then the cold box energy efficiency sample does not meet the corresponding disturbed temporary state generation conditions. S3 generates the stable admission state.

[0093] S3 outputs a sample admission status record, which includes at least the target molecular sieve adsorber number, the current commissioning boundary, the observation window number, the cold box energy efficiency sample number, the purification disturbance status, the cold box residual direction, the direction assignment quantity, the hysteresis assignment quantity, the continuous assignment quantity, the number of cold box side response quantities that satisfy the cold box residual direction, the assignment judgment boundary, and the sample admission status. The sample admission status record is used by S4 to perform permission gating processing. S3 does not update the multivariate energy efficiency model and does not output air separation operation control quantity adjustment commands.

[0094] S4. Perform permission gating processing according to the sample access status. When the sample access status is in the disturbed and suspended state, suspend the corresponding cold box energy efficiency sample from participating in the multivariate energy efficiency model update. When the sample access status is in the stable access state, allow the corresponding cold box energy efficiency sample to participate in the multivariate energy efficiency model update, and output energy efficiency control commands for adjusting the operating control quantity of the air separation unit based on the multivariate energy efficiency model.

[0095] Following the sample admission status in S3, S4 uses the cold box energy efficiency sample number as the association index to associate the sample admission status with the cold box energy efficiency sample formed in S2. S4 does not recalculate the cold box operating residual, does not re-determine the direction of the cold box residual, and does not re-form continuous attribution quantities. The sample admission status serves as the basis for permitting the cold box energy efficiency sample to enter the model update chain.

[0096] When a cold box energy efficiency sample is not associated with a sample admission state, S4 maintains the temporary state of the cold box energy efficiency sample. After the corresponding sample admission state is generated, permission gating processing is performed. When the sample admission state is a disturbed and suspended state, S4 excludes the corresponding cold box energy efficiency sample from the model update sample set. The cold box energy efficiency sample does not participate in the current round of multivariate energy efficiency model update. When the sample admission state is a stable admission state, S4 writes the corresponding cold box energy efficiency sample into the model update sample set as a sample allowed to participate in the multivariate energy efficiency model update.

[0097] The cold box energy efficiency samples written into the model update sample set include the operating parameters formed by S2, unit product power consumption, and cold box side response, and retain the corresponding sample admission status. The multivariate energy efficiency model uses the operating parameters as input and unit product power consumption, main heat exchanger terminal difference, cold box pressure drop, and tower pressure deviation as prediction objects. Among them, unit product power consumption is used as the energy efficiency prediction object, and main heat exchanger terminal difference, cold box pressure drop, and tower pressure deviation are used as the cold box side response prediction objects.

[0098] When the number of cold box energy efficiency samples in the model update sample set reaches the lower limit of the model update sample, S4 generates a candidate model version based on the currently effective multivariate energy efficiency model and the model update sample set. The lower limit of the model update sample is preferably 8 to 20 cold box energy efficiency samples. When the sample segment length is 10 minutes, the unit product power consumption is statistically calculated according to the same metering cycle, and the cold box side response covers at least one adjustment response cycle, the lower limit of the model update sample can be 10 cold box energy efficiency samples. When the number of cold box energy efficiency samples in the model update sample set does not reach the lower limit of the model update sample, S4 does not generate a candidate model version and maintains the currently effective multivariate energy efficiency model.

[0099] When generating candidate model versions, S4 uses the currently effective multivariate energy efficiency model as the base model, organizes the operating parameters in the model update sample set into an input sample table, and organizes the unit product power consumption, main heat exchanger terminal temperature difference, cold box pressure drop, and tower pressure deviation into an output sample table. While maintaining consistency in input fields, prediction objects, and model structure, it updates the trainable parameters of the base model to generate candidate model versions. The candidate model versions retain the corresponding model update round, model update sample set number, and base model version number. The trainable parameter update is completed using engineering supervised learning training, preferably using one of gradient boosting regression model training, partial least squares regression model training, and random forest regression model training. The training objective is to reduce the prediction bias of unit product power consumption, main heat exchanger terminal temperature difference, cold box pressure drop, and tower pressure deviation. The candidate model versions maintain the same input fields, prediction objects, and model structure as the base model, only updating the trainable parameters.

[0100] After generating the candidate model version, S4 selects cold box energy efficiency samples that have been archived before this model update round and whose sample admission status is stable admission status as validation samples. The validation samples are separated from the model update sample set. The validation samples do not participate in the generation of this candidate model version. When the number of validation samples does not reach the lower limit of the model update sample, S4 does not enable the candidate model version and maintains the currently effective multivariate energy efficiency model.

[0101] The validation of candidate model versions includes energy efficiency prediction deviation validation and cold box side response prediction deviation validation. The energy efficiency prediction deviation is formed by the absolute value of the difference between the predicted unit product power consumption and the measured unit product power consumption in the validation sample, and is normalized according to the engineering response limit of the unit product power consumption. The cold box side response prediction deviation is formed by the absolute value of the difference between the predicted main heat exchanger terminal difference, the predicted cold box pressure drop, and the predicted tower pressure deviation and the corresponding measured results in the validation sample, and is normalized according to the engineering response limit of the corresponding cold box side response.

[0102] S4 uses the same validation sample to calculate the prediction deviations of the candidate model version and the currently effective multivariate energy efficiency model. When the energy efficiency prediction deviation of the candidate model version on the same validation sample is not higher than the energy efficiency prediction deviation of the currently effective multivariate energy efficiency model, and the prediction deviations of each cold box side response quantity of the candidate model version on the same validation sample are not higher than the prediction deviations of each cold box side response quantity of the currently effective multivariate energy efficiency model, S4 switches the candidate model version to the new effective model. If the validation conditions are not met, S4 keeps the currently effective multivariate energy efficiency model.

[0103] When outputting energy efficiency control commands, S4 reads the current operating status and operating constraints of the air separation unit. The current operating status of the air separation unit is collected through DCS, and the collected objects include the current operating control quantity, product purity, output, and equipment operating status. The operating constraints are formed by product indicators, equipment allowable operating range, and DCS control limits. When any one of the current operating status and operating constraints of the air separation unit is missing, S4 generates an energy efficiency control command to maintain the current operating control quantity, with the adjustment direction being the maintenance direction and the adjustment range being 0.

[0104] When the current air separation unit is in operation and all operating constraints are complete, S4 takes the current operating control quantity as the center and forms upward candidate values, downward candidate values, and maintenance candidate values ​​for each adjustable operating control quantity according to the DCS's allowed single-step change range. Candidate values ​​that exceed the equipment's allowed operating range are deleted. S4 prioritizes the candidate values ​​of a single adjustable operating control quantity to form candidate operating control quantities. When the DCS control strategy allows the linkage adjustment of two or more adjustable operating control quantities, corresponding linkage candidate operating control quantities are formed. The permission relationship for the linkage adjustment of two or more adjustable operating control quantities is read from the linkage adjustment permission record in the DCS sequential control strategy table, interlocking logic table, and operating procedures. When no linkage adjustment permission is obtained, no linkage candidate operating control quantity is formed. The resulting set of candidate operating control quantities includes the current operating control quantity.

[0105] S4 inputs the set of candidate operating control quantities into the currently effective multivariate energy efficiency model to generate unit product power consumption prediction results and cold box side response prediction results. When either the unit product power consumption prediction result or the cold box side response prediction result corresponding to the candidate operating control quantity is missing, S4 will not include the candidate operating control quantity in the energy efficiency screening. Product purity constraints and output constraints are judged based on the current DCS measured values, product index boundaries, and adjustment permission conditions in the DCS sequential control strategy table. Equipment operating upper limit and DCS control limit are judged based on the equipment's allowable operating range, single-step adjustment limit, and interlock permission status corresponding to the candidate operating control quantity. The multivariate energy efficiency model does not directly predict product purity and output. Product purity constraints and output constraints serve as external operating permission conditions for the generation and screening of candidate operating control quantities. Subsequently, S4 removes candidate operating control quantities that do not meet the operating constraints, so that candidate operating control quantities that do not meet the product purity, output, equipment operating upper limit, and DCS control limit will not enter the energy efficiency screening.

[0106] For the remaining candidate operation control quantities, S4 generates energy efficiency control instructions according to the energy efficiency screening conditions. In specific processing, S4 first sorts the remaining candidate operation control quantities from low to high according to the unit product power consumption prediction results. When the difference between the unit product power consumption prediction results of the candidate operation control quantity with the lowest unit product power consumption prediction results and other candidate operation control quantities reaches the DCS distinguishable energy consumption difference, S4 selects the candidate operation control quantity with the lowest unit product power consumption prediction results.

[0107] When the difference in the predicted unit product power consumption between two or more candidate operating control quantities does not reach the DCS-resolvable energy consumption difference, S4 calculates the remaining distance from the predicted main heat exchanger terminal difference, cold box pressure drop, and tower pressure deviation to the corresponding operating constraint boundary under each candidate operating control quantity, and normalizes them according to the allowable range of the corresponding operating constraints to obtain the dimensionless residual ratio corresponding to each cold box side response quantity; then, the minimum value among the dimensionless residual ratios under the same candidate operating control quantity is taken as the constraint margin evaluation value of the candidate operating control quantity, and the candidate operating control quantity with the largest constraint margin evaluation value is selected among the two or more candidate operating control quantities.

[0108] The resolvable energy consumption difference in DCS is formed by the unit product power consumption metering resolution and the minimum identifiable power consumption change within the DCS control cycle. The minimum identifiable power consumption change within the DCS control cycle is formed by dividing the minimum resolution of the total power consumption metering within the DCS control cycle by the converted quantity of the product in the same cycle. When the unit product power consumption metering resolution is directly given by the energy consumption metering system, the resolvable energy consumption difference in DCS is taken as the larger value between the unit product power consumption metering resolution and the minimum identifiable power consumption change. When historical operation records are insufficient, the value corresponding to the unit product power consumption metering resolution is taken. This parameter is used to avoid forcibly selecting the candidate operation control quantity with the lowest energy consumption value but whose cold box side response is close to the constraint boundary when the difference in the unit product power consumption prediction result is not identifiable.

[0109] When no candidate operating control quantity remains after the candidate operating control quantity is eliminated, S4 generates an energy efficiency control command to maintain the current operating control quantity. This energy efficiency control command does not change the current operating control quantity of the air separation unit, the adjustment direction is the maintenance direction, and the adjustment range is 0.

[0110] The energy efficiency control instruction includes the target operating control quantity, adjustment direction, adjustment range, execution time, and model version number. The target operating control quantity specifies the operating control quantity of the air separation unit that needs to be adjusted. The adjustment direction specifies one of the upward, downward, or maintenance directions. The adjustment range limits the range of change of the control quantity. The execution time is used to align with the DCS control cycle. The model version number identifies the currently effective multivariate energy efficiency model used when generating the energy efficiency control instruction. After S4 outputs the energy efficiency control instruction, the energy efficiency control instruction, model version number, and corresponding sample access status are written into the operation record.

[0111] In this embodiment, the target molecular sieve adsorber is adsorber A. The time corresponding to the current commissioning boundary is 10:00, and the observation window is from 10:19 to 11:49. For the cold box energy efficiency sample generated in the disturbed temporary state in S3, the cold box energy efficiency sample is excluded from the model update sample set in S4. For the cold box energy efficiency sample generated in the stable admission state in S3, the cold box energy efficiency sample is written into the model update sample set in S4. When the number of cold box energy efficiency samples in the model update sample set reaches the lower limit of the model update sample and the candidate model version passes the verification, S4 switches the candidate model version to the new effective model and generates energy efficiency control instructions based on the new effective model. When the candidate model version fails the verification, S4 maintains the currently effective multivariate energy efficiency model and outputs energy efficiency control instructions based on the currently effective multivariate energy efficiency model.

[0112] With permission gating, cold box energy efficiency samples in the disturbed and suspended state do not enter the multivariate energy efficiency model update chain. Cold box energy efficiency samples in the stable access state participate in the effective model switching after quantity access and model version verification. Energy efficiency control instructions are jointly generated by the currently effective multivariate energy efficiency model and operating constraints, and are not directly generated by a single cold box energy efficiency sample.

[0113] The present invention also provides an energy efficiency optimization system for air separation unit based on multivariate data modeling, comprising: a purification cycle processing module, which acquires the purification cycle operation record of the air separation unit, generates purification cycle event record according to the cycle operation boundary of the molecular sieve adsorber, performs pairing processing of adsorption load carrying result and regeneration recovery gap, and generates purification disturbance state. The cold box residual processing module acquires the cold box side operation record after the molecular sieve switching is completed, configures the observation window associated with the molecular sieve switching event according to the purification disturbance state, temporarily stores the cold box side operation record in the observation window as a cold box energy efficiency sample, calls the similar working condition baseline without purification disturbance to compare the cold box energy efficiency sample, and generates cold box operation residual. The sample admission processing module performs lag attribution processing on the cold box operation residuals. It generates sample admission states with the direction of the cold box residuals corresponding to the purification disturbance state as a reference. The sample admission states include the disturbance suspension state and the stable admission state. The sample admission states are generated based on the lag attribution result of the cold box operation residuals relative to the direction of the cold box residuals and the attribution judgment result of the corresponding purification disturbance state. The permission gating processing module performs permission gating processing according to the sample admission status. When the sample admission status is in the disturbed and suspended state, it suspends the corresponding cold box energy efficiency sample from participating in the multivariate energy efficiency model update. When the sample admission status is in the stable admission state, it allows the corresponding cold box energy efficiency sample to participate in the multivariate energy efficiency model update and outputs energy efficiency control commands based on the multivariate energy efficiency model to adjust the operating control quantity of the air separation unit.

[0114] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0115] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0116] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for optimizing the energy efficiency of a space separation unit based on multivariate data modeling, characterized in that, Includes the following steps: Acquire the purification cycle operation record of the air separation unit, generate purification cycle event record according to the cycle operation boundary of the molecular sieve adsorber, perform pairing processing of adsorption load carrying result and regeneration recovery gap, and generate purification disturbance state. After the molecular sieve switching is completed, the cold box side operation record is obtained. According to the purification disturbance state, the observation window associated with the molecular sieve switching event is configured. The cold box side operation record in the observation window is temporarily stored as the cold box energy efficiency sample. The similar working condition baseline without purification disturbance is called to compare the cold box energy efficiency sample and generate the cold box operation residual. The cold box operation residual is subjected to lag assignment processing. The sample admission state is generated with the cold box residual direction corresponding to the purification disturbance state as a reference. The sample admission state includes the disturbance suspension state and the stable admission state. The sample admission state is generated based on the lag assignment result of the cold box operation residual relative to the cold box residual direction and the assignment judgment result of the corresponding purification disturbance state. The permission gating process is performed according to the sample access status. When the sample access status is in the disturbed and suspended state, the corresponding cold box energy efficiency sample is suspended from participating in the multivariate energy efficiency model update. When the sample access status is in the stable access state, the corresponding cold box energy efficiency sample is allowed to participate in the multivariate energy efficiency model update, and energy efficiency control commands for adjusting the operating control quantity of the air separation unit are output based on the multivariate energy efficiency model.

2. The method for optimizing the energy efficiency of a space separation unit based on multivariate data modeling according to claim 1, characterized in that, The specific steps for generating purification cycle event records based on the cyclic operation boundary of the molecular sieve adsorber are as follows: The cycle operation boundary is formed by the adsorption start-up boundary, adsorption exit boundary, regeneration completion boundary and current start-up boundary of the target molecular sieve adsorber before this operation. The target molecular sieve adsorber is the molecular sieve adsorber put into the cold box inlet channel after this switch. If the cycle operation boundary does not form a complete continuous time sequence after being verified by the commissioning confirmation status and the regeneration completion confirmation status, an observation-type purification disturbance status for that cycle is generated.

3. The method for optimizing the energy efficiency of a space separation unit based on multivariate data modeling according to claim 2, characterized in that, The specific steps for pairing the adsorption load carrying capacity result with the regeneration recovery gap to generate the purification disturbance state are as follows: The purification disturbance state is generated by pairing the adsorption load carrying result and the regeneration recovery gap; when the adsorption load carrying result is a normal carrying state and the regeneration recovery gap is a low gap state, a low purification disturbance state is generated. When the adsorption load is at a normal load state and the regeneration recovery gap is at a medium gap state, an observation-type purification disturbance state is generated; when the adsorption load is at an increased load state and the regeneration recovery gap is at a low gap state, an observation-type purification disturbance state is generated; when the adsorption load is at a normal load state and the regeneration recovery gap is at a high gap state, a high purification disturbance state is generated; when the adsorption load is at an increased load state and the regeneration recovery gap is at a medium gap state, a high purification disturbance state is generated; when the adsorption load is at an increased load state and the regeneration recovery gap is at a high gap state, a high purification disturbance state is generated.

4. The method for optimizing the energy efficiency of a space separation unit based on multivariate data modeling according to claim 3, characterized in that, The specific steps to temporarily save the cold box side operation record in the observation window as a cold box energy efficiency sample are as follows: Read the current commissioning boundary and purification disturbance status from the purification cycle event record; superimpose the gas path transmission delay from the target molecular sieve adsorber outlet to the cold box side response measurement point on the time corresponding to the current commissioning boundary to generate the starting point of the observation window; An observation window is formed according to the window length corresponding to the purification disturbance state, wherein the window lengths corresponding to the low purification disturbance state, the observation-type purification disturbance state, and the high purification disturbance state increase sequentially; the cold box side operation records within the observation window are packaged into cold box energy efficiency samples according to sample segments and kept in a temporary storage state.

5. The method for optimizing the energy efficiency of a space separation unit based on multivariate data modeling according to claim 4, characterized in that, The specific steps for generating cold box operating residuals are as follows: A baseline of similar operating conditions without purification disturbances is used to compare the cold box energy efficiency samples. A set of similar segments is selected from historical observation segments that were archived before the time corresponding to the current commissioning boundary and whose purification disturbance state was low; based on the difference in operating condition fit between the current cold box energy efficiency sample and the set of similar segments and the data completeness, a similar operating condition baseline without purification disturbance and its validity label are generated. The cold box side response quantity in the current cold box energy efficiency sample is compared with the baseline of similar operating conditions without purification disturbance. The difference is then processed according to the engineering response limit to generate the cold box operation residual. When any of the following conditions are met: the number of similar segments does not meet the baseline generation requirement, the cold box side response quantity does not meet the effective coverage requirement, or the validity of the similar operating condition baseline is marked as low confidence, the cold box energy efficiency sample remains in a temporary state.

6. The method for optimizing the energy efficiency of a space separation unit based on multivariate data modeling according to claim 1, characterized in that, The specific steps for performing lag attribution processing on the cold box operation residuals are as follows: Read the cold box energy efficiency sample that has generated cold box operating residuals; when the purification disturbance state belongs to either the observational purification disturbance state or the high purification disturbance state, filter the cold box operating residuals located on the cold box residual direction side and exceeding the residual no-action boundary, and form the direction assignment quantity based on the filtered cold box operating residuals. Based on the time distance between the midpoint of the current cold box energy efficiency sample segment and the starting point of the observation window, hysteresis processing is performed on the directional assignment quantity to generate hysteresis assignment quantity; the hysteresis assignment quantities of the current cold box energy efficiency sample and its completed predecessor cold box energy efficiency samples are read to generate continuous assignment quantity, and the number of cold box side response quantities that satisfy the cold box residual direction is recorded.

7. The energy efficiency optimization method for a space separation unit based on multivariate data modeling according to claim 6, characterized in that, The specific steps for generating sample admission status are as follows: When the purification disturbance state is an observation-type purification disturbance state, the continuous attribution quantity reaches the attribution judgment boundary corresponding to the observation-type purification disturbance state, and the number of cold box side response quantities in the direction of cold box residual is not less than two, a disturbance suspension state is generated. When the purification disturbance state is a high purification disturbance state, the continuous attribution quantity reaches the attribution judgment boundary corresponding to the high purification disturbance state, the number of cold box side response quantities that satisfy the cold box residual direction is not less than two, and at least one of the main heat exchanger terminal differential residual and the cold box pressure drop residual satisfies the cold box residual direction, a disturbance pause state is generated. When the purification disturbance state belongs to either the observational purification disturbance state or the high purification disturbance state, and the corresponding disturbance suspension state generation condition is not met, a stable admission state is generated; when the purification disturbance state is the low purification disturbance state, no purification lag attribution direction is configured, and a stable admission state is generated when the corresponding cold box energy efficiency sample has generated the cold box operation residual.

8. The method for optimizing the energy efficiency of a space separation unit based on multivariate data modeling according to claim 7, characterized in that, The specific steps for performing permission gating based on sample access status are as follows: Read the sample admission status and associate the cold box energy efficiency samples according to the cold box energy efficiency sample number; exclude the cold box energy efficiency samples whose sample admission status is in the disturbed and suspended state from the model update sample set. Write the cold box energy efficiency samples with a stable admission state into the model update sample set.

9. The method for optimizing the energy efficiency of a space separation unit based on multivariate data modeling according to claim 8, characterized in that, The specific operation of allowing corresponding cold box energy efficiency samples to participate in the multivariate energy efficiency model update, and outputting energy efficiency control commands for adjusting the operating control quantities of the air separation unit based on the multivariate energy efficiency model is as follows: When the number of cold box energy efficiency samples in the model update sample set reaches the lower limit of the model update sample set, a candidate model version is generated based on the currently effective multivariate energy efficiency model and the model update sample set; cold box energy efficiency samples that have been archived before this model update round and whose sample admission status is stable admission status are selected as verification samples to verify the candidate model version; When the energy efficiency prediction deviation and cold box response prediction deviation of the candidate model version on the same validation sample are not higher than the corresponding prediction deviation of the currently effective multivariate energy efficiency model on the same validation sample, the candidate model version will be switched to the new effective model; when the number of cold box energy efficiency samples in the model update sample set does not reach the lower limit of the model update sample or the candidate model version fails the validation, the currently effective multivariate energy efficiency model will be maintained. Read the current operating status and operating constraints of the air separation unit to form a set of candidate operating control variables; Input the candidate set of operational control variables into the currently effective multivariate energy efficiency model to generate unit product power consumption prediction results and cold box side response prediction results; Candidate operational control variables that do not meet operational constraints are eliminated; among the remaining candidate operational control variables, candidate operational control variables are screened based on the unit product power consumption prediction results and the cold box side response prediction results, and energy efficiency control instructions are generated.

10. An energy efficiency optimization system for an air separation unit based on multivariate data modeling, employing the energy efficiency optimization method for an air separation unit based on multivariate data modeling as described in any one of claims 1-9, characterized in that, include: The purification cycle processing module acquires the purification cycle operation records of the air separation unit, generates purification cycle event records according to the cycle operation boundary of the molecular sieve adsorber, performs pairing processing of adsorption load carrying results and regeneration recovery gap, and generates purification disturbance state. The cold box residual processing module acquires the cold box side operation record after the molecular sieve switching is completed, configures the observation window associated with the molecular sieve switching event according to the purification disturbance state, temporarily stores the cold box side operation record in the observation window as a cold box energy efficiency sample, calls the similar working condition baseline without purification disturbance to compare the cold box energy efficiency sample, and generates cold box operation residual. The sample admission processing module performs lag attribution processing on the cold box operation residuals. It generates sample admission states with the direction of the cold box residuals corresponding to the purification disturbance state as a reference. The sample admission states include the disturbance suspension state and the stable admission state. The sample admission states are generated based on the lag attribution result of the cold box operation residuals relative to the direction of the cold box residuals and the attribution judgment result of the corresponding purification disturbance state. The permission gating processing module performs permission gating processing according to the sample admission status. When the sample admission status is in the disturbed and suspended state, it suspends the corresponding cold box energy efficiency sample from participating in the multivariate energy efficiency model update. When the sample admission status is in the stable admission state, it allows the corresponding cold box energy efficiency sample to participate in the multivariate energy efficiency model update and outputs energy efficiency control commands based on the multivariate energy efficiency model to adjust the operating control quantity of the air separation unit.

Images